diff --git a/parse/train/0NXUSlb6oEu/images/0b5677bc7e5a3bbf200b8fc702fec82fae37b0d605c6084e4e31257142276015.jpg b/parse/train/0NXUSlb6oEu/images/0b5677bc7e5a3bbf200b8fc702fec82fae37b0d605c6084e4e31257142276015.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a3cef4e13561a6b2bd14c2f4ace1be5638ef4589 --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/0b5677bc7e5a3bbf200b8fc702fec82fae37b0d605c6084e4e31257142276015.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fac0318cd799e25fdabf2d148344c13b98a972e20aa52d2c1a49fa2676289e5a +size 7429 diff --git a/parse/train/0NXUSlb6oEu/images/3e6487378e9debb436c615ff8311fb38dd237c64fb4e3dc9118341aa066fb749.jpg b/parse/train/0NXUSlb6oEu/images/3e6487378e9debb436c615ff8311fb38dd237c64fb4e3dc9118341aa066fb749.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f32a12fdc6b26acafd8f9bebd4dde9cb5005cfe --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/3e6487378e9debb436c615ff8311fb38dd237c64fb4e3dc9118341aa066fb749.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a472fbbec38875d3ad619563eb21c756e067a77236815f7c7a911cb8335c385 +size 19634 diff --git a/parse/train/0NXUSlb6oEu/images/5e1e8166420c62722485c6a7820dfe98653f2088972c7e84afa5280b7a8886d0.jpg b/parse/train/0NXUSlb6oEu/images/5e1e8166420c62722485c6a7820dfe98653f2088972c7e84afa5280b7a8886d0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4325ecef7d3bceb37f7f27898bd8c6217dff0e7f --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/5e1e8166420c62722485c6a7820dfe98653f2088972c7e84afa5280b7a8886d0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5ac374f813fe39683a8ebefb86888fcd45566ad6353a73cd145e76640321b7a +size 56122 diff --git a/parse/train/0NXUSlb6oEu/images/5f8c12ad3886b79e61c74d1e271f56df7ddab4dadb0270c4828ddb7eb8df55b4.jpg b/parse/train/0NXUSlb6oEu/images/5f8c12ad3886b79e61c74d1e271f56df7ddab4dadb0270c4828ddb7eb8df55b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b05e1fe6e3dfe478a6f6d8c96b9e162671694d73 --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/5f8c12ad3886b79e61c74d1e271f56df7ddab4dadb0270c4828ddb7eb8df55b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:552cfde3eed31e1c785493b25d8ad55b6d4be80f51cb636ddbae015f62e6440e +size 14075 diff --git a/parse/train/0NXUSlb6oEu/images/6d286d590fda27e0fbae3d5276222ff0a5236f19045b6bc7382a942a04e631ca.jpg b/parse/train/0NXUSlb6oEu/images/6d286d590fda27e0fbae3d5276222ff0a5236f19045b6bc7382a942a04e631ca.jpg new file mode 100644 index 0000000000000000000000000000000000000000..02cf7080f836abb0edddcd726fcb63e44fe166ec --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/6d286d590fda27e0fbae3d5276222ff0a5236f19045b6bc7382a942a04e631ca.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65f2d1e3aceceb220cf9dbf00e71de1684891091cb8e504cf52faf1343c8a18f +size 62737 diff --git a/parse/train/0NXUSlb6oEu/images/7ee27a6f2cb5172661304ea51430cd343ec719189143dbc98738b2764dbfe3e9.jpg b/parse/train/0NXUSlb6oEu/images/7ee27a6f2cb5172661304ea51430cd343ec719189143dbc98738b2764dbfe3e9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7016027d9a65da2c82edf104fc61b9e8a302e2ce --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/7ee27a6f2cb5172661304ea51430cd343ec719189143dbc98738b2764dbfe3e9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64d10892780d10b8a67823fd860c842955dc57bd8f35e237b4633649599ebb51 +size 7908 diff --git a/parse/train/0NXUSlb6oEu/images/8eb9ee6d3777e7d8166f9e6f212649a7a9831e7486d3e37219d94bb171c3f31b.jpg b/parse/train/0NXUSlb6oEu/images/8eb9ee6d3777e7d8166f9e6f212649a7a9831e7486d3e37219d94bb171c3f31b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..332b7618b6f7cf945fd6f7c6eabaedb89f76ea3f --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/8eb9ee6d3777e7d8166f9e6f212649a7a9831e7486d3e37219d94bb171c3f31b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f751b59be0559ce9a506746ac919cf0028ff55845eefa1192f3301d1794bd1e +size 24771 diff --git a/parse/train/0NXUSlb6oEu/images/8f10f51d06f3441bc5bb0521230257f6883c15ba0f829dfc03c3dfb4e01db041.jpg b/parse/train/0NXUSlb6oEu/images/8f10f51d06f3441bc5bb0521230257f6883c15ba0f829dfc03c3dfb4e01db041.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f4da30703bcbf6de3395cace8711c8287511dfd1 --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/8f10f51d06f3441bc5bb0521230257f6883c15ba0f829dfc03c3dfb4e01db041.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b1fa497c86bb0e1b72fdfde9f3256306aa35f377a7e3da03fa9169312ad2e05 +size 7310 diff --git a/parse/train/0NXUSlb6oEu/images/9250b3b0cafb63ebf6bc7675356e2df9650772e0cca3d83f6c4f1a8bda836553.jpg b/parse/train/0NXUSlb6oEu/images/9250b3b0cafb63ebf6bc7675356e2df9650772e0cca3d83f6c4f1a8bda836553.jpg new file mode 100644 index 0000000000000000000000000000000000000000..068668e8772c418a8b5bb6d653e82215c2f3960c --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/9250b3b0cafb63ebf6bc7675356e2df9650772e0cca3d83f6c4f1a8bda836553.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6b9c6c938e0300360c63a89b012786f55a063b3fe467e33320da727fcd86aa1 +size 69490 diff --git a/parse/train/0NXUSlb6oEu/images/a258345ad42ce663855c9b9f41cda34f90e8fb3d6c743c8733aecb768f7c50f9.jpg b/parse/train/0NXUSlb6oEu/images/a258345ad42ce663855c9b9f41cda34f90e8fb3d6c743c8733aecb768f7c50f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0f520013454dd414d404731d86952f984369e273 --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/a258345ad42ce663855c9b9f41cda34f90e8fb3d6c743c8733aecb768f7c50f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:faab8cf63af7c2f54cb0aacef491b4f2dcaa29dd37dcba33b4db8ad8c1791e90 +size 64446 diff --git a/parse/train/0NXUSlb6oEu/images/ae444178f536b7d1189a65c00f2c5be9d41a215baebe793de3b527af52a200d2.jpg b/parse/train/0NXUSlb6oEu/images/ae444178f536b7d1189a65c00f2c5be9d41a215baebe793de3b527af52a200d2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ec9bbe5a89003e5ea3df88ea8cd8b4cd902059f3 --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/ae444178f536b7d1189a65c00f2c5be9d41a215baebe793de3b527af52a200d2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:090c50959c3bb6a72ddbfd4e4738956e608d7669796e9ffa46b389055b94e7b4 +size 30736 diff --git a/parse/train/0NXUSlb6oEu/images/f08bfdeec2b2376d091d17de2edb57f50277f1460687ff71f9b6b3adb42c627e.jpg b/parse/train/0NXUSlb6oEu/images/f08bfdeec2b2376d091d17de2edb57f50277f1460687ff71f9b6b3adb42c627e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9488e3fb1da19548967bbd48d3c9b380053ccd2a --- /dev/null +++ b/parse/train/0NXUSlb6oEu/images/f08bfdeec2b2376d091d17de2edb57f50277f1460687ff71f9b6b3adb42c627e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d11e29eef2cfa88ee8b900e91da7f45a12f3e3d5875488457baf1a3e9e2b5c0f +size 6532 diff --git a/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9.md b/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9.md new file mode 100644 index 0000000000000000000000000000000000000000..bd56adc9f2c4357b91beec72de2ba7a534520269 --- /dev/null +++ b/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9.md @@ -0,0 +1,314 @@ +# BARTSCORE: Evaluating Generated Text as Text Generation + +Weizhe Yuan Carnegie Mellon University weizhey@cs.cmu.edu + +Graham Neubig Carnegie Mellon University gneubig@cs.cmu.edu + +Pengfei Liu ∗ Carnegie Mellon University pliu3@cs.cmu.edu + +# Abstract + +A wide variety of NLP applications, such as machine translation, summarization, and dialog, involve text generation. One major challenge for these applications is how to evaluate whether such generated texts are actually fluent, accurate, or effective. In this work, we conceptualize the evaluation of generated text as a text generation problem, modeled using pre-trained sequence-to-sequence models. The general idea is that models trained to convert the generated text to/from a reference output or the source text will achieve higher scores when the generated text is better. We operationalize this idea using BART [32], an encoder-decoder based pre-trained model, and propose a metric BARTSCORE with a number of variants that can be flexibly applied in an unsupervised fashion to evaluation of text from different perspectives (e.g. informativeness, fluency, or factuality). BARTSCORE is conceptually simple and empirically effective. It can outperform existing top-scoring metrics in 16 of 22 test settings, covering evaluation of 16 datasets (e.g., machine translation, text summarization) and 7 different perspectives (e.g., informativeness, factuality). Code to calculate BARTScore is available at https://github.com/neulab/BARTScore, and we have released an interactive leaderboard for meta-evaluation at http: //explainaboard.nlpedia.ai/leaderboard/task-meval/ on the EXPLAINABOARD platform [38], which allows us to interactively understand the strengths, weaknesses, and complementarity of each metric. + +# 1 Introduction + +One defining feature of recent NLP models is the use of neural representations trained on raw text, using unsupervised objectives such as language modeling [6,53], or denoising autoencoding [9,32,54]. By learning to predict the words or sentences in natural text, these models simultaneously learn to extract features that not only benefit mainstream NLP tasks such as information extraction [23, 37], question answering [1, 26], text summarization [40, 77] but also have proven effective in development of automatic metrics for evaluation of text generation itself [62, 65]. For example, BERTScore [75] and MoverScore [76] take features extracted by BERT [9] and apply unsupervised matching functions to compare system outputs against references. Other works build supervised frameworks that use the extracted features to learn to rank [56] or regress [62] to human evaluation scores. + +However, in the context of generation evaluation, one may note that there is a decided disconnect between how models are pre-trained using text generation objectives and how they are used as down-stream feature extractors. This leads to potential under-utilization of the pre-trained model parameters. For example, the output prediction layer is not used at all in this case. This disconnect is particularly striking because of the close connection between the pre-training objectives and the generation tasks we want to evaluate. + +In this paper, we instead argue for a formulation of evaluation of generated text as a text generation problem, directly evaluating text through the lens of its probability of being generated from or generating other textual inputs and outputs. This is a better match with the underlying pre-training tasks and allows us to more fully take advantage of the parameters learned during the pre-training phase. We solve the modeling problem with a pre-trained sequence-to-sequence (seq2seq) model, specifically BART [32], and devise a metric named BARTSCORE, which has the following characteristics: (1) BARTSCORE is parameter- and data-efficient. Architecturally there are no extra parameters beyond those used in pre-training itself, and it is an unsupervised metric that doesn’t require human judgments to train. (2) BARTSCORE can better support evaluation of generated text from different perspectives (e.g., informativeness, coherence, factuality, $\ S 4$ ) by adjusting the inputs and outputs of the conditional text generation problem, as we demonstrate in $\ S 3 . 2$ . This is in contrast to most previous work, which mostly examines correlation of the devised metrics with output quality from a limited number of perspectives. (3) BARTSCORE can be further enhanced by (i) providing textual prompts that bring the evaluation task closer to the pre-training task, or (ii) updating the underlying model by fine-tuning BART based on downstream generation tasks (e.g., text summarization). + +Experimentally, we evaluate different variants of BARTSCORE from 7 perspectives on 16 datasets. BARTSCORE achieves the best performance in 16 of 22 test settings against existing top-scoring metrics. Empirical results also show the effectiveness of the prompting strategy supported by BARTSCORE. For example, simply adding the phrase “such as” to the translated text when using BARTSCORE can lead to a $3 \%$ point absolute improvement in correlation on “German-English” machine translation (MT) evaluation. Additional analysis shows that BARTSCORE is more robust when dealing with high-quality texts generated by top-performing systems. + +# 2 Preliminaries + +# 2.1 Problem Formulation + +As stated above, our goal is to assess the quality of generated text [3, 46]. In this work, we focus on conditional text generation (e.g., machine translation), where the goal is to generate a hypothesis $( \boldsymbol { h } = h _ { 1 } , \cdots , h _ { m } )$ based on a given source text $( s = s _ { 1 } , \cdots , s _ { n } )$ . Commonly, one or multiple human-created references $( r = r _ { 1 } , \cdots , r _ { l } )$ are provided to aid this evaluation. + +# 2.2 Gold-standard Human Evaluation + +In general, the gold-standard method for evaluating such texts is still human evaluation, where human annotators assess the generated texts’ quality. This evaluation can be done from perspectives, and we list a few common varieties below (all are investigated in $\ S 4$ ): + +1. Informativeness (INFO): How well the generated hypothesis captures the key ideas of the source text [18]. +2. Relevance (REL): How consistent the generated hypothesis is with respect to the source text [19]. +3. Fluency (FLU): Whether the text has no formatting problems, capitalization errors or obviously ungrammatical sentences (e.g., fragments, missing components) that make the text difficult to read [13]. +4. Coherence (COH): Whether the text builds from sentence to sentence to a coherent body of information about a topic [7]. +5. Factuality (FAC): Whether the generated hypothesis contains only statements entailed by the source text [30]. +6. Semantic Coverage (COV): How many semantic content units from reference texts are covered by the generated hypothesis [49]. +7. Adequacy (ADE): Whether the output conveys the same meaning as the input sentence, and none of the message is lost, added, or distorted [29]. + +Most existing evaluation metrics were designed to cover a small subset of these perspectives. For example, BLEU [50] aims to capture the adequacy and fluency of translations, while ROUGE [36] was designed to match the semantic coverage metric. Some metrics, particularly trainable ones, can perform evaluation from different perspectives but generally require maximizing correlation with each type of judgment separately [8]. + +![](images/6dc5bc0e6cf11512055650092647de02eabd130d83aa9f95c1a0baa74e5cfd9f.jpg) +Figure 1: Evaluation metrics as different tasks, where $s _ { i }$ , $h _ { i }$ and $r _ { j }$ represent source, hypothesis and reference words respectively. + +As we describe more in $\ S 4$ , BARTSCORE can evaluate text from the great majority of these perspecFactuality Relevance Factuality Relevance tives, significantly expanding its applicability compared to these metrics. + +# s r2.3 Evaluation as Different Tasks + +There is a recent trend that leverages neural models for automated evaluation in different ways, as shown in Fig. 1. We first elaborate on their characteristics by highlighting differences in task formulation and evaluation perspectives. + +T1: Unsupervised Matching. Unsupervised matching metrics aim to measure the semantic equivalence between the reference and hypothesis by using a token-level matching functions in distributed representation space, such as BERTScore [75], MoverScore [76] or discrete string space like ROUGE [35], BLEU [50], CHRF [52]. Although similar matching functions can be used to assess the quality beyond semantic equivalence (e.g, factuality, a relationship between source text and hypothesis), to our knowledge prior research has not attested to the capability of unsupervised matching methods in this regard; we explore this further in our experiments (Tab. 5). + +T2: Supervised Regression. Regression-based models introduce a parameterized regression layer, which would be learned in a supervised fashion to accurately predict human judgments. Examples include recent metrics BLEURT [62], COMET [56] and traditional metrics like ${ \bar { S } } ^ { 3 }$ [51], VRM [21]. + +T3: Supervised Ranking. Evaluation can also be conceived as a ranking problem, where the main idea is to learn a scoring function that assigns a higher score to better hypotheses than to worse ones. Examples include COMET [56] and BEER [64], where COMET focuses the machine translation task and relies on human judgments to tune parameters in ranking or regression layers, and BEER combines many simple features in a tunable linear model of MT evaluation metrics. + +T4: Text Generation. In this work, we formulate evaluating generated text as a text generation task from pre-trained language models. The basic idea is that a high-quality hypothesis will be easily generated based on source or reference text or vice-versa. This has not been covered as extensively in previous work, with one notable exception being PRISM [65]. Our work differs from PRISM in several ways: (i) PRISM formulates evaluation as a paraphrasing task, whose definition that two texts are with the same meaning limits its applicable scenarios, like factuality evaluation in text summarization that takes source documents and generated summaries as input whose semantic space are different. (ii) PRISM trained a model from scratch on parallel data while BARTSCORE is based on open-sourced pre-trained seq2seq models. (iii) BARTSCORE supports prompt-based learning [59, 63] which hasn’t been examined in PRISM. + +# 3 BARTScore + +# 3.1 Sequence-to-Sequence Pre-trained Models + +Although pre-trained models differ along different axes, one of the main axes of variation is the training objective, with two main variants: language modeling objectives (e.g., masked language modeling [9]) and seq2seq objectives [54]. In particular, seq2seq pre-trained models are particularly well-suited to conditioned generation tasks since they consist of both an encoder and a decoder, and predictions are made auto-regressively [32]. In this work, we operationalize our idea by using + +BART [32] as our backbone due to its superior performance in text generation [12, 42, 71]. We also report preliminary experiments comparing BART with T5 [54] and PEGASUS [74] in the Appendix. + +Given a seq2seq model parameterized by $\theta$ , a source sequence containing $n$ tokens $\mathbf { x } = \{ x _ { 1 } , \cdots , x _ { n } \}$ and a target sequence containing $m$ tokens $\mathbf { y } = \{ y _ { 1 } , \dots , y _ { m } \}$ . We can factorize the generation probability of $\mathbf { y }$ conditioned on $\mathbf { x }$ as follows: + +$$ +p ( \mathbf { y } | \mathbf { x } , \theta ) = \prod _ { t = 1 } ^ { m } p ( \mathbf { y } _ { t } | \mathbf { y } _ { < t } , \mathbf { x } , \theta ) +$$ + +By exploring these probabilities, we design metrics that can gauge the quality of the generated text. + +# 3.2 BARTScore + +The most general form of our proposed BARTSCORE is shown in Eq. 2, where we use the weighted log probability of one text y given another text $\mathbf { x }$ . The weights are used to put different emphasis on different tokens, which can be instantiated using different methods like Inverse Document Frequency (IDF) [25] etc. In our work, we weigh each token equally.2 + +$$ +\mathbf { B A R T S C O R E } = \sum _ { t = 1 } ^ { m } \omega _ { t } \log p ( \mathbf { y } _ { t } | \mathbf { y } _ { < t } , \mathbf { x } , \theta ) +$$ + +Due to its generation task-based formulation and ability to utilize the entirety of BART’s pre-trained parameters, BARTSCORE can be flexibly used in different evaluation scenarios. We specifically present four methods for using BARTSCORE based on different generation directions, which are, + +• Faithfulness $s \to h$ ): from source document to hypothesis $p ( \boldsymbol { h } | \boldsymbol { s } , \boldsymbol { \theta } )$ . This direction measures how likely it is that the hypothesis could be generated based on the source text. Potential application scenarios are factuality and relevance introduced in $\ S 2 . 2$ . This measure can also be used for estimating measures of the quality of only the target text, such as coherence and fluency $( \ S 2 . 2 )$ . +• Precision $( r h )$ ): from reference text to system-generated text $p ( \boldsymbol { h } | \boldsymbol { r } , \boldsymbol { \theta } )$ . This direction assesses how likely the hypothesis could be constructed based on the gold reference and is suitable for the +precision-focused scenario. +• Recall $( h \to r$ ): from system-generated text to reference text $p ( \pmb { r } | \pmb { h } , \theta )$ . This version quantifies how easily a gold reference could be generated by the hypothesis and is suitable for pyramid-based evaluation (i.e., semantic coverage introduced in $\ S 2 . 2$ ) in summarization task since pyramid score measures fine-grained Semantic Content Units (SCUs) [49] covered by system-generated texts. +• $\mathcal { F }$ score $( r h$ ): Consider both directions and use the arithmetic average of Precision and Recall ones. This version can be broadly used to evaluate the semantic overlap (informativeness, adequacy detailed in $\ S 2 . 2 \AA ,$ ) between reference texts and generated texts. + +# 3.3 BARTScore Variants + +We also investigate two extensions to BARTSCORE: (i) changing $\mathbf { x }$ and $\mathbf { y }$ through prompting, which can bring the evaluation task closer to the pre-training task. (ii) changing $\theta$ by considering different fine-tuning tasks, which can bring the pre-training domain closer to the evaluation task. + +# 3.3.1 Prompt + +Prompting is a practice of adding short phrases to the input or output to encourage pre-trained models to perform specific tasks, which has been proven effective in several other NLP scenarios [24,57,58,60,63]. The generative formulation of BARTSCORE makes it relatively easy to incorporate these insights here as well; we name this variant BARTSCORE-PROMPT. + +Given a prompt of $l$ tokens $\mathbf { z } = \{ z _ { 1 } , \cdots , z _ { l } \}$ , we can either (i) append it to the source text, in which case we get $\mathbf { x } ^ { \prime } = \{ x _ { 1 } , \cdot \cdot \cdot , x _ { n } , \dot { z } _ { 1 } , \cdot \cdot \cdot , z _ { l } \}$ , and calculate the score based on this new source text using Eq.2. or (ii) prepend it to the target text, getting $\mathbf { y } ^ { \prime } = \{ z _ { 1 } , \cdot \cdot \cdot , z _ { l } , y _ { 1 } , \cdot \cdot \cdot , y _ { m } \}$ . Then we can also use Eq.2 given the new target text. + +# 3.3.2 Fine-tuning Task + +Different from BERT-based metrics, which typically use classification-based tasks (e.g., natural language inference) [67] to fine-tune, BARTSCORE can be fine-tuned using generation-based tasks, which will make the pre-training domain closer to the evaluation task. In this paper, we explore two downstream tasks. (1) Summarization. We use BART fine-tuned on CNNDM dataset [20], which is available off-the-shelf in Huggingface Transformers [70]. (2) Paraphrasing. We continue fine-tuning BART from (1) on ParaBank2 dataset [22], which contains a large paraphrase collection. We used a random subset of 30,000 data and fine-tuned for one epoch with a batch size of 20 and a learning rate of $5 e ^ { - 5 }$ . We used two 2080Ti GPUs, and the training time is less than one hour. + +# 4 Experiment + +This section aims to evaluate the reliability of different automated metrics, which is commonly achieved by quantifying how well different metrics correlate with human judgments using measures (e.g., Spearman Correlation [72]) defined below (§4.1.2). + +# 4.1 Baselines and Datasets + +# 4.1.1 Evaluation Metrics + +We comprehensively examine metrics outlined in $\ S 2 . 3$ , which either require human judgments to train (i.e., supervised metrics): COMET [56], BLEURT [62], or are human judgment-free (i.e., unsupervised): BLEU [50] ROUGE-1 and ROUGE-2, ROUGE-L, CHRF [52], PRISM [65], MoverScore [76], BERTScore [75]. The detailed comparisons of those metrics can be found in Appendix. We use the official code for each metric. + +# 4.1.2 Measures for Meta Evaluation + +Pearson Correlation [15] measures the linear correlation between two sets of data. Spearman Correlation [72] assesses the monotonic relationships between two variables. Kendall’s Tau [27] measures the ordinal association between two measured quantities. Accuracy, in our experiments, measures the percentage of correct ranking between factual texts and non-factual texts. We follow previous works in the choices of measures for different datasets to make a fair comparison. + +# 4.1.3 Datasets + +The datasets we use are summarized in Tab. 1. We consider three different tasks: summarization (SUM), machine translation (MT), and data-to-text (D2T). + +Machine Translation We obtain the source language sentences, machine-translated texts and reference texts from the WMT19 metrics shared task [44]. We use the DARR corpus and consider 7 language pairs, which are de-en, fi-en, gu-en, kk-en, lt-en, ru-en, zh-en. + +Text Summarization (1) REALSumm [4] is a metaevaluation dataset for text summarization which measures pyramid recall of each system-generated summary. (2) SummEval [13] is a collection of human judgments of model-generated summaries on the CNNDM dataset annotated by both expert judges and crowd-source workers. Each system generated summary is gauged through the lens of coherence, consistency, fluency and relevance.3 (3) NeR18 The + +Table 1: A summary of tasks, datasets, and evaluation perspectives that we have covered in our experiments. Explanation of evaluation perspectives can be found in $\ S 2 . 2$ . + +
TasksDatasetsEval. Perspectives
SUMREALSUMCov
SummEvalCOH FAC FLU INFO
NeR18COH FLU REL INFO
Rank19 QAGS-C QAGS-XFAC
DE FI GU KK IT RU ZHADE FLU
MT D2TBAGEL SFHOT SFRESINFO
+ +NEWSROOM dataset [18] contains 60 articles with summaries generated by 7 different methods are annotated with human scores in terms of coherence, fluency, informativeness, relevance. + +Factuality (1) Rank19 [14] is used to meta-evaluate factuality metrics. It is a collection of 373 triples of a source sentence with two summary sentences, one correct and one incorrect. (2) QAGS20 [66] collected 235 test outputs on CNNDM dataset from [16] and 239 test outputs on XSUM dataset [47] from BART fine-tuned on XSUM. Sentences in each summary are annotated with correctness scores w.r.t. factuality. + +Data to Text We consider the following datasets which target utterance generation for spoken dialogue systems. (1) BAGEL [45] provides information about restaurants. (2) SFHOT [69] provides information about hotels in San Francisco. (3) SFRES [69] provides information about restaurants in San Francisco. They contain 202, 398, and 581 samples respectively, each sample consists of one meaning representation, multiple references, and utterances generated by different systems. + +# 4.2 Setup + +# 4.2.1 Prompt Design + +To perform prompting, we first need to find proper prompts within a search space. Instead of considering a large discrete search space $[ 6 3 ] ^ { 4 }$ or continuous search space [34], we use simple heuristics to narrow our search space. In particular, we use manually devised seed prompts and gather paraphrases to construct our prompt set.5 The seed prompts and some examples of paraphrased prompts are shown in Tab. 2. Details are listed in the Appendix. + +Table 2: Seed prompts and examples of final prompts. “Number” denotes the size of our final prompt set that was acquired from the seed prompts. + +
UsageNumberSeedExample
s→h70in summaryin short，inword， atosum up
h←r34inother wordstorephrase it，that istosay，i.e.
+ +# 4.2.2 Settings + +Variants. We consider four variants of BARTSCORE, which are (1) BARTSCORE, which uses the vanilla BART; (2) BARTSCORE-CNN, which uses the BART fine-tuned on the summarization dataset CNNDM; (3) BARTSCORE-CNN-PARA, where BART is first fine-tuned on CNNDM, then fine-tuned on ParaBank2. (4) BARTSCORE-PROMPT, which is enhanced by adding prompts. + +Selection of Prompts. For the summarization and data-to-text tasks, we use all entries (either all prompts designed for $s h$ or all prompts designed for $h r$ depending on the BARTScore usage chosen) in the prompt set by prefixing the decoder input and getting different generation scores (calculated by Eq.2) for each hypothesis based on different prompts. We finally get the score for one hypothesis by taking the average of all its generation scores using different prompts ( [24]; details about prompt ensembling can be found in the Appendix). For the machine translation task, due to the more expensive computational cost brought by larger text sets, we first use WMT18 [43] as a development set to search for one best prompt and obtain the phrase “Such as”, which is then used for the test language pairs. + +Selection of BARTScore Usage. Although BARTSCORE can be used in different ways (shown in $\ S 3 . 2 )$ ), in different tasks, they can be chosen based on how targeted evaluation perspectives are defined (described in $\ S 2 . 2 \AA$ ) as well as the types of tasks. Specifically, (i) For those datasets whose gold standard human evaluation are obtained based on recall-based pyramid method, we adopt recall-based BARTSCORE $( h \to r$ ). (ii) For those datasets whose human judgments focus on linguistic quality (coherence, fluency) and factual correctness (factuality), or the source and hypothesis texts are in the same modality (i.e., language), we use faithfulness-based BARTSCORE $s h$ ). (iii) For data-to-text and machine translation tasks, to make a fair comparison, we use BARTSCORE with the F-score version that other existing works [65] have adopted when evaluating generated texts. + +Table 3: Kendall’s Tau correlation of different metrics on WMT19 dataset. The highest correlation for each language pair achieved by unsupervised method is bold, and the highest correlation overall is underlined. Avg. denotes the average correlation achieved by a metric across all language pairs. + +
de-enfi-en gu-enkk-enlt-enru-enzh-enAvg.
SUPERVISED METHODS
BLEURT0.1740.3740.3130.3720.3880.2200.4360.325
COMET0.2190.3690.3160.3780.4050.2260.4620.339
UNSUPERVISED METHODS
BLEU0.0540.2360.1940.2760.2490.1150.3210.206
CHRF0.1230.2920.2400.3230.3040.1770.3710.261
PRISM0.1990.3660.3200.3620.3820.2200.4340.326
BERTScore0.1900.3540.2920.3510.3810.2210.4300.317
BARTSCORE0.1560.3350.2730.3240.3220.1670.3890.281
+CNN0.1900.3650.3000.3480.3840.2080.4250.317
+ CNN+Para0.205t0.370t0.3160.378t0.386†0.219_0.442t0.331
+ CNN +Para +Prompt0.2380.3740.3180.376t0.386+0.2190.4470.337
+ +Table 4: Spearman correlation of different metrics on three human judgement datasets. For promptbased learning, we consider adding prompts to the best-performing BARTSCORE $( \Omega )$ on each dataset. The highest correlation overall for each aspect on each dataset is bold. + +
REALSummSummEvalNeR18Avg.
CovCOHFACFLUINFOCOHFLUINFOREL
ROUGE-10.4980.1670.1600.1150.3260.0950.1040.1300.1470.194
ROUGE-20.4230.1840.1870.1590.2900.0260.0480.0790.0910.165
ROUGE-L0.4880.1280.1150.1050.3110.0640.0720.0890.1060.164
BERTScore0.4400.2840.1100.1930.3120.1470.1700.1310.1630.217
MoverScore0.3720.1590.1570.1290.3180.1610.1200.1880.1950.200
PRISM0.4110.2490.3450.2540.2120.5730.5320.5610.5530.410
BARTSCORE0.4410.322†0.3110.2480.2640.679†0.670t0.646†0.604t0.465
+ CNN0.4750.448‡0.382†0.356t0.356t0.653†0.640t0.616†0.5670.499
+ CNN+Para0.4710.424†0.401‡0.378t0.3130.657t0.652t0.614†0.5620.497
+Ω+Prompt0.4880.4070.3780.338f0.368f0.701t0.679t0.686t0.6200.518
+ +Significance Tests. To perform rigorous analysis, we adopt the bootstrapping method (p-value $<$ 0.05) [28] for pair-wise significance tests. In all tables, we use $\dagger$ on BARTSCORE if it significantly $( p < 0 . 0 5 )$ outperforms other unsupervised metrics excluding BARTSCORE variants. We use $\ddagger$ on BARTSCORE if it significantly outperforms all other unsupervised metrics including BARTSCORE variants. + +# 4.3 Experimental Results + +# 4.3.1 Machine Translation + +Tab. 3 illustrates Kendall’s Tau correlation of diverse metrics on different language pairs. We can observe that: (1) BARTSCORE enhanced by fine-tuning tasks $\left( \mathrm { C N N + P a r a } \right)$ can significantly outperform all other unsupervised methods on five language pairs and achieve comparable results on the other two. (2) The performance of BARTSCORE can be further improved by simply adding a prompt (i.e., such as) without any other overhead. Notably, on the language pair ${ \tt d e } \mathrm { - } \in \mathrm { n }$ , using the prompt results in a 0.033 improvement, which even significantly surpasses existing state-of-the-art supervised metrics BLEURT and COMET. This suggests a promising future direction for metric design: searching for proper prompts to better leverage knowledge stored in pre-trained language models instead of training on human judgment data [31]. + +# 4.3.2 Text Summarization + +Tab. 4 shows the meta-evaluation results of different metrics on the summarization task. We can observe that: (1) Simply vanilla BARTSCORE can outperform BERTScore and MoverScore by a large margin on 8 settings except the INFO perspective on SummEval. Strikingly, it achieves improvements of 0.251 and 0.265 over BERTScore and MoverScore respectively. (2) The improvement on REALSum and SummEval datasets can be further improved when introducing fine-tuning tasks. However, fine-tuning does not improve on the NeR18 dataset, likely because this dataset only contains 7 systems with easily distinguishable quality, and vanilla BARTSCORE can already achieve a high level of correlation $( > 0 . 6$ on average). (3) Our prompt combination strategy can consistently improve the performance on informativeness, up to 0.072 Spearman correlation on the NeR18 dataset and 0.055 on SummEval. However, the performance from other perspectives such as fluency and factuality do not show consistent improvements, which we will elaborate on later (§4.4.2). + +Analysis on Factuality Datasets The goal of these datasets is to judge whether a short generated summary is faithful to the original long documents. As shown in Tab. 5, we observe that (1) BARTSCORE $+ \thinspace C N N$ can almost match human baseline on Rank19 and outperform all other metrics, including the most recent top-performing factuality metrics FactCC and QAGS by a large margin. (2) Using paraphrase as a fine-tuning task will reduce BARTSCORE’s performance, which is reasonable since these two texts (i.e., the summary and document) shouldn’t maintain the paraphrased relationship in general. (3) Introducing prompts does not bring an improvement, even resulting in a performance decrease. + +Table 5: Results on Rank19 and QAGS datasets. where “Q” represents QAGS. Metrics achieve highest correlation are bold. + +
Rank19 Q-CNNQ-XSUM
Acc.Pearson
ROUGE-10.5680.338-0.008
ROUGE-20.6300.4590.097
ROUGE-L0.5870.3570.024
BERTScore0.7130.5760.024
MoverScore0.7130.4140.054
PRISM0.7800.4790.025
FactCC [30] QAGS [66]0.700 0.7211 0.5451 0.175
Human [14] BARTSCORE0.83911 0.009
+CNN0.684 0.836‡0.661† 0.735±0.184‡
+ CNN+Para0.680t0.074
0.788
+CNN +Prompt70.7960.719f0.094
+ +# 4.3.3 Data-to-text + +The experiment results on data-to-text datasets are shown in Tab. 6. We observe that (1) finetuning on the CNNDM dataset can consistently boost the correlation, for example, up to 0.056 gain on BAGEL. (2) Additionally, further finetuning on paraphrase datasets results in even higher performance compared to the version without any fine-tuning, up to 0.083 Spearman correlation on BAGEL dataset. These results surpass all existing top-performing metrics. (3) Our proposed prompt combination strategy can consistently improve correlation, on average 0.028 Spearman correlation. This is consistent with the findings in $\ S 4 . 3 . 2$ that we can improve the aspect of informativeness through proper prompting. + +Table 6: Results on data-to-text datasets. We report Spearman correlation. Metrics achieve highest correlation are bold. + +
BAGELSFRESSFHOTAvg.
ROUGE-10.2340.1150.1180.156
ROUGE-20.1990.1160.0880.134
ROUGE-L0.1890.1030.1100.134
BERTScore0.2890.1560.1350.193
MoverScore0.2840.1530.1720.203
PRISM0.3050.1550.1960.219
BARTSCORE0.2470.164†0.1580.190
+ CNN0.3030.191†0.1900.228
+ CNN+Para0.330t0.185t0.211†0.242
+Ω+Prompt0.3360.238t0.235t0.270
+ +# 4.4 Analysis + +We design experiments to better understand the mechanism by which BARTSCORE obtains these promising results, specifically asking three questions: Q1: Compared to other unsupervised metrics, where does BARTSCORE outperform them? Q2: How does adding prompts benefit evaluation? Q3: Will BARTScore introduce biases in unpredictable ways? + +![](images/78f8c1e31ba1ab4fbe0920e68cff9da08fddef39259e3f11e995865e34618d98.jpg) +Figure 2: Fine-grained analysis (a,b) and prompt analysis (c). In (a, b), BE, PR, BL, CO, BA represent BERTScore, PRISM, BLEURT, COMET and BARTSCORE respectively. In (c), SEM, LIN, FAC denote semantic overlap, linguistic quality and factual correctness respectively. + +# 4.4.1 Fine-grained Analysis + +To answer Q1, we choose the MT task and break down the performance of each metric into different buckets based on different axes. + +Top-k Systems We report the average correlation across all language pairs achieved by each metric given only translations from top- $k$ systems. We vary the number of $k$ , and the results are shown in Fig. 2-(a). We can see that BARTSCORE can outperform all other metrics (including one supervised metric BLEURT) except the existing state-of-the-art supervised metric COMET for different $k$ , and the decrease in correlation becomes smoother than others when considering top-scoring systems. This indicates that BARTSCORE is robust to high-quality generated texts. + +Reference Length We break down each test set into four buckets based on the reference length, which are [15, 25), [25, 35), [35, 45), [45, 54] and compute the Kendall’s Tau average correlation of different metrics across all language pairs within each bucket.6 The results are shown in Fig. 2-(b). We observe that BARTSCORE can outperform or tie with other unsupervised metrics over different reference lengths. Also, its correlation with human judgments is more stable compared to all other metrics. This indicates its robustness to different input lengths. More other analyses can be found in Appendix. + +# 4.4.2 Prompt Analysis + +For Q2, we choose the summarization and data-to-text tasks for analysis where we used all prompts from our prompt set. We first group all the evaluation perspectives into three categories: (1) semantic overlap (informativeness, pyramid score, and relevance) (2) linguistic quality (fluency, coherence) (3) factual correctness (factuality). We then calculate the percentage of prompts that result in performance improvements for each perspective within a dataset. Finally, we compute the average percentage of prompts that can lead to performance gains for each category. The results are shown in Tab. 2-(c). We can see that for semantic overlap, almost all prompts can lead to the performance increase, while for factuality only a few prompts can improve the performance. This also explains the results in $\ S 4 . 3 . 2$ where we found that combining the results of different prompts can lead to consistent increases in semantic overlap but worse performance in factuality. Regarding linguistic quality, the effect of adding a prompt is not that predictive, which is also consistent with our findings in $\ S 4 . 3 . 2$ . + +# 4.4.3 Bias Analysis + +To answer Q3, we conduct bias analysis. Bias would indicate that the scores are too high or too low compared to the scores they are given by human annotators. Therefore, to see whether such biases exist, we inspected the rank differences given by human annotators and BARTScore (fine-tuned on CNNDM dataset) on the REALSumm dataset where 24 systems are considered, including both abstractive models and extractive models as well as models based on pre-trained models and models that are trained from scratch. We list all the systems below. And the resulting rank difference is shown in Fig. 3. + +![](images/71ce21efad222c710fbe5f7a584a046b9f08143d9d25180689982349d95a3ce5.jpg) +Figure 3: Bias analysis of BARTScore. The “Rank Difference" is the rank obtained using human judgements minus the rank got from BARTScore. Systems beginning with letter “E" are extractive systems while systems beginning with letter “A" are abstractive systems. + +Extractive Systems E1: BanditSum [11]; E2: Refresh [48]; E3: NeuSum [80]; E4: LSTMPN-RL [79]; E5: BERT-TF-SL [79]; E6: BERT-TF-PN [79]; E7: BERT-LSTM-PN-RL [79]; E8: BERT-LSTM-PN [79]; E9: HeterGraph [68]; E10: MatchSum [78]. + +Abstractive Systems A1: Ptr-Gen [61]; A2: Bottom-up [17]; A3: Fast-Abs-RL [5]; A4: Two-stageRL [73]; A5: BERT-Ext-Abs [41]; A6: BERT-Abs [41]; A7: Trans-Abs [41]; A8: UniLM-1 [10]; A9: UniLM-2 [2]; A10: T5-base [55]; A11: T5-large [55]; A12: T5-11B [55]; A13: BART [33]; A14: SemSim [42]. + +As shown in Fig. 3, BARTScore is less effective at distinguishing the quality of extractive summarization systems while much better at distinguishing the quality of abstractive summarization systems. However, given that there is a trend for using abstractive systems as more and more pre-trained sequence-to-sequence models being proposed, BARTScore’s weaknesses on extractive systems will be mitigated. + +# 5 Implications and Future Directions + +In this paper, we proposed a metric BARTSCORE that formulates evaluation of generated text as a text generation task, and empirically demonstrated its efficacy. Without the supervision of human judgments, BARTSCORE can effectively evaluate texts from 7 perspectives and achieve the best performance on 16 of 22 settings against existing top-scoring metrics. We highlight potential future directions based on what we have learned. + +Prompt-augmented metrics As an easy-to-use but powerful method, prompting [39] has achieved impressive performance particularly on semantic overlap-based evaluation perspectives. However, its effectiveness in factuality and linguistic quality-based perspectives has not been fully demonstrated in this paper. In the future, more works can explore how to make better use of prompts for these and other evaluation scenarios. + +Co-evolving evaluation metrics and systems BARTSCORE builds the connection between metric design and system design, which allows them to share their technological advances, thereby progressing together. For example, a better BART-based summarization system may be directly used as a more reliable automated metric for evaluating summaries, and this work makes them connected. + +# Acknowledgments + +The authors would like to thank the anonymous reviewers for their insightful comments and suggestions. The authors also thank Wei Zhao for assisting with reproducing baseline results. This work was supported by the Air Force Research Laboratory under agreement number FA8750-19-2-0200. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright notation thereon. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the Air Force Research Laboratory or the U.S. Government. + +References +[1] Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong. Learning to retrieve reasoning paths over wikipedia graph for question answering. arXiv preprint arXiv:1911.10470, 2019. +[2] Hangbo Bao, Li Dong, Furu Wei, Wenhui Wang, Nan Yang, Xiaodong Liu, Yu Wang, Jianfeng Gao, Songhao Piao, Ming Zhou, and Hsiao-Wuen Hon. Unilmv2: Pseudo-masked language models for unified language model pre-training. In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, volume 119 of Proceedings of Machine Learning Research, pages 642–652. PMLR, 2020. +[3] Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, and Pengfei Liu. Metrics also disagree in the low scoring range: Revisiting summarization evaluation metrics. In Proceedings of the 28th International Conference on Computational Linguistics, pages 5702–5711, Barcelona, Spain (Online), December 2020. International Committee on Computational Linguistics. +[4] Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, Pengfei Liu, and Graham Neubig. Re-evaluating evaluation in text summarization. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9347–9359, Online, November 2020. Association for Computational Linguistics. +[5] Yen-Chun Chen and Mohit Bansal. Fast abstractive summarization with reinforce-selected sentence rewriting. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 675–686, Melbourne, Australia, July 2018. Association for Computational Linguistics. +[6] Andrew M Dai and Quoc V Le. Semi-supervised sequence learning. In C. Cortes, N. Lawrence, D. Lee, M. Sugiyama, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 28. Curran Associates, Inc., 2015. +[7] Hoa Trang Dang. Overview of duc 2005. In Proceedings of the document understanding conference, volume 2005, pages 1–12, 2005. +[8] Michael Denkowski and Alon Lavie. Extending the METEOR machine translation evaluation metric to the phrase level. In Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, pages 250–253, Los Angeles, California, June 2010. Association for Computational Linguistics. +[9] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, 2019. +[10] Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming Zhou, and Hsiao-Wuen Hon. Unified language model pre-training for natural language understanding and generation. In Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence d’Alché-Buc, Emily B. Fox, and Roman Garnett, editors, Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019, NeurIPS 2019, December 8-14, 2019, Vancouver, BC, Canada, pages 13042–13054, 2019. +[11] Yue Dong, Yikang Shen, Eric Crawford, Herke van Hoof, and Jackie Chi Kit Cheung. BanditSum: Extractive summarization as a contextual bandit. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 3739–3748, Brussels, Belgium, October-November 2018. Association for Computational Linguistics. +[12] Zi-Yi Dou, Pengfei Liu, Hiroaki Hayashi, Zhengbao Jiang, and Graham Neubig. GSum: A general framework for guided neural abstractive summarization. In Meeting of the North American Chapter of the Association for Computational Linguistics (NAACL), Mexico City, June 2021. +[13] A. R. Fabbri, Wojciech Kryscinski, Bryan McCann, R. Socher, and Dragomir Radev. Summeval: Re-evaluating summarization evaluation. Trans. Assoc. Comput. Linguistics, 9:391–409, 2021. +[14] Tobias Falke, Leonardo FR Ribeiro, Prasetya Ajie Utama, Ido Dagan, and Iryna Gurevych. Ranking generated summaries by correctness: An interesting but challenging application for natural language inference. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2214–2220, 2019. +[15] David Freedman, Robert Pisani, and Roger Purves. Statistics (international student edition). Pisani, R. Purves, 4th edn. WW Norton & Company, New York, 2007. +[16] Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. Bottom-up abstractive summarization. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4098–4109, 2018. +[17] Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. Bottom-up abstractive summarization. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4098–4109, Brussels, Belgium, October-November 2018. Association for Computational Linguistics. +[18] Max Grusky, Mor Naaman, and Yoav Artzi. Newsroom: A dataset of 1.3 million summaries with diverse extractive strategies. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 708–719, New Orleans, Louisiana, June 2018. Association for Computational Linguistics. +[19] Max Grusky, Mor Naaman, and Yoav Artzi. Newsroom: A dataset of 1.3 million summaries with diverse extractive strategies. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), volume 1, pages 708–719, 2018. +[20] K. Hermann, Tomás Kociský, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and P. Blunsom. Teaching machines to read and comprehend. In NIPS, 2015. +[21] Tsutomu Hirao, Manabu Okumura, Norihito Yasuda, and Hideki Isozaki. Supervised automatic evaluation for summarization with voted regression model. Information Processing & Management, 43(6):1521–1535, 2007. +[22] J. Edward Hu, Abhinav Singh, Nils Holzenberger, Matt Post, and Benjamin Van Durme. Largescale, diverse, paraphrastic bitexts via sampling and clustering. In Proceedings of the $2 3 r d$ Conference on Computational Natural Language Learning (CoNLL), pages 44–54, Hong Kong, China, November 2019. Association for Computational Linguistics. +[23] Sarthak Jain, Madeleine van Zuylen, Hannaneh Hajishirzi, and Iz Beltagy. Scirex: A challenge dataset for document-level information extraction. arXiv preprint arXiv:2005.00512, 2020. +[24] Zhengbao Jiang, Frank F. Xu, Jun Araki, and Graham Neubig. How can we know what language models know? Transactions of the Association for Computational Linguistics, 8:423–438, 2020. +[25] Karen Sparck Jones. A statistical interpretation of term specificity and its application in retrieval. Journal of documentation, 1972. +[26] Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, ˘ Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering. arXiv preprint arXiv:2004.04906, 2020. +[27] M. G. Kendall. A new measure of rank correlation. Biometrika, 30(1/2):81–93, 1938. +[28] Philipp Koehn. Statistical significance tests for machine translation evaluation. In Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing, pages 388–395, Barcelona, Spain, July 2004. Association for Computational Linguistics. +[29] Philipp Koehn. Statistical machine translation. Cambridge University Press, 2009. +[30] Wojciech Kryscinski, Bryan McCann, Caiming Xiong, and Richard Socher. Evaluating the factual consistency of abstractive text summarization. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9332–9346, Online, November 2020. Association for Computational Linguistics. +[31] Teven Le Scao and Alexander Rush. How many data points is a prompt worth? In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2627–2636, Online, June 2021. Association for Computational Linguistics. +[32] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov, and Luke Zettlemoyer. Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. arXiv preprint arXiv:1910.13461, 2019. +[33] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. BART: Denoising sequence-to-sequence pretraining for natural language generation, translation, and comprehension. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7871–7880, Online, July 2020. Association for Computational Linguistics. +[34] Xiang Lisa Li and Percy Liang. Prefix-tuning: Optimizing continuous prompts for generation. ArXiv, abs/2101.00190, 2021. +[35] Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. Text Summarization Branches Out, 2004. +[36] Chin-Yew Lin and Eduard Hovy. Automatic evaluation of summaries using n-gram cooccurrence statistics. In Proceedings of the 2003 Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics, pages 150–157, 2003. +[37] Ying Lin, Heng Ji, Fei Huang, and Lingfei Wu. A joint neural model for information extraction with global features. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7999–8009, 2020. +[38] Pengfei Liu, Jinlan Fu, Yang Xiao, Weizhe Yuan, Shuaicheng Chang, Junqi Dai, Yixin Liu, Zihuiwen Ye, and Graham Neubig. Explainaboard: An explainable leaderboard for nlp. arXiv preprint arXiv:2104.06387, 2021. +[39] Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig. Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing, 2021. +[40] Yang Liu and Mirella Lapata. Text summarization with pretrained encoders. arXiv preprint arXiv:1908.08345, 2019. +[41] Yang Liu and Mirella Lapata. Text summarization with pretrained encoders. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3730–3740, Hong Kong, China, November 2019. Association for Computational Linguistics. +[42] Yixin Liu and Pengfei Liu. Simcls: A simple framework for contrastive learning of abstractive summarization. arXiv preprint arXiv:2106.01890, 2021. +[43] Qingsong Ma, Ondˇrej Bojar, and Yvette Graham. Results of the WMT18 metrics shared task: Both characters and embeddings achieve good performance. In Proceedings of the Third Conference on Machine Translation: Shared Task Papers, pages 671–688, Belgium, Brussels, October 2018. Association for Computational Linguistics. +[44] Qingsong Ma, Johnny Wei, Ondˇrej Bojar, and Yvette Graham. Results of the WMT19 metrics shared task: Segment-level and strong MT systems pose big challenges. In Proceedings of the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages 62–90, Florence, Italy, August 2019. Association for Computational Linguistics. +[45] François Mairesse, Milica Gasic, Filip Jurcicek, Simon Keizer, Blaise Thomson, Kai Yu, and Steve Young. Phrase-based statistical language generation using graphical models and active learning. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 1552–1561, 2010. +[46] Nitika Mathur, Timothy Baldwin, and Trevor Cohn. Tangled up in BLEU: Reevaluating the evaluation of automatic machine translation evaluation metrics. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4984–4997, Online, July 2020. Association for Computational Linguistics. +[47] Shashi Narayan, Shay B Cohen, and Mirella Lapata. Don’t give me the details, just the summary! topic-aware convolutional neural networks for extreme summarization. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1797–1807, 2018. +[48] Shashi Narayan, Shay B. Cohen, and Mirella Lapata. Ranking sentences for extractive summarization with reinforcement learning. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 1747–1759, New Orleans, Louisiana, June 2018. Association for Computational Linguistics. +[49] Ani Nenkova and Rebecca Passonneau. Evaluating content selection in summarization: The pyramid method. In Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics: HLT-NAACL 2004, pages 145–152, Boston, Massachusetts, USA, May 2 - May 7 2004. Association for Computational Linguistics. +[50] Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA, July 2002. Association for Computational Linguistics. +[51] Maxime Peyrard, Teresa Botschen, and Iryna Gurevych. Learning to score system summaries for better content selection evaluation. In Proceedings of the Workshop on New Frontiers in Summarization, pages 74–84, Copenhagen, Denmark, September 2017. Association for Computational Linguistics. +[52] Maja Popovic. chrF: character n-gram F-score for automatic MT evaluation. In ´ Proceedings of the Tenth Workshop on Statistical Machine Translation, pages 392–395, Lisbon, Portugal, September 2015. Association for Computational Linguistics. +[53] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019. +[54] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67, 2020. +[55] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. J. Mach. Learn. Res., 21:140:1–140:67, 2020. +[56] Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. COMET: A neural framework for MT evaluation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2685–2702, Online, November 2020. Association for Computational Linguistics. +[57] Laria Reynolds and Kyle McDonell. Prompt programming for large language models: Beyond the few-shot paradigm. In Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems, pages 1–7, 2021. +[58] Timo Schick and Hinrich Schütze. Exploiting cloze questions for few-shot text classification and natural language inference. arXiv preprint arXiv:2001.07676, 2020. +[59] Timo Schick and Hinrich Schütze. Few-shot text generation with pattern-exploiting training. arXiv preprint arXiv:2012.11926, 2020. +[60] Timo Schick and Hinrich Schütze. It’s not just size that matters: Small language models are also few-shot learners. arXiv preprint arXiv:2009.07118, 2020. +[61] Abigail See, Peter J. Liu, and Christopher D. Manning. Get to the point: Summarization with pointer-generator networks. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1073–1083, Vancouver, Canada, July 2017. Association for Computational Linguistics. +[62] Thibault Sellam, Dipanjan Das, and Ankur Parikh. BLEURT: Learning robust metrics for text generation. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7881–7892, Online, July 2020. Association for Computational Linguistics. +[63] Taylor Shin, Yasaman Razeghi, Robert L Logan IV, Eric Wallace, and Sameer Singh. Autoprompt: Eliciting knowledge from language models with automatically generated prompts. arXiv preprint arXiv:2010.15980, 2020. +[64] Miloš Stanojevic and Khalil Sima’an. BEER: BEtter evaluation as ranking. In ´ Proceedings of the Ninth Workshop on Statistical Machine Translation, pages 414–419, Baltimore, Maryland, USA, June 2014. Association for Computational Linguistics. +[65] Brian Thompson and Matt Post. Automatic machine translation evaluation in many languages via zero-shot paraphrasing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 90–121, Online, November 2020. Association for Computational Linguistics. +[66] Alex Wang, Kyunghyun Cho, and M. Lewis. Asking and answering questions to evaluate the factual consistency of summaries. In ACL, 2020. +[67] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel Bowman. GLUE: A multi-task benchmark and analysis platform for natural language understanding. In Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 353–355, Brussels, Belgium, November 2018. Association for Computational Linguistics. +[68] Danqing Wang, Pengfei Liu, Yining Zheng, Xipeng Qiu, and Xuanjing Huang. Heterogeneous graph neural networks for extractive document summarization. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6209–6219, Online, July 2020. Association for Computational Linguistics. +[69] Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Pei-Hao Su, David Vandyke, and Steve Young. Semantically conditioned lstm-based natural language generation for spoken dialogue systems. arXiv preprint arXiv:1508.01745, 2015. +[70] Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 38–45, Online, October 2020. Association for Computational Linguistics. +[71] Weizhe Yuan, Pengfei Liu, and Graham Neubig. Can we automate scientific reviewing? arXiv preprint arXiv:2102.00176, 2021. +[72] Jerrold H Zar. Spearman rank correlation. Encyclopedia of Biostatistics, 7, 2005. +[73] Haoyu Zhang, Jingjing Cai, Jianjun Xu, and Ji Wang. Pretraining-based natural language generation for text summarization. In Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL), pages 789–797, Hong Kong, China, November 2019. Association for Computational Linguistics. +[74] Jingqing Zhang, Yao Zhao, Mohammad Saleh, and Peter Liu. Pegasus: Pre-training with extracted gap-sentences for abstractive summarization. In International Conference on Machine Learning, pages 11328–11339. PMLR, 2020. +[75] Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. Bertscore: Evaluating text generation with bert. In International Conference on Learning Representations, 2020. +[76] Wei Zhao, Maxime Peyrard, Fei Liu, Yang Gao, Christian M. Meyer, and Steffen Eger. MoverScore: Text generation evaluating with contextualized embeddings and earth mover distance. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 563–578, Hong Kong, China, November 2019. Association for Computational Linguistics. +[77] Ming Zhong, Pengfei Liu, Yiran Chen, Danqing Wang, Xipeng Qiu, and Xuanjing Huang. Extractive summarization as text matching. arXiv preprint arXiv:2004.08795, 2020. +[78] Ming Zhong, Pengfei Liu, Yiran Chen, Danqing Wang, Xipeng Qiu, and Xuanjing Huang. Extractive summarization as text matching. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6197–6208, Online, July 2020. Association for Computational Linguistics. +[79] Ming Zhong, Pengfei Liu, Danqing Wang, Xipeng Qiu, and Xuanjing Huang. Searching for effective neural extractive summarization: What works and what’s next. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1049–1058, Florence, Italy, July 2019. Association for Computational Linguistics. +[80] Qingyu Zhou, Nan Yang, Furu Wei, Shaohan Huang, Ming Zhou, and Tiejun Zhao. Neural document summarization by jointly learning to score and select sentences. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 654–663, Melbourne, Australia, July 2018. Association for Computational Linguistics. \ No newline at end of file diff --git a/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_content_list.json b/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..e6e67a682b1b2896ac966d4b1beecf3ac304cf44 --- /dev/null +++ b/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_content_list.json @@ -0,0 +1,1278 @@ +[ + { + "type": "text", + "text": "BARTSCORE: Evaluating Generated Text as Text Generation ", + "text_level": 1, + "bbox": [ + 218, + 122, + 781, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Weizhe Yuan Carnegie Mellon University weizhey@cs.cmu.edu ", + "bbox": [ + 194, + 226, + 377, + 267 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Graham Neubig Carnegie Mellon University gneubig@cs.cmu.edu ", + "bbox": [ + 406, + 226, + 589, + 268 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Pengfei Liu ∗ Carnegie Mellon University pliu3@cs.cmu.edu ", + "bbox": [ + 617, + 226, + 803, + 268 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 304, + 535, + 320 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "A wide variety of NLP applications, such as machine translation, summarization, and dialog, involve text generation. One major challenge for these applications is how to evaluate whether such generated texts are actually fluent, accurate, or effective. In this work, we conceptualize the evaluation of generated text as a text generation problem, modeled using pre-trained sequence-to-sequence models. The general idea is that models trained to convert the generated text to/from a reference output or the source text will achieve higher scores when the generated text is better. We operationalize this idea using BART [32], an encoder-decoder based pre-trained model, and propose a metric BARTSCORE with a number of variants that can be flexibly applied in an unsupervised fashion to evaluation of text from different perspectives (e.g. informativeness, fluency, or factuality). BARTSCORE is conceptually simple and empirically effective. It can outperform existing top-scoring metrics in 16 of 22 test settings, covering evaluation of 16 datasets (e.g., machine translation, text summarization) and 7 different perspectives (e.g., informativeness, factuality). Code to calculate BARTScore is available at https://github.com/neulab/BARTScore, and we have released an interactive leaderboard for meta-evaluation at http: //explainaboard.nlpedia.ai/leaderboard/task-meval/ on the EXPLAINABOARD platform [38], which allows us to interactively understand the strengths, weaknesses, and complementarity of each metric. ", + "bbox": [ + 232, + 334, + 766, + 609 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 Introduction ", + "text_level": 1, + "bbox": [ + 174, + 633, + 310, + 651 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "One defining feature of recent NLP models is the use of neural representations trained on raw text, using unsupervised objectives such as language modeling [6,53], or denoising autoencoding [9,32,54]. By learning to predict the words or sentences in natural text, these models simultaneously learn to extract features that not only benefit mainstream NLP tasks such as information extraction [23, 37], question answering [1, 26], text summarization [40, 77] but also have proven effective in development of automatic metrics for evaluation of text generation itself [62, 65]. For example, BERTScore [75] and MoverScore [76] take features extracted by BERT [9] and apply unsupervised matching functions to compare system outputs against references. Other works build supervised frameworks that use the extracted features to learn to rank [56] or regress [62] to human evaluation scores. ", + "bbox": [ + 174, + 665, + 825, + 790 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "However, in the context of generation evaluation, one may note that there is a decided disconnect between how models are pre-trained using text generation objectives and how they are used as down-stream feature extractors. This leads to potential under-utilization of the pre-trained model parameters. For example, the output prediction layer is not used at all in this case. This disconnect is particularly striking because of the close connection between the pre-training objectives and the generation tasks we want to evaluate. ", + "bbox": [ + 174, + 796, + 825, + 878 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In this paper, we instead argue for a formulation of evaluation of generated text as a text generation problem, directly evaluating text through the lens of its probability of being generated from or generating other textual inputs and outputs. This is a better match with the underlying pre-training tasks and allows us to more fully take advantage of the parameters learned during the pre-training phase. We solve the modeling problem with a pre-trained sequence-to-sequence (seq2seq) model, specifically BART [32], and devise a metric named BARTSCORE, which has the following characteristics: (1) BARTSCORE is parameter- and data-efficient. Architecturally there are no extra parameters beyond those used in pre-training itself, and it is an unsupervised metric that doesn’t require human judgments to train. (2) BARTSCORE can better support evaluation of generated text from different perspectives (e.g., informativeness, coherence, factuality, $\\ S 4$ ) by adjusting the inputs and outputs of the conditional text generation problem, as we demonstrate in $\\ S 3 . 2$ . This is in contrast to most previous work, which mostly examines correlation of the devised metrics with output quality from a limited number of perspectives. (3) BARTSCORE can be further enhanced by (i) providing textual prompts that bring the evaluation task closer to the pre-training task, or (ii) updating the underlying model by fine-tuning BART based on downstream generation tasks (e.g., text summarization). ", + "bbox": [ + 173, + 92, + 825, + 299 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Experimentally, we evaluate different variants of BARTSCORE from 7 perspectives on 16 datasets. BARTSCORE achieves the best performance in 16 of 22 test settings against existing top-scoring metrics. Empirical results also show the effectiveness of the prompting strategy supported by BARTSCORE. For example, simply adding the phrase “such as” to the translated text when using BARTSCORE can lead to a $3 \\%$ point absolute improvement in correlation on “German-English” machine translation (MT) evaluation. Additional analysis shows that BARTSCORE is more robust when dealing with high-quality texts generated by top-performing systems. ", + "bbox": [ + 174, + 304, + 825, + 402 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 Preliminaries ", + "text_level": 1, + "bbox": [ + 174, + 420, + 318, + 438 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.1 Problem Formulation ", + "text_level": 1, + "bbox": [ + 174, + 450, + 362, + 465 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "As stated above, our goal is to assess the quality of generated text [3, 46]. In this work, we focus on conditional text generation (e.g., machine translation), where the goal is to generate a hypothesis $( \\boldsymbol { h } = h _ { 1 } , \\cdots , h _ { m } )$ based on a given source text $( s = s _ { 1 } , \\cdots , s _ { n } )$ . Commonly, one or multiple human-created references $( r = r _ { 1 } , \\cdots , r _ { l } )$ are provided to aid this evaluation. ", + "bbox": [ + 173, + 477, + 825, + 532 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.2 Gold-standard Human Evaluation ", + "text_level": 1, + "bbox": [ + 174, + 547, + 452, + 563 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In general, the gold-standard method for evaluating such texts is still human evaluation, where human annotators assess the generated texts’ quality. This evaluation can be done from perspectives, and we list a few common varieties below (all are investigated in $\\ S 4$ ): ", + "bbox": [ + 176, + 573, + 825, + 614 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "1. Informativeness (INFO): How well the generated hypothesis captures the key ideas of the source text [18]. \n2. Relevance (REL): How consistent the generated hypothesis is with respect to the source text [19]. \n3. Fluency (FLU): Whether the text has no formatting problems, capitalization errors or obviously ungrammatical sentences (e.g., fragments, missing components) that make the text difficult to read [13]. \n4. Coherence (COH): Whether the text builds from sentence to sentence to a coherent body of information about a topic [7]. \n5. Factuality (FAC): Whether the generated hypothesis contains only statements entailed by the source text [30]. \n6. Semantic Coverage (COV): How many semantic content units from reference texts are covered by the generated hypothesis [49]. \n7. Adequacy (ADE): Whether the output conveys the same meaning as the input sentence, and none of the message is lost, added, or distorted [29]. ", + "bbox": [ + 210, + 625, + 825, + 833 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Most existing evaluation metrics were designed to cover a small subset of these perspectives. For example, BLEU [50] aims to capture the adequacy and fluency of translations, while ROUGE [36] was designed to match the semantic coverage metric. Some metrics, particularly trainable ones, can perform evaluation from different perspectives but generally require maximizing correlation with each type of judgment separately [8]. ", + "bbox": [ + 174, + 842, + 825, + 911 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/6dc5bc0e6cf11512055650092647de02eabd130d83aa9f95c1a0baa74e5cfd9f.jpg", + "image_caption": [ + "Figure 1: Evaluation metrics as different tasks, where $s _ { i }$ , $h _ { i }$ and $r _ { j }$ represent source, hypothesis and reference words respectively. " + ], + "image_footnote": [], + "bbox": [ + 194, + 0, + 797, + 195 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "As we describe more in $\\ S 4$ , BARTSCORE can evaluate text from the great majority of these perspecFactuality Relevance Factuality Relevance tives, significantly expanding its applicability compared to these metrics. ", + "bbox": [ + 173, + 262, + 823, + 291 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "s r2.3 Evaluation as Different Tasks ", + "text_level": 1, + "bbox": [ + 176, + 313, + 416, + 327 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "There is a recent trend that leverages neural models for automated evaluation in different ways, as shown in Fig. 1. We first elaborate on their characteristics by highlighting differences in task formulation and evaluation perspectives. ", + "bbox": [ + 176, + 340, + 823, + 382 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "T1: Unsupervised Matching. Unsupervised matching metrics aim to measure the semantic equivalence between the reference and hypothesis by using a token-level matching functions in distributed representation space, such as BERTScore [75], MoverScore [76] or discrete string space like ROUGE [35], BLEU [50], CHRF [52]. Although similar matching functions can be used to assess the quality beyond semantic equivalence (e.g, factuality, a relationship between source text and hypothesis), to our knowledge prior research has not attested to the capability of unsupervised matching methods in this regard; we explore this further in our experiments (Tab. 5). ", + "bbox": [ + 174, + 388, + 825, + 486 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "T2: Supervised Regression. Regression-based models introduce a parameterized regression layer, which would be learned in a supervised fashion to accurately predict human judgments. Examples include recent metrics BLEURT [62], COMET [56] and traditional metrics like ${ \\bar { S } } ^ { 3 }$ [51], VRM [21]. ", + "bbox": [ + 174, + 492, + 823, + 535 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "T3: Supervised Ranking. Evaluation can also be conceived as a ranking problem, where the main idea is to learn a scoring function that assigns a higher score to better hypotheses than to worse ones. Examples include COMET [56] and BEER [64], where COMET focuses the machine translation task and relies on human judgments to tune parameters in ranking or regression layers, and BEER combines many simple features in a tunable linear model of MT evaluation metrics. ", + "bbox": [ + 174, + 540, + 825, + 609 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "T4: Text Generation. In this work, we formulate evaluating generated text as a text generation task from pre-trained language models. The basic idea is that a high-quality hypothesis will be easily generated based on source or reference text or vice-versa. This has not been covered as extensively in previous work, with one notable exception being PRISM [65]. Our work differs from PRISM in several ways: (i) PRISM formulates evaluation as a paraphrasing task, whose definition that two texts are with the same meaning limits its applicable scenarios, like factuality evaluation in text summarization that takes source documents and generated summaries as input whose semantic space are different. (ii) PRISM trained a model from scratch on parallel data while BARTSCORE is based on open-sourced pre-trained seq2seq models. (iii) BARTSCORE supports prompt-based learning [59, 63] which hasn’t been examined in PRISM. ", + "bbox": [ + 173, + 616, + 825, + 755 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 BARTScore ", + "text_level": 1, + "bbox": [ + 174, + 779, + 303, + 796 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1 Sequence-to-Sequence Pre-trained Models ", + "text_level": 1, + "bbox": [ + 174, + 814, + 506, + 829 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Although pre-trained models differ along different axes, one of the main axes of variation is the training objective, with two main variants: language modeling objectives (e.g., masked language modeling [9]) and seq2seq objectives [54]. In particular, seq2seq pre-trained models are particularly well-suited to conditioned generation tasks since they consist of both an encoder and a decoder, and predictions are made auto-regressively [32]. In this work, we operationalize our idea by using ", + "bbox": [ + 174, + 842, + 825, + 911 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "BART [32] as our backbone due to its superior performance in text generation [12, 42, 71]. We also report preliminary experiments comparing BART with T5 [54] and PEGASUS [74] in the Appendix. ", + "bbox": [ + 171, + 90, + 825, + 119 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Given a seq2seq model parameterized by $\\theta$ , a source sequence containing $n$ tokens $\\mathbf { x } = \\{ x _ { 1 } , \\cdots , x _ { n } \\}$ and a target sequence containing $m$ tokens $\\mathbf { y } = \\{ y _ { 1 } , \\dots , y _ { m } \\}$ . We can factorize the generation probability of $\\mathbf { y }$ conditioned on $\\mathbf { x }$ as follows: ", + "bbox": [ + 174, + 126, + 823, + 167 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/3d003f060de88a32edea6412d277ef24062aa11f4ba7b6d0a401b82261881f32.jpg", + "text": "$$\np ( \\mathbf { y } | \\mathbf { x } , \\theta ) = \\prod _ { t = 1 } ^ { m } p ( \\mathbf { y } _ { t } | \\mathbf { y } _ { < t } , \\mathbf { x } , \\theta )\n$$", + "text_format": "latex", + "bbox": [ + 393, + 166, + 604, + 207 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "By exploring these probabilities, we design metrics that can gauge the quality of the generated text. ", + "bbox": [ + 173, + 217, + 823, + 233 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2 BARTScore ", + "text_level": 1, + "bbox": [ + 173, + 247, + 295, + 262 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The most general form of our proposed BARTSCORE is shown in Eq. 2, where we use the weighted log probability of one text y given another text $\\mathbf { x }$ . The weights are used to put different emphasis on different tokens, which can be instantiated using different methods like Inverse Document Frequency (IDF) [25] etc. In our work, we weigh each token equally.2 ", + "bbox": [ + 173, + 272, + 825, + 329 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/bfa67b3ee2347ec4fa7a57ebb4b52d83186124f1544c141e530ce54a2d1cb637.jpg", + "text": "$$\n\\mathbf { B A R T S C O R E } = \\sum _ { t = 1 } ^ { m } \\omega _ { t } \\log p ( \\mathbf { y } _ { t } | \\mathbf { y } _ { < t } , \\mathbf { x } , \\theta )\n$$", + "text_format": "latex", + "bbox": [ + 354, + 333, + 642, + 375 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Due to its generation task-based formulation and ability to utilize the entirety of BART’s pre-trained parameters, BARTSCORE can be flexibly used in different evaluation scenarios. We specifically present four methods for using BARTSCORE based on different generation directions, which are, ", + "bbox": [ + 174, + 377, + 825, + 421 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "• Faithfulness $s \\to h$ ): from source document to hypothesis $p ( \\boldsymbol { h } | \\boldsymbol { s } , \\boldsymbol { \\theta } )$ . This direction measures how likely it is that the hypothesis could be generated based on the source text. Potential application scenarios are factuality and relevance introduced in $\\ S 2 . 2$ . This measure can also be used for estimating measures of the quality of only the target text, such as coherence and fluency $( \\ S 2 . 2 )$ . \n• Precision $( r h )$ ): from reference text to system-generated text $p ( \\boldsymbol { h } | \\boldsymbol { r } , \\boldsymbol { \\theta } )$ . This direction assesses how likely the hypothesis could be constructed based on the gold reference and is suitable for the \nprecision-focused scenario. \n• Recall $( h \\to r$ ): from system-generated text to reference text $p ( \\pmb { r } | \\pmb { h } , \\theta )$ . This version quantifies how easily a gold reference could be generated by the hypothesis and is suitable for pyramid-based evaluation (i.e., semantic coverage introduced in $\\ S 2 . 2$ ) in summarization task since pyramid score measures fine-grained Semantic Content Units (SCUs) [49] covered by system-generated texts. \n• $\\mathcal { F }$ score $( r h$ ): Consider both directions and use the arithmetic average of Precision and Recall ones. This version can be broadly used to evaluate the semantic overlap (informativeness, adequacy detailed in $\\ S 2 . 2 \\AA ,$ ) between reference texts and generated texts. ", + "bbox": [ + 176, + 431, + 826, + 625 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.3 BARTScore Variants ", + "text_level": 1, + "bbox": [ + 174, + 640, + 359, + 655 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We also investigate two extensions to BARTSCORE: (i) changing $\\mathbf { x }$ and $\\mathbf { y }$ through prompting, which can bring the evaluation task closer to the pre-training task. (ii) changing $\\theta$ by considering different fine-tuning tasks, which can bring the pre-training domain closer to the evaluation task. ", + "bbox": [ + 176, + 665, + 823, + 708 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.3.1 Prompt ", + "text_level": 1, + "bbox": [ + 174, + 722, + 279, + 736 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Prompting is a practice of adding short phrases to the input or output to encourage pre-trained models to perform specific tasks, which has been proven effective in several other NLP scenarios [24,57,58,60,63]. The generative formulation of BARTSCORE makes it relatively easy to incorporate these insights here as well; we name this variant BARTSCORE-PROMPT. ", + "bbox": [ + 174, + 746, + 825, + 801 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Given a prompt of $l$ tokens $\\mathbf { z } = \\{ z _ { 1 } , \\cdots , z _ { l } \\}$ , we can either (i) append it to the source text, in which case we get $\\mathbf { x } ^ { \\prime } = \\{ x _ { 1 } , \\cdot \\cdot \\cdot , x _ { n } , \\dot { z } _ { 1 } , \\cdot \\cdot \\cdot , z _ { l } \\}$ , and calculate the score based on this new source text using Eq.2. or (ii) prepend it to the target text, getting $\\mathbf { y } ^ { \\prime } = \\{ z _ { 1 } , \\cdot \\cdot \\cdot , z _ { l } , y _ { 1 } , \\cdot \\cdot \\cdot , y _ { m } \\}$ . Then we can also use Eq.2 given the new target text. ", + "bbox": [ + 174, + 806, + 825, + 863 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.3.2 Fine-tuning Task ", + "text_level": 1, + "bbox": [ + 174, + 90, + 343, + 106 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Different from BERT-based metrics, which typically use classification-based tasks (e.g., natural language inference) [67] to fine-tune, BARTSCORE can be fine-tuned using generation-based tasks, which will make the pre-training domain closer to the evaluation task. In this paper, we explore two downstream tasks. (1) Summarization. We use BART fine-tuned on CNNDM dataset [20], which is available off-the-shelf in Huggingface Transformers [70]. (2) Paraphrasing. We continue fine-tuning BART from (1) on ParaBank2 dataset [22], which contains a large paraphrase collection. We used a random subset of 30,000 data and fine-tuned for one epoch with a batch size of 20 and a learning rate of $5 e ^ { - 5 }$ . We used two 2080Ti GPUs, and the training time is less than one hour. ", + "bbox": [ + 174, + 114, + 825, + 227 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4 Experiment ", + "text_level": 1, + "bbox": [ + 174, + 247, + 303, + 263 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "This section aims to evaluate the reliability of different automated metrics, which is commonly achieved by quantifying how well different metrics correlate with human judgments using measures (e.g., Spearman Correlation [72]) defined below (§4.1.2). ", + "bbox": [ + 174, + 279, + 823, + 321 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.1 Baselines and Datasets ", + "text_level": 1, + "bbox": [ + 174, + 339, + 370, + 353 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.1.1 Evaluation Metrics ", + "text_level": 1, + "bbox": [ + 174, + 364, + 357, + 380 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We comprehensively examine metrics outlined in $\\ S 2 . 3$ , which either require human judgments to train (i.e., supervised metrics): COMET [56], BLEURT [62], or are human judgment-free (i.e., unsupervised): BLEU [50] ROUGE-1 and ROUGE-2, ROUGE-L, CHRF [52], PRISM [65], MoverScore [76], BERTScore [75]. The detailed comparisons of those metrics can be found in Appendix. We use the official code for each metric. ", + "bbox": [ + 174, + 390, + 825, + 458 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.1.2 Measures for Meta Evaluation ", + "text_level": 1, + "bbox": [ + 176, + 474, + 436, + 489 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Pearson Correlation [15] measures the linear correlation between two sets of data. Spearman Correlation [72] assesses the monotonic relationships between two variables. Kendall’s Tau [27] measures the ordinal association between two measured quantities. Accuracy, in our experiments, measures the percentage of correct ranking between factual texts and non-factual texts. We follow previous works in the choices of measures for different datasets to make a fair comparison. ", + "bbox": [ + 174, + 500, + 825, + 569 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.1.3 Datasets ", + "text_level": 1, + "bbox": [ + 174, + 585, + 284, + 599 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The datasets we use are summarized in Tab. 1. We consider three different tasks: summarization (SUM), machine translation (MT), and data-to-text (D2T). ", + "bbox": [ + 174, + 609, + 516, + 651 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Machine Translation We obtain the source language sentences, machine-translated texts and reference texts from the WMT19 metrics shared task [44]. We use the DARR corpus and consider 7 language pairs, which are de-en, fi-en, gu-en, kk-en, lt-en, ru-en, zh-en. ", + "bbox": [ + 174, + 657, + 517, + 741 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Text Summarization (1) REALSumm [4] is a metaevaluation dataset for text summarization which measures pyramid recall of each system-generated summary. (2) SummEval [13] is a collection of human judgments of model-generated summaries on the CNNDM dataset annotated by both expert judges and crowd-source workers. Each system generated summary is gauged through the lens of coherence, consistency, fluency and relevance.3 (3) NeR18 The ", + "bbox": [ + 174, + 747, + 517, + 872 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/2e7285a37838112a4182c3dc65fb5f53419dcbe2dd78fa7f3712dfe6244fe154.jpg", + "table_caption": [ + "Table 1: A summary of tasks, datasets, and evaluation perspectives that we have covered in our experiments. Explanation of evaluation perspectives can be found in $\\ S 2 . 2$ . " + ], + "table_footnote": [], + "table_body": "
TasksDatasetsEval. Perspectives
SUMREALSUMCov
SummEvalCOH FAC FLU INFO
NeR18COH FLU REL INFO
Rank19 QAGS-C QAGS-XFAC
DE FI GU KK IT RU ZHADE FLU
MT D2TBAGEL SFHOT SFRESINFO
", + "bbox": [ + 529, + 674, + 830, + 867 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "NEWSROOM dataset [18] contains 60 articles with summaries generated by 7 different methods are annotated with human scores in terms of coherence, fluency, informativeness, relevance. ", + "bbox": [ + 173, + 90, + 823, + 119 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Factuality (1) Rank19 [14] is used to meta-evaluate factuality metrics. It is a collection of 373 triples of a source sentence with two summary sentences, one correct and one incorrect. (2) QAGS20 [66] collected 235 test outputs on CNNDM dataset from [16] and 239 test outputs on XSUM dataset [47] from BART fine-tuned on XSUM. Sentences in each summary are annotated with correctness scores w.r.t. factuality. ", + "bbox": [ + 174, + 133, + 825, + 203 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Data to Text We consider the following datasets which target utterance generation for spoken dialogue systems. (1) BAGEL [45] provides information about restaurants. (2) SFHOT [69] provides information about hotels in San Francisco. (3) SFRES [69] provides information about restaurants in San Francisco. They contain 202, 398, and 581 samples respectively, each sample consists of one meaning representation, multiple references, and utterances generated by different systems. ", + "bbox": [ + 174, + 209, + 825, + 279 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.2 Setup", + "text_level": 1, + "bbox": [ + 174, + 294, + 253, + 309 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.2.1 Prompt Design ", + "text_level": 1, + "bbox": [ + 174, + 319, + 330, + 334 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "To perform prompting, we first need to find proper prompts within a search space. Instead of considering a large discrete search space $[ 6 3 ] ^ { 4 }$ or continuous search space [34], we use simple heuristics to narrow our search space. In particular, we use manually devised seed prompts and gather paraphrases to construct our prompt set.5 The seed prompts and some examples of paraphrased prompts are shown in Tab. 2. Details are listed in the Appendix. ", + "bbox": [ + 174, + 343, + 825, + 412 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/afdc30151920352456f85417d4afcaf483115183510ae8210ad965e53ccb985e.jpg", + "table_caption": [ + "Table 2: Seed prompts and examples of final prompts. “Number” denotes the size of our final prompt set that was acquired from the seed prompts. " + ], + "table_footnote": [], + "table_body": "
UsageNumberSeedExample
s→h70in summaryin short，inword， atosum up
h←r34inother wordstorephrase it，that istosay，i.e.
", + "bbox": [ + 187, + 460, + 808, + 515 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.2.2 Settings ", + "text_level": 1, + "bbox": [ + 174, + 534, + 281, + 549 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Variants. We consider four variants of BARTSCORE, which are (1) BARTSCORE, which uses the vanilla BART; (2) BARTSCORE-CNN, which uses the BART fine-tuned on the summarization dataset CNNDM; (3) BARTSCORE-CNN-PARA, where BART is first fine-tuned on CNNDM, then fine-tuned on ParaBank2. (4) BARTSCORE-PROMPT, which is enhanced by adding prompts. ", + "bbox": [ + 174, + 558, + 823, + 613 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Selection of Prompts. For the summarization and data-to-text tasks, we use all entries (either all prompts designed for $s h$ or all prompts designed for $h r$ depending on the BARTScore usage chosen) in the prompt set by prefixing the decoder input and getting different generation scores (calculated by Eq.2) for each hypothesis based on different prompts. We finally get the score for one hypothesis by taking the average of all its generation scores using different prompts ( [24]; details about prompt ensembling can be found in the Appendix). For the machine translation task, due to the more expensive computational cost brought by larger text sets, we first use WMT18 [43] as a development set to search for one best prompt and obtain the phrase “Such as”, which is then used for the test language pairs. ", + "bbox": [ + 173, + 619, + 825, + 744 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Selection of BARTScore Usage. Although BARTSCORE can be used in different ways (shown in $\\ S 3 . 2 )$ ), in different tasks, they can be chosen based on how targeted evaluation perspectives are defined (described in $\\ S 2 . 2 \\AA$ ) as well as the types of tasks. Specifically, (i) For those datasets whose gold standard human evaluation are obtained based on recall-based pyramid method, we adopt recall-based BARTSCORE $( h \\to r$ ). (ii) For those datasets whose human judgments focus on linguistic quality (coherence, fluency) and factual correctness (factuality), or the source and hypothesis texts are in the same modality (i.e., language), we use faithfulness-based BARTSCORE $s h$ ). (iii) For data-to-text and machine translation tasks, to make a fair comparison, we use BARTSCORE with the F-score version that other existing works [65] have adopted when evaluating generated texts. ", + "bbox": [ + 173, + 751, + 825, + 876 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/10078ed3879a01510492b203ac9ef173d092fe67d6150bd64664257da154ae67.jpg", + "table_caption": [ + "Table 3: Kendall’s Tau correlation of different metrics on WMT19 dataset. The highest correlation for each language pair achieved by unsupervised method is bold, and the highest correlation overall is underlined. Avg. denotes the average correlation achieved by a metric across all language pairs. " + ], + "table_footnote": [], + "table_body": "
de-enfi-en gu-enkk-enlt-enru-enzh-enAvg.
SUPERVISED METHODS
BLEURT0.1740.3740.3130.3720.3880.2200.4360.325
COMET0.2190.3690.3160.3780.4050.2260.4620.339
UNSUPERVISED METHODS
BLEU0.0540.2360.1940.2760.2490.1150.3210.206
CHRF0.1230.2920.2400.3230.3040.1770.3710.261
PRISM0.1990.3660.3200.3620.3820.2200.4340.326
BERTScore0.1900.3540.2920.3510.3810.2210.4300.317
BARTSCORE0.1560.3350.2730.3240.3220.1670.3890.281
+CNN0.1900.3650.3000.3480.3840.2080.4250.317
+ CNN+Para0.205t0.370t0.3160.378t0.386†0.219_0.442t0.331
+ CNN +Para +Prompt0.2380.3740.3180.376t0.386+0.2190.4470.337
", + "bbox": [ + 183, + 137, + 813, + 345 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/975ee48dd3e22e76b11d2026accb30e8c6cebbef0d1d3bce2e29af12c12df5c3.jpg", + "table_caption": [ + "Table 4: Spearman correlation of different metrics on three human judgement datasets. For promptbased learning, we consider adding prompts to the best-performing BARTSCORE $( \\Omega )$ on each dataset. The highest correlation overall for each aspect on each dataset is bold. " + ], + "table_footnote": [], + "table_body": "
REALSummSummEvalNeR18Avg.
CovCOHFACFLUINFOCOHFLUINFOREL
ROUGE-10.4980.1670.1600.1150.3260.0950.1040.1300.1470.194
ROUGE-20.4230.1840.1870.1590.2900.0260.0480.0790.0910.165
ROUGE-L0.4880.1280.1150.1050.3110.0640.0720.0890.1060.164
BERTScore0.4400.2840.1100.1930.3120.1470.1700.1310.1630.217
MoverScore0.3720.1590.1570.1290.3180.1610.1200.1880.1950.200
PRISM0.4110.2490.3450.2540.2120.5730.5320.5610.5530.410
BARTSCORE0.4410.322†0.3110.2480.2640.679†0.670t0.646†0.604t0.465
+ CNN0.4750.448‡0.382†0.356t0.356t0.653†0.640t0.616†0.5670.499
+ CNN+Para0.4710.424†0.401‡0.378t0.3130.657t0.652t0.614†0.5620.497
+Ω+Prompt0.4880.4070.3780.338f0.368f0.701t0.679t0.686t0.6200.518
", + "bbox": [ + 178, + 416, + 820, + 598 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Significance Tests. To perform rigorous analysis, we adopt the bootstrapping method (p-value $<$ 0.05) [28] for pair-wise significance tests. In all tables, we use $\\dagger$ on BARTSCORE if it significantly $( p < 0 . 0 5 )$ outperforms other unsupervised metrics excluding BARTSCORE variants. We use $\\ddagger$ on BARTSCORE if it significantly outperforms all other unsupervised metrics including BARTSCORE variants. ", + "bbox": [ + 173, + 633, + 825, + 703 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.3 Experimental Results ", + "text_level": 1, + "bbox": [ + 176, + 729, + 362, + 744 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.3.1 Machine Translation ", + "text_level": 1, + "bbox": [ + 176, + 758, + 369, + 773 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Tab. 3 illustrates Kendall’s Tau correlation of diverse metrics on different language pairs. We can observe that: (1) BARTSCORE enhanced by fine-tuning tasks $\\left( \\mathrm { C N N + P a r a } \\right)$ can significantly outperform all other unsupervised methods on five language pairs and achieve comparable results on the other two. (2) The performance of BARTSCORE can be further improved by simply adding a prompt (i.e., such as) without any other overhead. Notably, on the language pair ${ \\tt d e } \\mathrm { - } \\in \\mathrm { n }$ , using the prompt results in a 0.033 improvement, which even significantly surpasses existing state-of-the-art supervised metrics BLEURT and COMET. This suggests a promising future direction for metric design: searching for proper prompts to better leverage knowledge stored in pre-trained language models instead of training on human judgment data [31]. ", + "bbox": [ + 173, + 786, + 825, + 911 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.3.2 Text Summarization ", + "text_level": 1, + "bbox": [ + 176, + 92, + 367, + 106 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Tab. 4 shows the meta-evaluation results of different metrics on the summarization task. We can observe that: (1) Simply vanilla BARTSCORE can outperform BERTScore and MoverScore by a large margin on 8 settings except the INFO perspective on SummEval. Strikingly, it achieves improvements of 0.251 and 0.265 over BERTScore and MoverScore respectively. (2) The improvement on REALSum and SummEval datasets can be further improved when introducing fine-tuning tasks. However, fine-tuning does not improve on the NeR18 dataset, likely because this dataset only contains 7 systems with easily distinguishable quality, and vanilla BARTSCORE can already achieve a high level of correlation $( > 0 . 6$ on average). (3) Our prompt combination strategy can consistently improve the performance on informativeness, up to 0.072 Spearman correlation on the NeR18 dataset and 0.055 on SummEval. However, the performance from other perspectives such as fluency and factuality do not show consistent improvements, which we will elaborate on later (§4.4.2). ", + "bbox": [ + 173, + 121, + 825, + 247 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 252, + 508, + 308 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Analysis on Factuality Datasets The goal of these datasets is to judge whether a short generated summary is faithful to the original long documents. As shown in Tab. 5, we observe that (1) BARTSCORE $+ \\thinspace C N N$ can almost match human baseline on Rank19 and outperform all other metrics, including the most recent top-performing factuality metrics FactCC and QAGS by a large margin. (2) Using paraphrase as a fine-tuning task will reduce BARTSCORE’s performance, which is reasonable since these two texts (i.e., the summary and document) shouldn’t maintain the paraphrased relationship in general. (3) Introducing prompts does not bring an improvement, even resulting in a performance decrease. ", + "bbox": [ + 174, + 314, + 508, + 507 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/17bdfd18aa5de5a0b5c0e6b2d1231e0adb998a06d3721c28e735fea9fd3f7f93.jpg", + "table_caption": [ + "Table 5: Results on Rank19 and QAGS datasets. where “Q” represents QAGS. Metrics achieve highest correlation are bold. " + ], + "table_footnote": [], + "table_body": "
Rank19 Q-CNNQ-XSUM
Acc.Pearson
ROUGE-10.5680.338-0.008
ROUGE-20.6300.4590.097
ROUGE-L0.5870.3570.024
BERTScore0.7130.5760.024
MoverScore0.7130.4140.054
PRISM0.7800.4790.025
FactCC [30] QAGS [66]0.700 0.7211 0.5451 0.175
Human [14] BARTSCORE0.83911 0.009
+CNN0.684 0.836‡0.661† 0.735±0.184‡
+ CNN+Para0.680t0.074
0.788
+CNN +Prompt70.7960.719f0.094
", + "bbox": [ + 521, + 313, + 823, + 539 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.3.3 Data-to-text ", + "text_level": 1, + "bbox": [ + 174, + 539, + 308, + 553 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "The experiment results on data-to-text datasets are shown in Tab. 6. We observe that (1) finetuning on the CNNDM dataset can consistently boost the correlation, for example, up to 0.056 gain on BAGEL. (2) Additionally, further finetuning on paraphrase datasets results in even higher performance compared to the version without any fine-tuning, up to 0.083 Spearman correlation on BAGEL dataset. These results surpass all existing top-performing metrics. (3) Our proposed prompt combination strategy can consistently improve correlation, on average 0.028 Spearman correlation. This is consistent with the findings in $\\ S 4 . 3 . 2$ that we can improve the aspect of informativeness through proper prompting. ", + "bbox": [ + 174, + 569, + 483, + 791 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/796d087ca8e9733af385701c9958226a7d869290614949a2929e782de3bdc02e.jpg", + "table_caption": [ + "Table 6: Results on data-to-text datasets. We report Spearman correlation. Metrics achieve highest correlation are bold. " + ], + "table_footnote": [], + "table_body": "
BAGELSFRESSFHOTAvg.
ROUGE-10.2340.1150.1180.156
ROUGE-20.1990.1160.0880.134
ROUGE-L0.1890.1030.1100.134
BERTScore0.2890.1560.1350.193
MoverScore0.2840.1530.1720.203
PRISM0.3050.1550.1960.219
BARTSCORE0.2470.164†0.1580.190
+ CNN0.3030.191†0.1900.228
+ CNN+Para0.330t0.185t0.211†0.242
+Ω+Prompt0.3360.238t0.235t0.270
", + "bbox": [ + 496, + 617, + 823, + 792 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.4 Analysis ", + "text_level": 1, + "bbox": [ + 174, + 824, + 271, + 838 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We design experiments to better understand the mechanism by which BARTSCORE obtains these promising results, specifically asking three questions: Q1: Compared to other unsupervised metrics, where does BARTSCORE outperform them? Q2: How does adding prompts benefit evaluation? Q3: Will BARTScore introduce biases in unpredictable ways? ", + "bbox": [ + 174, + 856, + 825, + 911 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/78f8c1e31ba1ab4fbe0920e68cff9da08fddef39259e3f11e995865e34618d98.jpg", + "image_caption": [ + "Figure 2: Fine-grained analysis (a,b) and prompt analysis (c). In (a, b), BE, PR, BL, CO, BA represent BERTScore, PRISM, BLEURT, COMET and BARTSCORE respectively. In (c), SEM, LIN, FAC denote semantic overlap, linguistic quality and factual correctness respectively. " + ], + "image_footnote": [], + "bbox": [ + 178, + 97, + 818, + 226 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.4.1 Fine-grained Analysis ", + "text_level": 1, + "bbox": [ + 174, + 304, + 379, + 319 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "To answer Q1, we choose the MT task and break down the performance of each metric into different buckets based on different axes. ", + "bbox": [ + 173, + 328, + 823, + 356 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Top-k Systems We report the average correlation across all language pairs achieved by each metric given only translations from top- $k$ systems. We vary the number of $k$ , and the results are shown in Fig. 2-(a). We can see that BARTSCORE can outperform all other metrics (including one supervised metric BLEURT) except the existing state-of-the-art supervised metric COMET for different $k$ , and the decrease in correlation becomes smoother than others when considering top-scoring systems. This indicates that BARTSCORE is robust to high-quality generated texts. ", + "bbox": [ + 174, + 372, + 825, + 455 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Reference Length We break down each test set into four buckets based on the reference length, which are [15, 25), [25, 35), [35, 45), [45, 54] and compute the Kendall’s Tau average correlation of different metrics across all language pairs within each bucket.6 The results are shown in Fig. 2-(b). We observe that BARTSCORE can outperform or tie with other unsupervised metrics over different reference lengths. Also, its correlation with human judgments is more stable compared to all other metrics. This indicates its robustness to different input lengths. More other analyses can be found in Appendix. ", + "bbox": [ + 174, + 472, + 825, + 569 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.4.2 Prompt Analysis ", + "text_level": 1, + "bbox": [ + 174, + 585, + 341, + 599 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "For Q2, we choose the summarization and data-to-text tasks for analysis where we used all prompts from our prompt set. We first group all the evaluation perspectives into three categories: (1) semantic overlap (informativeness, pyramid score, and relevance) (2) linguistic quality (fluency, coherence) (3) factual correctness (factuality). We then calculate the percentage of prompts that result in performance improvements for each perspective within a dataset. Finally, we compute the average percentage of prompts that can lead to performance gains for each category. The results are shown in Tab. 2-(c). We can see that for semantic overlap, almost all prompts can lead to the performance increase, while for factuality only a few prompts can improve the performance. This also explains the results in $\\ S 4 . 3 . 2$ where we found that combining the results of different prompts can lead to consistent increases in semantic overlap but worse performance in factuality. Regarding linguistic quality, the effect of adding a prompt is not that predictive, which is also consistent with our findings in $\\ S 4 . 3 . 2$ . ", + "bbox": [ + 174, + 609, + 825, + 762 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.4.3 Bias Analysis ", + "text_level": 1, + "bbox": [ + 174, + 777, + 318, + 792 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "To answer Q3, we conduct bias analysis. Bias would indicate that the scores are too high or too low compared to the scores they are given by human annotators. Therefore, to see whether such biases exist, we inspected the rank differences given by human annotators and BARTScore (fine-tuned on CNNDM dataset) on the REALSumm dataset where 24 systems are considered, including both abstractive models and extractive models as well as models based on pre-trained models and models that are trained from scratch. We list all the systems below. And the resulting rank difference is shown in Fig. 3. ", + "bbox": [ + 174, + 803, + 825, + 872 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/71ce21efad222c710fbe5f7a584a046b9f08143d9d25180689982349d95a3ce5.jpg", + "image_caption": [ + "Figure 3: Bias analysis of BARTScore. The “Rank Difference\" is the rank obtained using human judgements minus the rank got from BARTScore. Systems beginning with letter “E\" are extractive systems while systems beginning with letter “A\" are abstractive systems. " + ], + "image_footnote": [], + "bbox": [ + 178, + 87, + 826, + 180 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 256, + 823, + 285 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Extractive Systems E1: BanditSum [11]; E2: Refresh [48]; E3: NeuSum [80]; E4: LSTMPN-RL [79]; E5: BERT-TF-SL [79]; E6: BERT-TF-PN [79]; E7: BERT-LSTM-PN-RL [79]; E8: BERT-LSTM-PN [79]; E9: HeterGraph [68]; E10: MatchSum [78]. ", + "bbox": [ + 174, + 299, + 826, + 342 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Abstractive Systems A1: Ptr-Gen [61]; A2: Bottom-up [17]; A3: Fast-Abs-RL [5]; A4: Two-stageRL [73]; A5: BERT-Ext-Abs [41]; A6: BERT-Abs [41]; A7: Trans-Abs [41]; A8: UniLM-1 [10]; A9: UniLM-2 [2]; A10: T5-base [55]; A11: T5-large [55]; A12: T5-11B [55]; A13: BART [33]; A14: SemSim [42]. ", + "bbox": [ + 174, + 357, + 825, + 414 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "As shown in Fig. 3, BARTScore is less effective at distinguishing the quality of extractive summarization systems while much better at distinguishing the quality of abstractive summarization systems. However, given that there is a trend for using abstractive systems as more and more pre-trained sequence-to-sequence models being proposed, BARTScore’s weaknesses on extractive systems will be mitigated. ", + "bbox": [ + 174, + 419, + 826, + 489 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "5 Implications and Future Directions ", + "text_level": 1, + "bbox": [ + 173, + 508, + 500, + 526 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this paper, we proposed a metric BARTSCORE that formulates evaluation of generated text as a text generation task, and empirically demonstrated its efficacy. Without the supervision of human judgments, BARTSCORE can effectively evaluate texts from 7 perspectives and achieve the best performance on 16 of 22 settings against existing top-scoring metrics. We highlight potential future directions based on what we have learned. ", + "bbox": [ + 174, + 541, + 825, + 609 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Prompt-augmented metrics As an easy-to-use but powerful method, prompting [39] has achieved impressive performance particularly on semantic overlap-based evaluation perspectives. However, its effectiveness in factuality and linguistic quality-based perspectives has not been fully demonstrated in this paper. In the future, more works can explore how to make better use of prompts for these and other evaluation scenarios. ", + "bbox": [ + 174, + 617, + 825, + 685 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Co-evolving evaluation metrics and systems BARTSCORE builds the connection between metric design and system design, which allows them to share their technological advances, thereby progressing together. For example, a better BART-based summarization system may be directly used as a more reliable automated metric for evaluating summaries, and this work makes them connected. ", + "bbox": [ + 174, + 693, + 825, + 748 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Acknowledgments ", + "text_level": 1, + "bbox": [ + 176, + 768, + 328, + 786 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "The authors would like to thank the anonymous reviewers for their insightful comments and suggestions. The authors also thank Wei Zhao for assisting with reproducing baseline results. This work was supported by the Air Force Research Laboratory under agreement number FA8750-19-2-0200. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright notation thereon. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the Air Force Research Laboratory or the U.S. Government. ", + "bbox": [ + 174, + 800, + 826, + 911 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "References \n[1] Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong. Learning to retrieve reasoning paths over wikipedia graph for question answering. arXiv preprint arXiv:1911.10470, 2019. \n[2] Hangbo Bao, Li Dong, Furu Wei, Wenhui Wang, Nan Yang, Xiaodong Liu, Yu Wang, Jianfeng Gao, Songhao Piao, Ming Zhou, and Hsiao-Wuen Hon. Unilmv2: Pseudo-masked language models for unified language model pre-training. In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, volume 119 of Proceedings of Machine Learning Research, pages 642–652. PMLR, 2020. \n[3] Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, and Pengfei Liu. Metrics also disagree in the low scoring range: Revisiting summarization evaluation metrics. In Proceedings of the 28th International Conference on Computational Linguistics, pages 5702–5711, Barcelona, Spain (Online), December 2020. International Committee on Computational Linguistics. \n[4] Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, Pengfei Liu, and Graham Neubig. Re-evaluating evaluation in text summarization. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9347–9359, Online, November 2020. Association for Computational Linguistics. \n[5] Yen-Chun Chen and Mohit Bansal. Fast abstractive summarization with reinforce-selected sentence rewriting. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 675–686, Melbourne, Australia, July 2018. Association for Computational Linguistics. \n[6] Andrew M Dai and Quoc V Le. Semi-supervised sequence learning. In C. Cortes, N. Lawrence, D. Lee, M. Sugiyama, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 28. Curran Associates, Inc., 2015. \n[7] Hoa Trang Dang. Overview of duc 2005. In Proceedings of the document understanding conference, volume 2005, pages 1–12, 2005. \n[8] Michael Denkowski and Alon Lavie. Extending the METEOR machine translation evaluation metric to the phrase level. In Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, pages 250–253, Los Angeles, California, June 2010. Association for Computational Linguistics. \n[9] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, 2019. \n[10] Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming Zhou, and Hsiao-Wuen Hon. Unified language model pre-training for natural language understanding and generation. In Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence d’Alché-Buc, Emily B. Fox, and Roman Garnett, editors, Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019, NeurIPS 2019, December 8-14, 2019, Vancouver, BC, Canada, pages 13042–13054, 2019. \n[11] Yue Dong, Yikang Shen, Eric Crawford, Herke van Hoof, and Jackie Chi Kit Cheung. BanditSum: Extractive summarization as a contextual bandit. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 3739–3748, Brussels, Belgium, October-November 2018. Association for Computational Linguistics. \n[12] Zi-Yi Dou, Pengfei Liu, Hiroaki Hayashi, Zhengbao Jiang, and Graham Neubig. GSum: A general framework for guided neural abstractive summarization. In Meeting of the North American Chapter of the Association for Computational Linguistics (NAACL), Mexico City, June 2021. \n[13] A. R. Fabbri, Wojciech Kryscinski, Bryan McCann, R. Socher, and Dragomir Radev. Summeval: Re-evaluating summarization evaluation. Trans. Assoc. Comput. Linguistics, 9:391–409, 2021. \n[14] Tobias Falke, Leonardo FR Ribeiro, Prasetya Ajie Utama, Ido Dagan, and Iryna Gurevych. Ranking generated summaries by correctness: An interesting but challenging application for natural language inference. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2214–2220, 2019. \n[15] David Freedman, Robert Pisani, and Roger Purves. Statistics (international student edition). Pisani, R. Purves, 4th edn. WW Norton & Company, New York, 2007. \n[16] Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. Bottom-up abstractive summarization. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4098–4109, 2018. \n[17] Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. Bottom-up abstractive summarization. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4098–4109, Brussels, Belgium, October-November 2018. Association for Computational Linguistics. \n[18] Max Grusky, Mor Naaman, and Yoav Artzi. Newsroom: A dataset of 1.3 million summaries with diverse extractive strategies. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 708–719, New Orleans, Louisiana, June 2018. Association for Computational Linguistics. \n[19] Max Grusky, Mor Naaman, and Yoav Artzi. Newsroom: A dataset of 1.3 million summaries with diverse extractive strategies. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), volume 1, pages 708–719, 2018. \n[20] K. Hermann, Tomás Kociský, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and P. Blunsom. Teaching machines to read and comprehend. In NIPS, 2015. \n[21] Tsutomu Hirao, Manabu Okumura, Norihito Yasuda, and Hideki Isozaki. Supervised automatic evaluation for summarization with voted regression model. Information Processing & Management, 43(6):1521–1535, 2007. \n[22] J. Edward Hu, Abhinav Singh, Nils Holzenberger, Matt Post, and Benjamin Van Durme. Largescale, diverse, paraphrastic bitexts via sampling and clustering. In Proceedings of the $2 3 r d$ Conference on Computational Natural Language Learning (CoNLL), pages 44–54, Hong Kong, China, November 2019. Association for Computational Linguistics. \n[23] Sarthak Jain, Madeleine van Zuylen, Hannaneh Hajishirzi, and Iz Beltagy. Scirex: A challenge dataset for document-level information extraction. arXiv preprint arXiv:2005.00512, 2020. \n[24] Zhengbao Jiang, Frank F. Xu, Jun Araki, and Graham Neubig. How can we know what language models know? Transactions of the Association for Computational Linguistics, 8:423–438, 2020. \n[25] Karen Sparck Jones. A statistical interpretation of term specificity and its application in retrieval. Journal of documentation, 1972. \n[26] Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, ˘ Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering. arXiv preprint arXiv:2004.04906, 2020. \n[27] M. G. Kendall. A new measure of rank correlation. Biometrika, 30(1/2):81–93, 1938. \n[28] Philipp Koehn. Statistical significance tests for machine translation evaluation. In Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing, pages 388–395, Barcelona, Spain, July 2004. Association for Computational Linguistics. \n[29] Philipp Koehn. Statistical machine translation. Cambridge University Press, 2009. \n[30] Wojciech Kryscinski, Bryan McCann, Caiming Xiong, and Richard Socher. Evaluating the factual consistency of abstractive text summarization. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9332–9346, Online, November 2020. Association for Computational Linguistics. \n[31] Teven Le Scao and Alexander Rush. How many data points is a prompt worth? In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2627–2636, Online, June 2021. Association for Computational Linguistics. \n[32] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov, and Luke Zettlemoyer. Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. arXiv preprint arXiv:1910.13461, 2019. \n[33] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. BART: Denoising sequence-to-sequence pretraining for natural language generation, translation, and comprehension. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7871–7880, Online, July 2020. Association for Computational Linguistics. \n[34] Xiang Lisa Li and Percy Liang. Prefix-tuning: Optimizing continuous prompts for generation. ArXiv, abs/2101.00190, 2021. \n[35] Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. Text Summarization Branches Out, 2004. \n[36] Chin-Yew Lin and Eduard Hovy. Automatic evaluation of summaries using n-gram cooccurrence statistics. In Proceedings of the 2003 Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics, pages 150–157, 2003. \n[37] Ying Lin, Heng Ji, Fei Huang, and Lingfei Wu. A joint neural model for information extraction with global features. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7999–8009, 2020. \n[38] Pengfei Liu, Jinlan Fu, Yang Xiao, Weizhe Yuan, Shuaicheng Chang, Junqi Dai, Yixin Liu, Zihuiwen Ye, and Graham Neubig. Explainaboard: An explainable leaderboard for nlp. arXiv preprint arXiv:2104.06387, 2021. \n[39] Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig. Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing, 2021. \n[40] Yang Liu and Mirella Lapata. Text summarization with pretrained encoders. arXiv preprint arXiv:1908.08345, 2019. \n[41] Yang Liu and Mirella Lapata. Text summarization with pretrained encoders. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3730–3740, Hong Kong, China, November 2019. Association for Computational Linguistics. \n[42] Yixin Liu and Pengfei Liu. Simcls: A simple framework for contrastive learning of abstractive summarization. arXiv preprint arXiv:2106.01890, 2021. \n[43] Qingsong Ma, Ondˇrej Bojar, and Yvette Graham. Results of the WMT18 metrics shared task: Both characters and embeddings achieve good performance. In Proceedings of the Third Conference on Machine Translation: Shared Task Papers, pages 671–688, Belgium, Brussels, October 2018. Association for Computational Linguistics. \n[44] Qingsong Ma, Johnny Wei, Ondˇrej Bojar, and Yvette Graham. Results of the WMT19 metrics shared task: Segment-level and strong MT systems pose big challenges. In Proceedings of the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages 62–90, Florence, Italy, August 2019. Association for Computational Linguistics. \n[45] François Mairesse, Milica Gasic, Filip Jurcicek, Simon Keizer, Blaise Thomson, Kai Yu, and Steve Young. Phrase-based statistical language generation using graphical models and active learning. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 1552–1561, 2010. \n[46] Nitika Mathur, Timothy Baldwin, and Trevor Cohn. Tangled up in BLEU: Reevaluating the evaluation of automatic machine translation evaluation metrics. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4984–4997, Online, July 2020. Association for Computational Linguistics. \n[47] Shashi Narayan, Shay B Cohen, and Mirella Lapata. Don’t give me the details, just the summary! topic-aware convolutional neural networks for extreme summarization. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1797–1807, 2018. \n[48] Shashi Narayan, Shay B. Cohen, and Mirella Lapata. Ranking sentences for extractive summarization with reinforcement learning. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 1747–1759, New Orleans, Louisiana, June 2018. Association for Computational Linguistics. \n[49] Ani Nenkova and Rebecca Passonneau. Evaluating content selection in summarization: The pyramid method. In Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics: HLT-NAACL 2004, pages 145–152, Boston, Massachusetts, USA, May 2 - May 7 2004. Association for Computational Linguistics. \n[50] Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA, July 2002. Association for Computational Linguistics. \n[51] Maxime Peyrard, Teresa Botschen, and Iryna Gurevych. Learning to score system summaries for better content selection evaluation. In Proceedings of the Workshop on New Frontiers in Summarization, pages 74–84, Copenhagen, Denmark, September 2017. Association for Computational Linguistics. \n[52] Maja Popovic. chrF: character n-gram F-score for automatic MT evaluation. In ´ Proceedings of the Tenth Workshop on Statistical Machine Translation, pages 392–395, Lisbon, Portugal, September 2015. Association for Computational Linguistics. \n[53] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019. \n[54] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67, 2020. \n[55] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. J. Mach. Learn. Res., 21:140:1–140:67, 2020. \n[56] Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. COMET: A neural framework for MT evaluation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2685–2702, Online, November 2020. Association for Computational Linguistics. \n[57] Laria Reynolds and Kyle McDonell. Prompt programming for large language models: Beyond the few-shot paradigm. In Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems, pages 1–7, 2021. \n[58] Timo Schick and Hinrich Schütze. Exploiting cloze questions for few-shot text classification and natural language inference. arXiv preprint arXiv:2001.07676, 2020. \n[59] Timo Schick and Hinrich Schütze. Few-shot text generation with pattern-exploiting training. arXiv preprint arXiv:2012.11926, 2020. \n[60] Timo Schick and Hinrich Schütze. It’s not just size that matters: Small language models are also few-shot learners. arXiv preprint arXiv:2009.07118, 2020. \n[61] Abigail See, Peter J. Liu, and Christopher D. Manning. Get to the point: Summarization with pointer-generator networks. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1073–1083, Vancouver, Canada, July 2017. Association for Computational Linguistics. \n[62] Thibault Sellam, Dipanjan Das, and Ankur Parikh. BLEURT: Learning robust metrics for text generation. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7881–7892, Online, July 2020. Association for Computational Linguistics. \n[63] Taylor Shin, Yasaman Razeghi, Robert L Logan IV, Eric Wallace, and Sameer Singh. Autoprompt: Eliciting knowledge from language models with automatically generated prompts. arXiv preprint arXiv:2010.15980, 2020. \n[64] Miloš Stanojevic and Khalil Sima’an. BEER: BEtter evaluation as ranking. In ´ Proceedings of the Ninth Workshop on Statistical Machine Translation, pages 414–419, Baltimore, Maryland, USA, June 2014. Association for Computational Linguistics. \n[65] Brian Thompson and Matt Post. Automatic machine translation evaluation in many languages via zero-shot paraphrasing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 90–121, Online, November 2020. Association for Computational Linguistics. \n[66] Alex Wang, Kyunghyun Cho, and M. Lewis. Asking and answering questions to evaluate the factual consistency of summaries. In ACL, 2020. \n[67] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel Bowman. GLUE: A multi-task benchmark and analysis platform for natural language understanding. In Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 353–355, Brussels, Belgium, November 2018. Association for Computational Linguistics. \n[68] Danqing Wang, Pengfei Liu, Yining Zheng, Xipeng Qiu, and Xuanjing Huang. Heterogeneous graph neural networks for extractive document summarization. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6209–6219, Online, July 2020. Association for Computational Linguistics. \n[69] Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Pei-Hao Su, David Vandyke, and Steve Young. Semantically conditioned lstm-based natural language generation for spoken dialogue systems. arXiv preprint arXiv:1508.01745, 2015. \n[70] Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 38–45, Online, October 2020. Association for Computational Linguistics. \n[71] Weizhe Yuan, Pengfei Liu, and Graham Neubig. Can we automate scientific reviewing? arXiv preprint arXiv:2102.00176, 2021. \n[72] Jerrold H Zar. Spearman rank correlation. Encyclopedia of Biostatistics, 7, 2005. \n[73] Haoyu Zhang, Jingjing Cai, Jianjun Xu, and Ji Wang. Pretraining-based natural language generation for text summarization. In Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL), pages 789–797, Hong Kong, China, November 2019. Association for Computational Linguistics. \n[74] Jingqing Zhang, Yao Zhao, Mohammad Saleh, and Peter Liu. Pegasus: Pre-training with extracted gap-sentences for abstractive summarization. In International Conference on Machine Learning, pages 11328–11339. PMLR, 2020. \n[75] Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. Bertscore: Evaluating text generation with bert. In International Conference on Learning Representations, 2020. \n[76] Wei Zhao, Maxime Peyrard, Fei Liu, Yang Gao, Christian M. Meyer, and Steffen Eger. MoverScore: Text generation evaluating with contextualized embeddings and earth mover distance. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 563–578, Hong Kong, China, November 2019. Association for Computational Linguistics. \n[77] Ming Zhong, Pengfei Liu, Yiran Chen, Danqing Wang, Xipeng Qiu, and Xuanjing Huang. Extractive summarization as text matching. arXiv preprint arXiv:2004.08795, 2020. \n[78] Ming Zhong, Pengfei Liu, Yiran Chen, Danqing Wang, Xipeng Qiu, and Xuanjing Huang. Extractive summarization as text matching. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6197–6208, Online, July 2020. Association for Computational Linguistics. \n[79] Ming Zhong, Pengfei Liu, Danqing Wang, Xipeng Qiu, and Xuanjing Huang. Searching for effective neural extractive summarization: What works and what’s next. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1049–1058, Florence, Italy, July 2019. Association for Computational Linguistics. \n[80] Qingyu Zhou, Nan Yang, Furu Wei, Shaohan Huang, Ming Zhou, and Tiejun Zhao. Neural document summarization by jointly learning to score and select sentences. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 654–663, Melbourne, Australia, July 2018. Association for Computational Linguistics. ", + "bbox": [ + 169, + 88, + 828, + 912 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 42, + 828, + 920 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 45, + 828, + 916 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 85, + 828, + 917 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 51, + 828, + 898 + ], + "page_idx": 14 + } +] \ No newline at end of file diff --git a/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_middle.json b/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..a4bf08fdd9e72932361586023907e62059b00374 --- /dev/null +++ b/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_middle.json @@ -0,0 +1,40064 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 134, + 97, + 478, + 137 + ], + "lines": [ + { + "bbox": [ + 250, + 97, + 364, + 117 + ], + "spans": [ + { + "bbox": [ + 250, + 97, + 364, + 117 + ], + "score": 1.0, + "content": "BARTSCORE:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 132, + 116, + 479, + 138 + ], + "spans": [ + { + "bbox": [ + 132, + 116, + 479, + 138 + ], + "score": 1.0, + "content": "Evaluating Generated Text as Text Generation", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 119, + 179, + 231, + 212 + ], + "lines": [ + { + "bbox": [ + 147, + 178, + 205, + 191 + ], + "spans": [ + { + "bbox": [ + 147, + 178, + 205, + 191 + ], + "score": 1.0, + "content": "Weizhe Yuan", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 118, + 189, + 233, + 204 + ], + "spans": [ + { + "bbox": [ + 118, + 189, + 233, + 204 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 121, + 202, + 230, + 213 + ], + "spans": [ + { + "bbox": [ + 121, + 202, + 230, + 213 + ], + "score": 1.0, + "content": "weizhey@cs.cmu.edu", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 249, + 179, + 361, + 213 + ], + "lines": [ + { + "bbox": [ + 270, + 178, + 342, + 192 + ], + "spans": [ + { + "bbox": [ + 270, + 178, + 342, + 192 + ], + "score": 1.0, + "content": "Graham Neubig", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 248, + 189, + 363, + 204 + ], + "spans": [ + { + "bbox": [ + 248, + 189, + 363, + 204 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 250, + 202, + 361, + 214 + ], + "spans": [ + { + "bbox": [ + 250, + 202, + 361, + 214 + ], + "score": 1.0, + "content": "gneubig@cs.cmu.edu", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 378, + 179, + 492, + 213 + ], + "lines": [ + { + "bbox": [ + 407, + 179, + 467, + 191 + ], + "spans": [ + { + "bbox": [ + 407, + 179, + 467, + 191 + ], + "score": 1.0, + "content": "Pengfei Liu ∗", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 378, + 189, + 492, + 204 + ], + "spans": [ + { + "bbox": [ + 378, + 189, + 492, + 204 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 385, + 201, + 484, + 213 + ], + "spans": [ + { + "bbox": [ + 385, + 201, + 484, + 213 + ], + "score": 1.0, + "content": "pliu3@cs.cmu.edu", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 283, + 241, + 328, + 254 + ], + "lines": [ + { + "bbox": [ + 281, + 240, + 331, + 255 + ], + "spans": [ + { + "bbox": [ + 281, + 240, + 331, + 255 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 142, + 265, + 469, + 483 + ], + "lines": [ + { + "bbox": [ + 141, + 265, + 470, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 265, + 470, + 277 + ], + "score": 1.0, + "content": "A wide variety of NLP applications, such as machine translation, summarization,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 276, + 470, + 289 + ], + "spans": [ + { + "bbox": [ + 141, + 276, + 470, + 289 + ], + "score": 1.0, + "content": "and dialog, involve text generation. One major challenge for these applications", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 286, + 470, + 299 + ], + "spans": [ + { + "bbox": [ + 141, + 286, + 470, + 299 + ], + "score": 1.0, + "content": "is how to evaluate whether such generated texts are actually fluent, accurate,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "score": 1.0, + "content": "or effective. In this work, we conceptualize the evaluation of generated text", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 309, + 470, + 322 + ], + "spans": [ + { + "bbox": [ + 141, + 309, + 470, + 322 + ], + "score": 1.0, + "content": "as a text generation problem, modeled using pre-trained sequence-to-sequence", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 321, + 469, + 331 + ], + "spans": [ + { + "bbox": [ + 142, + 321, + 469, + 331 + ], + "score": 1.0, + "content": "models. The general idea is that models trained to convert the generated text", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 330, + 470, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 330, + 470, + 343 + ], + "score": 1.0, + "content": "to/from a reference output or the source text will achieve higher scores when", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 142, + 342, + 469, + 354 + ], + "spans": [ + { + "bbox": [ + 142, + 342, + 469, + 354 + ], + "score": 1.0, + "content": "the generated text is better. We operationalize this idea using BART [32], an", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 353, + 469, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 353, + 469, + 365 + ], + "score": 1.0, + "content": "encoder-decoder based pre-trained model, and propose a metric BARTSCORE", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 363, + 469, + 375 + ], + "spans": [ + { + "bbox": [ + 141, + 363, + 469, + 375 + ], + "score": 1.0, + "content": "with a number of variants that can be flexibly applied in an unsupervised fashion", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 373, + 471, + 388 + ], + "spans": [ + { + "bbox": [ + 141, + 373, + 471, + 388 + ], + "score": 1.0, + "content": "to evaluation of text from different perspectives (e.g. informativeness, fluency,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 385, + 471, + 398 + ], + "spans": [ + { + "bbox": [ + 141, + 385, + 471, + 398 + ], + "score": 1.0, + "content": "or factuality). BARTSCORE is conceptually simple and empirically effective.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 396, + 471, + 409 + ], + "spans": [ + { + "bbox": [ + 141, + 396, + 471, + 409 + ], + "score": 1.0, + "content": "It can outperform existing top-scoring metrics in 16 of 22 test settings, cov-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 407, + 470, + 420 + ], + "spans": [ + { + "bbox": [ + 141, + 407, + 470, + 420 + ], + "score": 1.0, + "content": "ering evaluation of 16 datasets (e.g., machine translation, text summarization)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 419, + 469, + 430 + ], + "spans": [ + { + "bbox": [ + 141, + 419, + 469, + 430 + ], + "score": 1.0, + "content": "and 7 different perspectives (e.g., informativeness, factuality). Code to calculate", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 429, + 471, + 441 + ], + "spans": [ + { + "bbox": [ + 141, + 429, + 471, + 441 + ], + "score": 1.0, + "content": "BARTScore is available at https://github.com/neulab/BARTScore,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 439, + 471, + 452 + ], + "spans": [ + { + "bbox": [ + 141, + 439, + 471, + 452 + ], + "score": 1.0, + "content": "and we have released an interactive leaderboard for meta-evaluation at http:", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 143, + 451, + 469, + 463 + ], + "spans": [ + { + "bbox": [ + 143, + 451, + 469, + 463 + ], + "score": 1.0, + "content": "//explainaboard.nlpedia.ai/leaderboard/task-meval/ on the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 142, + 462, + 469, + 473 + ], + "spans": [ + { + "bbox": [ + 142, + 462, + 469, + 473 + ], + "score": 1.0, + "content": "EXPLAINABOARD platform [38], which allows us to interactively understand the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 473, + 383, + 484 + ], + "spans": [ + { + "bbox": [ + 141, + 473, + 383, + 484 + ], + "score": 1.0, + "content": "strengths, weaknesses, and complementarity of each metric.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 107, + 502, + 190, + 516 + ], + "lines": [ + { + "bbox": [ + 105, + 501, + 192, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 192, + 519 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 527, + 505, + 626 + ], + "lines": [ + { + "bbox": [ + 106, + 527, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 506, + 540 + ], + "score": 1.0, + "content": "One defining feature of recent NLP models is the use of neural representations trained on raw text,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "using unsupervised objectives such as language modeling [6,53], or denoising autoencoding [9,32,54].", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 550, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 505, + 561 + ], + "score": 1.0, + "content": "By learning to predict the words or sentences in natural text, these models simultaneously learn to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 560, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 506, + 573 + ], + "score": 1.0, + "content": "extract features that not only benefit mainstream NLP tasks such as information extraction [23, 37],", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 571, + 506, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 506, + 584 + ], + "score": 1.0, + "content": "question answering [1, 26], text summarization [40, 77] but also have proven effective in development", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 505, + 595 + ], + "score": 1.0, + "content": "of automatic metrics for evaluation of text generation itself [62, 65]. For example, BERTScore [75]", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 594, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 605 + ], + "score": 1.0, + "content": "and MoverScore [76] take features extracted by BERT [9] and apply unsupervised matching functions", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 604, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 506, + 616 + ], + "score": 1.0, + "content": "to compare system outputs against references. Other works build supervised frameworks that use the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 615, + 435, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 435, + 627 + ], + "score": 1.0, + "content": "extracted features to learn to rank [56] or regress [62] to human evaluation scores.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 631, + 505, + 696 + ], + "lines": [ + { + "bbox": [ + 106, + 632, + 505, + 643 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 505, + 643 + ], + "score": 1.0, + "content": "However, in the context of generation evaluation, one may note that there is a decided disconnect", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "between how models are pre-trained using text generation objectives and how they are used as", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 653, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 106, + 653, + 505, + 665 + ], + "score": 1.0, + "content": "down-stream feature extractors. This leads to potential under-utilization of the pre-trained model", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 664, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 506, + 677 + ], + "score": 1.0, + "content": "parameters. For example, the output prediction layer is not used at all in this case. This disconnect", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 675, + 505, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 505, + 687 + ], + "score": 1.0, + "content": "is particularly striking because of the close connection between the pre-training objectives and the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 687, + 257, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 257, + 698 + ], + "score": 1.0, + "content": "generation tasks we want to evaluate.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 731, + 385, + 742 + ], + "lines": [ + { + "bbox": [ + 106, + 730, + 386, + 743 + ], + "spans": [ + { + "bbox": [ + 106, + 730, + 386, + 743 + ], + "score": 1.0, + "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 118, + 704, + 208, + 714 + ], + "lines": [ + { + "bbox": [ + 117, + 701, + 210, + 716 + ], + "spans": [ + { + "bbox": [ + 117, + 701, + 210, + 716 + ], + "score": 1.0, + "content": "∗ Corresponding author.", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 134, + 97, + 478, + 137 + ], + "lines": [ + { + "bbox": [ + 250, + 97, + 364, + 117 + ], + "spans": [ + { + "bbox": [ + 250, + 97, + 364, + 117 + ], + "score": 1.0, + "content": "BARTSCORE:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 132, + 116, + 479, + 138 + ], + "spans": [ + { + "bbox": [ + 132, + 116, + 479, + 138 + ], + "score": 1.0, + "content": "Evaluating Generated Text as Text Generation", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 119, + 179, + 231, + 212 + ], + "lines": [ + { + "bbox": [ + 147, + 178, + 205, + 191 + ], + "spans": [ + { + "bbox": [ + 147, + 178, + 205, + 191 + ], + "score": 1.0, + "content": "Weizhe Yuan", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 118, + 189, + 233, + 204 + ], + "spans": [ + { + "bbox": [ + 118, + 189, + 233, + 204 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 121, + 202, + 230, + 213 + ], + "spans": [ + { + "bbox": [ + 121, + 202, + 230, + 213 + ], + "score": 1.0, + "content": "weizhey@cs.cmu.edu", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 118, + 178, + 233, + 213 + ] + }, + { + "type": "text", + "bbox": [ + 249, + 179, + 361, + 213 + ], + "lines": [ + { + "bbox": [ + 270, + 178, + 342, + 192 + ], + "spans": [ + { + "bbox": [ + 270, + 178, + 342, + 192 + ], + "score": 1.0, + "content": "Graham Neubig", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 248, + 189, + 363, + 204 + ], + "spans": [ + { + "bbox": [ + 248, + 189, + 363, + 204 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 250, + 202, + 361, + 214 + ], + "spans": [ + { + "bbox": [ + 250, + 202, + 361, + 214 + ], + "score": 1.0, + "content": "gneubig@cs.cmu.edu", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 248, + 178, + 363, + 214 + ] + }, + { + "type": "text", + "bbox": [ + 378, + 179, + 492, + 213 + ], + "lines": [ + { + "bbox": [ + 407, + 179, + 467, + 191 + ], + "spans": [ + { + "bbox": [ + 407, + 179, + 467, + 191 + ], + "score": 1.0, + "content": "Pengfei Liu ∗", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 378, + 189, + 492, + 204 + ], + "spans": [ + { + "bbox": [ + 378, + 189, + 492, + 204 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 385, + 201, + 484, + 213 + ], + "spans": [ + { + "bbox": [ + 385, + 201, + 484, + 213 + ], + "score": 1.0, + "content": "pliu3@cs.cmu.edu", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 378, + 179, + 492, + 213 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 241, + 328, + 254 + ], + "lines": [ + { + "bbox": [ + 281, + 240, + 331, + 255 + ], + "spans": [ + { + "bbox": [ + 281, + 240, + 331, + 255 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 142, + 265, + 469, + 483 + ], + "lines": [ + { + "bbox": [ + 141, + 265, + 470, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 265, + 470, + 277 + ], + "score": 1.0, + "content": "A wide variety of NLP applications, such as machine translation, summarization,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 276, + 470, + 289 + ], + "spans": [ + { + "bbox": [ + 141, + 276, + 470, + 289 + ], + "score": 1.0, + "content": "and dialog, involve text generation. One major challenge for these applications", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 286, + 470, + 299 + ], + "spans": [ + { + "bbox": [ + 141, + 286, + 470, + 299 + ], + "score": 1.0, + "content": "is how to evaluate whether such generated texts are actually fluent, accurate,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "score": 1.0, + "content": "or effective. In this work, we conceptualize the evaluation of generated text", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 309, + 470, + 322 + ], + "spans": [ + { + "bbox": [ + 141, + 309, + 470, + 322 + ], + "score": 1.0, + "content": "as a text generation problem, modeled using pre-trained sequence-to-sequence", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 321, + 469, + 331 + ], + "spans": [ + { + "bbox": [ + 142, + 321, + 469, + 331 + ], + "score": 1.0, + "content": "models. The general idea is that models trained to convert the generated text", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 330, + 470, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 330, + 470, + 343 + ], + "score": 1.0, + "content": "to/from a reference output or the source text will achieve higher scores when", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 142, + 342, + 469, + 354 + ], + "spans": [ + { + "bbox": [ + 142, + 342, + 469, + 354 + ], + "score": 1.0, + "content": "the generated text is better. We operationalize this idea using BART [32], an", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 353, + 469, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 353, + 469, + 365 + ], + "score": 1.0, + "content": "encoder-decoder based pre-trained model, and propose a metric BARTSCORE", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 363, + 469, + 375 + ], + "spans": [ + { + "bbox": [ + 141, + 363, + 469, + 375 + ], + "score": 1.0, + "content": "with a number of variants that can be flexibly applied in an unsupervised fashion", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 373, + 471, + 388 + ], + "spans": [ + { + "bbox": [ + 141, + 373, + 471, + 388 + ], + "score": 1.0, + "content": "to evaluation of text from different perspectives (e.g. informativeness, fluency,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 385, + 471, + 398 + ], + "spans": [ + { + "bbox": [ + 141, + 385, + 471, + 398 + ], + "score": 1.0, + "content": "or factuality). BARTSCORE is conceptually simple and empirically effective.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 396, + 471, + 409 + ], + "spans": [ + { + "bbox": [ + 141, + 396, + 471, + 409 + ], + "score": 1.0, + "content": "It can outperform existing top-scoring metrics in 16 of 22 test settings, cov-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 407, + 470, + 420 + ], + "spans": [ + { + "bbox": [ + 141, + 407, + 470, + 420 + ], + "score": 1.0, + "content": "ering evaluation of 16 datasets (e.g., machine translation, text summarization)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 419, + 469, + 430 + ], + "spans": [ + { + "bbox": [ + 141, + 419, + 469, + 430 + ], + "score": 1.0, + "content": "and 7 different perspectives (e.g., informativeness, factuality). Code to calculate", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 429, + 471, + 441 + ], + "spans": [ + { + "bbox": [ + 141, + 429, + 471, + 441 + ], + "score": 1.0, + "content": "BARTScore is available at https://github.com/neulab/BARTScore,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 439, + 471, + 452 + ], + "spans": [ + { + "bbox": [ + 141, + 439, + 471, + 452 + ], + "score": 1.0, + "content": "and we have released an interactive leaderboard for meta-evaluation at http:", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 143, + 451, + 469, + 463 + ], + "spans": [ + { + "bbox": [ + 143, + 451, + 469, + 463 + ], + "score": 1.0, + "content": "//explainaboard.nlpedia.ai/leaderboard/task-meval/ on the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 142, + 462, + 469, + 473 + ], + "spans": [ + { + "bbox": [ + 142, + 462, + 469, + 473 + ], + "score": 1.0, + "content": "EXPLAINABOARD platform [38], which allows us to interactively understand the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 473, + 383, + 484 + ], + "spans": [ + { + "bbox": [ + 141, + 473, + 383, + 484 + ], + "score": 1.0, + "content": "strengths, weaknesses, and complementarity of each metric.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 21.5, + "bbox_fs": [ + 141, + 265, + 471, + 484 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 502, + 190, + 516 + ], + "lines": [ + { + "bbox": [ + 105, + 501, + 192, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 192, + 519 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 527, + 505, + 626 + ], + "lines": [ + { + "bbox": [ + 106, + 527, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 506, + 540 + ], + "score": 1.0, + "content": "One defining feature of recent NLP models is the use of neural representations trained on raw text,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "using unsupervised objectives such as language modeling [6,53], or denoising autoencoding [9,32,54].", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 550, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 505, + 561 + ], + "score": 1.0, + "content": "By learning to predict the words or sentences in natural text, these models simultaneously learn to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 560, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 506, + 573 + ], + "score": 1.0, + "content": "extract features that not only benefit mainstream NLP tasks such as information extraction [23, 37],", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 571, + 506, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 506, + 584 + ], + "score": 1.0, + "content": "question answering [1, 26], text summarization [40, 77] but also have proven effective in development", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 505, + 595 + ], + "score": 1.0, + "content": "of automatic metrics for evaluation of text generation itself [62, 65]. For example, BERTScore [75]", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 594, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 605 + ], + "score": 1.0, + "content": "and MoverScore [76] take features extracted by BERT [9] and apply unsupervised matching functions", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 604, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 506, + 616 + ], + "score": 1.0, + "content": "to compare system outputs against references. Other works build supervised frameworks that use the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 615, + 435, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 435, + 627 + ], + "score": 1.0, + "content": "extracted features to learn to rank [56] or regress [62] to human evaluation scores.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 527, + 506, + 627 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 631, + 505, + 696 + ], + "lines": [ + { + "bbox": [ + 106, + 632, + 505, + 643 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 505, + 643 + ], + "score": 1.0, + "content": "However, in the context of generation evaluation, one may note that there is a decided disconnect", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "between how models are pre-trained using text generation objectives and how they are used as", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 653, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 106, + 653, + 505, + 665 + ], + "score": 1.0, + "content": "down-stream feature extractors. This leads to potential under-utilization of the pre-trained model", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 664, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 506, + 677 + ], + "score": 1.0, + "content": "parameters. For example, the output prediction layer is not used at all in this case. This disconnect", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 675, + 505, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 505, + 687 + ], + "score": 1.0, + "content": "is particularly striking because of the close connection between the pre-training objectives and the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 687, + 257, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 257, + 698 + ], + "score": 1.0, + "content": "generation tasks we want to evaluate.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 632, + 506, + 698 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 73, + 505, + 237 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "In this paper, we instead argue for a formulation of evaluation of generated text as a text generation", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 83, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 104, + 83, + 506, + 97 + ], + "score": 1.0, + "content": "problem, directly evaluating text through the lens of its probability of being generated from or generat-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "ing other textual inputs and outputs. This is a better match with the underlying pre-training tasks and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "allows us to more fully take advantage of the parameters learned during the pre-training phase. We", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 116, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 506, + 129 + ], + "score": 1.0, + "content": "solve the modeling problem with a pre-trained sequence-to-sequence (seq2seq) model, specifically", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 127, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 506, + 140 + ], + "score": 1.0, + "content": "BART [32], and devise a metric named BARTSCORE, which has the following characteristics: (1)", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 138, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 505, + 150 + ], + "score": 1.0, + "content": "BARTSCORE is parameter- and data-efficient. Architecturally there are no extra parameters beyond", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "those used in pre-training itself, and it is an unsupervised metric that doesn’t require human judgments", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 104, + 159, + 506, + 173 + ], + "score": 1.0, + "content": "to train. (2) BARTSCORE can better support evaluation of generated text from different perspectives", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 280, + 183 + ], + "score": 1.0, + "content": "(e.g., informativeness, coherence, factuality,", + "type": "text" + }, + { + "bbox": [ + 280, + 171, + 292, + 182 + ], + "score": 0.6, + "content": "\\ S 4", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 171, + 505, + 183 + ], + "score": 1.0, + "content": ") by adjusting the inputs and outputs of the conditional", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 182, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 290, + 194 + ], + "score": 1.0, + "content": "text generation problem, as we demonstrate in", + "type": "text" + }, + { + "bbox": [ + 290, + 182, + 309, + 193 + ], + "score": 0.82, + "content": "\\ S 3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 182, + 506, + 194 + ], + "score": 1.0, + "content": ". This is in contrast to most previous work, which", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 192, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 506, + 205 + ], + "score": 1.0, + "content": "mostly examines correlation of the devised metrics with output quality from a limited number of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 203, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 506, + 217 + ], + "score": 1.0, + "content": "perspectives. (3) BARTSCORE can be further enhanced by (i) providing textual prompts that bring", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "the evaluation task closer to the pre-training task, or (ii) updating the underlying model by fine-tuning", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 225, + 398, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 398, + 238 + ], + "score": 1.0, + "content": "BART based on downstream generation tasks (e.g., text summarization).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 241, + 505, + 319 + ], + "lines": [ + { + "bbox": [ + 106, + 242, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 506, + 254 + ], + "score": 1.0, + "content": "Experimentally, we evaluate different variants of BARTSCORE from 7 perspectives on 16 datasets.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 252, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 252, + 505, + 266 + ], + "score": 1.0, + "content": "BARTSCORE achieves the best performance in 16 of 22 test settings against existing top-scoring", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "metrics. Empirical results also show the effectiveness of the prompting strategy supported by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 273, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 506, + 288 + ], + "score": 1.0, + "content": "BARTSCORE. For example, simply adding the phrase “such as” to the translated text when using", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 285, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 222, + 297 + ], + "score": 1.0, + "content": "BARTSCORE can lead to a", + "type": "text" + }, + { + "bbox": [ + 223, + 286, + 238, + 296 + ], + "score": 0.84, + "content": "3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 285, + 506, + 297 + ], + "score": 1.0, + "content": "point absolute improvement in correlation on “German-English”", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "score": 1.0, + "content": "machine translation (MT) evaluation. Additional analysis shows that BARTSCORE is more robust", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 308, + 408, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 408, + 320 + ], + "score": 1.0, + "content": "when dealing with high-quality texts generated by top-performing systems.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 107, + 333, + 195, + 347 + ], + "lines": [ + { + "bbox": [ + 104, + 331, + 196, + 350 + ], + "spans": [ + { + "bbox": [ + 104, + 331, + 196, + 350 + ], + "score": 1.0, + "content": "2 Preliminaries", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 107, + 357, + 222, + 369 + ], + "lines": [ + { + "bbox": [ + 105, + 356, + 223, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 223, + 370 + ], + "score": 1.0, + "content": "2.1 Problem Formulation", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 378, + 505, + 422 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "As stated above, our goal is to assess the quality of generated text [3, 46]. In this work, we focus on", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "conditional text generation (e.g., machine translation), where the goal is to generate a hypothesis", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 107, + 399, + 185, + 411 + ], + "score": 0.9, + "content": "( \\boldsymbol { h } = h _ { 1 } , \\cdots , h _ { m } )", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 399, + 309, + 412 + ], + "score": 1.0, + "content": "based on a given source text", + "type": "text" + }, + { + "bbox": [ + 309, + 400, + 380, + 411 + ], + "score": 0.92, + "content": "( s = s _ { 1 } , \\cdots , s _ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 399, + 505, + 412 + ], + "score": 1.0, + "content": ". Commonly, one or multiple", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 411, + 420, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 213, + 423 + ], + "score": 1.0, + "content": "human-created references", + "type": "text" + }, + { + "bbox": [ + 214, + 411, + 277, + 422 + ], + "score": 0.89, + "content": "( r = r _ { 1 } , \\cdots , r _ { l } )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 411, + 420, + 423 + ], + "score": 1.0, + "content": "are provided to aid this evaluation.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "title", + "bbox": [ + 107, + 434, + 277, + 446 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 277, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 277, + 447 + ], + "score": 1.0, + "content": "2.2 Gold-standard Human Evaluation", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 108, + 454, + 505, + 487 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "score": 1.0, + "content": "In general, the gold-standard method for evaluating such texts is still human evaluation, where human", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "annotators assess the generated texts’ quality. This evaluation can be done from perspectives, and we", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 475, + 355, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 336, + 488 + ], + "score": 1.0, + "content": "list a few common varieties below (all are investigated in", + "type": "text" + }, + { + "bbox": [ + 336, + 477, + 347, + 487 + ], + "score": 0.55, + "content": "\\ S 4", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 475, + 355, + 488 + ], + "score": 1.0, + "content": "):", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 129, + 495, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 129, + 493, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 129, + 493, + 505, + 509 + ], + "score": 1.0, + "content": "1. Informativeness (INFO): How well the generated hypothesis captures the key ideas of the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 141, + 506, + 210, + 518 + ], + "spans": [ + { + "bbox": [ + 141, + 506, + 210, + 518 + ], + "score": 1.0, + "content": "source text [18].", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 127, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 127, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "2. Relevance (REL): How consistent the generated hypothesis is with respect to the source", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 141, + 528, + 181, + 540 + ], + "spans": [ + { + "bbox": [ + 141, + 528, + 181, + 540 + ], + "score": 1.0, + "content": "text [19].", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 128, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 128, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "3. Fluency (FLU): Whether the text has no formatting problems, capitalization errors or", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 141, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 141, + 550, + 505, + 563 + ], + "score": 1.0, + "content": "obviously ungrammatical sentences (e.g., fragments, missing components) that make the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 141, + 560, + 244, + 573 + ], + "spans": [ + { + "bbox": [ + 141, + 560, + 244, + 573 + ], + "score": 1.0, + "content": "text difficult to read [13].", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 130, + 572, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 130, + 572, + 505, + 583 + ], + "score": 1.0, + "content": "4. Coherence (COH): Whether the text builds from sentence to sentence to a coherent body of", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 141, + 582, + 262, + 595 + ], + "spans": [ + { + "bbox": [ + 141, + 582, + 262, + 595 + ], + "score": 1.0, + "content": "information about a topic [7].", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 128, + 592, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 128, + 592, + 505, + 606 + ], + "score": 1.0, + "content": "5. Factuality (FAC): Whether the generated hypothesis contains only statements entailed by", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 141, + 604, + 224, + 616 + ], + "spans": [ + { + "bbox": [ + 141, + 604, + 224, + 616 + ], + "score": 1.0, + "content": "the source text [30].", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 131, + 614, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 131, + 614, + 505, + 628 + ], + "score": 1.0, + "content": "6. Semantic Coverage (COV): How many semantic content units from reference texts are", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 141, + 626, + 311, + 639 + ], + "spans": [ + { + "bbox": [ + 141, + 626, + 311, + 639 + ], + "score": 1.0, + "content": "covered by the generated hypothesis [49].", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 128, + 636, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 128, + 636, + 505, + 650 + ], + "score": 1.0, + "content": "7. Adequacy (ADE): Whether the output conveys the same meaning as the input sentence, and", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 141, + 648, + 353, + 661 + ], + "spans": [ + { + "bbox": [ + 141, + 648, + 353, + 661 + ], + "score": 1.0, + "content": "none of the message is lost, added, or distorted [29].", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "score": 1.0, + "content": "Most existing evaluation metrics were designed to cover a small subset of these perspectives. For", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "example, BLEU [50] aims to capture the adequacy and fluency of translations, while ROUGE [36]", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 689, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 505, + 701 + ], + "score": 1.0, + "content": "was designed to match the semantic coverage metric. Some metrics, particularly trainable ones, can", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "perform evaluation from different perspectives but generally require maximizing correlation with", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 711, + 257, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 257, + 724 + ], + "score": 1.0, + "content": "each type of judgment separately [8].", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 49 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 73, + 505, + 237 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "In this paper, we instead argue for a formulation of evaluation of generated text as a text generation", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 83, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 104, + 83, + 506, + 97 + ], + "score": 1.0, + "content": "problem, directly evaluating text through the lens of its probability of being generated from or generat-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "ing other textual inputs and outputs. This is a better match with the underlying pre-training tasks and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "allows us to more fully take advantage of the parameters learned during the pre-training phase. We", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 116, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 506, + 129 + ], + "score": 1.0, + "content": "solve the modeling problem with a pre-trained sequence-to-sequence (seq2seq) model, specifically", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 127, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 506, + 140 + ], + "score": 1.0, + "content": "BART [32], and devise a metric named BARTSCORE, which has the following characteristics: (1)", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 138, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 505, + 150 + ], + "score": 1.0, + "content": "BARTSCORE is parameter- and data-efficient. Architecturally there are no extra parameters beyond", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "those used in pre-training itself, and it is an unsupervised metric that doesn’t require human judgments", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 104, + 159, + 506, + 173 + ], + "score": 1.0, + "content": "to train. (2) BARTSCORE can better support evaluation of generated text from different perspectives", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 280, + 183 + ], + "score": 1.0, + "content": "(e.g., informativeness, coherence, factuality,", + "type": "text" + }, + { + "bbox": [ + 280, + 171, + 292, + 182 + ], + "score": 0.6, + "content": "\\ S 4", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 171, + 505, + 183 + ], + "score": 1.0, + "content": ") by adjusting the inputs and outputs of the conditional", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 182, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 290, + 194 + ], + "score": 1.0, + "content": "text generation problem, as we demonstrate in", + "type": "text" + }, + { + "bbox": [ + 290, + 182, + 309, + 193 + ], + "score": 0.82, + "content": "\\ S 3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 182, + 506, + 194 + ], + "score": 1.0, + "content": ". This is in contrast to most previous work, which", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 192, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 506, + 205 + ], + "score": 1.0, + "content": "mostly examines correlation of the devised metrics with output quality from a limited number of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 203, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 506, + 217 + ], + "score": 1.0, + "content": "perspectives. (3) BARTSCORE can be further enhanced by (i) providing textual prompts that bring", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "the evaluation task closer to the pre-training task, or (ii) updating the underlying model by fine-tuning", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 225, + 398, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 398, + 238 + ], + "score": 1.0, + "content": "BART based on downstream generation tasks (e.g., text summarization).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 7, + "bbox_fs": [ + 104, + 72, + 506, + 238 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 241, + 505, + 319 + ], + "lines": [ + { + "bbox": [ + 106, + 242, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 506, + 254 + ], + "score": 1.0, + "content": "Experimentally, we evaluate different variants of BARTSCORE from 7 perspectives on 16 datasets.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 252, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 252, + 505, + 266 + ], + "score": 1.0, + "content": "BARTSCORE achieves the best performance in 16 of 22 test settings against existing top-scoring", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "metrics. Empirical results also show the effectiveness of the prompting strategy supported by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 273, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 506, + 288 + ], + "score": 1.0, + "content": "BARTSCORE. For example, simply adding the phrase “such as” to the translated text when using", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 285, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 222, + 297 + ], + "score": 1.0, + "content": "BARTSCORE can lead to a", + "type": "text" + }, + { + "bbox": [ + 223, + 286, + 238, + 296 + ], + "score": 0.84, + "content": "3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 285, + 506, + 297 + ], + "score": 1.0, + "content": "point absolute improvement in correlation on “German-English”", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "score": 1.0, + "content": "machine translation (MT) evaluation. Additional analysis shows that BARTSCORE is more robust", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 308, + 408, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 408, + 320 + ], + "score": 1.0, + "content": "when dealing with high-quality texts generated by top-performing systems.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 242, + 506, + 320 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 333, + 195, + 347 + ], + "lines": [ + { + "bbox": [ + 104, + 331, + 196, + 350 + ], + "spans": [ + { + "bbox": [ + 104, + 331, + 196, + 350 + ], + "score": 1.0, + "content": "2 Preliminaries", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 107, + 357, + 222, + 369 + ], + "lines": [ + { + "bbox": [ + 105, + 356, + 223, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 223, + 370 + ], + "score": 1.0, + "content": "2.1 Problem Formulation", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 378, + 505, + 422 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "As stated above, our goal is to assess the quality of generated text [3, 46]. In this work, we focus on", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "conditional text generation (e.g., machine translation), where the goal is to generate a hypothesis", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 107, + 399, + 185, + 411 + ], + "score": 0.9, + "content": "( \\boldsymbol { h } = h _ { 1 } , \\cdots , h _ { m } )", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 399, + 309, + 412 + ], + "score": 1.0, + "content": "based on a given source text", + "type": "text" + }, + { + "bbox": [ + 309, + 400, + 380, + 411 + ], + "score": 0.92, + "content": "( s = s _ { 1 } , \\cdots , s _ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 399, + 505, + 412 + ], + "score": 1.0, + "content": ". Commonly, one or multiple", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 411, + 420, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 213, + 423 + ], + "score": 1.0, + "content": "human-created references", + "type": "text" + }, + { + "bbox": [ + 214, + 411, + 277, + 422 + ], + "score": 0.89, + "content": "( r = r _ { 1 } , \\cdots , r _ { l } )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 411, + 420, + 423 + ], + "score": 1.0, + "content": "are provided to aid this evaluation.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 378, + 506, + 423 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 434, + 277, + 446 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 277, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 277, + 447 + ], + "score": 1.0, + "content": "2.2 Gold-standard Human Evaluation", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 108, + 454, + 505, + 487 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "score": 1.0, + "content": "In general, the gold-standard method for evaluating such texts is still human evaluation, where human", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "annotators assess the generated texts’ quality. This evaluation can be done from perspectives, and we", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 475, + 355, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 336, + 488 + ], + "score": 1.0, + "content": "list a few common varieties below (all are investigated in", + "type": "text" + }, + { + "bbox": [ + 336, + 477, + 347, + 487 + ], + "score": 0.55, + "content": "\\ S 4", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 475, + 355, + 488 + ], + "score": 1.0, + "content": "):", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 454, + 505, + 488 + ] + }, + { + "type": "list", + "bbox": [ + 129, + 495, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 129, + 493, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 129, + 493, + 505, + 509 + ], + "score": 1.0, + "content": "1. Informativeness (INFO): How well the generated hypothesis captures the key ideas of the", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 506, + 210, + 518 + ], + "spans": [ + { + "bbox": [ + 141, + 506, + 210, + 518 + ], + "score": 1.0, + "content": "source text [18].", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 127, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 127, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "2. Relevance (REL): How consistent the generated hypothesis is with respect to the source", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 528, + 181, + 540 + ], + "spans": [ + { + "bbox": [ + 141, + 528, + 181, + 540 + ], + "score": 1.0, + "content": "text [19].", + "type": "text" + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 128, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "3. Fluency (FLU): Whether the text has no formatting problems, capitalization errors or", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 141, + 550, + 505, + 563 + ], + "score": 1.0, + "content": "obviously ungrammatical sentences (e.g., fragments, missing components) that make the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 141, + 560, + 244, + 573 + ], + "spans": [ + { + "bbox": [ + 141, + 560, + 244, + 573 + ], + "score": 1.0, + "content": "text difficult to read [13].", + "type": "text" + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 130, + 572, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 130, + 572, + 505, + 583 + ], + "score": 1.0, + "content": "4. Coherence (COH): Whether the text builds from sentence to sentence to a coherent body of", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 582, + 262, + 595 + ], + "spans": [ + { + "bbox": [ + 141, + 582, + 262, + 595 + ], + "score": 1.0, + "content": "information about a topic [7].", + "type": "text" + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 592, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 128, + 592, + 505, + 606 + ], + "score": 1.0, + "content": "5. Factuality (FAC): Whether the generated hypothesis contains only statements entailed by", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 604, + 224, + 616 + ], + "spans": [ + { + "bbox": [ + 141, + 604, + 224, + 616 + ], + "score": 1.0, + "content": "the source text [30].", + "type": "text" + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 614, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 131, + 614, + 505, + 628 + ], + "score": 1.0, + "content": "6. Semantic Coverage (COV): How many semantic content units from reference texts are", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 626, + 311, + 639 + ], + "spans": [ + { + "bbox": [ + 141, + 626, + 311, + 639 + ], + "score": 1.0, + "content": "covered by the generated hypothesis [49].", + "type": "text" + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 636, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 128, + 636, + 505, + 650 + ], + "score": 1.0, + "content": "7. Adequacy (ADE): Whether the output conveys the same meaning as the input sentence, and", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 648, + 353, + 661 + ], + "spans": [ + { + "bbox": [ + 141, + 648, + 353, + 661 + ], + "score": 1.0, + "content": "none of the message is lost, added, or distorted [29].", + "type": "text" + } + ], + "index": 46, + "is_list_end_line": true + } + ], + "index": 39, + "bbox_fs": [ + 127, + 493, + 506, + 661 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "score": 1.0, + "content": "Most existing evaluation metrics were designed to cover a small subset of these perspectives. For", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "example, BLEU [50] aims to capture the adequacy and fluency of translations, while ROUGE [36]", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 689, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 505, + 701 + ], + "score": 1.0, + "content": "was designed to match the semantic coverage metric. Some metrics, particularly trainable ones, can", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "perform evaluation from different perspectives but generally require maximizing correlation with", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 711, + 257, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 257, + 724 + ], + "score": 1.0, + "content": "each type of judgment separately [8].", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 49, + "bbox_fs": [ + 105, + 667, + 505, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 0, + 488, + 155 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 0, + 488, + 155 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 0, + 488, + 155 + ], + "spans": [ + { + "bbox": [ + 119, + 0, + 488, + 155 + ], + "score": 0.126, + "type": "image", + "image_path": "6dc5bc0e6cf11512055650092647de02eabd130d83aa9f95c1a0baa74e5cfd9f.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 0, + 488, + 51.666666666666664 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 51.666666666666664, + 488, + 103.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 103.33333333333333, + 488, + 155.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 104, + 160, + 505, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 159, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 321, + 173 + ], + "score": 1.0, + "content": "Figure 1: Evaluation metrics as different tasks, where", + "type": "text" + }, + { + "bbox": [ + 321, + 162, + 330, + 172 + ], + "score": 0.74, + "content": "s _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 159, + 334, + 173 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 334, + 161, + 344, + 172 + ], + "score": 0.79, + "content": "h _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 159, + 362, + 173 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 362, + 163, + 372, + 173 + ], + "score": 0.84, + "content": "r _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 159, + 505, + 173 + ], + "score": 1.0, + "content": "represent source, hypothesis and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 171, + 225, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 225, + 184 + ], + "score": 1.0, + "content": "reference words respectively.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 208, + 504, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 206, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 201, + 223 + ], + "score": 1.0, + "content": "As we describe more in", + "type": "text" + }, + { + "bbox": [ + 202, + 209, + 213, + 219 + ], + "score": 0.66, + "content": "\\ S 4", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 206, + 505, + 223 + ], + "score": 1.0, + "content": ", BARTSCORE can evaluate text from the great majority of these perspec-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 219, + 399, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 399, + 232 + ], + "score": 1.0, + "content": "Factuality Relevance Factuality Relevance tives, significantly expanding its applicability compared to these metrics.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "title", + "bbox": [ + 108, + 248, + 255, + 259 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 257, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 257, + 261 + ], + "score": 1.0, + "content": "s r2.3 Evaluation as Different Tasks", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 108, + 270, + 504, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 268, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 506, + 284 + ], + "score": 1.0, + "content": "There is a recent trend that leverages neural models for automated evaluation in different ways,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 281, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 506, + 293 + ], + "score": 1.0, + "content": "as shown in Fig. 1. We first elaborate on their characteristics by highlighting differences in task", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 291, + 270, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 270, + 304 + ], + "score": 1.0, + "content": "formulation and evaluation perspectives.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 308, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 308, + 507, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 507, + 322 + ], + "score": 1.0, + "content": "T1: Unsupervised Matching. Unsupervised matching metrics aim to measure the semantic equiv-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 320, + 507, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 507, + 332 + ], + "score": 1.0, + "content": "alence between the reference and hypothesis by using a token-level matching functions in dis-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "score": 1.0, + "content": "tributed representation space, such as BERTScore [75], MoverScore [76] or discrete string space", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "score": 1.0, + "content": "like ROUGE [35], BLEU [50], CHRF [52]. Although similar matching functions can be used to", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "score": 1.0, + "content": "assess the quality beyond semantic equivalence (e.g, factuality, a relationship between source text", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 363, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 376 + ], + "score": 1.0, + "content": "and hypothesis), to our knowledge prior research has not attested to the capability of unsupervised", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 374, + 446, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 446, + 387 + ], + "score": 1.0, + "content": "matching methods in this regard; we explore this further in our experiments (Tab. 5).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 390, + 504, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "T2: Supervised Regression. Regression-based models introduce a parameterized regression layer,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 401, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 505, + 414 + ], + "score": 1.0, + "content": "which would be learned in a supervised fashion to accurately predict human judgments. Examples", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 411, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 425, + 425 + ], + "score": 1.0, + "content": "include recent metrics BLEURT [62], COMET [56] and traditional metrics like", + "type": "text" + }, + { + "bbox": [ + 425, + 412, + 437, + 423 + ], + "score": 0.86, + "content": "{ \\bar { S } } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 411, + 506, + 425 + ], + "score": 1.0, + "content": "[51], VRM [21].", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 428, + 505, + 483 + ], + "lines": [ + { + "bbox": [ + 105, + 427, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 505, + 441 + ], + "score": 1.0, + "content": "T3: Supervised Ranking. Evaluation can also be conceived as a ranking problem, where the main", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "idea is to learn a scoring function that assigns a higher score to better hypotheses than to worse ones.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "score": 1.0, + "content": "Examples include COMET [56] and BEER [64], where COMET focuses the machine translation", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "task and relies on human judgments to tune parameters in ranking or regression layers, and BEER", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 472, + 442, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 442, + 484 + ], + "score": 1.0, + "content": "combines many simple features in a tunable linear model of MT evaluation metrics.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 488, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "T4: Text Generation. In this work, we formulate evaluating generated text as a text generation task", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "from pre-trained language models. The basic idea is that a high-quality hypothesis will be easily", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 104, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "generated based on source or reference text or vice-versa. This has not been covered as extensively", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 521, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 533 + ], + "score": 1.0, + "content": "in previous work, with one notable exception being PRISM [65]. Our work differs from PRISM", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 532, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 506, + 545 + ], + "score": 1.0, + "content": "in several ways: (i) PRISM formulates evaluation as a paraphrasing task, whose definition that", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "score": 1.0, + "content": "two texts are with the same meaning limits its applicable scenarios, like factuality evaluation in", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 555, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 566 + ], + "score": 1.0, + "content": "text summarization that takes source documents and generated summaries as input whose semantic", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 564, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 104, + 564, + 506, + 578 + ], + "score": 1.0, + "content": "space are different. (ii) PRISM trained a model from scratch on parallel data while BARTSCORE", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "is based on open-sourced pre-trained seq2seq models. (iii) BARTSCORE supports prompt-based", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 587, + 335, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 335, + 599 + ], + "score": 1.0, + "content": "learning [59, 63] which hasn’t been examined in PRISM.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5 + }, + { + "type": "title", + "bbox": [ + 107, + 617, + 186, + 631 + ], + "lines": [ + { + "bbox": [ + 104, + 617, + 189, + 634 + ], + "spans": [ + { + "bbox": [ + 104, + 617, + 189, + 634 + ], + "score": 1.0, + "content": "3 BARTScore", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "title", + "bbox": [ + 107, + 645, + 310, + 657 + ], + "lines": [ + { + "bbox": [ + 104, + 644, + 312, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 644, + 312, + 659 + ], + "score": 1.0, + "content": "3.1 Sequence-to-Sequence Pre-trained Models", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "score": 1.0, + "content": "Although pre-trained models differ along different axes, one of the main axes of variation is the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "score": 1.0, + "content": "training objective, with two main variants: language modeling objectives (e.g., masked language", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "score": 1.0, + "content": "modeling [9]) and seq2seq objectives [54]. In particular, seq2seq pre-trained models are particularly", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "well-suited to conditioned generation tasks since they consist of both an encoder and a decoder,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 710, + 505, + 725 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 725 + ], + "score": 1.0, + "content": "and predictions are made auto-regressively [32]. In this work, we operationalize our idea by using", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 0, + 488, + 155 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 0, + 488, + 155 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 0, + 488, + 155 + ], + "spans": [ + { + "bbox": [ + 119, + 0, + 488, + 155 + ], + "score": 0.126, + "type": "image", + "image_path": "6dc5bc0e6cf11512055650092647de02eabd130d83aa9f95c1a0baa74e5cfd9f.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 0, + 488, + 51.666666666666664 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 51.666666666666664, + 488, + 103.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 103.33333333333333, + 488, + 155.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 104, + 160, + 505, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 159, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 321, + 173 + ], + "score": 1.0, + "content": "Figure 1: Evaluation metrics as different tasks, where", + "type": "text" + }, + { + "bbox": [ + 321, + 162, + 330, + 172 + ], + "score": 0.74, + "content": "s _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 159, + 334, + 173 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 334, + 161, + 344, + 172 + ], + "score": 0.79, + "content": "h _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 159, + 362, + 173 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 362, + 163, + 372, + 173 + ], + "score": 0.84, + "content": "r _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 159, + 505, + 173 + ], + "score": 1.0, + "content": "represent source, hypothesis and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 171, + 225, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 225, + 184 + ], + "score": 1.0, + "content": "reference words respectively.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 208, + 504, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 206, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 201, + 223 + ], + "score": 1.0, + "content": "As we describe more in", + "type": "text" + }, + { + "bbox": [ + 202, + 209, + 213, + 219 + ], + "score": 0.66, + "content": "\\ S 4", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 206, + 505, + 223 + ], + "score": 1.0, + "content": ", BARTSCORE can evaluate text from the great majority of these perspec-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 219, + 399, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 399, + 232 + ], + "score": 1.0, + "content": "Factuality Relevance Factuality Relevance tives, significantly expanding its applicability compared to these metrics.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 105, + 206, + 505, + 232 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 248, + 255, + 259 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 257, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 257, + 261 + ], + "score": 1.0, + "content": "s r2.3 Evaluation as Different Tasks", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 108, + 270, + 504, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 268, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 506, + 284 + ], + "score": 1.0, + "content": "There is a recent trend that leverages neural models for automated evaluation in different ways,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 281, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 506, + 293 + ], + "score": 1.0, + "content": "as shown in Fig. 1. We first elaborate on their characteristics by highlighting differences in task", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 291, + 270, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 270, + 304 + ], + "score": 1.0, + "content": "formulation and evaluation perspectives.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 106, + 268, + 506, + 304 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 308, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 308, + 507, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 507, + 322 + ], + "score": 1.0, + "content": "T1: Unsupervised Matching. Unsupervised matching metrics aim to measure the semantic equiv-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 320, + 507, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 507, + 332 + ], + "score": 1.0, + "content": "alence between the reference and hypothesis by using a token-level matching functions in dis-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "score": 1.0, + "content": "tributed representation space, such as BERTScore [75], MoverScore [76] or discrete string space", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "score": 1.0, + "content": "like ROUGE [35], BLEU [50], CHRF [52]. Although similar matching functions can be used to", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "score": 1.0, + "content": "assess the quality beyond semantic equivalence (e.g, factuality, a relationship between source text", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 363, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 376 + ], + "score": 1.0, + "content": "and hypothesis), to our knowledge prior research has not attested to the capability of unsupervised", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 374, + 446, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 446, + 387 + ], + "score": 1.0, + "content": "matching methods in this regard; we explore this further in our experiments (Tab. 5).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 308, + 507, + 387 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 390, + 504, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "T2: Supervised Regression. Regression-based models introduce a parameterized regression layer,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 401, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 505, + 414 + ], + "score": 1.0, + "content": "which would be learned in a supervised fashion to accurately predict human judgments. Examples", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 411, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 425, + 425 + ], + "score": 1.0, + "content": "include recent metrics BLEURT [62], COMET [56] and traditional metrics like", + "type": "text" + }, + { + "bbox": [ + 425, + 412, + 437, + 423 + ], + "score": 0.86, + "content": "{ \\bar { S } } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 411, + 506, + 425 + ], + "score": 1.0, + "content": "[51], VRM [21].", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 390, + 506, + 425 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 428, + 505, + 483 + ], + "lines": [ + { + "bbox": [ + 105, + 427, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 505, + 441 + ], + "score": 1.0, + "content": "T3: Supervised Ranking. Evaluation can also be conceived as a ranking problem, where the main", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "idea is to learn a scoring function that assigns a higher score to better hypotheses than to worse ones.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "score": 1.0, + "content": "Examples include COMET [56] and BEER [64], where COMET focuses the machine translation", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "task and relies on human judgments to tune parameters in ranking or regression layers, and BEER", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 472, + 442, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 442, + 484 + ], + "score": 1.0, + "content": "combines many simple features in a tunable linear model of MT evaluation metrics.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 427, + 506, + 484 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 488, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "T4: Text Generation. In this work, we formulate evaluating generated text as a text generation task", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "from pre-trained language models. The basic idea is that a high-quality hypothesis will be easily", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 104, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "generated based on source or reference text or vice-versa. This has not been covered as extensively", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 521, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 533 + ], + "score": 1.0, + "content": "in previous work, with one notable exception being PRISM [65]. Our work differs from PRISM", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 532, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 506, + 545 + ], + "score": 1.0, + "content": "in several ways: (i) PRISM formulates evaluation as a paraphrasing task, whose definition that", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "score": 1.0, + "content": "two texts are with the same meaning limits its applicable scenarios, like factuality evaluation in", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 555, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 566 + ], + "score": 1.0, + "content": "text summarization that takes source documents and generated summaries as input whose semantic", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 564, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 104, + 564, + 506, + 578 + ], + "score": 1.0, + "content": "space are different. (ii) PRISM trained a model from scratch on parallel data while BARTSCORE", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "is based on open-sourced pre-trained seq2seq models. (iii) BARTSCORE supports prompt-based", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 587, + 335, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 335, + 599 + ], + "score": 1.0, + "content": "learning [59, 63] which hasn’t been examined in PRISM.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5, + "bbox_fs": [ + 104, + 488, + 506, + 599 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 617, + 186, + 631 + ], + "lines": [ + { + "bbox": [ + 104, + 617, + 189, + 634 + ], + "spans": [ + { + "bbox": [ + 104, + 617, + 189, + 634 + ], + "score": 1.0, + "content": "3 BARTScore", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "title", + "bbox": [ + 107, + 645, + 310, + 657 + ], + "lines": [ + { + "bbox": [ + 104, + 644, + 312, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 644, + 312, + 659 + ], + "score": 1.0, + "content": "3.1 Sequence-to-Sequence Pre-trained Models", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "score": 1.0, + "content": "Although pre-trained models differ along different axes, one of the main axes of variation is the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "score": 1.0, + "content": "training objective, with two main variants: language modeling objectives (e.g., masked language", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "score": 1.0, + "content": "modeling [9]) and seq2seq objectives [54]. In particular, seq2seq pre-trained models are particularly", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "well-suited to conditioned generation tasks since they consist of both an encoder and a decoder,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 710, + 505, + 725 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 725 + ], + "score": 1.0, + "content": "and predictions are made auto-regressively [32]. In this work, we operationalize our idea by using", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 667, + 506, + 725 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 72, + 505, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "BART [32] as our backbone due to its superior performance in text generation [12, 42, 71]. We also", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "score": 1.0, + "content": "report preliminary experiments comparing BART with T5 [54] and PEGASUS [74] in the Appendix.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 504, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 504, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 267, + 113 + ], + "score": 1.0, + "content": "Given a seq2seq model parameterized by", + "type": "text" + }, + { + "bbox": [ + 268, + 101, + 273, + 110 + ], + "score": 0.78, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 99, + 392, + 113 + ], + "score": 1.0, + "content": ", a source sequence containing", + "type": "text" + }, + { + "bbox": [ + 393, + 102, + 400, + 110 + ], + "score": 0.75, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 99, + 428, + 113 + ], + "score": 1.0, + "content": "tokens", + "type": "text" + }, + { + "bbox": [ + 428, + 100, + 504, + 112 + ], + "score": 0.92, + "content": "\\mathbf { x } = \\{ x _ { 1 } , \\cdots , x _ { n } \\}", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 244, + 123 + ], + "score": 1.0, + "content": "and a target sequence containing", + "type": "text" + }, + { + "bbox": [ + 245, + 113, + 255, + 121 + ], + "score": 0.75, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 110, + 286, + 123 + ], + "score": 1.0, + "content": "tokens", + "type": "text" + }, + { + "bbox": [ + 286, + 111, + 365, + 123 + ], + "score": 0.93, + "content": "\\mathbf { y } = \\{ y _ { 1 } , \\dots , y _ { m } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 110, + 506, + 123 + ], + "score": 1.0, + "content": ". We can factorize the generation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 122, + 289, + 133 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 163, + 133 + ], + "score": 1.0, + "content": "probability of", + "type": "text" + }, + { + "bbox": [ + 164, + 123, + 171, + 133 + ], + "score": 0.47, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 122, + 234, + 133 + ], + "score": 1.0, + "content": "conditioned on", + "type": "text" + }, + { + "bbox": [ + 234, + 124, + 242, + 132 + ], + "score": 0.55, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 122, + 289, + 133 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 132, + 370, + 164 + ], + "lines": [ + { + "bbox": [ + 241, + 132, + 370, + 164 + ], + "spans": [ + { + "bbox": [ + 241, + 132, + 370, + 164 + ], + "score": 0.95, + "content": "p ( \\mathbf { y } | \\mathbf { x } , \\theta ) = \\prod _ { t = 1 } ^ { m } p ( \\mathbf { y } _ { t } | \\mathbf { y } _ { < t } , \\mathbf { x } , \\theta )", + "type": "interline_equation", + "image_path": "3d003f060de88a32edea6412d277ef24062aa11f4ba7b6d0a401b82261881f32.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 241, + 132, + 370, + 148.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 241, + 148.0, + 370, + 164.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 504, + 185 + ], + "lines": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "score": 1.0, + "content": "By exploring these probabilities, we design metrics that can gauge the quality of the generated text.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 106, + 196, + 181, + 208 + ], + "lines": [ + { + "bbox": [ + 105, + 194, + 182, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 182, + 211 + ], + "score": 1.0, + "content": "3.2 BARTScore", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 216, + 505, + 261 + ], + "lines": [ + { + "bbox": [ + 105, + 216, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 505, + 230 + ], + "score": 1.0, + "content": "The most general form of our proposed BARTSCORE is shown in Eq. 2, where we use the weighted", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 227, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 294, + 240 + ], + "score": 1.0, + "content": "log probability of one text y given another text", + "type": "text" + }, + { + "bbox": [ + 294, + 230, + 301, + 237 + ], + "score": 0.56, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 227, + 506, + 240 + ], + "score": 1.0, + "content": ". The weights are used to put different emphasis on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 237, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 252 + ], + "score": 1.0, + "content": "different tokens, which can be instantiated using different methods like Inverse Document Frequency", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 249, + 343, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 343, + 261 + ], + "score": 1.0, + "content": "(IDF) [25] etc. In our work, we weigh each token equally.2", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "interline_equation", + "bbox": [ + 217, + 264, + 393, + 297 + ], + "lines": [ + { + "bbox": [ + 217, + 264, + 393, + 297 + ], + "spans": [ + { + "bbox": [ + 217, + 264, + 393, + 297 + ], + "score": 0.95, + "content": "\\mathbf { B A R T S C O R E } = \\sum _ { t = 1 } ^ { m } \\omega _ { t } \\log p ( \\mathbf { y } _ { t } | \\mathbf { y } _ { < t } , \\mathbf { x } , \\theta )", + "type": "interline_equation", + "image_path": "bfa67b3ee2347ec4fa7a57ebb4b52d83186124f1544c141e530ce54a2d1cb637.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 217, + 264, + 393, + 280.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 217, + 280.5, + 393, + 297.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 299, + 505, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 299, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 506, + 313 + ], + "score": 1.0, + "content": "Due to its generation task-based formulation and ability to utilize the entirety of BART’s pre-trained", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "score": 1.0, + "content": "parameters, BARTSCORE can be flexibly used in different evaluation scenarios. We specifically", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 322, + 499, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 499, + 335 + ], + "score": 1.0, + "content": "present four methods for using BARTSCORE based on different generation directions, which are,", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 108, + 342, + 506, + 495 + ], + "lines": [ + { + "bbox": [ + 107, + 341, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 107, + 341, + 169, + 354 + ], + "score": 1.0, + "content": "• Faithfulness", + "type": "text" + }, + { + "bbox": [ + 169, + 342, + 199, + 352 + ], + "score": 0.87, + "content": "s \\to h", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 341, + 349, + 354 + ], + "score": 1.0, + "content": "): from source document to hypothesis", + "type": "text" + }, + { + "bbox": [ + 349, + 342, + 387, + 354 + ], + "score": 0.93, + "content": "p ( \\boldsymbol { h } | \\boldsymbol { s } , \\boldsymbol { \\theta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 341, + 506, + 354 + ], + "score": 1.0, + "content": ". This direction measures how", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 352, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 115, + 352, + 506, + 365 + ], + "score": 1.0, + "content": "likely it is that the hypothesis could be generated based on the source text. Potential application", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 114, + 363, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 114, + 363, + 333, + 375 + ], + "score": 1.0, + "content": "scenarios are factuality and relevance introduced in", + "type": "text" + }, + { + "bbox": [ + 333, + 363, + 352, + 374 + ], + "score": 0.87, + "content": "\\ S 2 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 363, + 506, + 375 + ], + "score": 1.0, + "content": ". This measure can also be used for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 374, + 498, + 387 + ], + "spans": [ + { + "bbox": [ + 115, + 374, + 469, + 387 + ], + "score": 1.0, + "content": "estimating measures of the quality of only the target text, such as coherence and fluency", + "type": "text" + }, + { + "bbox": [ + 470, + 374, + 494, + 385 + ], + "score": 0.32, + "content": "( \\ S 2 . 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 374, + 498, + 387 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 109, + 384, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 109, + 384, + 158, + 398 + ], + "score": 1.0, + "content": "• Precision", + "type": "text" + }, + { + "bbox": [ + 158, + 385, + 189, + 396 + ], + "score": 0.87, + "content": "( r h )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 384, + 371, + 398 + ], + "score": 1.0, + "content": "): from reference text to system-generated text", + "type": "text" + }, + { + "bbox": [ + 371, + 385, + 409, + 397 + ], + "score": 0.94, + "content": "p ( \\boldsymbol { h } | \\boldsymbol { r } , \\boldsymbol { \\theta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 384, + 506, + 398 + ], + "score": 1.0, + "content": ". This direction assesses", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "how likely the hypothesis could be constructed based on the gold reference and is suitable for the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 113, + 407, + 228, + 419 + ], + "spans": [ + { + "bbox": [ + 113, + 407, + 228, + 419 + ], + "score": 1.0, + "content": "precision-focused scenario.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 108, + 417, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 108, + 417, + 147, + 432 + ], + "score": 1.0, + "content": "• Recall", + "type": "text" + }, + { + "bbox": [ + 147, + 418, + 178, + 429 + ], + "score": 0.86, + "content": "( h \\to r", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 417, + 367, + 432 + ], + "score": 1.0, + "content": "): from system-generated text to reference text", + "type": "text" + }, + { + "bbox": [ + 367, + 418, + 405, + 430 + ], + "score": 0.93, + "content": "p ( \\pmb { r } | \\pmb { h } , \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 417, + 506, + 432 + ], + "score": 1.0, + "content": ". This version quantifies", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 115, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "how easily a gold reference could be generated by the hypothesis and is suitable for pyramid-based", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 439, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 115, + 439, + 310, + 452 + ], + "score": 1.0, + "content": "evaluation (i.e., semantic coverage introduced in", + "type": "text" + }, + { + "bbox": [ + 311, + 440, + 330, + 451 + ], + "score": 0.82, + "content": "\\ S 2 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 439, + 506, + 452 + ], + "score": 1.0, + "content": ") in summarization task since pyramid score", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 114, + 451, + 498, + 463 + ], + "spans": [ + { + "bbox": [ + 114, + 451, + 498, + 463 + ], + "score": 1.0, + "content": "measures fine-grained Semantic Content Units (SCUs) [49] covered by system-generated texts.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 109, + 461, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 109, + 461, + 116, + 474 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 117, + 462, + 126, + 472 + ], + "score": 0.78, + "content": "\\mathcal { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 461, + 152, + 474 + ], + "score": 1.0, + "content": "score", + "type": "text" + }, + { + "bbox": [ + 153, + 462, + 183, + 473 + ], + "score": 0.85, + "content": "( r h", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 461, + 506, + 474 + ], + "score": 1.0, + "content": "): Consider both directions and use the arithmetic average of Precision and Recall", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 471, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 115, + 471, + 505, + 486 + ], + "score": 1.0, + "content": "ones. This version can be broadly used to evaluate the semantic overlap (informativeness, adequacy", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 482, + 362, + 496 + ], + "spans": [ + { + "bbox": [ + 115, + 482, + 160, + 496 + ], + "score": 1.0, + "content": "detailed in", + "type": "text" + }, + { + "bbox": [ + 161, + 483, + 181, + 494 + ], + "score": 0.76, + "content": "\\ S 2 . 2 \\AA ,", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 482, + 362, + 496 + ], + "score": 1.0, + "content": ") between reference texts and generated texts.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 107, + 507, + 220, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 221, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 221, + 520 + ], + "score": 1.0, + "content": "3.3 BARTScore Variants", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 108, + 527, + 504, + 561 + ], + "lines": [ + { + "bbox": [ + 106, + 527, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 366, + 541 + ], + "score": 1.0, + "content": "We also investigate two extensions to BARTSCORE: (i) changing", + "type": "text" + }, + { + "bbox": [ + 366, + 529, + 374, + 538 + ], + "score": 0.38, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 527, + 391, + 541 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 391, + 529, + 399, + 539 + ], + "score": 0.39, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 527, + 505, + 541 + ], + "score": 1.0, + "content": "through prompting, which", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 400, + 551 + ], + "score": 1.0, + "content": "can bring the evaluation task closer to the pre-training task. (ii) changing", + "type": "text" + }, + { + "bbox": [ + 400, + 539, + 407, + 548 + ], + "score": 0.74, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "by considering different", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 549, + 460, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 460, + 562 + ], + "score": 1.0, + "content": "fine-tuning tasks, which can bring the pre-training domain closer to the evaluation task.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "title", + "bbox": [ + 107, + 572, + 171, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 172, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 172, + 587 + ], + "score": 1.0, + "content": "3.3.1 Prompt", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 505, + 635 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 506, + 604 + ], + "score": 1.0, + "content": "Prompting is a practice of adding short phrases to the input or output to encourage pre-trained", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 601, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 506, + 614 + ], + "score": 1.0, + "content": "models to perform specific tasks, which has been proven effective in several other NLP scenarios", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 612, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 505, + 625 + ], + "score": 1.0, + "content": "[24,57,58,60,63]. The generative formulation of BARTSCORE makes it relatively easy to incorporate", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 624, + 402, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 402, + 635 + ], + "score": 1.0, + "content": "these insights here as well; we name this variant BARTSCORE-PROMPT.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 107, + 639, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 106, + 639, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 181, + 652 + ], + "score": 1.0, + "content": "Given a prompt of", + "type": "text" + }, + { + "bbox": [ + 182, + 641, + 186, + 649 + ], + "score": 0.53, + "content": "l", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 639, + 215, + 652 + ], + "score": 1.0, + "content": "tokens", + "type": "text" + }, + { + "bbox": [ + 216, + 640, + 285, + 651 + ], + "score": 0.9, + "content": "\\mathbf { z } = \\{ z _ { 1 } , \\cdots , z _ { l } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 639, + 505, + 652 + ], + "score": 1.0, + "content": ", we can either (i) append it to the source text, in which", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 650, + 507, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 157, + 664 + ], + "score": 1.0, + "content": "case we get", + "type": "text" + }, + { + "bbox": [ + 157, + 651, + 282, + 663 + ], + "score": 0.9, + "content": "\\mathbf { x } ^ { \\prime } = \\{ x _ { 1 } , \\cdot \\cdot \\cdot , x _ { n } , \\dot { z } _ { 1 } , \\cdot \\cdot \\cdot , z _ { l } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 650, + 507, + 664 + ], + "score": 1.0, + "content": ", and calculate the score based on this new source text", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 661, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 322, + 675 + ], + "score": 1.0, + "content": "using Eq.2. or (ii) prepend it to the target text, getting", + "type": "text" + }, + { + "bbox": [ + 323, + 662, + 447, + 674 + ], + "score": 0.93, + "content": "\\mathbf { y } ^ { \\prime } = \\{ z _ { 1 } , \\cdot \\cdot \\cdot , z _ { l } , y _ { 1 } , \\cdot \\cdot \\cdot , y _ { m } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 661, + 505, + 675 + ], + "score": 1.0, + "content": ". Then we can", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 672, + 264, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 672, + 264, + 685 + ], + "score": 1.0, + "content": "also use Eq.2 given the new target text.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 691, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 117, + 689, + 507, + 705 + ], + "spans": [ + { + "bbox": [ + 117, + 689, + 507, + 705 + ], + "score": 1.0, + "content": "2We have tried several other weighting schemes, including: (i) uniform weighting while ignoring stop words.", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 701, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 506, + 713 + ], + "score": 1.0, + "content": "(ii) IDF weighting. (iii) using the prior probability of each target token (calculated within the target sequence) as", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 711, + 435, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 435, + 723 + ], + "score": 1.0, + "content": "the weighting factor. However, none of those outperformed the uniform weighting scheme.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "4", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 72, + 505, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "BART [32] as our backbone due to its superior performance in text generation [12, 42, 71]. We also", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "score": 1.0, + "content": "report preliminary experiments comparing BART with T5 [54] and PEGASUS [74] in the Appendix.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 72, + 506, + 96 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 504, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 504, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 267, + 113 + ], + "score": 1.0, + "content": "Given a seq2seq model parameterized by", + "type": "text" + }, + { + "bbox": [ + 268, + 101, + 273, + 110 + ], + "score": 0.78, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 99, + 392, + 113 + ], + "score": 1.0, + "content": ", a source sequence containing", + "type": "text" + }, + { + "bbox": [ + 393, + 102, + 400, + 110 + ], + "score": 0.75, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 99, + 428, + 113 + ], + "score": 1.0, + "content": "tokens", + "type": "text" + }, + { + "bbox": [ + 428, + 100, + 504, + 112 + ], + "score": 0.92, + "content": "\\mathbf { x } = \\{ x _ { 1 } , \\cdots , x _ { n } \\}", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 244, + 123 + ], + "score": 1.0, + "content": "and a target sequence containing", + "type": "text" + }, + { + "bbox": [ + 245, + 113, + 255, + 121 + ], + "score": 0.75, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 110, + 286, + 123 + ], + "score": 1.0, + "content": "tokens", + "type": "text" + }, + { + "bbox": [ + 286, + 111, + 365, + 123 + ], + "score": 0.93, + "content": "\\mathbf { y } = \\{ y _ { 1 } , \\dots , y _ { m } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 110, + 506, + 123 + ], + "score": 1.0, + "content": ". We can factorize the generation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 122, + 289, + 133 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 163, + 133 + ], + "score": 1.0, + "content": "probability of", + "type": "text" + }, + { + "bbox": [ + 164, + 123, + 171, + 133 + ], + "score": 0.47, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 122, + 234, + 133 + ], + "score": 1.0, + "content": "conditioned on", + "type": "text" + }, + { + "bbox": [ + 234, + 124, + 242, + 132 + ], + "score": 0.55, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 122, + 289, + 133 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 99, + 506, + 133 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 132, + 370, + 164 + ], + "lines": [ + { + "bbox": [ + 241, + 132, + 370, + 164 + ], + "spans": [ + { + "bbox": [ + 241, + 132, + 370, + 164 + ], + "score": 0.95, + "content": "p ( \\mathbf { y } | \\mathbf { x } , \\theta ) = \\prod _ { t = 1 } ^ { m } p ( \\mathbf { y } _ { t } | \\mathbf { y } _ { < t } , \\mathbf { x } , \\theta )", + "type": "interline_equation", + "image_path": "3d003f060de88a32edea6412d277ef24062aa11f4ba7b6d0a401b82261881f32.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 241, + 132, + 370, + 148.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 241, + 148.0, + 370, + 164.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 504, + 185 + ], + "lines": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "score": 1.0, + "content": "By exploring these probabilities, we design metrics that can gauge the quality of the generated text.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 171, + 505, + 186 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 196, + 181, + 208 + ], + "lines": [ + { + "bbox": [ + 105, + 194, + 182, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 182, + 211 + ], + "score": 1.0, + "content": "3.2 BARTScore", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 216, + 505, + 261 + ], + "lines": [ + { + "bbox": [ + 105, + 216, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 505, + 230 + ], + "score": 1.0, + "content": "The most general form of our proposed BARTSCORE is shown in Eq. 2, where we use the weighted", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 227, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 294, + 240 + ], + "score": 1.0, + "content": "log probability of one text y given another text", + "type": "text" + }, + { + "bbox": [ + 294, + 230, + 301, + 237 + ], + "score": 0.56, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 227, + 506, + 240 + ], + "score": 1.0, + "content": ". The weights are used to put different emphasis on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 237, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 252 + ], + "score": 1.0, + "content": "different tokens, which can be instantiated using different methods like Inverse Document Frequency", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 249, + 343, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 343, + 261 + ], + "score": 1.0, + "content": "(IDF) [25] etc. In our work, we weigh each token equally.2", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 216, + 506, + 261 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 217, + 264, + 393, + 297 + ], + "lines": [ + { + "bbox": [ + 217, + 264, + 393, + 297 + ], + "spans": [ + { + "bbox": [ + 217, + 264, + 393, + 297 + ], + "score": 0.95, + "content": "\\mathbf { B A R T S C O R E } = \\sum _ { t = 1 } ^ { m } \\omega _ { t } \\log p ( \\mathbf { y } _ { t } | \\mathbf { y } _ { < t } , \\mathbf { x } , \\theta )", + "type": "interline_equation", + "image_path": "bfa67b3ee2347ec4fa7a57ebb4b52d83186124f1544c141e530ce54a2d1cb637.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 217, + 264, + 393, + 280.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 217, + 280.5, + 393, + 297.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 299, + 505, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 299, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 506, + 313 + ], + "score": 1.0, + "content": "Due to its generation task-based formulation and ability to utilize the entirety of BART’s pre-trained", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "score": 1.0, + "content": "parameters, BARTSCORE can be flexibly used in different evaluation scenarios. We specifically", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 322, + 499, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 499, + 335 + ], + "score": 1.0, + "content": "present four methods for using BARTSCORE based on different generation directions, which are,", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 299, + 506, + 335 + ] + }, + { + "type": "list", + "bbox": [ + 108, + 342, + 506, + 495 + ], + "lines": [ + { + "bbox": [ + 107, + 341, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 107, + 341, + 169, + 354 + ], + "score": 1.0, + "content": "• Faithfulness", + "type": "text" + }, + { + "bbox": [ + 169, + 342, + 199, + 352 + ], + "score": 0.87, + "content": "s \\to h", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 341, + 349, + 354 + ], + "score": 1.0, + "content": "): from source document to hypothesis", + "type": "text" + }, + { + "bbox": [ + 349, + 342, + 387, + 354 + ], + "score": 0.93, + "content": "p ( \\boldsymbol { h } | \\boldsymbol { s } , \\boldsymbol { \\theta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 341, + 506, + 354 + ], + "score": 1.0, + "content": ". This direction measures how", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 352, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 115, + 352, + 506, + 365 + ], + "score": 1.0, + "content": "likely it is that the hypothesis could be generated based on the source text. Potential application", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 114, + 363, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 114, + 363, + 333, + 375 + ], + "score": 1.0, + "content": "scenarios are factuality and relevance introduced in", + "type": "text" + }, + { + "bbox": [ + 333, + 363, + 352, + 374 + ], + "score": 0.87, + "content": "\\ S 2 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 363, + 506, + 375 + ], + "score": 1.0, + "content": ". This measure can also be used for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 374, + 498, + 387 + ], + "spans": [ + { + "bbox": [ + 115, + 374, + 469, + 387 + ], + "score": 1.0, + "content": "estimating measures of the quality of only the target text, such as coherence and fluency", + "type": "text" + }, + { + "bbox": [ + 470, + 374, + 494, + 385 + ], + "score": 0.32, + "content": "( \\ S 2 . 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 374, + 498, + 387 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 109, + 384, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 109, + 384, + 158, + 398 + ], + "score": 1.0, + "content": "• Precision", + "type": "text" + }, + { + "bbox": [ + 158, + 385, + 189, + 396 + ], + "score": 0.87, + "content": "( r h )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 384, + 371, + 398 + ], + "score": 1.0, + "content": "): from reference text to system-generated text", + "type": "text" + }, + { + "bbox": [ + 371, + 385, + 409, + 397 + ], + "score": 0.94, + "content": "p ( \\boldsymbol { h } | \\boldsymbol { r } , \\boldsymbol { \\theta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 384, + 506, + 398 + ], + "score": 1.0, + "content": ". This direction assesses", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "how likely the hypothesis could be constructed based on the gold reference and is suitable for the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 113, + 407, + 228, + 419 + ], + "spans": [ + { + "bbox": [ + 113, + 407, + 228, + 419 + ], + "score": 1.0, + "content": "precision-focused scenario.", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 417, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 108, + 417, + 147, + 432 + ], + "score": 1.0, + "content": "• Recall", + "type": "text" + }, + { + "bbox": [ + 147, + 418, + 178, + 429 + ], + "score": 0.86, + "content": "( h \\to r", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 417, + 367, + 432 + ], + "score": 1.0, + "content": "): from system-generated text to reference text", + "type": "text" + }, + { + "bbox": [ + 367, + 418, + 405, + 430 + ], + "score": 0.93, + "content": "p ( \\pmb { r } | \\pmb { h } , \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 417, + 506, + 432 + ], + "score": 1.0, + "content": ". This version quantifies", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 115, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "how easily a gold reference could be generated by the hypothesis and is suitable for pyramid-based", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 439, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 115, + 439, + 310, + 452 + ], + "score": 1.0, + "content": "evaluation (i.e., semantic coverage introduced in", + "type": "text" + }, + { + "bbox": [ + 311, + 440, + 330, + 451 + ], + "score": 0.82, + "content": "\\ S 2 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 439, + 506, + 452 + ], + "score": 1.0, + "content": ") in summarization task since pyramid score", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 114, + 451, + 498, + 463 + ], + "spans": [ + { + "bbox": [ + 114, + 451, + 498, + 463 + ], + "score": 1.0, + "content": "measures fine-grained Semantic Content Units (SCUs) [49] covered by system-generated texts.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 109, + 461, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 109, + 461, + 116, + 474 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 117, + 462, + 126, + 472 + ], + "score": 0.78, + "content": "\\mathcal { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 461, + 152, + 474 + ], + "score": 1.0, + "content": "score", + "type": "text" + }, + { + "bbox": [ + 153, + 462, + 183, + 473 + ], + "score": 0.85, + "content": "( r h", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 461, + 506, + 474 + ], + "score": 1.0, + "content": "): Consider both directions and use the arithmetic average of Precision and Recall", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 471, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 115, + 471, + 505, + 486 + ], + "score": 1.0, + "content": "ones. This version can be broadly used to evaluate the semantic overlap (informativeness, adequacy", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 482, + 362, + 496 + ], + "spans": [ + { + "bbox": [ + 115, + 482, + 160, + 496 + ], + "score": 1.0, + "content": "detailed in", + "type": "text" + }, + { + "bbox": [ + 161, + 483, + 181, + 494 + ], + "score": 0.76, + "content": "\\ S 2 . 2 \\AA ,", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 482, + 362, + 496 + ], + "score": 1.0, + "content": ") between reference texts and generated texts.", + "type": "text" + } + ], + "index": 31, + "is_list_end_line": true + } + ], + "index": 24.5, + "bbox_fs": [ + 107, + 341, + 506, + 496 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 507, + 220, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 221, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 221, + 520 + ], + "score": 1.0, + "content": "3.3 BARTScore Variants", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 108, + 527, + 504, + 561 + ], + "lines": [ + { + "bbox": [ + 106, + 527, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 366, + 541 + ], + "score": 1.0, + "content": "We also investigate two extensions to BARTSCORE: (i) changing", + "type": "text" + }, + { + "bbox": [ + 366, + 529, + 374, + 538 + ], + "score": 0.38, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 527, + 391, + 541 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 391, + 529, + 399, + 539 + ], + "score": 0.39, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 527, + 505, + 541 + ], + "score": 1.0, + "content": "through prompting, which", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 400, + 551 + ], + "score": 1.0, + "content": "can bring the evaluation task closer to the pre-training task. (ii) changing", + "type": "text" + }, + { + "bbox": [ + 400, + 539, + 407, + 548 + ], + "score": 0.74, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "by considering different", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 549, + 460, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 460, + 562 + ], + "score": 1.0, + "content": "fine-tuning tasks, which can bring the pre-training domain closer to the evaluation task.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 527, + 506, + 562 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 572, + 171, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 172, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 172, + 587 + ], + "score": 1.0, + "content": "3.3.1 Prompt", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 505, + 635 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 506, + 604 + ], + "score": 1.0, + "content": "Prompting is a practice of adding short phrases to the input or output to encourage pre-trained", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 601, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 506, + 614 + ], + "score": 1.0, + "content": "models to perform specific tasks, which has been proven effective in several other NLP scenarios", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 612, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 505, + 625 + ], + "score": 1.0, + "content": "[24,57,58,60,63]. The generative formulation of BARTSCORE makes it relatively easy to incorporate", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 624, + 402, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 402, + 635 + ], + "score": 1.0, + "content": "these insights here as well; we name this variant BARTSCORE-PROMPT.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 590, + 506, + 635 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 639, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 106, + 639, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 181, + 652 + ], + "score": 1.0, + "content": "Given a prompt of", + "type": "text" + }, + { + "bbox": [ + 182, + 641, + 186, + 649 + ], + "score": 0.53, + "content": "l", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 639, + 215, + 652 + ], + "score": 1.0, + "content": "tokens", + "type": "text" + }, + { + "bbox": [ + 216, + 640, + 285, + 651 + ], + "score": 0.9, + "content": "\\mathbf { z } = \\{ z _ { 1 } , \\cdots , z _ { l } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 639, + 505, + 652 + ], + "score": 1.0, + "content": ", we can either (i) append it to the source text, in which", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 650, + 507, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 157, + 664 + ], + "score": 1.0, + "content": "case we get", + "type": "text" + }, + { + "bbox": [ + 157, + 651, + 282, + 663 + ], + "score": 0.9, + "content": "\\mathbf { x } ^ { \\prime } = \\{ x _ { 1 } , \\cdot \\cdot \\cdot , x _ { n } , \\dot { z } _ { 1 } , \\cdot \\cdot \\cdot , z _ { l } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 650, + 507, + 664 + ], + "score": 1.0, + "content": ", and calculate the score based on this new source text", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 661, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 322, + 675 + ], + "score": 1.0, + "content": "using Eq.2. or (ii) prepend it to the target text, getting", + "type": "text" + }, + { + "bbox": [ + 323, + 662, + 447, + 674 + ], + "score": 0.93, + "content": "\\mathbf { y } ^ { \\prime } = \\{ z _ { 1 } , \\cdot \\cdot \\cdot , z _ { l } , y _ { 1 } , \\cdot \\cdot \\cdot , y _ { m } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 661, + 505, + 675 + ], + "score": 1.0, + "content": ". Then we can", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 672, + 264, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 672, + 264, + 685 + ], + "score": 1.0, + "content": "also use Eq.2 given the new target text.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 639, + 507, + 685 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 72, + 210, + 84 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 212, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 212, + 86 + ], + "score": 1.0, + "content": "3.3.2 Fine-tuning Task", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 91, + 505, + 180 + ], + "lines": [ + { + "bbox": [ + 105, + 91, + 505, + 104 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 505, + 104 + ], + "score": 1.0, + "content": "Different from BERT-based metrics, which typically use classification-based tasks (e.g., natural", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 506, + 115 + ], + "score": 1.0, + "content": "language inference) [67] to fine-tune, BARTSCORE can be fine-tuned using generation-based tasks,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 113, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 505, + 126 + ], + "score": 1.0, + "content": "which will make the pre-training domain closer to the evaluation task. In this paper, we explore two", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "score": 1.0, + "content": "downstream tasks. (1) Summarization. We use BART fine-tuned on CNNDM dataset [20], which is", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "available off-the-shelf in Huggingface Transformers [70]. (2) Paraphrasing. We continue fine-tuning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 146, + 506, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 506, + 158 + ], + "score": 1.0, + "content": "BART from (1) on ParaBank2 dataset [22], which contains a large paraphrase collection. We used", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 156, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 505, + 171 + ], + "score": 1.0, + "content": "a random subset of 30,000 data and fine-tuned for one epoch with a batch size of 20 and a learning", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 167, + 446, + 181 + ], + "spans": [ + { + "bbox": [ + 104, + 167, + 135, + 181 + ], + "score": 1.0, + "content": "rate of", + "type": "text" + }, + { + "bbox": [ + 135, + 168, + 156, + 178 + ], + "score": 0.89, + "content": "5 e ^ { - 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 167, + 446, + 181 + ], + "score": 1.0, + "content": ". We used two 2080Ti GPUs, and the training time is less than one hour.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4.5 + }, + { + "type": "title", + "bbox": [ + 107, + 196, + 186, + 209 + ], + "lines": [ + { + "bbox": [ + 104, + 193, + 189, + 213 + ], + "spans": [ + { + "bbox": [ + 104, + 193, + 189, + 213 + ], + "score": 1.0, + "content": "4 Experiment", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 221, + 504, + 255 + ], + "lines": [ + { + "bbox": [ + 105, + 220, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 235 + ], + "score": 1.0, + "content": "This section aims to evaluate the reliability of different automated metrics, which is commonly", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 233, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 506, + 245 + ], + "score": 1.0, + "content": "achieved by quantifying how well different metrics correlate with human judgments using measures", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 244, + 336, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 336, + 255 + ], + "score": 1.0, + "content": "(e.g., Spearman Correlation [72]) defined below (§4.1.2).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 107, + 269, + 227, + 280 + ], + "lines": [ + { + "bbox": [ + 105, + 269, + 228, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 228, + 282 + ], + "score": 1.0, + "content": "4.1 Baselines and Datasets", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 107, + 289, + 219, + 301 + ], + "lines": [ + { + "bbox": [ + 106, + 289, + 221, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 221, + 303 + ], + "score": 1.0, + "content": "4.1.1 Evaluation Metrics", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 309, + 505, + 363 + ], + "lines": [ + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 310, + 322 + ], + "score": 1.0, + "content": "We comprehensively examine metrics outlined in", + "type": "text" + }, + { + "bbox": [ + 311, + 309, + 329, + 320 + ], + "score": 0.85, + "content": "\\ S 2 . 3", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 308, + 505, + 322 + ], + "score": 1.0, + "content": ", which either require human judgments to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 318, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 506, + 333 + ], + "score": 1.0, + "content": "train (i.e., supervised metrics): COMET [56], BLEURT [62], or are human judgment-free (i.e.,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 330, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 506, + 344 + ], + "score": 1.0, + "content": "unsupervised): BLEU [50] ROUGE-1 and ROUGE-2, ROUGE-L, CHRF [52], PRISM [65],", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 340, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 354 + ], + "score": 1.0, + "content": "MoverScore [76], BERTScore [75]. The detailed comparisons of those metrics can be found in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 352, + 314, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 314, + 365 + ], + "score": 1.0, + "content": "Appendix. We use the official code for each metric.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 108, + 376, + 267, + 388 + ], + "lines": [ + { + "bbox": [ + 106, + 376, + 268, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 268, + 389 + ], + "score": 1.0, + "content": "4.1.2 Measures for Meta Evaluation", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 396, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "Pearson Correlation [15] measures the linear correlation between two sets of data. Spearman", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 407, + 504, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 504, + 419 + ], + "score": 1.0, + "content": "Correlation [72] assesses the monotonic relationships between two variables. Kendall’s Tau [27]", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 417, + 506, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 506, + 431 + ], + "score": 1.0, + "content": "measures the ordinal association between two measured quantities. Accuracy, in our experiments,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "measures the percentage of correct ranking between factual texts and non-factual texts. We follow", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 439, + 470, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 470, + 452 + ], + "score": 1.0, + "content": "previous works in the choices of measures for different datasets to make a fair comparison.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 107, + 464, + 174, + 475 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 175, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 175, + 477 + ], + "score": 1.0, + "content": "4.1.3 Datasets", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 483, + 316, + 516 + ], + "lines": [ + { + "bbox": [ + 106, + 483, + 317, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 317, + 495 + ], + "score": 1.0, + "content": "The datasets we use are summarized in Tab. 1. We", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 493, + 318, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 318, + 507 + ], + "score": 1.0, + "content": "consider three different tasks: summarization (SUM),", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 505, + 309, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 309, + 518 + ], + "score": 1.0, + "content": "machine translation (MT), and data-to-text (D2T).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 521, + 317, + 587 + ], + "lines": [ + { + "bbox": [ + 106, + 521, + 318, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 318, + 533 + ], + "score": 1.0, + "content": "Machine Translation We obtain the source lan-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 318, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 318, + 543 + ], + "score": 1.0, + "content": "guage sentences, machine-translated texts and refer-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 543, + 317, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 317, + 554 + ], + "score": 1.0, + "content": "ence texts from the WMT19 metrics shared task [44].", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 553, + 317, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 317, + 566 + ], + "score": 1.0, + "content": "We use the DARR corpus and consider 7 language", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 565, + 318, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 318, + 577 + ], + "score": 1.0, + "content": "pairs, which are de-en, fi-en, gu-en, kk-en,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 576, + 211, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 211, + 587 + ], + "score": 1.0, + "content": "lt-en, ru-en, zh-en.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 107, + 592, + 317, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 592, + 318, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 318, + 604 + ], + "score": 1.0, + "content": "Text Summarization (1) REALSumm [4] is a meta-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 603, + 318, + 615 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 318, + 615 + ], + "score": 1.0, + "content": "evaluation dataset for text summarization which mea-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 614, + 317, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 317, + 626 + ], + "score": 1.0, + "content": "sures pyramid recall of each system-generated sum-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 625, + 318, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 318, + 637 + ], + "score": 1.0, + "content": "mary. (2) SummEval [13] is a collection of hu-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 636, + 316, + 647 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 316, + 647 + ], + "score": 1.0, + "content": "man judgments of model-generated summaries on", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 646, + 316, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 316, + 659 + ], + "score": 1.0, + "content": "the CNNDM dataset annotated by both expert judges", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 658, + 316, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 316, + 669 + ], + "score": 1.0, + "content": "and crowd-source workers. Each system generated", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 669, + 317, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 317, + 681 + ], + "score": 1.0, + "content": "summary is gauged through the lens of coherence,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 680, + 316, + 692 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 316, + 692 + ], + "score": 1.0, + "content": "consistency, fluency and relevance.3 (3) NeR18 The", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 40 + }, + { + "type": "table", + "bbox": [ + 324, + 534, + 508, + 687 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 324, + 483, + 505, + 528 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 324, + 484, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 324, + 484, + 505, + 495 + ], + "score": 1.0, + "content": "Table 1: A summary of tasks, datasets, and", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 324, + 495, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 324, + 495, + 505, + 506 + ], + "score": 1.0, + "content": "evaluation perspectives that we have covered", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 323, + 506, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 323, + 506, + 505, + 517 + ], + "score": 1.0, + "content": "in our experiments. Explanation of evaluation", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 322, + 516, + 462, + 528 + ], + "spans": [ + { + "bbox": [ + 322, + 517, + 439, + 528 + ], + "score": 1.0, + "content": "perspectives can be found in", + "type": "text" + }, + { + "bbox": [ + 439, + 516, + 458, + 527 + ], + "score": 0.85, + "content": "\\ S 2 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 517, + 462, + 528 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 46.5 + }, + { + "type": "table_body", + "bbox": [ + 324, + 534, + 508, + 687 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 324, + 534, + 508, + 687 + ], + "spans": [ + { + "bbox": [ + 324, + 534, + 508, + 687 + ], + "score": 0.98, + "html": "
TasksDatasetsEval. Perspectives
SUMREALSUMCov
SummEvalCOH FAC FLU INFO
NeR18COH FLU REL INFO
Rank19 QAGS-C QAGS-XFAC
DE FI GU KK IT RU ZHADE FLU
MT D2TBAGEL SFHOT SFRESINFO
", + "type": "table", + "image_path": "2e7285a37838112a4182c3dc65fb5f53419dcbe2dd78fa7f3712dfe6244fe154.jpg" + } + ] + } + ], + "index": 54.5, + "virtual_lines": [ + { + "bbox": [ + 324, + 534, + 508, + 546.75 + ], + "spans": [], + "index": 49 + }, + { + "bbox": [ + 324, + 546.75, + 508, + 559.5 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 324, + 559.5, + 508, + 572.25 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 324, + 572.25, + 508, + 585.0 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 324, + 585.0, + 508, + 597.75 + ], + "spans": [], + "index": 53 + }, + { + "bbox": [ + 324, + 597.75, + 508, + 610.5 + ], + "spans": [], + "index": 54 + }, + { + "bbox": [ + 324, + 610.5, + 508, + 623.25 + ], + "spans": [], + "index": 55 + }, + { + "bbox": [ + 324, + 623.25, + 508, + 636.0 + ], + "spans": [], + "index": 56 + }, + { + "bbox": [ + 324, + 636.0, + 508, + 648.75 + ], + "spans": [], + "index": 57 + }, + { + "bbox": [ + 324, + 648.75, + 508, + 661.5 + ], + "spans": [], + "index": 58 + }, + { + "bbox": [ + 324, + 661.5, + 508, + 674.25 + ], + "spans": [], + "index": 59 + }, + { + "bbox": [ + 324, + 674.25, + 508, + 687.0 + ], + "spans": [], + "index": 60 + } + ] + } + ], + "index": 50.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 105, + 701, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 118, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 118, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "3We rephrase the original “relevance” into “informativeness” and “consistency” into “factuality” based on", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 712, + 314, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 712, + 294, + 723 + ], + "score": 1.0, + "content": "the descriptions in their paper and our definitions in", + "type": "text" + }, + { + "bbox": [ + 294, + 712, + 311, + 722 + ], + "score": 0.82, + "content": "\\ S 2 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 712, + 314, + 723 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 72, + 210, + 84 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 212, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 212, + 86 + ], + "score": 1.0, + "content": "3.3.2 Fine-tuning Task", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 91, + 505, + 180 + ], + "lines": [ + { + "bbox": [ + 105, + 91, + 505, + 104 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 505, + 104 + ], + "score": 1.0, + "content": "Different from BERT-based metrics, which typically use classification-based tasks (e.g., natural", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 506, + 115 + ], + "score": 1.0, + "content": "language inference) [67] to fine-tune, BARTSCORE can be fine-tuned using generation-based tasks,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 113, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 505, + 126 + ], + "score": 1.0, + "content": "which will make the pre-training domain closer to the evaluation task. In this paper, we explore two", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "score": 1.0, + "content": "downstream tasks. (1) Summarization. We use BART fine-tuned on CNNDM dataset [20], which is", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "available off-the-shelf in Huggingface Transformers [70]. (2) Paraphrasing. We continue fine-tuning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 146, + 506, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 506, + 158 + ], + "score": 1.0, + "content": "BART from (1) on ParaBank2 dataset [22], which contains a large paraphrase collection. We used", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 156, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 505, + 171 + ], + "score": 1.0, + "content": "a random subset of 30,000 data and fine-tuned for one epoch with a batch size of 20 and a learning", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 167, + 446, + 181 + ], + "spans": [ + { + "bbox": [ + 104, + 167, + 135, + 181 + ], + "score": 1.0, + "content": "rate of", + "type": "text" + }, + { + "bbox": [ + 135, + 168, + 156, + 178 + ], + "score": 0.89, + "content": "5 e ^ { - 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 167, + 446, + 181 + ], + "score": 1.0, + "content": ". We used two 2080Ti GPUs, and the training time is less than one hour.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4.5, + "bbox_fs": [ + 104, + 91, + 506, + 181 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 196, + 186, + 209 + ], + "lines": [ + { + "bbox": [ + 104, + 193, + 189, + 213 + ], + "spans": [ + { + "bbox": [ + 104, + 193, + 189, + 213 + ], + "score": 1.0, + "content": "4 Experiment", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 221, + 504, + 255 + ], + "lines": [ + { + "bbox": [ + 105, + 220, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 235 + ], + "score": 1.0, + "content": "This section aims to evaluate the reliability of different automated metrics, which is commonly", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 233, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 506, + 245 + ], + "score": 1.0, + "content": "achieved by quantifying how well different metrics correlate with human judgments using measures", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 244, + 336, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 336, + 255 + ], + "score": 1.0, + "content": "(e.g., Spearman Correlation [72]) defined below (§4.1.2).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 220, + 506, + 255 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 269, + 227, + 280 + ], + "lines": [ + { + "bbox": [ + 105, + 269, + 228, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 228, + 282 + ], + "score": 1.0, + "content": "4.1 Baselines and Datasets", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 107, + 289, + 219, + 301 + ], + "lines": [ + { + "bbox": [ + 106, + 289, + 221, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 221, + 303 + ], + "score": 1.0, + "content": "4.1.1 Evaluation Metrics", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 309, + 505, + 363 + ], + "lines": [ + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 310, + 322 + ], + "score": 1.0, + "content": "We comprehensively examine metrics outlined in", + "type": "text" + }, + { + "bbox": [ + 311, + 309, + 329, + 320 + ], + "score": 0.85, + "content": "\\ S 2 . 3", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 308, + 505, + 322 + ], + "score": 1.0, + "content": ", which either require human judgments to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 318, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 506, + 333 + ], + "score": 1.0, + "content": "train (i.e., supervised metrics): COMET [56], BLEURT [62], or are human judgment-free (i.e.,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 330, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 506, + 344 + ], + "score": 1.0, + "content": "unsupervised): BLEU [50] ROUGE-1 and ROUGE-2, ROUGE-L, CHRF [52], PRISM [65],", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 340, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 354 + ], + "score": 1.0, + "content": "MoverScore [76], BERTScore [75]. The detailed comparisons of those metrics can be found in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 352, + 314, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 314, + 365 + ], + "score": 1.0, + "content": "Appendix. We use the official code for each metric.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 308, + 506, + 365 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 376, + 267, + 388 + ], + "lines": [ + { + "bbox": [ + 106, + 376, + 268, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 268, + 389 + ], + "score": 1.0, + "content": "4.1.2 Measures for Meta Evaluation", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 396, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "Pearson Correlation [15] measures the linear correlation between two sets of data. Spearman", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 407, + 504, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 504, + 419 + ], + "score": 1.0, + "content": "Correlation [72] assesses the monotonic relationships between two variables. Kendall’s Tau [27]", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 417, + 506, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 506, + 431 + ], + "score": 1.0, + "content": "measures the ordinal association between two measured quantities. Accuracy, in our experiments,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "measures the percentage of correct ranking between factual texts and non-factual texts. We follow", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 439, + 470, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 470, + 452 + ], + "score": 1.0, + "content": "previous works in the choices of measures for different datasets to make a fair comparison.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 396, + 506, + 452 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 464, + 174, + 475 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 175, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 175, + 477 + ], + "score": 1.0, + "content": "4.1.3 Datasets", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 483, + 316, + 516 + ], + "lines": [ + { + "bbox": [ + 106, + 483, + 317, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 317, + 495 + ], + "score": 1.0, + "content": "The datasets we use are summarized in Tab. 1. We", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 493, + 318, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 318, + 507 + ], + "score": 1.0, + "content": "consider three different tasks: summarization (SUM),", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 505, + 309, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 309, + 518 + ], + "score": 1.0, + "content": "machine translation (MT), and data-to-text (D2T).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 483, + 318, + 518 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 521, + 317, + 587 + ], + "lines": [ + { + "bbox": [ + 106, + 521, + 318, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 318, + 533 + ], + "score": 1.0, + "content": "Machine Translation We obtain the source lan-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 318, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 318, + 543 + ], + "score": 1.0, + "content": "guage sentences, machine-translated texts and refer-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 543, + 317, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 317, + 554 + ], + "score": 1.0, + "content": "ence texts from the WMT19 metrics shared task [44].", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 553, + 317, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 317, + 566 + ], + "score": 1.0, + "content": "We use the DARR corpus and consider 7 language", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 565, + 318, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 318, + 577 + ], + "score": 1.0, + "content": "pairs, which are de-en, fi-en, gu-en, kk-en,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 576, + 211, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 211, + 587 + ], + "score": 1.0, + "content": "lt-en, ru-en, zh-en.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 521, + 318, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 592, + 317, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 592, + 318, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 318, + 604 + ], + "score": 1.0, + "content": "Text Summarization (1) REALSumm [4] is a meta-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 603, + 318, + 615 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 318, + 615 + ], + "score": 1.0, + "content": "evaluation dataset for text summarization which mea-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 614, + 317, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 317, + 626 + ], + "score": 1.0, + "content": "sures pyramid recall of each system-generated sum-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 625, + 318, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 318, + 637 + ], + "score": 1.0, + "content": "mary. (2) SummEval [13] is a collection of hu-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 636, + 316, + 647 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 316, + 647 + ], + "score": 1.0, + "content": "man judgments of model-generated summaries on", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 646, + 316, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 316, + 659 + ], + "score": 1.0, + "content": "the CNNDM dataset annotated by both expert judges", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 658, + 316, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 316, + 669 + ], + "score": 1.0, + "content": "and crowd-source workers. Each system generated", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 669, + 317, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 317, + 681 + ], + "score": 1.0, + "content": "summary is gauged through the lens of coherence,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 680, + 316, + 692 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 316, + 692 + ], + "score": 1.0, + "content": "consistency, fluency and relevance.3 (3) NeR18 The", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 592, + 318, + 692 + ] + }, + { + "type": "table", + "bbox": [ + 324, + 534, + 508, + 687 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 324, + 483, + 505, + 528 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 324, + 484, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 324, + 484, + 505, + 495 + ], + "score": 1.0, + "content": "Table 1: A summary of tasks, datasets, and", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 324, + 495, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 324, + 495, + 505, + 506 + ], + "score": 1.0, + "content": "evaluation perspectives that we have covered", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 323, + 506, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 323, + 506, + 505, + 517 + ], + "score": 1.0, + "content": "in our experiments. Explanation of evaluation", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 322, + 516, + 462, + 528 + ], + "spans": [ + { + "bbox": [ + 322, + 517, + 439, + 528 + ], + "score": 1.0, + "content": "perspectives can be found in", + "type": "text" + }, + { + "bbox": [ + 439, + 516, + 458, + 527 + ], + "score": 0.85, + "content": "\\ S 2 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 517, + 462, + 528 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 46.5 + }, + { + "type": "table_body", + "bbox": [ + 324, + 534, + 508, + 687 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 324, + 534, + 508, + 687 + ], + "spans": [ + { + "bbox": [ + 324, + 534, + 508, + 687 + ], + "score": 0.98, + "html": "
TasksDatasetsEval. Perspectives
SUMREALSUMCov
SummEvalCOH FAC FLU INFO
NeR18COH FLU REL INFO
Rank19 QAGS-C QAGS-XFAC
DE FI GU KK IT RU ZHADE FLU
MT D2TBAGEL SFHOT SFRESINFO
", + "type": "table", + "image_path": "2e7285a37838112a4182c3dc65fb5f53419dcbe2dd78fa7f3712dfe6244fe154.jpg" + } + ] + } + ], + "index": 54.5, + "virtual_lines": [ + { + "bbox": [ + 324, + 534, + 508, + 546.75 + ], + "spans": [], + "index": 49 + }, + { + "bbox": [ + 324, + 546.75, + 508, + 559.5 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 324, + 559.5, + 508, + 572.25 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 324, + 572.25, + 508, + 585.0 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 324, + 585.0, + 508, + 597.75 + ], + "spans": [], + "index": 53 + }, + { + "bbox": [ + 324, + 597.75, + 508, + 610.5 + ], + "spans": [], + "index": 54 + }, + { + "bbox": [ + 324, + 610.5, + 508, + 623.25 + ], + "spans": [], + "index": 55 + }, + { + "bbox": [ + 324, + 623.25, + 508, + 636.0 + ], + "spans": [], + "index": 56 + }, + { + "bbox": [ + 324, + 636.0, + 508, + 648.75 + ], + "spans": [], + "index": 57 + }, + { + "bbox": [ + 324, + 648.75, + 508, + 661.5 + ], + "spans": [], + "index": 58 + }, + { + "bbox": [ + 324, + 661.5, + 508, + 674.25 + ], + "spans": [], + "index": 59 + }, + { + "bbox": [ + 324, + 674.25, + 508, + 687.0 + ], + "spans": [], + "index": 60 + } + ] + } + ], + "index": 50.5 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 504, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "NEWSROOM dataset [18] contains 60 articles with summaries generated by 7 different methods are", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 461, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 461, + 96 + ], + "score": 1.0, + "content": "annotated with human scores in terms of coherence, fluency, informativeness, relevance.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 161 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "Factuality (1) Rank19 [14] is used to meta-evaluate factuality metrics. It is a collection of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 116, + 506, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 506, + 130 + ], + "score": 1.0, + "content": "373 triples of a source sentence with two summary sentences, one correct and one incorrect. (2)", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 506, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 506, + 141 + ], + "score": 1.0, + "content": "QAGS20 [66] collected 235 test outputs on CNNDM dataset from [16] and 239 test outputs on", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "score": 1.0, + "content": "XSUM dataset [47] from BART fine-tuned on XSUM. Sentences in each summary are annotated with", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 150, + 245, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 245, + 162 + ], + "score": 1.0, + "content": "correctness scores w.r.t. factuality.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 166, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 165, + 505, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 505, + 179 + ], + "score": 1.0, + "content": "Data to Text We consider the following datasets which target utterance generation for spoken", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 505, + 190 + ], + "score": 1.0, + "content": "dialogue systems. (1) BAGEL [45] provides information about restaurants. (2) SFHOT [69] provides", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 188, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 188, + 505, + 201 + ], + "score": 1.0, + "content": "information about hotels in San Francisco. (3) SFRES [69] provides information about restaurants in", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 199, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 211 + ], + "score": 1.0, + "content": "San Francisco. They contain 202, 398, and 581 samples respectively, each sample consists of one", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 473, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 473, + 223 + ], + "score": 1.0, + "content": "meaning representation, multiple references, and utterances generated by different systems.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 107, + 233, + 155, + 245 + ], + "lines": [ + { + "bbox": [ + 105, + 230, + 157, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 157, + 248 + ], + "score": 1.0, + "content": "4.2 Setup", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 107, + 253, + 202, + 265 + ], + "lines": [ + { + "bbox": [ + 105, + 251, + 203, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 203, + 268 + ], + "score": 1.0, + "content": "4.2.1 Prompt Design", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 272, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 272, + 506, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 506, + 285 + ], + "score": 1.0, + "content": "To perform prompting, we first need to find proper prompts within a search space. Instead of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 282, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 280, + 296 + ], + "score": 1.0, + "content": "considering a large discrete search space", + "type": "text" + }, + { + "bbox": [ + 280, + 283, + 302, + 294 + ], + "score": 0.58, + "content": "[ 6 3 ] ^ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 282, + 505, + 296 + ], + "score": 1.0, + "content": "or continuous search space [34], we use simple", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 293, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 506, + 306 + ], + "score": 1.0, + "content": "heuristics to narrow our search space. In particular, we use manually devised seed prompts and gather", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 303, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 506, + 318 + ], + "score": 1.0, + "content": "paraphrases to construct our prompt set.5 The seed prompts and some examples of paraphrased", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 316, + 363, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 363, + 329 + ], + "score": 1.0, + "content": "prompts are shown in Tab. 2. Details are listed in the Appendix.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16 + }, + { + "type": "table", + "bbox": [ + 115, + 365, + 495, + 408 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 336, + 502, + 358 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 334, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 505, + 349 + ], + "score": 1.0, + "content": "Table 2: Seed prompts and examples of final prompts. “Number” denotes the size of our final prompt", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 347, + 286, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 286, + 360 + ], + "score": 1.0, + "content": "set that was acquired from the seed prompts.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "table_body", + "bbox": [ + 115, + 365, + 495, + 408 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 365, + 495, + 408 + ], + "spans": [ + { + "bbox": [ + 115, + 365, + 495, + 408 + ], + "score": 0.977, + "html": "
UsageNumberSeedExample
s→h70in summaryin short，inword， atosum up
h←r34inother wordstorephrase it，that istosay，i.e.
", + "type": "table", + "image_path": "afdc30151920352456f85417d4afcaf483115183510ae8210ad965e53ccb985e.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 115, + 365, + 495, + 379.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 115, + 379.3333333333333, + 495, + 393.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 115, + 393.66666666666663, + 495, + 407.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 20.75 + }, + { + "type": "title", + "bbox": [ + 107, + 423, + 172, + 435 + ], + "lines": [ + { + "bbox": [ + 105, + 421, + 173, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 173, + 438 + ], + "score": 1.0, + "content": "4.2.2 Settings", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 442, + 504, + 486 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 454 + ], + "score": 1.0, + "content": "Variants. We consider four variants of BARTSCORE, which are (1) BARTSCORE, which uses the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 454, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 506, + 465 + ], + "score": 1.0, + "content": "vanilla BART; (2) BARTSCORE-CNN, which uses the BART fine-tuned on the summarization dataset", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 465, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 506, + 476 + ], + "score": 1.0, + "content": "CNNDM; (3) BARTSCORE-CNN-PARA, where BART is first fine-tuned on CNNDM, then fine-tuned", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 474, + 449, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 449, + 488 + ], + "score": 1.0, + "content": "on ParaBank2. (4) BARTSCORE-PROMPT, which is enhanced by adding prompts.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "Selection of Prompts. For the summarization and data-to-text tasks, we use all entries (either all", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 197, + 515 + ], + "score": 1.0, + "content": "prompts designed for", + "type": "text" + }, + { + "bbox": [ + 197, + 502, + 228, + 513 + ], + "score": 0.9, + "content": "s h", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 502, + 345, + 515 + ], + "score": 1.0, + "content": "or all prompts designed for", + "type": "text" + }, + { + "bbox": [ + 346, + 502, + 377, + 513 + ], + "score": 0.9, + "content": "h r", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 502, + 505, + 515 + ], + "score": 1.0, + "content": "depending on the BARTScore", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 513, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 505, + 526 + ], + "score": 1.0, + "content": "usage chosen) in the prompt set by prefixing the decoder input and getting different generation scores", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "(calculated by Eq.2) for each hypothesis based on different prompts. We finally get the score for one", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 535, + 506, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 506, + 548 + ], + "score": 1.0, + "content": "hypothesis by taking the average of all its generation scores using different prompts ( [24]; details", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 546, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 506, + 559 + ], + "score": 1.0, + "content": "about prompt ensembling can be found in the Appendix). For the machine translation task, due to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "score": 1.0, + "content": "the more expensive computational cost brought by larger text sets, we first use WMT18 [43] as a", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "score": 1.0, + "content": "development set to search for one best prompt and obtain the phrase “Such as”, which is then used", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 578, + 214, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 214, + 592 + ], + "score": 1.0, + "content": "for the test language pairs.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 608 + ], + "score": 1.0, + "content": "Selection of BARTScore Usage. Although BARTSCORE can be used in different ways (shown", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 605, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 117, + 619 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 117, + 606, + 139, + 617 + ], + "score": 0.77, + "content": "\\ S 3 . 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 605, + 506, + 619 + ], + "score": 1.0, + "content": "), in different tasks, they can be chosen based on how targeted evaluation perspectives are", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 616, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 190, + 630 + ], + "score": 1.0, + "content": "defined (described in", + "type": "text" + }, + { + "bbox": [ + 190, + 617, + 209, + 628 + ], + "score": 0.82, + "content": "\\ S 2 . 2 \\AA", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 616, + 506, + 630 + ], + "score": 1.0, + "content": ") as well as the types of tasks. Specifically, (i) For those datasets whose gold", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 628, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 506, + 641 + ], + "score": 1.0, + "content": "standard human evaluation are obtained based on recall-based pyramid method, we adopt recall-based", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 638, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 168, + 652 + ], + "score": 1.0, + "content": "BARTSCORE", + "type": "text" + }, + { + "bbox": [ + 168, + 639, + 200, + 650 + ], + "score": 0.86, + "content": "( h \\to r", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 638, + 505, + 652 + ], + "score": 1.0, + "content": "). (ii) For those datasets whose human judgments focus on linguistic quality", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "(coherence, fluency) and factual correctness (factuality), or the source and hypothesis texts are in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 660, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 427, + 674 + ], + "score": 1.0, + "content": "the same modality (i.e., language), we use faithfulness-based BARTSCORE", + "type": "text" + }, + { + "bbox": [ + 427, + 661, + 461, + 671 + ], + "score": 0.85, + "content": "s h", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 660, + 505, + 674 + ], + "score": 1.0, + "content": "). (iii) For", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 672, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 672, + 505, + 684 + ], + "score": 1.0, + "content": "data-to-text and machine translation tasks, to make a fair comparison, we use BARTSCORE with the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 682, + 478, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 478, + 695 + ], + "score": 1.0, + "content": "F-score version that other existing works [65] have adopted when evaluating generated texts.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 42 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 700, + 424, + 722 + ], + "lines": [ + { + "bbox": [ + 118, + 698, + 425, + 713 + ], + "spans": [ + { + "bbox": [ + 118, + 698, + 425, + 713 + ], + "score": 1.0, + "content": "4We explored this first and found that discovered prompts led to worse performance.", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 709, + 424, + 725 + ], + "spans": [ + { + "bbox": [ + 118, + 709, + 424, + 725 + ], + "score": 1.0, + "content": "5We use the website https://www.wordhippo.com/ to search for synonyms.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 504, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "NEWSROOM dataset [18] contains 60 articles with summaries generated by 7 different methods are", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 461, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 461, + 96 + ], + "score": 1.0, + "content": "annotated with human scores in terms of coherence, fluency, informativeness, relevance.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 72, + 505, + 96 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 161 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "Factuality (1) Rank19 [14] is used to meta-evaluate factuality metrics. It is a collection of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 116, + 506, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 506, + 130 + ], + "score": 1.0, + "content": "373 triples of a source sentence with two summary sentences, one correct and one incorrect. (2)", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 506, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 506, + 141 + ], + "score": 1.0, + "content": "QAGS20 [66] collected 235 test outputs on CNNDM dataset from [16] and 239 test outputs on", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "score": 1.0, + "content": "XSUM dataset [47] from BART fine-tuned on XSUM. Sentences in each summary are annotated with", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 150, + 245, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 245, + 162 + ], + "score": 1.0, + "content": "correctness scores w.r.t. factuality.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 105, + 506, + 162 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 166, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 165, + 505, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 505, + 179 + ], + "score": 1.0, + "content": "Data to Text We consider the following datasets which target utterance generation for spoken", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 505, + 190 + ], + "score": 1.0, + "content": "dialogue systems. (1) BAGEL [45] provides information about restaurants. (2) SFHOT [69] provides", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 188, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 188, + 505, + 201 + ], + "score": 1.0, + "content": "information about hotels in San Francisco. (3) SFRES [69] provides information about restaurants in", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 199, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 211 + ], + "score": 1.0, + "content": "San Francisco. They contain 202, 398, and 581 samples respectively, each sample consists of one", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 473, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 473, + 223 + ], + "score": 1.0, + "content": "meaning representation, multiple references, and utterances generated by different systems.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 165, + 505, + 223 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 233, + 155, + 245 + ], + "lines": [ + { + "bbox": [ + 105, + 230, + 157, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 157, + 248 + ], + "score": 1.0, + "content": "4.2 Setup", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 107, + 253, + 202, + 265 + ], + "lines": [ + { + "bbox": [ + 105, + 251, + 203, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 203, + 268 + ], + "score": 1.0, + "content": "4.2.1 Prompt Design", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 272, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 272, + 506, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 506, + 285 + ], + "score": 1.0, + "content": "To perform prompting, we first need to find proper prompts within a search space. Instead of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 282, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 280, + 296 + ], + "score": 1.0, + "content": "considering a large discrete search space", + "type": "text" + }, + { + "bbox": [ + 280, + 283, + 302, + 294 + ], + "score": 0.58, + "content": "[ 6 3 ] ^ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 282, + 505, + 296 + ], + "score": 1.0, + "content": "or continuous search space [34], we use simple", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 293, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 506, + 306 + ], + "score": 1.0, + "content": "heuristics to narrow our search space. In particular, we use manually devised seed prompts and gather", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 303, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 506, + 318 + ], + "score": 1.0, + "content": "paraphrases to construct our prompt set.5 The seed prompts and some examples of paraphrased", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 316, + 363, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 363, + 329 + ], + "score": 1.0, + "content": "prompts are shown in Tab. 2. Details are listed in the Appendix.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 272, + 506, + 329 + ] + }, + { + "type": "table", + "bbox": [ + 115, + 365, + 495, + 408 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 336, + 502, + 358 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 334, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 505, + 349 + ], + "score": 1.0, + "content": "Table 2: Seed prompts and examples of final prompts. “Number” denotes the size of our final prompt", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 347, + 286, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 286, + 360 + ], + "score": 1.0, + "content": "set that was acquired from the seed prompts.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "table_body", + "bbox": [ + 115, + 365, + 495, + 408 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 365, + 495, + 408 + ], + "spans": [ + { + "bbox": [ + 115, + 365, + 495, + 408 + ], + "score": 0.977, + "html": "
UsageNumberSeedExample
s→h70in summaryin short，inword， atosum up
h←r34inother wordstorephrase it，that istosay，i.e.
", + "type": "table", + "image_path": "afdc30151920352456f85417d4afcaf483115183510ae8210ad965e53ccb985e.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 115, + 365, + 495, + 379.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 115, + 379.3333333333333, + 495, + 393.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 115, + 393.66666666666663, + 495, + 407.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 20.75 + }, + { + "type": "title", + "bbox": [ + 107, + 423, + 172, + 435 + ], + "lines": [ + { + "bbox": [ + 105, + 421, + 173, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 173, + 438 + ], + "score": 1.0, + "content": "4.2.2 Settings", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 442, + 504, + 486 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 454 + ], + "score": 1.0, + "content": "Variants. We consider four variants of BARTSCORE, which are (1) BARTSCORE, which uses the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 454, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 506, + 465 + ], + "score": 1.0, + "content": "vanilla BART; (2) BARTSCORE-CNN, which uses the BART fine-tuned on the summarization dataset", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 465, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 506, + 476 + ], + "score": 1.0, + "content": "CNNDM; (3) BARTSCORE-CNN-PARA, where BART is first fine-tuned on CNNDM, then fine-tuned", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 474, + 449, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 449, + 488 + ], + "score": 1.0, + "content": "on ParaBank2. (4) BARTSCORE-PROMPT, which is enhanced by adding prompts.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 443, + 506, + 488 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "Selection of Prompts. For the summarization and data-to-text tasks, we use all entries (either all", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 197, + 515 + ], + "score": 1.0, + "content": "prompts designed for", + "type": "text" + }, + { + "bbox": [ + 197, + 502, + 228, + 513 + ], + "score": 0.9, + "content": "s h", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 502, + 345, + 515 + ], + "score": 1.0, + "content": "or all prompts designed for", + "type": "text" + }, + { + "bbox": [ + 346, + 502, + 377, + 513 + ], + "score": 0.9, + "content": "h r", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 502, + 505, + 515 + ], + "score": 1.0, + "content": "depending on the BARTScore", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 513, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 505, + 526 + ], + "score": 1.0, + "content": "usage chosen) in the prompt set by prefixing the decoder input and getting different generation scores", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "(calculated by Eq.2) for each hypothesis based on different prompts. We finally get the score for one", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 535, + 506, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 506, + 548 + ], + "score": 1.0, + "content": "hypothesis by taking the average of all its generation scores using different prompts ( [24]; details", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 546, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 506, + 559 + ], + "score": 1.0, + "content": "about prompt ensembling can be found in the Appendix). For the machine translation task, due to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "score": 1.0, + "content": "the more expensive computational cost brought by larger text sets, we first use WMT18 [43] as a", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "score": 1.0, + "content": "development set to search for one best prompt and obtain the phrase “Such as”, which is then used", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 578, + 214, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 214, + 592 + ], + "score": 1.0, + "content": "for the test language pairs.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 492, + 506, + 592 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 608 + ], + "score": 1.0, + "content": "Selection of BARTScore Usage. Although BARTSCORE can be used in different ways (shown", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 605, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 117, + 619 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 117, + 606, + 139, + 617 + ], + "score": 0.77, + "content": "\\ S 3 . 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 605, + 506, + 619 + ], + "score": 1.0, + "content": "), in different tasks, they can be chosen based on how targeted evaluation perspectives are", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 616, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 190, + 630 + ], + "score": 1.0, + "content": "defined (described in", + "type": "text" + }, + { + "bbox": [ + 190, + 617, + 209, + 628 + ], + "score": 0.82, + "content": "\\ S 2 . 2 \\AA", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 616, + 506, + 630 + ], + "score": 1.0, + "content": ") as well as the types of tasks. Specifically, (i) For those datasets whose gold", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 628, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 506, + 641 + ], + "score": 1.0, + "content": "standard human evaluation are obtained based on recall-based pyramid method, we adopt recall-based", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 638, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 168, + 652 + ], + "score": 1.0, + "content": "BARTSCORE", + "type": "text" + }, + { + "bbox": [ + 168, + 639, + 200, + 650 + ], + "score": 0.86, + "content": "( h \\to r", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 638, + 505, + 652 + ], + "score": 1.0, + "content": "). (ii) For those datasets whose human judgments focus on linguistic quality", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "(coherence, fluency) and factual correctness (factuality), or the source and hypothesis texts are in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 660, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 427, + 674 + ], + "score": 1.0, + "content": "the same modality (i.e., language), we use faithfulness-based BARTSCORE", + "type": "text" + }, + { + "bbox": [ + 427, + 661, + 461, + 671 + ], + "score": 0.85, + "content": "s h", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 660, + 505, + 674 + ], + "score": 1.0, + "content": "). (iii) For", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 672, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 672, + 505, + 684 + ], + "score": 1.0, + "content": "data-to-text and machine translation tasks, to make a fair comparison, we use BARTSCORE with the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 682, + 478, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 478, + 695 + ], + "score": 1.0, + "content": "F-score version that other existing works [65] have adopted when evaluating generated texts.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 594, + 506, + 695 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 112, + 109, + 498, + 274 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 70, + 505, + 104 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 70, + 505, + 82 + ], + "spans": [ + { + "bbox": [ + 105, + 70, + 505, + 82 + ], + "score": 1.0, + "content": "Table 3: Kendall’s Tau correlation of different metrics on WMT19 dataset. The highest correlation", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 81, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 93 + ], + "score": 1.0, + "content": "for each language pair achieved by unsupervised method is bold, and the highest correlation overall", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 92, + 499, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 499, + 105 + ], + "score": 1.0, + "content": "is underlined. Avg. denotes the average correlation achieved by a metric across all language pairs.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 112, + 109, + 498, + 274 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 109, + 498, + 274 + ], + "spans": [ + { + "bbox": [ + 112, + 109, + 498, + 274 + ], + "score": 0.983, + "html": "
de-enfi-en gu-enkk-enlt-enru-enzh-enAvg.
SUPERVISED METHODS
BLEURT0.1740.3740.3130.3720.3880.2200.4360.325
COMET0.2190.3690.3160.3780.4050.2260.4620.339
UNSUPERVISED METHODS
BLEU0.0540.2360.1940.2760.2490.1150.3210.206
CHRF0.1230.2920.2400.3230.3040.1770.3710.261
PRISM0.1990.3660.3200.3620.3820.2200.4340.326
BERTScore0.1900.3540.2920.3510.3810.2210.4300.317
BARTSCORE0.1560.3350.2730.3240.3220.1670.3890.281
+CNN0.1900.3650.3000.3480.3840.2080.4250.317
+ CNN+Para0.205t0.370t0.3160.378t0.386†0.219_0.442t0.331
+ CNN +Para +Prompt0.2380.3740.3180.376t0.386+0.2190.4470.337
", + "type": "table", + "image_path": "10078ed3879a01510492b203ac9ef173d092fe67d6150bd64664257da154ae67.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 112, + 109, + 498, + 164.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 112, + 164.0, + 498, + 219.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 112, + 219.0, + 498, + 274.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "table", + "bbox": [ + 109, + 330, + 502, + 474 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 291, + 505, + 325 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "score": 1.0, + "content": "Table 4: Spearman correlation of different metrics on three human judgement datasets. For prompt-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 303, + 507, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 427, + 315 + ], + "score": 1.0, + "content": "based learning, we consider adding prompts to the best-performing BARTSCORE", + "type": "text" + }, + { + "bbox": [ + 427, + 303, + 442, + 313 + ], + "score": 0.64, + "content": "( \\Omega )", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 303, + 507, + 315 + ], + "score": 1.0, + "content": "on each dataset.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 313, + 388, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 388, + 325 + ], + "score": 1.0, + "content": "The highest correlation overall for each aspect on each dataset is bold.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "table_body", + "bbox": [ + 109, + 330, + 502, + 474 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 109, + 330, + 502, + 474 + ], + "spans": [ + { + "bbox": [ + 109, + 330, + 502, + 474 + ], + "score": 0.985, + "html": "
REALSummSummEvalNeR18Avg.
CovCOHFACFLUINFOCOHFLUINFOREL
ROUGE-10.4980.1670.1600.1150.3260.0950.1040.1300.1470.194
ROUGE-20.4230.1840.1870.1590.2900.0260.0480.0790.0910.165
ROUGE-L0.4880.1280.1150.1050.3110.0640.0720.0890.1060.164
BERTScore0.4400.2840.1100.1930.3120.1470.1700.1310.1630.217
MoverScore0.3720.1590.1570.1290.3180.1610.1200.1880.1950.200
PRISM0.4110.2490.3450.2540.2120.5730.5320.5610.5530.410
BARTSCORE0.4410.322†0.3110.2480.2640.679†0.670t0.646†0.604t0.465
+ CNN0.4750.448‡0.382†0.356t0.356t0.653†0.640t0.616†0.5670.499
+ CNN+Para0.4710.424†0.401‡0.378t0.3130.657t0.652t0.614†0.5620.497
+Ω+Prompt0.4880.4070.3780.338f0.368f0.701t0.679t0.686t0.6200.518
", + "type": "table", + "image_path": "975ee48dd3e22e76b11d2026accb30e8c6cebbef0d1d3bce2e29af12c12df5c3.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 109, + 330, + 502, + 378.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 109, + 378.0, + 502, + 426.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 109, + 426.0, + 502, + 474.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 106, + 502, + 505, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 496, + 515 + ], + "score": 1.0, + "content": "Significance Tests. To perform rigorous analysis, we adopt the bootstrapping method (p-value", + "type": "text" + }, + { + "bbox": [ + 497, + 504, + 506, + 513 + ], + "score": 0.59, + "content": "<", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 357, + 526 + ], + "score": 1.0, + "content": "0.05) [28] for pair-wise significance tests. In all tables, we use", + "type": "text" + }, + { + "bbox": [ + 358, + 514, + 364, + 524 + ], + "score": 0.8, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 512, + 505, + 526 + ], + "score": 1.0, + "content": "on BARTSCORE if it significantly", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 108, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 108, + 524, + 149, + 535 + ], + "score": 0.86, + "content": "( p < 0 . 0 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 524, + 485, + 537 + ], + "score": 1.0, + "content": "outperforms other unsupervised metrics excluding BARTSCORE variants. We use", + "type": "text" + }, + { + "bbox": [ + 485, + 524, + 492, + 536 + ], + "score": 0.81, + "content": "\\ddagger", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 535, + 505, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 505, + 548 + ], + "score": 1.0, + "content": "BARTSCORE if it significantly outperforms all other unsupervised metrics including BARTSCORE", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 546, + 142, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 142, + 558 + ], + "score": 1.0, + "content": "variants.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 578, + 222, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 223, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 223, + 592 + ], + "score": 1.0, + "content": "4.3 Experimental Results", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 108, + 601, + 226, + 613 + ], + "lines": [ + { + "bbox": [ + 105, + 601, + 227, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 227, + 614 + ], + "score": 1.0, + "content": "4.3.1 Machine Translation", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 623, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "Tab. 3 illustrates Kendall’s Tau correlation of diverse metrics on different language pairs. We", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 634, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 380, + 648 + ], + "score": 1.0, + "content": "can observe that: (1) BARTSCORE enhanced by fine-tuning tasks", + "type": "text" + }, + { + "bbox": [ + 380, + 635, + 433, + 646 + ], + "score": 0.4, + "content": "\\left( \\mathrm { C N N + P a r a } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 634, + 505, + 648 + ], + "score": 1.0, + "content": "can significantly", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 645, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 659 + ], + "score": 1.0, + "content": "outperform all other unsupervised methods on five language pairs and achieve comparable results on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 654, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 670 + ], + "score": 1.0, + "content": "the other two. (2) The performance of BARTSCORE can be further improved by simply adding a", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 432, + 681 + ], + "score": 1.0, + "content": "prompt (i.e., such as) without any other overhead. Notably, on the language pair", + "type": "text" + }, + { + "bbox": [ + 433, + 668, + 463, + 678 + ], + "score": 0.66, + "content": "{ \\tt d e } \\mathrm { - } \\in \\mathrm { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 667, + 506, + 681 + ], + "score": 1.0, + "content": ", using the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "prompt results in a 0.033 improvement, which even significantly surpasses existing state-of-the-art", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "score": 1.0, + "content": "supervised metrics BLEURT and COMET. This suggests a promising future direction for metric", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 699, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 714 + ], + "score": 1.0, + "content": "design: searching for proper prompts to better leverage knowledge stored in pre-trained language", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 711, + 335, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 335, + 723 + ], + "score": 1.0, + "content": "models instead of training on human judgment data [31].", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 112, + 109, + 498, + 274 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 70, + 505, + 104 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 70, + 505, + 82 + ], + "spans": [ + { + "bbox": [ + 105, + 70, + 505, + 82 + ], + "score": 1.0, + "content": "Table 3: Kendall’s Tau correlation of different metrics on WMT19 dataset. The highest correlation", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 81, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 93 + ], + "score": 1.0, + "content": "for each language pair achieved by unsupervised method is bold, and the highest correlation overall", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 92, + 499, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 499, + 105 + ], + "score": 1.0, + "content": "is underlined. Avg. denotes the average correlation achieved by a metric across all language pairs.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 112, + 109, + 498, + 274 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 109, + 498, + 274 + ], + "spans": [ + { + "bbox": [ + 112, + 109, + 498, + 274 + ], + "score": 0.983, + "html": "
de-enfi-en gu-enkk-enlt-enru-enzh-enAvg.
SUPERVISED METHODS
BLEURT0.1740.3740.3130.3720.3880.2200.4360.325
COMET0.2190.3690.3160.3780.4050.2260.4620.339
UNSUPERVISED METHODS
BLEU0.0540.2360.1940.2760.2490.1150.3210.206
CHRF0.1230.2920.2400.3230.3040.1770.3710.261
PRISM0.1990.3660.3200.3620.3820.2200.4340.326
BERTScore0.1900.3540.2920.3510.3810.2210.4300.317
BARTSCORE0.1560.3350.2730.3240.3220.1670.3890.281
+CNN0.1900.3650.3000.3480.3840.2080.4250.317
+ CNN+Para0.205t0.370t0.3160.378t0.386†0.219_0.442t0.331
+ CNN +Para +Prompt0.2380.3740.3180.376t0.386+0.2190.4470.337
", + "type": "table", + "image_path": "10078ed3879a01510492b203ac9ef173d092fe67d6150bd64664257da154ae67.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 112, + 109, + 498, + 164.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 112, + 164.0, + 498, + 219.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 112, + 219.0, + 498, + 274.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "table", + "bbox": [ + 109, + 330, + 502, + 474 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 291, + 505, + 325 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "score": 1.0, + "content": "Table 4: Spearman correlation of different metrics on three human judgement datasets. For prompt-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 303, + 507, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 427, + 315 + ], + "score": 1.0, + "content": "based learning, we consider adding prompts to the best-performing BARTSCORE", + "type": "text" + }, + { + "bbox": [ + 427, + 303, + 442, + 313 + ], + "score": 0.64, + "content": "( \\Omega )", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 303, + 507, + 315 + ], + "score": 1.0, + "content": "on each dataset.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 313, + 388, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 388, + 325 + ], + "score": 1.0, + "content": "The highest correlation overall for each aspect on each dataset is bold.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "table_body", + "bbox": [ + 109, + 330, + 502, + 474 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 109, + 330, + 502, + 474 + ], + "spans": [ + { + "bbox": [ + 109, + 330, + 502, + 474 + ], + "score": 0.985, + "html": "
REALSummSummEvalNeR18Avg.
CovCOHFACFLUINFOCOHFLUINFOREL
ROUGE-10.4980.1670.1600.1150.3260.0950.1040.1300.1470.194
ROUGE-20.4230.1840.1870.1590.2900.0260.0480.0790.0910.165
ROUGE-L0.4880.1280.1150.1050.3110.0640.0720.0890.1060.164
BERTScore0.4400.2840.1100.1930.3120.1470.1700.1310.1630.217
MoverScore0.3720.1590.1570.1290.3180.1610.1200.1880.1950.200
PRISM0.4110.2490.3450.2540.2120.5730.5320.5610.5530.410
BARTSCORE0.4410.322†0.3110.2480.2640.679†0.670t0.646†0.604t0.465
+ CNN0.4750.448‡0.382†0.356t0.356t0.653†0.640t0.616†0.5670.499
+ CNN+Para0.4710.424†0.401‡0.378t0.3130.657t0.652t0.614†0.5620.497
+Ω+Prompt0.4880.4070.3780.338f0.368f0.701t0.679t0.686t0.6200.518
", + "type": "table", + "image_path": "975ee48dd3e22e76b11d2026accb30e8c6cebbef0d1d3bce2e29af12c12df5c3.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 109, + 330, + 502, + 378.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 109, + 378.0, + 502, + 426.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 109, + 426.0, + 502, + 474.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 106, + 502, + 505, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 496, + 515 + ], + "score": 1.0, + "content": "Significance Tests. To perform rigorous analysis, we adopt the bootstrapping method (p-value", + "type": "text" + }, + { + "bbox": [ + 497, + 504, + 506, + 513 + ], + "score": 0.59, + "content": "<", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 357, + 526 + ], + "score": 1.0, + "content": "0.05) [28] for pair-wise significance tests. In all tables, we use", + "type": "text" + }, + { + "bbox": [ + 358, + 514, + 364, + 524 + ], + "score": 0.8, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 512, + 505, + 526 + ], + "score": 1.0, + "content": "on BARTSCORE if it significantly", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 108, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 108, + 524, + 149, + 535 + ], + "score": 0.86, + "content": "( p < 0 . 0 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 524, + 485, + 537 + ], + "score": 1.0, + "content": "outperforms other unsupervised metrics excluding BARTSCORE variants. We use", + "type": "text" + }, + { + "bbox": [ + 485, + 524, + 492, + 536 + ], + "score": 0.81, + "content": "\\ddagger", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 535, + 505, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 505, + 548 + ], + "score": 1.0, + "content": "BARTSCORE if it significantly outperforms all other unsupervised metrics including BARTSCORE", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 546, + 142, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 142, + 558 + ], + "score": 1.0, + "content": "variants.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 502, + 506, + 558 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 578, + 222, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 223, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 223, + 592 + ], + "score": 1.0, + "content": "4.3 Experimental Results", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 108, + 601, + 226, + 613 + ], + "lines": [ + { + "bbox": [ + 105, + 601, + 227, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 227, + 614 + ], + "score": 1.0, + "content": "4.3.1 Machine Translation", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 623, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "Tab. 3 illustrates Kendall’s Tau correlation of diverse metrics on different language pairs. We", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 634, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 380, + 648 + ], + "score": 1.0, + "content": "can observe that: (1) BARTSCORE enhanced by fine-tuning tasks", + "type": "text" + }, + { + "bbox": [ + 380, + 635, + 433, + 646 + ], + "score": 0.4, + "content": "\\left( \\mathrm { C N N + P a r a } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 634, + 505, + 648 + ], + "score": 1.0, + "content": "can significantly", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 645, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 659 + ], + "score": 1.0, + "content": "outperform all other unsupervised methods on five language pairs and achieve comparable results on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 654, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 670 + ], + "score": 1.0, + "content": "the other two. (2) The performance of BARTSCORE can be further improved by simply adding a", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 432, + 681 + ], + "score": 1.0, + "content": "prompt (i.e., such as) without any other overhead. Notably, on the language pair", + "type": "text" + }, + { + "bbox": [ + 433, + 668, + 463, + 678 + ], + "score": 0.66, + "content": "{ \\tt d e } \\mathrm { - } \\in \\mathrm { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 667, + 506, + 681 + ], + "score": 1.0, + "content": ", using the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "prompt results in a 0.033 improvement, which even significantly surpasses existing state-of-the-art", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "score": 1.0, + "content": "supervised metrics BLEURT and COMET. This suggests a promising future direction for metric", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 699, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 714 + ], + "score": 1.0, + "content": "design: searching for proper prompts to better leverage knowledge stored in pre-trained language", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 711, + 335, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 335, + 723 + ], + "score": 1.0, + "content": "models instead of training on human judgment data [31].", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 624, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 73, + 225, + 84 + ], + "lines": [ + { + "bbox": [ + 106, + 72, + 226, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 226, + 85 + ], + "score": 1.0, + "content": "4.3.2 Text Summarization", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 96, + 505, + 196 + ], + "lines": [ + { + "bbox": [ + 105, + 96, + 505, + 109 + ], + "spans": [ + { + "bbox": [ + 105, + 96, + 505, + 109 + ], + "score": 1.0, + "content": "Tab. 4 shows the meta-evaluation results of different metrics on the summarization task. We can", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 106, + 506, + 121 + ], + "spans": [ + { + "bbox": [ + 104, + 106, + 506, + 121 + ], + "score": 1.0, + "content": "observe that: (1) Simply vanilla BARTSCORE can outperform BERTScore and MoverScore by", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 118, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 506, + 132 + ], + "score": 1.0, + "content": "a large margin on 8 settings except the INFO perspective on SummEval. Strikingly, it achieves", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 128, + 506, + 143 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 506, + 143 + ], + "score": 1.0, + "content": "improvements of 0.251 and 0.265 over BERTScore and MoverScore respectively. (2) The improve-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 138, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 138, + 506, + 155 + ], + "score": 1.0, + "content": "ment on REALSum and SummEval datasets can be further improved when introducing fine-tuning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 150, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 505, + 164 + ], + "score": 1.0, + "content": "tasks. However, fine-tuning does not improve on the NeR18 dataset, likely because this dataset only", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 162, + 505, + 175 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 505, + 175 + ], + "score": 1.0, + "content": "contains 7 systems with easily distinguishable quality, and vanilla BARTSCORE can already achieve", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 213, + 186 + ], + "score": 1.0, + "content": "a high level of correlation", + "type": "text" + }, + { + "bbox": [ + 213, + 173, + 239, + 184 + ], + "score": 0.84, + "content": "( > 0 . 6", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 172, + 505, + 186 + ], + "score": 1.0, + "content": "on average). (3) Our prompt combination strategy can consistently", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 183, + 506, + 197 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 506, + 197 + ], + "score": 1.0, + "content": "improve the performance on informativeness, up to 0.072 Spearman correlation on the NeR18 dataset", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 311, + 244 + ], + "lines": [ + { + "bbox": [ + 106, + 200, + 312, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 312, + 212 + ], + "score": 1.0, + "content": "and 0.055 on SummEval. However, the perfor-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 211, + 311, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 311, + 223 + ], + "score": 1.0, + "content": "mance from other perspectives such as fluency and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 222, + 312, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 312, + 235 + ], + "score": 1.0, + "content": "factuality do not show consistent improvements,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 232, + 273, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 273, + 245 + ], + "score": 1.0, + "content": "which we will elaborate on later (§4.4.2).", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 249, + 311, + 402 + ], + "lines": [ + { + "bbox": [ + 106, + 249, + 311, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 311, + 262 + ], + "score": 1.0, + "content": "Analysis on Factuality Datasets The goal of these", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 261, + 311, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 311, + 272 + ], + "score": 1.0, + "content": "datasets is to judge whether a short generated sum-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 272, + 311, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 311, + 283 + ], + "score": 1.0, + "content": "mary is faithful to the original long documents. As", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 282, + 310, + 293 + ], + "spans": [ + { + "bbox": [ + 107, + 282, + 310, + 293 + ], + "score": 1.0, + "content": "shown in Tab. 5, we observe that (1) BARTSCORE", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 293, + 310, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 132, + 304 + ], + "score": 0.49, + "content": "+ \\thinspace C N N", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 293, + 310, + 304 + ], + "score": 1.0, + "content": "can almost match human baseline on Rank19", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 304, + 311, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 311, + 316 + ], + "score": 1.0, + "content": "and outperform all other metrics, including the most", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 315, + 311, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 311, + 326 + ], + "score": 1.0, + "content": "recent top-performing factuality metrics FactCC", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 326, + 311, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 311, + 339 + ], + "score": 1.0, + "content": "and QAGS by a large margin. (2) Using paraphrase", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 337, + 311, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 311, + 348 + ], + "score": 1.0, + "content": "as a fine-tuning task will reduce BARTSCORE’s", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 348, + 311, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 311, + 359 + ], + "score": 1.0, + "content": "performance, which is reasonable since these two", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 358, + 311, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 311, + 371 + ], + "score": 1.0, + "content": "texts (i.e., the summary and document) shouldn’t", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 370, + 312, + 382 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 312, + 382 + ], + "score": 1.0, + "content": "maintain the paraphrased relationship in general.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 381, + 312, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 312, + 392 + ], + "score": 1.0, + "content": "(3) Introducing prompts does not bring an improve-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 392, + 299, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 299, + 403 + ], + "score": 1.0, + "content": "ment, even resulting in a performance decrease.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 23.5 + }, + { + "type": "table", + "bbox": [ + 319, + 248, + 504, + 427 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 318, + 209, + 505, + 242 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 317, + 207, + 507, + 221 + ], + "spans": [ + { + "bbox": [ + 317, + 207, + 507, + 221 + ], + "score": 1.0, + "content": "Table 5: Results on Rank19 and QAGS datasets.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 317, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 317, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "where “Q” represents QAGS. Metrics achieve", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 317, + 231, + 434, + 242 + ], + "spans": [ + { + "bbox": [ + 317, + 231, + 434, + 242 + ], + "score": 1.0, + "content": "highest correlation are bold.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "table_body", + "bbox": [ + 319, + 248, + 504, + 427 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 319, + 248, + 504, + 427 + ], + "spans": [ + { + "bbox": [ + 319, + 248, + 504, + 427 + ], + "score": 0.982, + "html": "
Rank19 Q-CNNQ-XSUM
Acc.Pearson
ROUGE-10.5680.338-0.008
ROUGE-20.6300.4590.097
ROUGE-L0.5870.3570.024
BERTScore0.7130.5760.024
MoverScore0.7130.4140.054
PRISM0.7800.4790.025
FactCC [30] QAGS [66]0.700 0.7211 0.5451 0.175
Human [14] BARTSCORE0.83911 0.009
+CNN0.684 0.836‡0.661† 0.735±0.184‡
+ CNN+Para0.680t0.074
0.788
+CNN +Prompt70.7960.719f0.094
", + "type": "table", + "image_path": "17bdfd18aa5de5a0b5c0e6b2d1231e0adb998a06d3721c28e735fea9fd3f7f93.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 319, + 248, + 504, + 261.7692307692308 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 319, + 261.7692307692308, + 504, + 275.53846153846155 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 319, + 275.53846153846155, + 504, + 289.3076923076923 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 319, + 289.3076923076923, + 504, + 303.0769230769231 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 319, + 303.0769230769231, + 504, + 316.84615384615387 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 319, + 316.84615384615387, + 504, + 330.61538461538464 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 319, + 330.61538461538464, + 504, + 344.3846153846154 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 319, + 344.3846153846154, + 504, + 358.1538461538462 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 319, + 358.1538461538462, + 504, + 371.92307692307696 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 319, + 371.92307692307696, + 504, + 385.69230769230774 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 319, + 385.69230769230774, + 504, + 399.4615384615385 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 319, + 399.4615384615385, + 504, + 413.2307692307693 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 319, + 413.2307692307693, + 504, + 427.00000000000006 + ], + "spans": [], + "index": 43 + } + ] + } + ], + "index": 26.0 + }, + { + "type": "title", + "bbox": [ + 107, + 427, + 189, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 191, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 191, + 439 + ], + "score": 1.0, + "content": "4.3.3 Data-to-text", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 451, + 296, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 451, + 297, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 297, + 464 + ], + "score": 1.0, + "content": "The experiment results on data-to-text datasets", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 462, + 297, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 297, + 474 + ], + "score": 1.0, + "content": "are shown in Tab. 6. We observe that (1) fine-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 473, + 297, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 297, + 486 + ], + "score": 1.0, + "content": "tuning on the CNNDM dataset can consistently", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 484, + 296, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 296, + 496 + ], + "score": 1.0, + "content": "boost the correlation, for example, up to 0.056", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 496, + 297, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 297, + 507 + ], + "score": 1.0, + "content": "gain on BAGEL. (2) Additionally, further fine-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 506, + 297, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 297, + 519 + ], + "score": 1.0, + "content": "tuning on paraphrase datasets results in even", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 517, + 297, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 297, + 529 + ], + "score": 1.0, + "content": "higher performance compared to the version", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 528, + 297, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 297, + 540 + ], + "score": 1.0, + "content": "without any fine-tuning, up to 0.083 Spearman", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 539, + 296, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 296, + 550 + ], + "score": 1.0, + "content": "correlation on BAGEL dataset. These results", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 550, + 297, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 297, + 562 + ], + "score": 1.0, + "content": "surpass all existing top-performing metrics. (3)", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 560, + 297, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 297, + 574 + ], + "score": 1.0, + "content": "Our proposed prompt combination strategy can", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 106, + 572, + 297, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 297, + 584 + ], + "score": 1.0, + "content": "consistently improve correlation, on average", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 581, + 297, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 297, + 595 + ], + "score": 1.0, + "content": "0.028 Spearman correlation. This is consistent", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 593, + 297, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 185, + 606 + ], + "score": 1.0, + "content": "with the findings in", + "type": "text" + }, + { + "bbox": [ + 186, + 594, + 212, + 605 + ], + "score": 0.86, + "content": "\\ S 4 . 3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 593, + 297, + 606 + ], + "score": 1.0, + "content": "that we can improve", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 604, + 297, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 297, + 616 + ], + "score": 1.0, + "content": "the aspect of informativeness through proper", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 615, + 153, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 153, + 628 + ], + "score": 1.0, + "content": "prompting.", + "type": "text" + } + ], + "index": 60 + } + ], + "index": 52.5 + }, + { + "type": "table", + "bbox": [ + 304, + 489, + 504, + 628 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 304, + 450, + 505, + 483 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 303, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 303, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "Table 6: Results on data-to-text datasets. We report", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 304, + 460, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 304, + 460, + 506, + 473 + ], + "score": 1.0, + "content": "Spearman correlation. Metrics achieve highest cor-", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 303, + 471, + 375, + 483 + ], + "spans": [ + { + "bbox": [ + 303, + 471, + 375, + 483 + ], + "score": 1.0, + "content": "relation are bold.", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 62 + }, + { + "type": "table_body", + "bbox": [ + 304, + 489, + 504, + 628 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 304, + 489, + 504, + 628 + ], + "spans": [ + { + "bbox": [ + 304, + 489, + 504, + 628 + ], + "score": 0.982, + "html": "
BAGELSFRESSFHOTAvg.
ROUGE-10.2340.1150.1180.156
ROUGE-20.1990.1160.0880.134
ROUGE-L0.1890.1030.1100.134
BERTScore0.2890.1560.1350.193
MoverScore0.2840.1530.1720.203
PRISM0.3050.1550.1960.219
BARTSCORE0.2470.164†0.1580.190
+ CNN0.3030.191†0.1900.228
+ CNN+Para0.330t0.185t0.211†0.242
+Ω+Prompt0.3360.238t0.235t0.270
", + "type": "table", + "image_path": "796d087ca8e9733af385701c9958226a7d869290614949a2929e782de3bdc02e.jpg" + } + ] + } + ], + "index": 68.5, + "virtual_lines": [ + { + "bbox": [ + 304, + 489, + 504, + 502.9 + ], + "spans": [], + "index": 64 + }, + { + "bbox": [ + 304, + 502.9, + 504, + 516.8 + ], + "spans": [], + "index": 65 + }, + { + "bbox": [ + 304, + 516.8, + 504, + 530.6999999999999 + ], + "spans": [], + "index": 66 + }, + { + "bbox": [ + 304, + 530.6999999999999, + 504, + 544.5999999999999 + ], + "spans": [], + "index": 67 + }, + { + "bbox": [ + 304, + 544.5999999999999, + 504, + 558.4999999999999 + ], + "spans": [], + "index": 68 + }, + { + "bbox": [ + 304, + 558.4999999999999, + 504, + 572.3999999999999 + ], + "spans": [], + "index": 69 + }, + { + "bbox": [ + 304, + 572.3999999999999, + 504, + 586.2999999999998 + ], + "spans": [], + "index": 70 + }, + { + "bbox": [ + 304, + 586.2999999999998, + 504, + 600.1999999999998 + ], + "spans": [], + "index": 71 + }, + { + "bbox": [ + 304, + 600.1999999999998, + 504, + 614.0999999999998 + ], + "spans": [], + "index": 72 + }, + { + "bbox": [ + 304, + 614.0999999999998, + 504, + 627.9999999999998 + ], + "spans": [], + "index": 73 + } + ] + } + ], + "index": 65.25 + }, + { + "type": "title", + "bbox": [ + 107, + 653, + 166, + 664 + ], + "lines": [ + { + "bbox": [ + 105, + 651, + 168, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 168, + 666 + ], + "score": 1.0, + "content": "4.4 Analysis", + "type": "text" + } + ], + "index": 74 + } + ], + "index": 74 + }, + { + "type": "text", + "bbox": [ + 107, + 678, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "We design experiments to better understand the mechanism by which BARTSCORE obtains these", + "type": "text" + } + ], + "index": 75 + }, + { + "bbox": [ + 105, + 689, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 507, + 702 + ], + "score": 1.0, + "content": "promising results, specifically asking three questions: Q1: Compared to other unsupervised metrics,", + "type": "text" + } + ], + "index": 76 + }, + { + "bbox": [ + 105, + 700, + 507, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 507, + 713 + ], + "score": 1.0, + "content": "where does BARTSCORE outperform them? Q2: How does adding prompts benefit evaluation? Q3:", + "type": "text" + } + ], + "index": 77 + }, + { + "bbox": [ + 105, + 711, + 338, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 338, + 723 + ], + "score": 1.0, + "content": "Will BARTScore introduce biases in unpredictable ways?", + "type": "text" + } + ], + "index": 78 + } + ], + "index": 76.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 73, + 225, + 84 + ], + "lines": [ + { + "bbox": [ + 106, + 72, + 226, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 226, + 85 + ], + "score": 1.0, + "content": "4.3.2 Text Summarization", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 96, + 505, + 196 + ], + "lines": [ + { + "bbox": [ + 105, + 96, + 505, + 109 + ], + "spans": [ + { + "bbox": [ + 105, + 96, + 505, + 109 + ], + "score": 1.0, + "content": "Tab. 4 shows the meta-evaluation results of different metrics on the summarization task. We can", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 106, + 506, + 121 + ], + "spans": [ + { + "bbox": [ + 104, + 106, + 506, + 121 + ], + "score": 1.0, + "content": "observe that: (1) Simply vanilla BARTSCORE can outperform BERTScore and MoverScore by", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 118, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 506, + 132 + ], + "score": 1.0, + "content": "a large margin on 8 settings except the INFO perspective on SummEval. Strikingly, it achieves", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 128, + 506, + 143 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 506, + 143 + ], + "score": 1.0, + "content": "improvements of 0.251 and 0.265 over BERTScore and MoverScore respectively. (2) The improve-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 138, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 138, + 506, + 155 + ], + "score": 1.0, + "content": "ment on REALSum and SummEval datasets can be further improved when introducing fine-tuning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 150, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 505, + 164 + ], + "score": 1.0, + "content": "tasks. However, fine-tuning does not improve on the NeR18 dataset, likely because this dataset only", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 162, + 505, + 175 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 505, + 175 + ], + "score": 1.0, + "content": "contains 7 systems with easily distinguishable quality, and vanilla BARTSCORE can already achieve", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 213, + 186 + ], + "score": 1.0, + "content": "a high level of correlation", + "type": "text" + }, + { + "bbox": [ + 213, + 173, + 239, + 184 + ], + "score": 0.84, + "content": "( > 0 . 6", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 172, + 505, + 186 + ], + "score": 1.0, + "content": "on average). (3) Our prompt combination strategy can consistently", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 183, + 506, + 197 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 506, + 197 + ], + "score": 1.0, + "content": "improve the performance on informativeness, up to 0.072 Spearman correlation on the NeR18 dataset", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 200, + 312, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 312, + 212 + ], + "score": 1.0, + "content": "and 0.055 on SummEval. However, the perfor-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 211, + 311, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 311, + 223 + ], + "score": 1.0, + "content": "mance from other perspectives such as fluency and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 222, + 312, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 312, + 235 + ], + "score": 1.0, + "content": "factuality do not show consistent improvements,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 232, + 273, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 273, + 245 + ], + "score": 1.0, + "content": "which we will elaborate on later (§4.4.2).", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 5, + "bbox_fs": [ + 104, + 96, + 506, + 197 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 311, + 244 + ], + "lines": [], + "index": 11.5, + "bbox_fs": [ + 105, + 200, + 312, + 245 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 249, + 311, + 402 + ], + "lines": [ + { + "bbox": [ + 106, + 249, + 311, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 311, + 262 + ], + "score": 1.0, + "content": "Analysis on Factuality Datasets The goal of these", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 261, + 311, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 311, + 272 + ], + "score": 1.0, + "content": "datasets is to judge whether a short generated sum-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 272, + 311, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 311, + 283 + ], + "score": 1.0, + "content": "mary is faithful to the original long documents. As", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 282, + 310, + 293 + ], + "spans": [ + { + "bbox": [ + 107, + 282, + 310, + 293 + ], + "score": 1.0, + "content": "shown in Tab. 5, we observe that (1) BARTSCORE", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 293, + 310, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 132, + 304 + ], + "score": 0.49, + "content": "+ \\thinspace C N N", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 293, + 310, + 304 + ], + "score": 1.0, + "content": "can almost match human baseline on Rank19", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 304, + 311, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 311, + 316 + ], + "score": 1.0, + "content": "and outperform all other metrics, including the most", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 315, + 311, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 311, + 326 + ], + "score": 1.0, + "content": "recent top-performing factuality metrics FactCC", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 326, + 311, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 311, + 339 + ], + "score": 1.0, + "content": "and QAGS by a large margin. (2) Using paraphrase", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 337, + 311, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 311, + 348 + ], + "score": 1.0, + "content": "as a fine-tuning task will reduce BARTSCORE’s", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 348, + 311, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 311, + 359 + ], + "score": 1.0, + "content": "performance, which is reasonable since these two", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 358, + 311, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 311, + 371 + ], + "score": 1.0, + "content": "texts (i.e., the summary and document) shouldn’t", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 370, + 312, + 382 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 312, + 382 + ], + "score": 1.0, + "content": "maintain the paraphrased relationship in general.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 381, + 312, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 312, + 392 + ], + "score": 1.0, + "content": "(3) Introducing prompts does not bring an improve-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 392, + 299, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 299, + 403 + ], + "score": 1.0, + "content": "ment, even resulting in a performance decrease.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 249, + 312, + 403 + ] + }, + { + "type": "table", + "bbox": [ + 319, + 248, + 504, + 427 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 318, + 209, + 505, + 242 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 317, + 207, + 507, + 221 + ], + "spans": [ + { + "bbox": [ + 317, + 207, + 507, + 221 + ], + "score": 1.0, + "content": "Table 5: Results on Rank19 and QAGS datasets.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 317, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 317, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "where “Q” represents QAGS. Metrics achieve", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 317, + 231, + 434, + 242 + ], + "spans": [ + { + "bbox": [ + 317, + 231, + 434, + 242 + ], + "score": 1.0, + "content": "highest correlation are bold.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "table_body", + "bbox": [ + 319, + 248, + 504, + 427 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 319, + 248, + 504, + 427 + ], + "spans": [ + { + "bbox": [ + 319, + 248, + 504, + 427 + ], + "score": 0.982, + "html": "
Rank19 Q-CNNQ-XSUM
Acc.Pearson
ROUGE-10.5680.338-0.008
ROUGE-20.6300.4590.097
ROUGE-L0.5870.3570.024
BERTScore0.7130.5760.024
MoverScore0.7130.4140.054
PRISM0.7800.4790.025
FactCC [30] QAGS [66]0.700 0.7211 0.5451 0.175
Human [14] BARTSCORE0.83911 0.009
+CNN0.684 0.836‡0.661† 0.735±0.184‡
+ CNN+Para0.680t0.074
0.788
+CNN +Prompt70.7960.719f0.094
", + "type": "table", + "image_path": "17bdfd18aa5de5a0b5c0e6b2d1231e0adb998a06d3721c28e735fea9fd3f7f93.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 319, + 248, + 504, + 261.7692307692308 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 319, + 261.7692307692308, + 504, + 275.53846153846155 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 319, + 275.53846153846155, + 504, + 289.3076923076923 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 319, + 289.3076923076923, + 504, + 303.0769230769231 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 319, + 303.0769230769231, + 504, + 316.84615384615387 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 319, + 316.84615384615387, + 504, + 330.61538461538464 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 319, + 330.61538461538464, + 504, + 344.3846153846154 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 319, + 344.3846153846154, + 504, + 358.1538461538462 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 319, + 358.1538461538462, + 504, + 371.92307692307696 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 319, + 371.92307692307696, + 504, + 385.69230769230774 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 319, + 385.69230769230774, + 504, + 399.4615384615385 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 319, + 399.4615384615385, + 504, + 413.2307692307693 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 319, + 413.2307692307693, + 504, + 427.00000000000006 + ], + "spans": [], + "index": 43 + } + ] + } + ], + "index": 26.0 + }, + { + "type": "title", + "bbox": [ + 107, + 427, + 189, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 191, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 191, + 439 + ], + "score": 1.0, + "content": "4.3.3 Data-to-text", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 451, + 296, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 451, + 297, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 297, + 464 + ], + "score": 1.0, + "content": "The experiment results on data-to-text datasets", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 462, + 297, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 297, + 474 + ], + "score": 1.0, + "content": "are shown in Tab. 6. We observe that (1) fine-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 473, + 297, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 297, + 486 + ], + "score": 1.0, + "content": "tuning on the CNNDM dataset can consistently", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 484, + 296, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 296, + 496 + ], + "score": 1.0, + "content": "boost the correlation, for example, up to 0.056", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 496, + 297, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 297, + 507 + ], + "score": 1.0, + "content": "gain on BAGEL. (2) Additionally, further fine-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 506, + 297, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 297, + 519 + ], + "score": 1.0, + "content": "tuning on paraphrase datasets results in even", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 517, + 297, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 297, + 529 + ], + "score": 1.0, + "content": "higher performance compared to the version", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 528, + 297, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 297, + 540 + ], + "score": 1.0, + "content": "without any fine-tuning, up to 0.083 Spearman", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 539, + 296, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 296, + 550 + ], + "score": 1.0, + "content": "correlation on BAGEL dataset. These results", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 550, + 297, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 297, + 562 + ], + "score": 1.0, + "content": "surpass all existing top-performing metrics. (3)", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 560, + 297, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 297, + 574 + ], + "score": 1.0, + "content": "Our proposed prompt combination strategy can", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 106, + 572, + 297, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 297, + 584 + ], + "score": 1.0, + "content": "consistently improve correlation, on average", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 581, + 297, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 297, + 595 + ], + "score": 1.0, + "content": "0.028 Spearman correlation. This is consistent", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 593, + 297, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 185, + 606 + ], + "score": 1.0, + "content": "with the findings in", + "type": "text" + }, + { + "bbox": [ + 186, + 594, + 212, + 605 + ], + "score": 0.86, + "content": "\\ S 4 . 3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 593, + 297, + 606 + ], + "score": 1.0, + "content": "that we can improve", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 604, + 297, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 297, + 616 + ], + "score": 1.0, + "content": "the aspect of informativeness through proper", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 615, + 153, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 153, + 628 + ], + "score": 1.0, + "content": "prompting.", + "type": "text" + } + ], + "index": 60 + } + ], + "index": 52.5, + "bbox_fs": [ + 105, + 451, + 297, + 628 + ] + }, + { + "type": "table", + "bbox": [ + 304, + 489, + 504, + 628 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 304, + 450, + 505, + 483 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 303, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 303, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "Table 6: Results on data-to-text datasets. We report", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 304, + 460, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 304, + 460, + 506, + 473 + ], + "score": 1.0, + "content": "Spearman correlation. Metrics achieve highest cor-", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 303, + 471, + 375, + 483 + ], + "spans": [ + { + "bbox": [ + 303, + 471, + 375, + 483 + ], + "score": 1.0, + "content": "relation are bold.", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 62 + }, + { + "type": "table_body", + "bbox": [ + 304, + 489, + 504, + 628 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 304, + 489, + 504, + 628 + ], + "spans": [ + { + "bbox": [ + 304, + 489, + 504, + 628 + ], + "score": 0.982, + "html": "
BAGELSFRESSFHOTAvg.
ROUGE-10.2340.1150.1180.156
ROUGE-20.1990.1160.0880.134
ROUGE-L0.1890.1030.1100.134
BERTScore0.2890.1560.1350.193
MoverScore0.2840.1530.1720.203
PRISM0.3050.1550.1960.219
BARTSCORE0.2470.164†0.1580.190
+ CNN0.3030.191†0.1900.228
+ CNN+Para0.330t0.185t0.211†0.242
+Ω+Prompt0.3360.238t0.235t0.270
", + "type": "table", + "image_path": "796d087ca8e9733af385701c9958226a7d869290614949a2929e782de3bdc02e.jpg" + } + ] + } + ], + "index": 68.5, + "virtual_lines": [ + { + "bbox": [ + 304, + 489, + 504, + 502.9 + ], + "spans": [], + "index": 64 + }, + { + "bbox": [ + 304, + 502.9, + 504, + 516.8 + ], + "spans": [], + "index": 65 + }, + { + "bbox": [ + 304, + 516.8, + 504, + 530.6999999999999 + ], + "spans": [], + "index": 66 + }, + { + "bbox": [ + 304, + 530.6999999999999, + 504, + 544.5999999999999 + ], + "spans": [], + "index": 67 + }, + { + "bbox": [ + 304, + 544.5999999999999, + 504, + 558.4999999999999 + ], + "spans": [], + "index": 68 + }, + { + "bbox": [ + 304, + 558.4999999999999, + 504, + 572.3999999999999 + ], + "spans": [], + "index": 69 + }, + { + "bbox": [ + 304, + 572.3999999999999, + 504, + 586.2999999999998 + ], + "spans": [], + "index": 70 + }, + { + "bbox": [ + 304, + 586.2999999999998, + 504, + 600.1999999999998 + ], + "spans": [], + "index": 71 + }, + { + "bbox": [ + 304, + 600.1999999999998, + 504, + 614.0999999999998 + ], + "spans": [], + "index": 72 + }, + { + "bbox": [ + 304, + 614.0999999999998, + 504, + 627.9999999999998 + ], + "spans": [], + "index": 73 + } + ] + } + ], + "index": 65.25 + }, + { + "type": "title", + "bbox": [ + 107, + 653, + 166, + 664 + ], + "lines": [ + { + "bbox": [ + 105, + 651, + 168, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 168, + 666 + ], + "score": 1.0, + "content": "4.4 Analysis", + "type": "text" + } + ], + "index": 74 + } + ], + "index": 74 + }, + { + "type": "text", + "bbox": [ + 107, + 678, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "We design experiments to better understand the mechanism by which BARTSCORE obtains these", + "type": "text" + } + ], + "index": 75 + }, + { + "bbox": [ + 105, + 689, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 507, + 702 + ], + "score": 1.0, + "content": "promising results, specifically asking three questions: Q1: Compared to other unsupervised metrics,", + "type": "text" + } + ], + "index": 76 + }, + { + "bbox": [ + 105, + 700, + 507, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 507, + 713 + ], + "score": 1.0, + "content": "where does BARTSCORE outperform them? Q2: How does adding prompts benefit evaluation? Q3:", + "type": "text" + } + ], + "index": 77 + }, + { + "bbox": [ + 105, + 711, + 338, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 338, + 723 + ], + "score": 1.0, + "content": "Will BARTScore introduce biases in unpredictable ways?", + "type": "text" + } + ], + "index": 78 + } + ], + "index": 76.5, + "bbox_fs": [ + 105, + 678, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 77, + 501, + 179 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 77, + 501, + 179 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 77, + 501, + 179 + ], + "spans": [ + { + "bbox": [ + 109, + 77, + 501, + 179 + ], + "score": 0.966, + "type": "image", + "image_path": "78f8c1e31ba1ab4fbe0920e68cff9da08fddef39259e3f11e995865e34618d98.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 77, + 501, + 111.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 111.0, + 501, + 145.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 145.0, + 501, + 179.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 185, + 505, + 219 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "Figure 2: Fine-grained analysis (a,b) and prompt analysis (c). In (a, b), BE, PR, BL, CO, BA represent", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 196, + 505, + 209 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 505, + 209 + ], + "score": 1.0, + "content": "BERTScore, PRISM, BLEURT, COMET and BARTSCORE respectively. In (c), SEM, LIN, FAC", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 207, + 426, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 426, + 221 + ], + "score": 1.0, + "content": "denote semantic overlap, linguistic quality and factual correctness respectively.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 107, + 241, + 232, + 253 + ], + "lines": [ + { + "bbox": [ + 105, + 240, + 232, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 232, + 255 + ], + "score": 1.0, + "content": "4.4.1 Fine-grained Analysis", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 504, + 282 + ], + "lines": [ + { + "bbox": [ + 106, + 260, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 505, + 272 + ], + "score": 1.0, + "content": "To answer Q1, we choose the MT task and break down the performance of each metric into different", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 271, + 236, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 236, + 283 + ], + "score": 1.0, + "content": "buckets based on different axes.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 107, + 295, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "score": 1.0, + "content": "Top-k Systems We report the average correlation across all language pairs achieved by each metric", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 306, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 239, + 319 + ], + "score": 1.0, + "content": "given only translations from top-", + "type": "text" + }, + { + "bbox": [ + 239, + 307, + 246, + 316 + ], + "score": 0.81, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 306, + 380, + 319 + ], + "score": 1.0, + "content": "systems. We vary the number of", + "type": "text" + }, + { + "bbox": [ + 380, + 307, + 387, + 317 + ], + "score": 0.79, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 306, + 506, + 319 + ], + "score": 1.0, + "content": ", and the results are shown in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "score": 1.0, + "content": "Fig. 2-(a). We can see that BARTSCORE can outperform all other metrics (including one supervised", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 477, + 340 + ], + "score": 1.0, + "content": "metric BLEURT) except the existing state-of-the-art supervised metric COMET for different", + "type": "text" + }, + { + "bbox": [ + 478, + 329, + 484, + 338 + ], + "score": 0.78, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 328, + 505, + 340 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 339, + 507, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 507, + 352 + ], + "score": 1.0, + "content": "the decrease in correlation becomes smoother than others when considering top-scoring systems.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 350, + 403, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 403, + 362 + ], + "score": 1.0, + "content": "This indicates that BARTSCORE is robust to high-quality generated texts.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 374, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 106, + 374, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 506, + 386 + ], + "score": 1.0, + "content": "Reference Length We break down each test set into four buckets based on the reference length,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 384, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 506, + 398 + ], + "score": 1.0, + "content": "which are [15, 25), [25, 35), [35, 45), [45, 54] and compute the Kendall’s Tau average correlation of", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "score": 1.0, + "content": "different metrics across all language pairs within each bucket.6 The results are shown in Fig. 2-(b).", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 407, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 506, + 419 + ], + "score": 1.0, + "content": "We observe that BARTSCORE can outperform or tie with other unsupervised metrics over different", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "score": 1.0, + "content": "reference lengths. Also, its correlation with human judgments is more stable compared to all other", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "metrics. This indicates its robustness to different input lengths. More other analyses can be found in", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 439, + 151, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 151, + 453 + ], + "score": 1.0, + "content": "Appendix.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 107, + 464, + 209, + 475 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 210, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 210, + 478 + ], + "score": 1.0, + "content": "4.4.2 Prompt Analysis", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 483, + 505, + 604 + ], + "lines": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "For Q2, we choose the summarization and data-to-text tasks for analysis where we used all prompts", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 494, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 506, + 507 + ], + "score": 1.0, + "content": "from our prompt set. We first group all the evaluation perspectives into three categories: (1) semantic", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "overlap (informativeness, pyramid score, and relevance) (2) linguistic quality (fluency, coherence)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "(3) factual correctness (factuality). We then calculate the percentage of prompts that result in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 526, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 505, + 541 + ], + "score": 1.0, + "content": "performance improvements for each perspective within a dataset. Finally, we compute the average", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 538, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 505, + 551 + ], + "score": 1.0, + "content": "percentage of prompts that can lead to performance gains for each category. The results are shown", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 548, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 548, + 505, + 561 + ], + "score": 1.0, + "content": "in Tab. 2-(c). We can see that for semantic overlap, almost all prompts can lead to the performance", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 560, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 573 + ], + "score": 1.0, + "content": "increase, while for factuality only a few prompts can improve the performance. This also explains the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 571, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 144, + 583 + ], + "score": 1.0, + "content": "results in", + "type": "text" + }, + { + "bbox": [ + 144, + 571, + 170, + 582 + ], + "score": 0.85, + "content": "\\ S 4 . 3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 571, + 506, + 583 + ], + "score": 1.0, + "content": "where we found that combining the results of different prompts can lead to consistent", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 582, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 594 + ], + "score": 1.0, + "content": "increases in semantic overlap but worse performance in factuality. Regarding linguistic quality, the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 592, + 504, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 474, + 605 + ], + "score": 1.0, + "content": "effect of adding a prompt is not that predictive, which is also consistent with our findings in", + "type": "text" + }, + { + "bbox": [ + 474, + 593, + 500, + 604 + ], + "score": 0.81, + "content": "\\ S 4 . 3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 500, + 592, + 504, + 605 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 28 + }, + { + "type": "title", + "bbox": [ + 107, + 616, + 195, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 196, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 196, + 630 + ], + "score": 1.0, + "content": "4.4.3 Bias Analysis", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 107, + 637, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 107, + 637, + 505, + 647 + ], + "score": 1.0, + "content": "To answer Q3, we conduct bias analysis. Bias would indicate that the scores are too high or too low", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "compared to the scores they are given by human annotators. Therefore, to see whether such biases", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 658, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 670 + ], + "score": 1.0, + "content": "exist, we inspected the rank differences given by human annotators and BARTScore (fine-tuned", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 669, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 505, + 680 + ], + "score": 1.0, + "content": "on CNNDM dataset) on the REALSumm dataset where 24 systems are considered, including both", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 679, + 505, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 505, + 692 + ], + "score": 1.0, + "content": "abstractive models and extractive models as well as models based on pre-trained models and models", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 702, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 118, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 118, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "6In each bucket, we remove the language pairs that do not contain over 500 samples. This results in the", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 711, + 419, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 147, + 723 + ], + "score": 1.0, + "content": "removal of", + "type": "text" + }, + { + "bbox": [ + 148, + 712, + 176, + 721 + ], + "score": 0.52, + "content": "\\operatorname { k k } - \\ e \\mathrm { e n }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 711, + 419, + 723 + ], + "score": 1.0, + "content": "in [35, 45) and the removal of gu-en, kk-en, lt-en in [45, 54].", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 77, + 501, + 179 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 77, + 501, + 179 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 77, + 501, + 179 + ], + "spans": [ + { + "bbox": [ + 109, + 77, + 501, + 179 + ], + "score": 0.966, + "type": "image", + "image_path": "78f8c1e31ba1ab4fbe0920e68cff9da08fddef39259e3f11e995865e34618d98.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 77, + 501, + 111.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 111.0, + 501, + 145.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 145.0, + 501, + 179.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 185, + 505, + 219 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "Figure 2: Fine-grained analysis (a,b) and prompt analysis (c). In (a, b), BE, PR, BL, CO, BA represent", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 196, + 505, + 209 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 505, + 209 + ], + "score": 1.0, + "content": "BERTScore, PRISM, BLEURT, COMET and BARTSCORE respectively. In (c), SEM, LIN, FAC", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 207, + 426, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 426, + 221 + ], + "score": 1.0, + "content": "denote semantic overlap, linguistic quality and factual correctness respectively.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 107, + 241, + 232, + 253 + ], + "lines": [ + { + "bbox": [ + 105, + 240, + 232, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 232, + 255 + ], + "score": 1.0, + "content": "4.4.1 Fine-grained Analysis", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 504, + 282 + ], + "lines": [ + { + "bbox": [ + 106, + 260, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 505, + 272 + ], + "score": 1.0, + "content": "To answer Q1, we choose the MT task and break down the performance of each metric into different", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 271, + 236, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 236, + 283 + ], + "score": 1.0, + "content": "buckets based on different axes.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 260, + 505, + 283 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 295, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "score": 1.0, + "content": "Top-k Systems We report the average correlation across all language pairs achieved by each metric", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 306, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 239, + 319 + ], + "score": 1.0, + "content": "given only translations from top-", + "type": "text" + }, + { + "bbox": [ + 239, + 307, + 246, + 316 + ], + "score": 0.81, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 306, + 380, + 319 + ], + "score": 1.0, + "content": "systems. We vary the number of", + "type": "text" + }, + { + "bbox": [ + 380, + 307, + 387, + 317 + ], + "score": 0.79, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 306, + 506, + 319 + ], + "score": 1.0, + "content": ", and the results are shown in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "score": 1.0, + "content": "Fig. 2-(a). We can see that BARTSCORE can outperform all other metrics (including one supervised", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 477, + 340 + ], + "score": 1.0, + "content": "metric BLEURT) except the existing state-of-the-art supervised metric COMET for different", + "type": "text" + }, + { + "bbox": [ + 478, + 329, + 484, + 338 + ], + "score": 0.78, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 328, + 505, + 340 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 339, + 507, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 507, + 352 + ], + "score": 1.0, + "content": "the decrease in correlation becomes smoother than others when considering top-scoring systems.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 350, + 403, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 403, + 362 + ], + "score": 1.0, + "content": "This indicates that BARTSCORE is robust to high-quality generated texts.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 295, + 507, + 362 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 374, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 106, + 374, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 506, + 386 + ], + "score": 1.0, + "content": "Reference Length We break down each test set into four buckets based on the reference length,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 384, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 506, + 398 + ], + "score": 1.0, + "content": "which are [15, 25), [25, 35), [35, 45), [45, 54] and compute the Kendall’s Tau average correlation of", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "score": 1.0, + "content": "different metrics across all language pairs within each bucket.6 The results are shown in Fig. 2-(b).", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 407, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 506, + 419 + ], + "score": 1.0, + "content": "We observe that BARTSCORE can outperform or tie with other unsupervised metrics over different", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "score": 1.0, + "content": "reference lengths. Also, its correlation with human judgments is more stable compared to all other", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "metrics. This indicates its robustness to different input lengths. More other analyses can be found in", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 439, + 151, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 151, + 453 + ], + "score": 1.0, + "content": "Appendix.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 374, + 506, + 453 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 464, + 209, + 475 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 210, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 210, + 478 + ], + "score": 1.0, + "content": "4.4.2 Prompt Analysis", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 483, + 505, + 604 + ], + "lines": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "For Q2, we choose the summarization and data-to-text tasks for analysis where we used all prompts", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 494, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 506, + 507 + ], + "score": 1.0, + "content": "from our prompt set. We first group all the evaluation perspectives into three categories: (1) semantic", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "overlap (informativeness, pyramid score, and relevance) (2) linguistic quality (fluency, coherence)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "(3) factual correctness (factuality). We then calculate the percentage of prompts that result in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 526, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 505, + 541 + ], + "score": 1.0, + "content": "performance improvements for each perspective within a dataset. Finally, we compute the average", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 538, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 505, + 551 + ], + "score": 1.0, + "content": "percentage of prompts that can lead to performance gains for each category. The results are shown", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 548, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 548, + 505, + 561 + ], + "score": 1.0, + "content": "in Tab. 2-(c). We can see that for semantic overlap, almost all prompts can lead to the performance", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 560, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 573 + ], + "score": 1.0, + "content": "increase, while for factuality only a few prompts can improve the performance. This also explains the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 571, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 144, + 583 + ], + "score": 1.0, + "content": "results in", + "type": "text" + }, + { + "bbox": [ + 144, + 571, + 170, + 582 + ], + "score": 0.85, + "content": "\\ S 4 . 3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 571, + 506, + 583 + ], + "score": 1.0, + "content": "where we found that combining the results of different prompts can lead to consistent", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 582, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 594 + ], + "score": 1.0, + "content": "increases in semantic overlap but worse performance in factuality. Regarding linguistic quality, the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 592, + 504, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 474, + 605 + ], + "score": 1.0, + "content": "effect of adding a prompt is not that predictive, which is also consistent with our findings in", + "type": "text" + }, + { + "bbox": [ + 474, + 593, + 500, + 604 + ], + "score": 0.81, + "content": "\\ S 4 . 3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 500, + 592, + 504, + 605 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 483, + 506, + 605 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 616, + 195, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 196, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 196, + 630 + ], + "score": 1.0, + "content": "4.4.3 Bias Analysis", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 107, + 637, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 107, + 637, + 505, + 647 + ], + "score": 1.0, + "content": "To answer Q3, we conduct bias analysis. Bias would indicate that the scores are too high or too low", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "compared to the scores they are given by human annotators. Therefore, to see whether such biases", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 658, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 670 + ], + "score": 1.0, + "content": "exist, we inspected the rank differences given by human annotators and BARTScore (fine-tuned", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 669, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 505, + 680 + ], + "score": 1.0, + "content": "on CNNDM dataset) on the REALSumm dataset where 24 systems are considered, including both", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 679, + 505, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 505, + 692 + ], + "score": 1.0, + "content": "abstractive models and extractive models as well as models based on pre-trained models and models", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 203, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 506, + 216 + ], + "score": 1.0, + "content": "that are trained from scratch. We list all the systems below. And the resulting rank difference is", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 213, + 173, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 173, + 227 + ], + "score": 1.0, + "content": "shown in Fig. 3.", + "type": "text", + "cross_page": true + } + ], + "index": 7 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 637, + 505, + 692 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 69, + 506, + 143 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 69, + 506, + 143 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 69, + 506, + 143 + ], + "spans": [ + { + "bbox": [ + 109, + 69, + 506, + 143 + ], + "score": 0.923, + "type": "image", + "image_path": "71ce21efad222c710fbe5f7a584a046b9f08143d9d25180689982349d95a3ce5.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 69, + 506, + 93.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 93.66666666666667, + 506, + 118.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 118.33333333333334, + 506, + 143.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 147, + 505, + 181 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 147, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 505, + 161 + ], + "score": 1.0, + "content": "Figure 3: Bias analysis of BARTScore. The “Rank Difference\" is the rank obtained using human", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "judgements minus the rank got from BARTScore. Systems beginning with letter “E\" are extractive", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 169, + 398, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 398, + 183 + ], + "score": 1.0, + "content": "systems while systems beginning with letter “A\" are abstractive systems.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 203, + 504, + 226 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 506, + 216 + ], + "score": 1.0, + "content": "that are trained from scratch. We list all the systems below. And the resulting rank difference is", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 213, + 173, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 173, + 227 + ], + "score": 1.0, + "content": "shown in Fig. 3.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 506, + 271 + ], + "lines": [ + { + "bbox": [ + 106, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 506, + 250 + ], + "score": 1.0, + "content": "Extractive Systems E1: BanditSum [11]; E2: Refresh [48]; E3: NeuSum [80]; E4: LSTM-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 249, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 506, + 261 + ], + "score": 1.0, + "content": "PN-RL [79]; E5: BERT-TF-SL [79]; E6: BERT-TF-PN [79]; E7: BERT-LSTM-PN-RL [79]; E8:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 259, + 379, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 379, + 272 + ], + "score": 1.0, + "content": "BERT-LSTM-PN [79]; E9: HeterGraph [68]; E10: MatchSum [78].", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 283, + 505, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 282, + 507, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 507, + 297 + ], + "score": 1.0, + "content": "Abstractive Systems A1: Ptr-Gen [61]; A2: Bottom-up [17]; A3: Fast-Abs-RL [5]; A4: Two-stage-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 295, + 507, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 507, + 306 + ], + "score": 1.0, + "content": "RL [73]; A5: BERT-Ext-Abs [41]; A6: BERT-Abs [41]; A7: Trans-Abs [41]; A8: UniLM-1 [10]; A9:", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 305, + 507, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 507, + 318 + ], + "score": 1.0, + "content": "UniLM-2 [2]; A10: T5-base [55]; A11: T5-large [55]; A12: T5-11B [55]; A13: BART [33]; A14:", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 316, + 165, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 165, + 329 + ], + "score": 1.0, + "content": "SemSim [42].", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 506, + 388 + ], + "lines": [ + { + "bbox": [ + 105, + 331, + 507, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 507, + 346 + ], + "score": 1.0, + "content": "As shown in Fig. 3, BARTScore is less effective at distinguishing the quality of extractive summariza-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 343, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 506, + 357 + ], + "score": 1.0, + "content": "tion systems while much better at distinguishing the quality of abstractive summarization systems.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 354, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 506, + 369 + ], + "score": 1.0, + "content": "However, given that there is a trend for using abstractive systems as more and more pre-trained", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "score": 1.0, + "content": "sequence-to-sequence models being proposed, BARTScore’s weaknesses on extractive systems will", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 377, + 160, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 160, + 389 + ], + "score": 1.0, + "content": "be mitigated.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 106, + 403, + 306, + 417 + ], + "lines": [ + { + "bbox": [ + 105, + 403, + 307, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 307, + 420 + ], + "score": 1.0, + "content": "5 Implications and Future Directions", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 429, + 505, + 483 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 506, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 506, + 442 + ], + "score": 1.0, + "content": "In this paper, we proposed a metric BARTSCORE that formulates evaluation of generated text as a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 439, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 505, + 453 + ], + "score": 1.0, + "content": "text generation task, and empirically demonstrated its efficacy. Without the supervision of human", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 451, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 463 + ], + "score": 1.0, + "content": "judgments, BARTSCORE can effectively evaluate texts from 7 perspectives and achieve the best", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "performance on 16 of 22 settings against existing top-scoring metrics. We highlight potential future", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 472, + 276, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 472, + 276, + 484 + ], + "score": 1.0, + "content": "directions based on what we have learned.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 489, + 505, + 543 + ], + "lines": [ + { + "bbox": [ + 106, + 489, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 505, + 501 + ], + "score": 1.0, + "content": "Prompt-augmented metrics As an easy-to-use but powerful method, prompting [39] has achieved", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 500, + 504, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 504, + 512 + ], + "score": 1.0, + "content": "impressive performance particularly on semantic overlap-based evaluation perspectives. However, its", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 510, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 505, + 524 + ], + "score": 1.0, + "content": "effectiveness in factuality and linguistic quality-based perspectives has not been fully demonstrated", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "in this paper. In the future, more works can explore how to make better use of prompts for these and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 533, + 215, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 215, + 545 + ], + "score": 1.0, + "content": "other evaluation scenarios.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 549, + 505, + 593 + ], + "lines": [ + { + "bbox": [ + 106, + 549, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 505, + 561 + ], + "score": 1.0, + "content": "Co-evolving evaluation metrics and systems BARTSCORE builds the connection between metric", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 560, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 506, + 572 + ], + "score": 1.0, + "content": "design and system design, which allows them to share their technological advances, thereby progress-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 572, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 506, + 583 + ], + "score": 1.0, + "content": "ing together. For example, a better BART-based summarization system may be directly used as a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 582, + 492, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 492, + 594 + ], + "score": 1.0, + "content": "more reliable automated metric for evaluating summaries, and this work makes them connected.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + }, + { + "type": "title", + "bbox": [ + 108, + 609, + 201, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 203, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 203, + 625 + ], + "score": 1.0, + "content": "Acknowledgments", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 634, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 506, + 647 + ], + "score": 1.0, + "content": "The authors would like to thank the anonymous reviewers for their insightful comments and sugges-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 645, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 657 + ], + "score": 1.0, + "content": "tions. The authors also thank Wei Zhao for assisting with reproducing baseline results. This work", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 656, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 506, + 668 + ], + "score": 1.0, + "content": "was supported by the Air Force Research Laboratory under agreement number FA8750-19-2-0200.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "notwithstanding any copyright notation thereon. The views and conclusions contained herein are", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "those of the authors and should not be interpreted as necessarily representing the official policies", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 700, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 507, + 712 + ], + "score": 1.0, + "content": "or endorsements, either expressed or implied, of the Air Force Research Laboratory or the U.S.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 710, + 160, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 160, + 723 + ], + "score": 1.0, + "content": "Government.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 39.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 69, + 506, + 143 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 69, + 506, + 143 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 69, + 506, + 143 + ], + "spans": [ + { + "bbox": [ + 109, + 69, + 506, + 143 + ], + "score": 0.923, + "type": "image", + "image_path": "71ce21efad222c710fbe5f7a584a046b9f08143d9d25180689982349d95a3ce5.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 69, + 506, + 93.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 93.66666666666667, + 506, + 118.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 118.33333333333334, + 506, + 143.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 147, + 505, + 181 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 147, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 505, + 161 + ], + "score": 1.0, + "content": "Figure 3: Bias analysis of BARTScore. The “Rank Difference\" is the rank obtained using human", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "judgements minus the rank got from BARTScore. Systems beginning with letter “E\" are extractive", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 169, + 398, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 398, + 183 + ], + "score": 1.0, + "content": "systems while systems beginning with letter “A\" are abstractive systems.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 203, + 504, + 226 + ], + "lines": [], + "index": 6.5, + "bbox_fs": [ + 105, + 203, + 506, + 227 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 506, + 271 + ], + "lines": [ + { + "bbox": [ + 106, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 506, + 250 + ], + "score": 1.0, + "content": "Extractive Systems E1: BanditSum [11]; E2: Refresh [48]; E3: NeuSum [80]; E4: LSTM-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 249, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 506, + 261 + ], + "score": 1.0, + "content": "PN-RL [79]; E5: BERT-TF-SL [79]; E6: BERT-TF-PN [79]; E7: BERT-LSTM-PN-RL [79]; E8:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 259, + 379, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 379, + 272 + ], + "score": 1.0, + "content": "BERT-LSTM-PN [79]; E9: HeterGraph [68]; E10: MatchSum [78].", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 237, + 506, + 272 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 283, + 505, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 282, + 507, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 507, + 297 + ], + "score": 1.0, + "content": "Abstractive Systems A1: Ptr-Gen [61]; A2: Bottom-up [17]; A3: Fast-Abs-RL [5]; A4: Two-stage-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 295, + 507, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 507, + 306 + ], + "score": 1.0, + "content": "RL [73]; A5: BERT-Ext-Abs [41]; A6: BERT-Abs [41]; A7: Trans-Abs [41]; A8: UniLM-1 [10]; A9:", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 305, + 507, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 507, + 318 + ], + "score": 1.0, + "content": "UniLM-2 [2]; A10: T5-base [55]; A11: T5-large [55]; A12: T5-11B [55]; A13: BART [33]; A14:", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 316, + 165, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 165, + 329 + ], + "score": 1.0, + "content": "SemSim [42].", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 282, + 507, + 329 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 506, + 388 + ], + "lines": [ + { + "bbox": [ + 105, + 331, + 507, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 507, + 346 + ], + "score": 1.0, + "content": "As shown in Fig. 3, BARTScore is less effective at distinguishing the quality of extractive summariza-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 343, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 506, + 357 + ], + "score": 1.0, + "content": "tion systems while much better at distinguishing the quality of abstractive summarization systems.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 354, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 506, + 369 + ], + "score": 1.0, + "content": "However, given that there is a trend for using abstractive systems as more and more pre-trained", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "score": 1.0, + "content": "sequence-to-sequence models being proposed, BARTScore’s weaknesses on extractive systems will", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 377, + 160, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 160, + 389 + ], + "score": 1.0, + "content": "be mitigated.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 331, + 507, + 389 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 403, + 306, + 417 + ], + "lines": [ + { + "bbox": [ + 105, + 403, + 307, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 307, + 420 + ], + "score": 1.0, + "content": "5 Implications and Future Directions", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 429, + 505, + 483 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 506, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 506, + 442 + ], + "score": 1.0, + "content": "In this paper, we proposed a metric BARTSCORE that formulates evaluation of generated text as a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 439, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 505, + 453 + ], + "score": 1.0, + "content": "text generation task, and empirically demonstrated its efficacy. Without the supervision of human", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 451, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 463 + ], + "score": 1.0, + "content": "judgments, BARTSCORE can effectively evaluate texts from 7 perspectives and achieve the best", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "performance on 16 of 22 settings against existing top-scoring metrics. We highlight potential future", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 472, + 276, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 472, + 276, + 484 + ], + "score": 1.0, + "content": "directions based on what we have learned.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 428, + 506, + 484 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 489, + 505, + 543 + ], + "lines": [ + { + "bbox": [ + 106, + 489, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 505, + 501 + ], + "score": 1.0, + "content": "Prompt-augmented metrics As an easy-to-use but powerful method, prompting [39] has achieved", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 500, + 504, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 504, + 512 + ], + "score": 1.0, + "content": "impressive performance particularly on semantic overlap-based evaluation perspectives. However, its", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 510, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 505, + 524 + ], + "score": 1.0, + "content": "effectiveness in factuality and linguistic quality-based perspectives has not been fully demonstrated", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "in this paper. In the future, more works can explore how to make better use of prompts for these and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 533, + 215, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 215, + 545 + ], + "score": 1.0, + "content": "other evaluation scenarios.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 489, + 505, + 545 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 549, + 505, + 593 + ], + "lines": [ + { + "bbox": [ + 106, + 549, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 505, + 561 + ], + "score": 1.0, + "content": "Co-evolving evaluation metrics and systems BARTSCORE builds the connection between metric", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 560, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 506, + 572 + ], + "score": 1.0, + "content": "design and system design, which allows them to share their technological advances, thereby progress-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 572, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 506, + 583 + ], + "score": 1.0, + "content": "ing together. For example, a better BART-based summarization system may be directly used as a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 582, + 492, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 492, + 594 + ], + "score": 1.0, + "content": "more reliable automated metric for evaluating summaries, and this work makes them connected.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5, + "bbox_fs": [ + 106, + 549, + 506, + 594 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 609, + 201, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 203, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 203, + 625 + ], + "score": 1.0, + "content": "Acknowledgments", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 634, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 506, + 647 + ], + "score": 1.0, + "content": "The authors would like to thank the anonymous reviewers for their insightful comments and sugges-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 645, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 657 + ], + "score": 1.0, + "content": "tions. The authors also thank Wei Zhao for assisting with reproducing baseline results. This work", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 656, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 506, + 668 + ], + "score": 1.0, + "content": "was supported by the Air Force Research Laboratory under agreement number FA8750-19-2-0200.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "notwithstanding any copyright notation thereon. The views and conclusions contained herein are", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "those of the authors and should not be interpreted as necessarily representing the official policies", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 700, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 507, + 712 + ], + "score": 1.0, + "content": "or endorsements, either expressed or implied, of the Air Force Research Laboratory or the U.S.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 710, + 160, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 160, + 723 + ], + "score": 1.0, + "content": "Government.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 633, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 70, + 507, + 723 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 109, + 91, + 507, + 107 + ], + "spans": [ + { + "bbox": [ + 109, + 91, + 507, + 107 + ], + "score": 1.0, + "content": "[1] Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 126, + 102, + 507, + 118 + ], + "spans": [ + { + "bbox": [ + 126, + 102, + 507, + 118 + ], + "score": 1.0, + "content": "Learning to retrieve reasoning paths over wikipedia graph for question answering. arXiv preprint", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 126, + 114, + 232, + 127 + ], + "spans": [ + { + "bbox": [ + 126, + 114, + 232, + 127 + ], + "score": 1.0, + "content": "arXiv:1911.10470, 2019.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 109, + 127, + 506, + 144 + ], + "spans": [ + { + "bbox": [ + 109, + 127, + 506, + 144 + ], + "score": 1.0, + "content": "[2] Hangbo Bao, Li Dong, Furu Wei, Wenhui Wang, Nan Yang, Xiaodong Liu, Yu Wang, Jianfeng", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 125, + 138, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 125, + 138, + 506, + 155 + ], + "score": 1.0, + "content": "Gao, Songhao Piao, Ming Zhou, and Hsiao-Wuen Hon. Unilmv2: Pseudo-masked language", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 127, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 127, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "models for unified language model pre-training. In Proceedings of the 37th International", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 126, + 160, + 507, + 176 + ], + "spans": [ + { + "bbox": [ + 126, + 160, + 507, + 176 + ], + "score": 1.0, + "content": "Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, volume 119 of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 171, + 430, + 186 + ], + "spans": [ + { + "bbox": [ + 126, + 171, + 430, + 186 + ], + "score": 1.0, + "content": "Proceedings of Machine Learning Research, pages 642–652. PMLR, 2020.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 110, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 110, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "[3] Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, and Pengfei Liu. Metrics also disagree", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 199, + 506, + 212 + ], + "spans": [ + { + "bbox": [ + 127, + 199, + 506, + 212 + ], + "score": 1.0, + "content": "in the low scoring range: Revisiting summarization evaluation metrics. In Proceedings of the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 209, + 506, + 223 + ], + "spans": [ + { + "bbox": [ + 127, + 209, + 506, + 223 + ], + "score": 1.0, + "content": "28th International Conference on Computational Linguistics, pages 5702–5711, Barcelona,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 220, + 484, + 234 + ], + "spans": [ + { + "bbox": [ + 127, + 220, + 484, + 234 + ], + "score": 1.0, + "content": "Spain (Online), December 2020. International Committee on Computational Linguistics.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 109, + 234, + 507, + 249 + ], + "spans": [ + { + "bbox": [ + 109, + 234, + 507, + 249 + ], + "score": 1.0, + "content": "[4] Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, Pengfei Liu, and Graham Neubig.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 127, + 246, + 506, + 259 + ], + "spans": [ + { + "bbox": [ + 127, + 246, + 506, + 259 + ], + "score": 1.0, + "content": "Re-evaluating evaluation in text summarization. In Proceedings of the 2020 Conference on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 257, + 507, + 270 + ], + "spans": [ + { + "bbox": [ + 127, + 257, + 507, + 270 + ], + "score": 1.0, + "content": "Empirical Methods in Natural Language Processing (EMNLP), pages 9347–9359, Online,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 267, + 372, + 281 + ], + "spans": [ + { + "bbox": [ + 126, + 267, + 372, + 281 + ], + "score": 1.0, + "content": "November 2020. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 110, + 282, + 506, + 296 + ], + "spans": [ + { + "bbox": [ + 110, + 282, + 506, + 296 + ], + "score": 1.0, + "content": "[5] Yen-Chun Chen and Mohit Bansal. Fast abstractive summarization with reinforce-selected", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 127, + 293, + 507, + 307 + ], + "spans": [ + { + "bbox": [ + 127, + 293, + 507, + 307 + ], + "score": 1.0, + "content": "sentence rewriting. In Proceedings of the 56th Annual Meeting of the Association for Computa-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 125, + 303, + 507, + 318 + ], + "spans": [ + { + "bbox": [ + 125, + 303, + 507, + 318 + ], + "score": 1.0, + "content": "tional Linguistics (Volume 1: Long Papers), pages 675–686, Melbourne, Australia, July 2018.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 127, + 315, + 303, + 329 + ], + "spans": [ + { + "bbox": [ + 127, + 315, + 303, + 329 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 109, + 328, + 507, + 345 + ], + "spans": [ + { + "bbox": [ + 109, + 328, + 507, + 345 + ], + "score": 1.0, + "content": "[6] Andrew M Dai and Quoc V Le. Semi-supervised sequence learning. In C. Cortes, N. Lawrence,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 125, + 339, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 125, + 339, + 506, + 356 + ], + "score": 1.0, + "content": "D. Lee, M. Sugiyama, and R. Garnett, editors, Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 127, + 352, + 335, + 366 + ], + "spans": [ + { + "bbox": [ + 127, + 352, + 335, + 366 + ], + "score": 1.0, + "content": "Systems, volume 28. Curran Associates, Inc., 2015.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 109, + 366, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 109, + 366, + 506, + 381 + ], + "score": 1.0, + "content": "[7] Hoa Trang Dang. Overview of duc 2005. In Proceedings of the document understanding", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 127, + 378, + 309, + 392 + ], + "spans": [ + { + "bbox": [ + 127, + 378, + 309, + 392 + ], + "score": 1.0, + "content": "conference, volume 2005, pages 1–12, 2005.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 110, + 392, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 110, + 392, + 506, + 406 + ], + "score": 1.0, + "content": "[8] Michael Denkowski and Alon Lavie. Extending the METEOR machine translation evaluation", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 126, + 402, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 126, + 402, + 506, + 418 + ], + "score": 1.0, + "content": "metric to the phrase level. In Human Language Technologies: The 2010 Annual Conference of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 127, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 127, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "the North American Chapter of the Association for Computational Linguistics, pages 250–253,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 127, + 425, + 449, + 439 + ], + "spans": [ + { + "bbox": [ + 127, + 425, + 449, + 439 + ], + "score": 1.0, + "content": "Los Angeles, California, June 2010. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 109, + 438, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 109, + 438, + 506, + 455 + ], + "score": 1.0, + "content": "[9] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 127, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 127, + 451, + 506, + 465 + ], + "score": 1.0, + "content": "bidirectional transformers for language understanding. In Proceedings of the 2019 Conference", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 127, + 463, + 505, + 475 + ], + "score": 1.0, + "content": "of the North American Chapter of the Association for Computational Linguistics: Human", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 124, + 472, + 470, + 487 + ], + "spans": [ + { + "bbox": [ + 124, + 472, + 470, + 487 + ], + "score": 1.0, + "content": "Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, 2019.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "[10] Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 127, + 498, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 127, + 498, + 506, + 512 + ], + "score": 1.0, + "content": "Zhou, and Hsiao-Wuen Hon. Unified language model pre-training for natural language under-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 509, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 127, + 509, + 506, + 523 + ], + "score": 1.0, + "content": "standing and generation. In Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 520, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 127, + 520, + 506, + 533 + ], + "score": 1.0, + "content": "d’Alché-Buc, Emily B. Fox, and Roman Garnett, editors, Advances in Neural Information", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 127, + 531, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 127, + 531, + 506, + 545 + ], + "score": 1.0, + "content": "Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 542, + 491, + 556 + ], + "spans": [ + { + "bbox": [ + 126, + 542, + 491, + 556 + ], + "score": 1.0, + "content": "NeurIPS 2019, December 8-14, 2019, Vancouver, BC, Canada, pages 13042–13054, 2019.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 558, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 506, + 569 + ], + "score": 1.0, + "content": "[11] Yue Dong, Yikang Shen, Eric Crawford, Herke van Hoof, and Jackie Chi Kit Cheung. Bandit-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 128, + 569, + 504, + 580 + ], + "spans": [ + { + "bbox": [ + 128, + 569, + 504, + 580 + ], + "score": 1.0, + "content": "Sum: Extractive summarization as a contextual bandit. In Proceedings of the 2018 Conference", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 125, + 578, + 507, + 593 + ], + "spans": [ + { + "bbox": [ + 125, + 578, + 507, + 593 + ], + "score": 1.0, + "content": "on Empirical Methods in Natural Language Processing, pages 3739–3748, Brussels, Belgium,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 126, + 589, + 407, + 603 + ], + "spans": [ + { + "bbox": [ + 126, + 589, + 407, + 603 + ], + "score": 1.0, + "content": "October-November 2018. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "[12] Zi-Yi Dou, Pengfei Liu, Hiroaki Hayashi, Zhengbao Jiang, and Graham Neubig. GSum: A", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 126, + 614, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 126, + 614, + 506, + 630 + ], + "score": 1.0, + "content": "general framework for guided neural abstractive summarization. In Meeting of the North", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 125, + 625, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 125, + 625, + 506, + 640 + ], + "score": 1.0, + "content": "American Chapter of the Association for Computational Linguistics (NAACL), Mexico City,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 126, + 637, + 174, + 650 + ], + "spans": [ + { + "bbox": [ + 126, + 637, + 174, + 650 + ], + "score": 1.0, + "content": "June 2021.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 104, + 651, + 507, + 666 + ], + "spans": [ + { + "bbox": [ + 104, + 651, + 507, + 666 + ], + "score": 1.0, + "content": "[13] A. R. Fabbri, Wojciech Kryscinski, Bryan McCann, R. Socher, and Dragomir Radev. Summeval:", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 125, + 662, + 508, + 677 + ], + "spans": [ + { + "bbox": [ + 125, + 662, + 508, + 677 + ], + "score": 1.0, + "content": "Re-evaluating summarization evaluation. Trans. Assoc. Comput. Linguistics, 9:391–409, 2021.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 678, + 507, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 507, + 691 + ], + "score": 1.0, + "content": "[14] Tobias Falke, Leonardo FR Ribeiro, Prasetya Ajie Utama, Ido Dagan, and Iryna Gurevych.", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 127, + 689, + 506, + 703 + ], + "spans": [ + { + "bbox": [ + 127, + 689, + 506, + 703 + ], + "score": 1.0, + "content": "Ranking generated summaries by correctness: An interesting but challenging application for", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 127, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 127, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "natural language inference. In Proceedings of the 57th Annual Meeting of the Association for", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 127, + 711, + 339, + 724 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 339, + 724 + ], + "score": 1.0, + "content": "Computational Linguistics, pages 2214–2220, 2019.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 26.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 310, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 104, + 70, + 507, + 723 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 91, + 507, + 107 + ], + "spans": [ + { + "bbox": [ + 109, + 91, + 507, + 107 + ], + "score": 1.0, + "content": "[1] Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong.", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 102, + 507, + 118 + ], + "spans": [ + { + "bbox": [ + 126, + 102, + 507, + 118 + ], + "score": 1.0, + "content": "Learning to retrieve reasoning paths over wikipedia graph for question answering. arXiv preprint", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 126, + 114, + 232, + 127 + ], + "spans": [ + { + "bbox": [ + 126, + 114, + 232, + 127 + ], + "score": 1.0, + "content": "arXiv:1911.10470, 2019.", + "type": "text" + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 127, + 506, + 144 + ], + "spans": [ + { + "bbox": [ + 109, + 127, + 506, + 144 + ], + "score": 1.0, + "content": "[2] Hangbo Bao, Li Dong, Furu Wei, Wenhui Wang, Nan Yang, Xiaodong Liu, Yu Wang, Jianfeng", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 138, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 125, + 138, + 506, + 155 + ], + "score": 1.0, + "content": "Gao, Songhao Piao, Ming Zhou, and Hsiao-Wuen Hon. Unilmv2: Pseudo-masked language", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 127, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 127, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "models for unified language model pre-training. In Proceedings of the 37th International", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 126, + 160, + 507, + 176 + ], + "spans": [ + { + "bbox": [ + 126, + 160, + 507, + 176 + ], + "score": 1.0, + "content": "Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, volume 119 of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 171, + 430, + 186 + ], + "spans": [ + { + "bbox": [ + 126, + 171, + 430, + 186 + ], + "score": 1.0, + "content": "Proceedings of Machine Learning Research, pages 642–652. PMLR, 2020.", + "type": "text" + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 110, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "[3] Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, and Pengfei Liu. Metrics also disagree", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 199, + 506, + 212 + ], + "spans": [ + { + "bbox": [ + 127, + 199, + 506, + 212 + ], + "score": 1.0, + "content": "in the low scoring range: Revisiting summarization evaluation metrics. In Proceedings of the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 209, + 506, + 223 + ], + "spans": [ + { + "bbox": [ + 127, + 209, + 506, + 223 + ], + "score": 1.0, + "content": "28th International Conference on Computational Linguistics, pages 5702–5711, Barcelona,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 220, + 484, + 234 + ], + "spans": [ + { + "bbox": [ + 127, + 220, + 484, + 234 + ], + "score": 1.0, + "content": "Spain (Online), December 2020. International Committee on Computational Linguistics.", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 234, + 507, + 249 + ], + "spans": [ + { + "bbox": [ + 109, + 234, + 507, + 249 + ], + "score": 1.0, + "content": "[4] Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, Pengfei Liu, and Graham Neubig.", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 246, + 506, + 259 + ], + "spans": [ + { + "bbox": [ + 127, + 246, + 506, + 259 + ], + "score": 1.0, + "content": "Re-evaluating evaluation in text summarization. In Proceedings of the 2020 Conference on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 257, + 507, + 270 + ], + "spans": [ + { + "bbox": [ + 127, + 257, + 507, + 270 + ], + "score": 1.0, + "content": "Empirical Methods in Natural Language Processing (EMNLP), pages 9347–9359, Online,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 267, + 372, + 281 + ], + "spans": [ + { + "bbox": [ + 126, + 267, + 372, + 281 + ], + "score": 1.0, + "content": "November 2020. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 282, + 506, + 296 + ], + "spans": [ + { + "bbox": [ + 110, + 282, + 506, + 296 + ], + "score": 1.0, + "content": "[5] Yen-Chun Chen and Mohit Bansal. Fast abstractive summarization with reinforce-selected", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 293, + 507, + 307 + ], + "spans": [ + { + "bbox": [ + 127, + 293, + 507, + 307 + ], + "score": 1.0, + "content": "sentence rewriting. In Proceedings of the 56th Annual Meeting of the Association for Computa-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 125, + 303, + 507, + 318 + ], + "spans": [ + { + "bbox": [ + 125, + 303, + 507, + 318 + ], + "score": 1.0, + "content": "tional Linguistics (Volume 1: Long Papers), pages 675–686, Melbourne, Australia, July 2018.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 127, + 315, + 303, + 329 + ], + "spans": [ + { + "bbox": [ + 127, + 315, + 303, + 329 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 328, + 507, + 345 + ], + "spans": [ + { + "bbox": [ + 109, + 328, + 507, + 345 + ], + "score": 1.0, + "content": "[6] Andrew M Dai and Quoc V Le. Semi-supervised sequence learning. In C. Cortes, N. Lawrence,", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 339, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 125, + 339, + 506, + 356 + ], + "score": 1.0, + "content": "D. Lee, M. Sugiyama, and R. Garnett, editors, Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 127, + 352, + 335, + 366 + ], + "spans": [ + { + "bbox": [ + 127, + 352, + 335, + 366 + ], + "score": 1.0, + "content": "Systems, volume 28. Curran Associates, Inc., 2015.", + "type": "text" + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 366, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 109, + 366, + 506, + 381 + ], + "score": 1.0, + "content": "[7] Hoa Trang Dang. Overview of duc 2005. In Proceedings of the document understanding", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 378, + 309, + 392 + ], + "spans": [ + { + "bbox": [ + 127, + 378, + 309, + 392 + ], + "score": 1.0, + "content": "conference, volume 2005, pages 1–12, 2005.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 392, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 110, + 392, + 506, + 406 + ], + "score": 1.0, + "content": "[8] Michael Denkowski and Alon Lavie. Extending the METEOR machine translation evaluation", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 402, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 126, + 402, + 506, + 418 + ], + "score": 1.0, + "content": "metric to the phrase level. In Human Language Technologies: The 2010 Annual Conference of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 127, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 127, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "the North American Chapter of the Association for Computational Linguistics, pages 250–253,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 127, + 425, + 449, + 439 + ], + "spans": [ + { + "bbox": [ + 127, + 425, + 449, + 439 + ], + "score": 1.0, + "content": "Los Angeles, California, June 2010. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 438, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 109, + 438, + 506, + 455 + ], + "score": 1.0, + "content": "[9] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 127, + 451, + 506, + 465 + ], + "score": 1.0, + "content": "bidirectional transformers for language understanding. In Proceedings of the 2019 Conference", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 127, + 463, + 505, + 475 + ], + "score": 1.0, + "content": "of the North American Chapter of the Association for Computational Linguistics: Human", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 124, + 472, + 470, + 487 + ], + "spans": [ + { + "bbox": [ + 124, + 472, + 470, + 487 + ], + "score": 1.0, + "content": "Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, 2019.", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "[10] Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 498, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 127, + 498, + 506, + 512 + ], + "score": 1.0, + "content": "Zhou, and Hsiao-Wuen Hon. Unified language model pre-training for natural language under-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 509, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 127, + 509, + 506, + 523 + ], + "score": 1.0, + "content": "standing and generation. In Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 520, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 127, + 520, + 506, + 533 + ], + "score": 1.0, + "content": "d’Alché-Buc, Emily B. Fox, and Roman Garnett, editors, Advances in Neural Information", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 127, + 531, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 127, + 531, + 506, + 545 + ], + "score": 1.0, + "content": "Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 542, + 491, + 556 + ], + "spans": [ + { + "bbox": [ + 126, + 542, + 491, + 556 + ], + "score": 1.0, + "content": "NeurIPS 2019, December 8-14, 2019, Vancouver, BC, Canada, pages 13042–13054, 2019.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 558, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 506, + 569 + ], + "score": 1.0, + "content": "[11] Yue Dong, Yikang Shen, Eric Crawford, Herke van Hoof, and Jackie Chi Kit Cheung. Bandit-", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 569, + 504, + 580 + ], + "spans": [ + { + "bbox": [ + 128, + 569, + 504, + 580 + ], + "score": 1.0, + "content": "Sum: Extractive summarization as a contextual bandit. In Proceedings of the 2018 Conference", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 125, + 578, + 507, + 593 + ], + "spans": [ + { + "bbox": [ + 125, + 578, + 507, + 593 + ], + "score": 1.0, + "content": "on Empirical Methods in Natural Language Processing, pages 3739–3748, Brussels, Belgium,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 126, + 589, + 407, + 603 + ], + "spans": [ + { + "bbox": [ + 126, + 589, + 407, + 603 + ], + "score": 1.0, + "content": "October-November 2018. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "[12] Zi-Yi Dou, Pengfei Liu, Hiroaki Hayashi, Zhengbao Jiang, and Graham Neubig. GSum: A", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 614, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 126, + 614, + 506, + 630 + ], + "score": 1.0, + "content": "general framework for guided neural abstractive summarization. In Meeting of the North", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 125, + 625, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 125, + 625, + 506, + 640 + ], + "score": 1.0, + "content": "American Chapter of the Association for Computational Linguistics (NAACL), Mexico City,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 126, + 637, + 174, + 650 + ], + "spans": [ + { + "bbox": [ + 126, + 637, + 174, + 650 + ], + "score": 1.0, + "content": "June 2021.", + "type": "text" + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 651, + 507, + 666 + ], + "spans": [ + { + "bbox": [ + 104, + 651, + 507, + 666 + ], + "score": 1.0, + "content": "[13] A. R. Fabbri, Wojciech Kryscinski, Bryan McCann, R. Socher, and Dragomir Radev. Summeval:", + "type": "text" + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 662, + 508, + 677 + ], + "spans": [ + { + "bbox": [ + 125, + 662, + 508, + 677 + ], + "score": 1.0, + "content": "Re-evaluating summarization evaluation. Trans. Assoc. Comput. Linguistics, 9:391–409, 2021.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 678, + 507, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 507, + 691 + ], + "score": 1.0, + "content": "[14] Tobias Falke, Leonardo FR Ribeiro, Prasetya Ajie Utama, Ido Dagan, and Iryna Gurevych.", + "type": "text" + } + ], + "index": 50, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 689, + 506, + 703 + ], + "spans": [ + { + "bbox": [ + 127, + 689, + 506, + 703 + ], + "score": 1.0, + "content": "Ranking generated summaries by correctness: An interesting but challenging application for", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 127, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 127, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "natural language inference. In Proceedings of the 57th Annual Meeting of the Association for", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 127, + 711, + 339, + 724 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 339, + 724 + ], + "score": 1.0, + "content": "Computational Linguistics, pages 2214–2220, 2019.", + "type": "text" + } + ], + "index": 53, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 72, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 507, + 87 + ], + "score": 1.0, + "content": "[15] David Freedman, Robert Pisani, and Roger Purves. Statistics (international student edition).", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 84, + 406, + 96 + ], + "spans": [ + { + "bbox": [ + 127, + 84, + 406, + 96 + ], + "score": 1.0, + "content": "Pisani, R. Purves, 4th edn. WW Norton & Company, New York, 2007.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 97, + 507, + 111 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 507, + 111 + ], + "score": 1.0, + "content": "[16] Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. Bottom-up abstractive summarization.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 107, + 508, + 124 + ], + "spans": [ + { + "bbox": [ + 125, + 107, + 508, + 124 + ], + "score": 1.0, + "content": "In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing,", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 118, + 230, + 134 + ], + "spans": [ + { + "bbox": [ + 126, + 118, + 230, + 134 + ], + "score": 1.0, + "content": "pages 4098–4109, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 134, + 507, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 507, + 148 + ], + "score": 1.0, + "content": "[17] Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. Bottom-up abstractive summarization.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 144, + 507, + 160 + ], + "spans": [ + { + "bbox": [ + 126, + 144, + 507, + 160 + ], + "score": 1.0, + "content": "In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing,", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 126, + 156, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 126, + 156, + 506, + 169 + ], + "score": 1.0, + "content": "pages 4098–4109, Brussels, Belgium, October-November 2018. Association for Computational", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 165, + 180, + 182 + ], + "spans": [ + { + "bbox": [ + 126, + 165, + 180, + 182 + ], + "score": 1.0, + "content": "Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 181, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 506, + 195 + ], + "score": 1.0, + "content": "[18] Max Grusky, Mor Naaman, and Yoav Artzi. Newsroom: A dataset of 1.3 million summaries", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 191, + 506, + 206 + ], + "spans": [ + { + "bbox": [ + 127, + 191, + 506, + 206 + ], + "score": 1.0, + "content": "with diverse extractive strategies. In Proceedings of the 2018 Conference of the North American", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 204, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 127, + 204, + 506, + 216 + ], + "score": 1.0, + "content": "Chapter of the Association for Computational Linguistics: Human Language Technologies,", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 127, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "Volume 1 (Long Papers), pages 708–719, New Orleans, Louisiana, June 2018. Association for", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 126, + 223, + 241, + 240 + ], + "spans": [ + { + "bbox": [ + 126, + 223, + 241, + 240 + ], + "score": 1.0, + "content": "Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 238, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 506, + 253 + ], + "score": 1.0, + "content": "[19] Max Grusky, Mor Naaman, and Yoav Artzi. Newsroom: A dataset of 1.3 million summaries", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 249, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 127, + 249, + 506, + 264 + ], + "score": 1.0, + "content": "with diverse extractive strategies. In Proceedings of the 2018 Conference of the North American", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 260, + 507, + 275 + ], + "spans": [ + { + "bbox": [ + 126, + 260, + 507, + 275 + ], + "score": 1.0, + "content": "Chapter of the Association for Computational Linguistics: Human Language Technologies,", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 127, + 271, + 360, + 286 + ], + "spans": [ + { + "bbox": [ + 127, + 271, + 360, + 286 + ], + "score": 1.0, + "content": "Volume 1 (Long Papers), volume 1, pages 708–719, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 284, + 507, + 302 + ], + "spans": [ + { + "bbox": [ + 104, + 284, + 507, + 302 + ], + "score": 1.0, + "content": "[20] K. Hermann, Tomás Kociský, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suley-", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 297, + 461, + 310 + ], + "spans": [ + { + "bbox": [ + 125, + 297, + 461, + 310 + ], + "score": 1.0, + "content": "man, and P. Blunsom. Teaching machines to read and comprehend. In NIPS, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "score": 1.0, + "content": "[21] Tsutomu Hirao, Manabu Okumura, Norihito Yasuda, and Hideki Isozaki. Supervised auto-", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 127, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "matic evaluation for summarization with voted regression model. Information Processing &", + "type": "text", + "cross_page": true + } + ], + "index": 21 + }, + { + "bbox": [ + 125, + 333, + 284, + 347 + ], + "spans": [ + { + "bbox": [ + 125, + 333, + 284, + 347 + ], + "score": 1.0, + "content": "Management, 43(6):1521–1535, 2007.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 348, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 506, + 362 + ], + "score": 1.0, + "content": "[22] J. Edward Hu, Abhinav Singh, Nils Holzenberger, Matt Post, and Benjamin Van Durme. Large-", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 358, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 127, + 358, + 483, + 373 + ], + "score": 1.0, + "content": "scale, diverse, paraphrastic bitexts via sampling and clustering. In Proceedings of the", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 483, + 359, + 505, + 370 + ], + "score": 0.33, + "content": "2 3 r d", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 24 + }, + { + "bbox": [ + 125, + 367, + 507, + 386 + ], + "spans": [ + { + "bbox": [ + 125, + 367, + 507, + 386 + ], + "score": 1.0, + "content": "Conference on Computational Natural Language Learning (CoNLL), pages 44–54, Hong Kong,", + "type": "text", + "cross_page": true + } + ], + "index": 25 + }, + { + "bbox": [ + 127, + 381, + 401, + 393 + ], + "spans": [ + { + "bbox": [ + 127, + 381, + 401, + 393 + ], + "score": 1.0, + "content": "China, November 2019. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 394, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 506, + 409 + ], + "score": 1.0, + "content": "[23] Sarthak Jain, Madeleine van Zuylen, Hannaneh Hajishirzi, and Iz Beltagy. Scirex: A challenge", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 406, + 495, + 419 + ], + "spans": [ + { + "bbox": [ + 126, + 406, + 495, + 419 + ], + "score": 1.0, + "content": "dataset for document-level information extraction. arXiv preprint arXiv:2005.00512, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "[24] Zhengbao Jiang, Frank F. Xu, Jun Araki, and Graham Neubig. How can we know what language", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 431, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 126, + 431, + 506, + 446 + ], + "score": 1.0, + "content": "models know? Transactions of the Association for Computational Linguistics, 8:423–438,", + "type": "text", + "cross_page": true + } + ], + "index": 30 + }, + { + "bbox": [ + 126, + 441, + 155, + 456 + ], + "spans": [ + { + "bbox": [ + 126, + 441, + 155, + 456 + ], + "score": 1.0, + "content": "2020.", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 456, + 507, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 507, + 471 + ], + "score": 1.0, + "content": "[25] Karen Sparck Jones. A statistical interpretation of term specificity and its application in retrieval.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 467, + 261, + 482 + ], + "spans": [ + { + "bbox": [ + 126, + 467, + 261, + 482 + ], + "score": 1.0, + "content": "Journal of documentation, 1972.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 482, + 507, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 507, + 496 + ], + "score": 1.0, + "content": "[26] Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, ˘", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 491, + 507, + 509 + ], + "spans": [ + { + "bbox": [ + 125, + 491, + 507, + 509 + ], + "score": 1.0, + "content": "Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering.", + "type": "text", + "cross_page": true + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 504, + 291, + 516 + ], + "spans": [ + { + "bbox": [ + 126, + 504, + 291, + 516 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2004.04906, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 516, + 473, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 473, + 533 + ], + "score": 1.0, + "content": "[27] M. G. Kendall. A new measure of rank correlation. Biometrika, 30(1/2):81–93, 1938.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 531, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 547 + ], + "score": 1.0, + "content": "[28] Philipp Koehn. Statistical significance tests for machine translation evaluation. In Proceedings", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 542, + 507, + 559 + ], + "spans": [ + { + "bbox": [ + 126, + 542, + 507, + 559 + ], + "score": 1.0, + "content": "of the 2004 Conference on Empirical Methods in Natural Language Processing, pages 388–395,", + "type": "text", + "cross_page": true + } + ], + "index": 39 + }, + { + "bbox": [ + 127, + 555, + 421, + 569 + ], + "spans": [ + { + "bbox": [ + 127, + 555, + 421, + 569 + ], + "score": 1.0, + "content": "Barcelona, Spain, July 2004. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 568, + 461, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 461, + 582 + ], + "score": 1.0, + "content": "[29] Philipp Koehn. Statistical machine translation. Cambridge University Press, 2009.", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 583, + 506, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 506, + 597 + ], + "score": 1.0, + "content": "[30] Wojciech Kryscinski, Bryan McCann, Caiming Xiong, and Richard Socher. Evaluating the", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 594, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 127, + 594, + 506, + 609 + ], + "score": 1.0, + "content": "factual consistency of abstractive text summarization. In Proceedings of the 2020 Conference", + "type": "text", + "cross_page": true + } + ], + "index": 43 + }, + { + "bbox": [ + 127, + 605, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 127, + 605, + 506, + 619 + ], + "score": 1.0, + "content": "on Empirical Methods in Natural Language Processing (EMNLP), pages 9332–9346, Online,", + "type": "text", + "cross_page": true + } + ], + "index": 44 + }, + { + "bbox": [ + 125, + 615, + 372, + 630 + ], + "spans": [ + { + "bbox": [ + 125, + 615, + 372, + 630 + ], + "score": 1.0, + "content": "November 2020. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 630, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 506, + 645 + ], + "score": 1.0, + "content": "[31] Teven Le Scao and Alexander Rush. How many data points is a prompt worth? In Proceedings", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 642, + 505, + 654 + ], + "spans": [ + { + "bbox": [ + 127, + 642, + 505, + 654 + ], + "score": 1.0, + "content": "of the 2021 Conference of the North American Chapter of the Association for Computational", + "type": "text", + "cross_page": true + } + ], + "index": 47 + }, + { + "bbox": [ + 127, + 652, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 127, + 652, + 506, + 667 + ], + "score": 1.0, + "content": "Linguistics: Human Language Technologies, pages 2627–2636, Online, June 2021. Association", + "type": "text", + "cross_page": true + } + ], + "index": 48 + }, + { + "bbox": [ + 125, + 662, + 254, + 677 + ], + "spans": [ + { + "bbox": [ + 125, + 662, + 254, + 677 + ], + "score": 1.0, + "content": "for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 677, + 507, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 507, + 692 + ], + "score": 1.0, + "content": "[32] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed,", + "type": "text", + "cross_page": true + } + ], + "index": 50, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 127, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Omer Levy, Ves Stoyanov, and Luke Zettlemoyer. Bart: Denoising sequence-to-sequence", + "type": "text", + "cross_page": true + } + ], + "index": 51 + }, + { + "bbox": [ + 127, + 699, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 127, + 699, + 506, + 714 + ], + "score": 1.0, + "content": "pre-training for natural language generation, translation, and comprehension. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 52 + }, + { + "bbox": [ + 127, + 711, + 230, + 723 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 230, + 723 + ], + "score": 1.0, + "content": "arXiv:1910.13461, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 53, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "[33] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 83, + 507, + 98 + ], + "spans": [ + { + "bbox": [ + 126, + 83, + 507, + 98 + ], + "score": 1.0, + "content": "Levy, Veselin Stoyanov, and Luke Zettlemoyer. BART: Denoising sequence-to-sequence pre-", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 127, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 127, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "training for natural language generation, translation, and comprehension. In Proceedings of", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 105, + 506, + 120 + ], + "spans": [ + { + "bbox": [ + 127, + 105, + 506, + 120 + ], + "score": 1.0, + "content": "the 58th Annual Meeting of the Association for Computational Linguistics, pages 7871–7880,", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 127, + 115, + 380, + 129 + ], + "spans": [ + { + "bbox": [ + 127, + 115, + 380, + 129 + ], + "score": 1.0, + "content": "Online, July 2020. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 130, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 506, + 145 + ], + "score": 1.0, + "content": "[34] Xiang Lisa Li and Percy Liang. Prefix-tuning: Optimizing continuous prompts for generation.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 141, + 251, + 155 + ], + "spans": [ + { + "bbox": [ + 126, + 141, + 251, + 155 + ], + "score": 1.0, + "content": "ArXiv, abs/2101.00190, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 154, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 107, + 154, + 506, + 169 + ], + "score": 1.0, + "content": "[35] Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. Text Summarization", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 167, + 213, + 179 + ], + "spans": [ + { + "bbox": [ + 127, + 167, + 213, + 179 + ], + "score": 1.0, + "content": "Branches Out, 2004.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 180, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 180, + 506, + 195 + ], + "score": 1.0, + "content": "[36] Chin-Yew Lin and Eduard Hovy. Automatic evaluation of summaries using n-gram co-", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 191, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 127, + 191, + 505, + 205 + ], + "score": 1.0, + "content": "occurrence statistics. In Proceedings of the 2003 Human Language Technology Conference of", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 202, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 127, + 202, + 506, + 217 + ], + "score": 1.0, + "content": "the North American Chapter of the Association for Computational Linguistics, pages 150–157,", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 212, + 155, + 227 + ], + "spans": [ + { + "bbox": [ + 126, + 212, + 155, + 227 + ], + "score": 1.0, + "content": "2003.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 227, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 506, + 241 + ], + "score": 1.0, + "content": "[37] Ying Lin, Heng Ji, Fei Huang, and Lingfei Wu. A joint neural model for information extrac-", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 238, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 127, + 238, + 506, + 252 + ], + "score": 1.0, + "content": "tion with global features. In Proceedings of the 58th Annual Meeting of the Association for", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 248, + 339, + 263 + ], + "spans": [ + { + "bbox": [ + 127, + 248, + 339, + 263 + ], + "score": 1.0, + "content": "Computational Linguistics, pages 7999–8009, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 263, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 506, + 277 + ], + "score": 1.0, + "content": "[38] Pengfei Liu, Jinlan Fu, Yang Xiao, Weizhe Yuan, Shuaicheng Chang, Junqi Dai, Yixin Liu,", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 127, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "Zihuiwen Ye, and Graham Neubig. Explainaboard: An explainable leaderboard for nlp. arXiv", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 125, + 285, + 268, + 298 + ], + "spans": [ + { + "bbox": [ + 125, + 285, + 268, + 298 + ], + "score": 1.0, + "content": "preprint arXiv:2104.06387, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 298, + 507, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 507, + 314 + ], + "score": 1.0, + "content": "[39] Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 309, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 126, + 309, + 506, + 325 + ], + "score": 1.0, + "content": "Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language", + "type": "text", + "cross_page": true + } + ], + "index": 20 + }, + { + "bbox": [ + 125, + 321, + 201, + 334 + ], + "spans": [ + { + "bbox": [ + 125, + 321, + 201, + 334 + ], + "score": 1.0, + "content": "processing, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 334, + 506, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 506, + 350 + ], + "score": 1.0, + "content": "[40] Yang Liu and Mirella Lapata. Text summarization with pretrained encoders. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 345, + 232, + 360 + ], + "spans": [ + { + "bbox": [ + 126, + 345, + 232, + 360 + ], + "score": 1.0, + "content": "arXiv:1908.08345, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 359, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 104, + 359, + 506, + 375 + ], + "score": 1.0, + "content": "[41] Yang Liu and Mirella Lapata. Text summarization with pretrained encoders. In Proceedings", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 371, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 127, + 371, + 505, + 385 + ], + "score": 1.0, + "content": "of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th", + "type": "text", + "cross_page": true + } + ], + "index": 25 + }, + { + "bbox": [ + 126, + 381, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 126, + 381, + 506, + 397 + ], + "score": 1.0, + "content": "International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages", + "type": "text", + "cross_page": true + } + ], + "index": 26 + }, + { + "bbox": [ + 127, + 394, + 502, + 406 + ], + "spans": [ + { + "bbox": [ + 127, + 394, + 502, + 406 + ], + "score": 1.0, + "content": "3730–3740, Hong Kong, China, November 2019. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 406, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 506, + 421 + ], + "score": 1.0, + "content": "[42] Yixin Liu and Pengfei Liu. Simcls: A simple framework for contrastive learning of abstractive", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 418, + 356, + 432 + ], + "spans": [ + { + "bbox": [ + 126, + 418, + 356, + 432 + ], + "score": 1.0, + "content": "summarization. arXiv preprint arXiv:2106.01890, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 432, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 445 + ], + "score": 1.0, + "content": "[43] Qingsong Ma, Ondˇrej Bojar, and Yvette Graham. Results of the WMT18 metrics shared", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 443, + 506, + 457 + ], + "spans": [ + { + "bbox": [ + 127, + 443, + 506, + 457 + ], + "score": 1.0, + "content": "task: Both characters and embeddings achieve good performance. In Proceedings of the Third", + "type": "text", + "cross_page": true + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 454, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 127, + 454, + 506, + 469 + ], + "score": 1.0, + "content": "Conference on Machine Translation: Shared Task Papers, pages 671–688, Belgium, Brussels,", + "type": "text", + "cross_page": true + } + ], + "index": 32 + }, + { + "bbox": [ + 127, + 465, + 362, + 477 + ], + "spans": [ + { + "bbox": [ + 127, + 465, + 362, + 477 + ], + "score": 1.0, + "content": "October 2018. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 478, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 104, + 478, + 506, + 493 + ], + "score": 1.0, + "content": "[44] Qingsong Ma, Johnny Wei, Ondˇrej Bojar, and Yvette Graham. Results of the WMT19 metrics", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 127, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "shared task: Segment-level and strong MT systems pose big challenges. In Proceedings of", + "type": "text", + "cross_page": true + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 501, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 127, + 501, + 506, + 515 + ], + "score": 1.0, + "content": "the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages", + "type": "text", + "cross_page": true + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 512, + 452, + 526 + ], + "spans": [ + { + "bbox": [ + 127, + 512, + 452, + 526 + ], + "score": 1.0, + "content": "62–90, Florence, Italy, August 2019. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 525, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 107, + 525, + 506, + 540 + ], + "score": 1.0, + "content": "[45] François Mairesse, Milica Gasic, Filip Jurcicek, Simon Keizer, Blaise Thomson, Kai Yu, and", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 537, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 127, + 537, + 506, + 551 + ], + "score": 1.0, + "content": "Steve Young. Phrase-based statistical language generation using graphical models and active", + "type": "text", + "cross_page": true + } + ], + "index": 39 + }, + { + "bbox": [ + 127, + 548, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 127, + 548, + 506, + 562 + ], + "score": 1.0, + "content": "learning. In Proceedings of the 48th Annual Meeting of the Association for Computational", + "type": "text", + "cross_page": true + } + ], + "index": 40 + }, + { + "bbox": [ + 126, + 559, + 277, + 572 + ], + "spans": [ + { + "bbox": [ + 126, + 559, + 277, + 572 + ], + "score": 1.0, + "content": "Linguistics, pages 1552–1561, 2010.", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 572, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 506, + 587 + ], + "score": 1.0, + "content": "[46] Nitika Mathur, Timothy Baldwin, and Trevor Cohn. Tangled up in BLEU: Reevaluating the", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 583, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 127, + 583, + 506, + 598 + ], + "score": 1.0, + "content": "evaluation of automatic machine translation evaluation metrics. In Proceedings of the 58th", + "type": "text", + "cross_page": true + } + ], + "index": 43 + }, + { + "bbox": [ + 126, + 595, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 126, + 595, + 506, + 609 + ], + "score": 1.0, + "content": "Annual Meeting of the Association for Computational Linguistics, pages 4984–4997, Online,", + "type": "text", + "cross_page": true + } + ], + "index": 44 + }, + { + "bbox": [ + 127, + 605, + 348, + 620 + ], + "spans": [ + { + "bbox": [ + 127, + 605, + 348, + 620 + ], + "score": 1.0, + "content": "July 2020. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 621, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 107, + 621, + 505, + 633 + ], + "score": 1.0, + "content": "[47] Shashi Narayan, Shay B Cohen, and Mirella Lapata. Don’t give me the details, just the summary!", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 631, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 127, + 631, + 506, + 645 + ], + "score": 1.0, + "content": "topic-aware convolutional neural networks for extreme summarization. In Proceedings of the", + "type": "text", + "cross_page": true + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 640, + 507, + 656 + ], + "spans": [ + { + "bbox": [ + 126, + 640, + 507, + 656 + ], + "score": 1.0, + "content": "2018 Conference on Empirical Methods in Natural Language Processing, pages 1797–1807,", + "type": "text", + "cross_page": true + } + ], + "index": 48 + }, + { + "bbox": [ + 127, + 653, + 155, + 665 + ], + "spans": [ + { + "bbox": [ + 127, + 653, + 155, + 665 + ], + "score": 1.0, + "content": "2018.", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 666, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 506, + 680 + ], + "score": 1.0, + "content": "[48] Shashi Narayan, Shay B. Cohen, and Mirella Lapata. Ranking sentences for extractive sum-", + "type": "text", + "cross_page": true + } + ], + "index": 50, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 677, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 127, + 677, + 506, + 692 + ], + "score": 1.0, + "content": "marization with reinforcement learning. In Proceedings of the 2018 Conference of the North", + "type": "text", + "cross_page": true + } + ], + "index": 51 + }, + { + "bbox": [ + 126, + 690, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 126, + 690, + 506, + 702 + ], + "score": 1.0, + "content": "American Chapter of the Association for Computational Linguistics: Human Language Tech-", + "type": "text", + "cross_page": true + } + ], + "index": 52 + }, + { + "bbox": [ + 127, + 699, + 507, + 714 + ], + "spans": [ + { + "bbox": [ + 127, + 699, + 507, + 714 + ], + "score": 1.0, + "content": "nologies, Volume 1 (Long Papers), pages 1747–1759, New Orleans, Louisiana, June 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 53 + }, + { + "bbox": [ + 127, + 712, + 302, + 723 + ], + "spans": [ + { + "bbox": [ + 127, + 712, + 302, + 723 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 54, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "[49] Ani Nenkova and Rebecca Passonneau. Evaluating content selection in summarization: The", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 84, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 126, + 84, + 506, + 96 + ], + "score": 1.0, + "content": "pyramid method. In Proceedings of the Human Language Technology Conference of the North", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 125, + 93, + 507, + 109 + ], + "spans": [ + { + "bbox": [ + 125, + 93, + 507, + 109 + ], + "score": 1.0, + "content": "American Chapter of the Association for Computational Linguistics: HLT-NAACL 2004, pages", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 106, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 127, + 106, + 505, + 118 + ], + "score": 1.0, + "content": "145–152, Boston, Massachusetts, USA, May 2 - May 7 2004. Association for Computational", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 117, + 179, + 130 + ], + "spans": [ + { + "bbox": [ + 126, + 117, + 179, + 130 + ], + "score": 1.0, + "content": "Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "score": 1.0, + "content": "[50] Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 142, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 126, + 142, + 506, + 156 + ], + "score": 1.0, + "content": "evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 127, + 154, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 127, + 154, + 506, + 165 + ], + "score": 1.0, + "content": "for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA, July 2002.", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 127, + 164, + 303, + 176 + ], + "spans": [ + { + "bbox": [ + 127, + 164, + 303, + 176 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 178, + 506, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 506, + 192 + ], + "score": 1.0, + "content": "[51] Maxime Peyrard, Teresa Botschen, and Iryna Gurevych. Learning to score system summaries", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 189, + 506, + 203 + ], + "spans": [ + { + "bbox": [ + 127, + 189, + 506, + 203 + ], + "score": 1.0, + "content": "for better content selection evaluation. In Proceedings of the Workshop on New Frontiers", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 200, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 126, + 200, + 506, + 214 + ], + "score": 1.0, + "content": "in Summarization, pages 74–84, Copenhagen, Denmark, September 2017. Association for", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 210, + 240, + 225 + ], + "spans": [ + { + "bbox": [ + 126, + 210, + 240, + 225 + ], + "score": 1.0, + "content": "Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 224, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 104, + 224, + 506, + 240 + ], + "score": 1.0, + "content": "[52] Maja Popovic. chrF: character n-gram F-score for automatic MT evaluation. In ´ Proceedings", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 235, + 507, + 251 + ], + "spans": [ + { + "bbox": [ + 126, + 235, + 507, + 251 + ], + "score": 1.0, + "content": "of the Tenth Workshop on Statistical Machine Translation, pages 392–395, Lisbon, Portugal,", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 247, + 373, + 261 + ], + "spans": [ + { + "bbox": [ + 127, + 247, + 373, + 261 + ], + "score": 1.0, + "content": "September 2015. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 262, + 507, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 507, + 276 + ], + "score": 1.0, + "content": "[53] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever.", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 273, + 459, + 287 + ], + "spans": [ + { + "bbox": [ + 126, + 273, + 459, + 287 + ], + "score": 1.0, + "content": "Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 288, + 507, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 507, + 301 + ], + "score": 1.0, + "content": "[54] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena,", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 298, + 506, + 312 + ], + "spans": [ + { + "bbox": [ + 128, + 298, + 506, + 312 + ], + "score": 1.0, + "content": "Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified", + "type": "text", + "cross_page": true + } + ], + "index": 19 + }, + { + "bbox": [ + 127, + 309, + 473, + 323 + ], + "spans": [ + { + "bbox": [ + 127, + 309, + 473, + 323 + ], + "score": 1.0, + "content": "text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 324, + 507, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 507, + 338 + ], + "score": 1.0, + "content": "[55] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena,", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 335, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 128, + 335, + 505, + 349 + ], + "score": 1.0, + "content": "Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified", + "type": "text", + "cross_page": true + } + ], + "index": 22 + }, + { + "bbox": [ + 126, + 345, + 412, + 358 + ], + "spans": [ + { + "bbox": [ + 126, + 345, + 412, + 358 + ], + "score": 1.0, + "content": "text-to-text transformer. J. Mach. Learn. Res., 21:140:1–140:67, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "[56] Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. COMET: A neural framework", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 371, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 127, + 371, + 506, + 385 + ], + "score": 1.0, + "content": "for MT evaluation. In Proceedings of the 2020 Conference on Empirical Methods in Natural", + "type": "text", + "cross_page": true + } + ], + "index": 25 + }, + { + "bbox": [ + 126, + 382, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 126, + 382, + 505, + 396 + ], + "score": 1.0, + "content": "Language Processing (EMNLP), pages 2685–2702, Online, November 2020. Association for", + "type": "text", + "cross_page": true + } + ], + "index": 26 + }, + { + "bbox": [ + 126, + 392, + 240, + 407 + ], + "spans": [ + { + "bbox": [ + 126, + 392, + 240, + 407 + ], + "score": 1.0, + "content": "Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 407, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 506, + 423 + ], + "score": 1.0, + "content": "[57] Laria Reynolds and Kyle McDonell. Prompt programming for large language models: Beyond", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 419, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 128, + 419, + 505, + 431 + ], + "score": 1.0, + "content": "the few-shot paradigm. In Extended Abstracts of the 2021 CHI Conference on Human Factors", + "type": "text", + "cross_page": true + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 429, + 291, + 444 + ], + "spans": [ + { + "bbox": [ + 125, + 429, + 291, + 444 + ], + "score": 1.0, + "content": "in Computing Systems, pages 1–7, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 444, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 458 + ], + "score": 1.0, + "content": "[58] Timo Schick and Hinrich Schütze. Exploiting cloze questions for few-shot text classification", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 455, + 418, + 469 + ], + "spans": [ + { + "bbox": [ + 127, + 455, + 418, + 469 + ], + "score": 1.0, + "content": "and natural language inference. arXiv preprint arXiv:2001.07676, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 469, + 507, + 486 + ], + "spans": [ + { + "bbox": [ + 104, + 469, + 507, + 486 + ], + "score": 1.0, + "content": "[59] Timo Schick and Hinrich Schütze. Few-shot text generation with pattern-exploiting training.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 480, + 290, + 494 + ], + "spans": [ + { + "bbox": [ + 127, + 480, + 290, + 494 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2012.11926, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 495, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 506, + 509 + ], + "score": 1.0, + "content": "[60] Timo Schick and Hinrich Schütze. It’s not just size that matters: Small language models are", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 507, + 384, + 520 + ], + "spans": [ + { + "bbox": [ + 126, + 507, + 384, + 520 + ], + "score": 1.0, + "content": "also few-shot learners. arXiv preprint arXiv:2009.07118, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 521, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 535 + ], + "score": 1.0, + "content": "[61] Abigail See, Peter J. Liu, and Christopher D. Manning. Get to the point: Summarization with", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 127, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "pointer-generator networks. In Proceedings of the 55th Annual Meeting of the Association for", + "type": "text", + "cross_page": true + } + ], + "index": 38 + }, + { + "bbox": [ + 127, + 543, + 507, + 557 + ], + "spans": [ + { + "bbox": [ + 127, + 543, + 507, + 557 + ], + "score": 1.0, + "content": "Computational Linguistics (Volume 1: Long Papers), pages 1073–1083, Vancouver, Canada,", + "type": "text", + "cross_page": true + } + ], + "index": 39 + }, + { + "bbox": [ + 127, + 555, + 346, + 566 + ], + "spans": [ + { + "bbox": [ + 127, + 555, + 346, + 566 + ], + "score": 1.0, + "content": "July 2017. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 569, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 569, + 505, + 581 + ], + "score": 1.0, + "content": "[62] Thibault Sellam, Dipanjan Das, and Ankur Parikh. BLEURT: Learning robust metrics for text", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 579, + 506, + 593 + ], + "spans": [ + { + "bbox": [ + 127, + 579, + 506, + 593 + ], + "score": 1.0, + "content": "generation. In Proceedings of the 58th Annual Meeting of the Association for Computational", + "type": "text", + "cross_page": true + } + ], + "index": 42 + }, + { + "bbox": [ + 126, + 590, + 502, + 604 + ], + "spans": [ + { + "bbox": [ + 126, + 590, + 502, + 604 + ], + "score": 1.0, + "content": "Linguistics, pages 7881–7892, Online, July 2020. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 605, + 507, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 507, + 618 + ], + "score": 1.0, + "content": "[63] Taylor Shin, Yasaman Razeghi, Robert L Logan IV, Eric Wallace, and Sameer Singh. Auto-", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 616, + 507, + 630 + ], + "spans": [ + { + "bbox": [ + 126, + 616, + 507, + 630 + ], + "score": 1.0, + "content": "prompt: Eliciting knowledge from language models with automatically generated prompts.", + "type": "text", + "cross_page": true + } + ], + "index": 45 + }, + { + "bbox": [ + 126, + 626, + 291, + 641 + ], + "spans": [ + { + "bbox": [ + 126, + 626, + 291, + 641 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2010.15980, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 641, + 507, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 507, + 655 + ], + "score": 1.0, + "content": "[64] Miloš Stanojevic and Khalil Sima’an. BEER: BEtter evaluation as ranking. In ´ Proceedings of", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 653, + 507, + 666 + ], + "spans": [ + { + "bbox": [ + 127, + 653, + 507, + 666 + ], + "score": 1.0, + "content": "the Ninth Workshop on Statistical Machine Translation, pages 414–419, Baltimore, Maryland,", + "type": "text", + "cross_page": true + } + ], + "index": 48 + }, + { + "bbox": [ + 126, + 662, + 374, + 677 + ], + "spans": [ + { + "bbox": [ + 126, + 662, + 374, + 677 + ], + "score": 1.0, + "content": "USA, June 2014. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "[65] Brian Thompson and Matt Post. Automatic machine translation evaluation in many languages", + "type": "text", + "cross_page": true + } + ], + "index": 50, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 689, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 127, + 689, + 506, + 702 + ], + "score": 1.0, + "content": "via zero-shot paraphrasing. In Proceedings of the 2020 Conference on Empirical Methods in", + "type": "text", + "cross_page": true + } + ], + "index": 51 + }, + { + "bbox": [ + 126, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 126, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "Natural Language Processing (EMNLP), pages 90–121, Online, November 2020. Association", + "type": "text", + "cross_page": true + } + ], + "index": 52 + }, + { + "bbox": [ + 126, + 709, + 254, + 724 + ], + "spans": [ + { + "bbox": [ + 126, + 709, + 254, + 724 + ], + "score": 1.0, + "content": "for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 53, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 70, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 104, + 70, + 507, + 87 + ], + "score": 1.0, + "content": "[66] Alex Wang, Kyunghyun Cho, and M. Lewis. Asking and answering questions to evaluate the", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 83, + 326, + 97 + ], + "spans": [ + { + "bbox": [ + 126, + 83, + 326, + 97 + ], + "score": 1.0, + "content": "factual consistency of summaries. In ACL, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 97, + 508, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 508, + 113 + ], + "score": 1.0, + "content": "[67] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel Bowman.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 106, + 508, + 124 + ], + "spans": [ + { + "bbox": [ + 126, + 106, + 508, + 124 + ], + "score": 1.0, + "content": "GLUE: A multi-task benchmark and analysis platform for natural language understanding.", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 125, + 118, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 125, + 118, + 506, + 135 + ], + "score": 1.0, + "content": "In Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 126, + 131, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 126, + 131, + 506, + 145 + ], + "score": 1.0, + "content": "Neural Networks for NLP, pages 353–355, Brussels, Belgium, November 2018. Association for", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 127, + 142, + 240, + 156 + ], + "spans": [ + { + "bbox": [ + 127, + 142, + 240, + 156 + ], + "score": 1.0, + "content": "Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 156, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 506, + 171 + ], + "score": 1.0, + "content": "[68] Danqing Wang, Pengfei Liu, Yining Zheng, Xipeng Qiu, and Xuanjing Huang. Heterogeneous", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 167, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 126, + 167, + 506, + 182 + ], + "score": 1.0, + "content": "graph neural networks for extractive document summarization. In Proceedings of the 58th", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 126, + 178, + 506, + 192 + ], + "spans": [ + { + "bbox": [ + 126, + 178, + 506, + 192 + ], + "score": 1.0, + "content": "Annual Meeting of the Association for Computational Linguistics, pages 6209–6219, Online,", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 125, + 188, + 348, + 204 + ], + "spans": [ + { + "bbox": [ + 125, + 188, + 348, + 204 + ], + "score": 1.0, + "content": "July 2020. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 202, + 507, + 219 + ], + "spans": [ + { + "bbox": [ + 107, + 202, + 507, + 219 + ], + "score": 1.0, + "content": "[69] Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Pei-Hao Su, David Vandyke, and Steve Young.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 215, + 508, + 231 + ], + "spans": [ + { + "bbox": [ + 126, + 215, + 508, + 231 + ], + "score": 1.0, + "content": "Semantically conditioned lstm-based natural language generation for spoken dialogue systems.", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 127, + 226, + 291, + 240 + ], + "spans": [ + { + "bbox": [ + 127, + 226, + 291, + 240 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1508.01745, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 239, + 506, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 506, + 256 + ], + "score": 1.0, + "content": "[70] Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 251, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 127, + 251, + 506, + 265 + ], + "score": 1.0, + "content": "Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer,", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 127, + 262, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 127, + 262, + 505, + 276 + ], + "score": 1.0, + "content": "Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 126, + 273, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 126, + 273, + 506, + 288 + ], + "score": 1.0, + "content": "Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. Transformers: State-of-the-art", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 126, + 285, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 126, + 285, + 506, + 299 + ], + "score": 1.0, + "content": "natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 126, + 296, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 126, + 296, + 506, + 309 + ], + "score": 1.0, + "content": "Natural Language Processing: System Demonstrations, pages 38–45, Online, October 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 19 + }, + { + "bbox": [ + 126, + 307, + 303, + 321 + ], + "spans": [ + { + "bbox": [ + 126, + 307, + 303, + 321 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "score": 1.0, + "content": "[71] Weizhe Yuan, Pengfei Liu, and Graham Neubig. Can we automate scientific reviewing? arXiv", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 333, + 266, + 345 + ], + "spans": [ + { + "bbox": [ + 126, + 333, + 266, + 345 + ], + "score": 1.0, + "content": "preprint arXiv:2102.00176, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 347, + 454, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 454, + 361 + ], + "score": 1.0, + "content": "[72] Jerrold H Zar. Spearman rank correlation. Encyclopedia of Biostatistics, 7, 2005.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 360, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 104, + 360, + 506, + 378 + ], + "score": 1.0, + "content": "[73] Haoyu Zhang, Jingjing Cai, Jianjun Xu, and Ji Wang. Pretraining-based natural language", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 372, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 126, + 372, + 506, + 388 + ], + "score": 1.0, + "content": "generation for text summarization. In Proceedings of the 23rd Conference on Computational", + "type": "text", + "cross_page": true + } + ], + "index": 25 + }, + { + "bbox": [ + 127, + 384, + 507, + 398 + ], + "spans": [ + { + "bbox": [ + 127, + 384, + 507, + 398 + ], + "score": 1.0, + "content": "Natural Language Learning (CoNLL), pages 789–797, Hong Kong, China, November 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 26 + }, + { + "bbox": [ + 128, + 396, + 302, + 407 + ], + "spans": [ + { + "bbox": [ + 128, + 396, + 302, + 407 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 410, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 506, + 424 + ], + "score": 1.0, + "content": "[74] Jingqing Zhang, Yao Zhao, Mohammad Saleh, and Peter Liu. Pegasus: Pre-training with", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 126, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "extracted gap-sentences for abstractive summarization. In International Conference on Machine", + "type": "text", + "cross_page": true + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 430, + 312, + 446 + ], + "spans": [ + { + "bbox": [ + 125, + 430, + 312, + 446 + ], + "score": 1.0, + "content": "Learning, pages 11328–11339. PMLR, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 447, + 507, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 507, + 461 + ], + "score": 1.0, + "content": "[75] Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. Bertscore:", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 457, + 507, + 471 + ], + "spans": [ + { + "bbox": [ + 127, + 457, + 507, + 471 + ], + "score": 1.0, + "content": "Evaluating text generation with bert. In International Conference on Learning Representations,", + "type": "text", + "cross_page": true + } + ], + "index": 32 + }, + { + "bbox": [ + 126, + 467, + 154, + 482 + ], + "spans": [ + { + "bbox": [ + 126, + 467, + 154, + 482 + ], + "score": 1.0, + "content": "2020.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 483, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 506, + 497 + ], + "score": 1.0, + "content": "[76] Wei Zhao, Maxime Peyrard, Fei Liu, Yang Gao, Christian M. Meyer, and Steffen Eger. Mover-", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 494, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 127, + 494, + 506, + 507 + ], + "score": 1.0, + "content": "Score: Text generation evaluating with contextualized embeddings and earth mover distance. In", + "type": "text", + "cross_page": true + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 504, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 126, + 504, + 506, + 519 + ], + "score": 1.0, + "content": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing", + "type": "text", + "cross_page": true + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 516, + 507, + 530 + ], + "spans": [ + { + "bbox": [ + 127, + 516, + 507, + 530 + ], + "score": 1.0, + "content": "and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP),", + "type": "text", + "cross_page": true + } + ], + "index": 37 + }, + { + "bbox": [ + 125, + 526, + 509, + 542 + ], + "spans": [ + { + "bbox": [ + 125, + 526, + 509, + 542 + ], + "score": 1.0, + "content": "pages 563–578, Hong Kong, China, November 2019. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 540, + 507, + 557 + ], + "spans": [ + { + "bbox": [ + 104, + 540, + 507, + 557 + ], + "score": 1.0, + "content": "[77] Ming Zhong, Pengfei Liu, Yiran Chen, Danqing Wang, Xipeng Qiu, and Xuanjing Huang.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 552, + 467, + 567 + ], + "spans": [ + { + "bbox": [ + 126, + 552, + 467, + 567 + ], + "score": 1.0, + "content": "Extractive summarization as text matching. arXiv preprint arXiv:2004.08795, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 566, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 104, + 566, + 506, + 583 + ], + "score": 1.0, + "content": "[78] Ming Zhong, Pengfei Liu, Yiran Chen, Danqing Wang, Xipeng Qiu, and Xuanjing Huang.", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 578, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 127, + 578, + 506, + 592 + ], + "score": 1.0, + "content": "Extractive summarization as text matching. In Proceedings of the 58th Annual Meeting of the", + "type": "text", + "cross_page": true + } + ], + "index": 42 + }, + { + "bbox": [ + 126, + 589, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 126, + 589, + 506, + 603 + ], + "score": 1.0, + "content": "Association for Computational Linguistics, pages 6197–6208, Online, July 2020. Association", + "type": "text", + "cross_page": true + } + ], + "index": 43 + }, + { + "bbox": [ + 126, + 599, + 254, + 614 + ], + "spans": [ + { + "bbox": [ + 126, + 599, + 254, + 614 + ], + "score": 1.0, + "content": "for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 614, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 104, + 614, + 506, + 629 + ], + "score": 1.0, + "content": "[79] Ming Zhong, Pengfei Liu, Danqing Wang, Xipeng Qiu, and Xuanjing Huang. Searching for", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 624, + 507, + 641 + ], + "spans": [ + { + "bbox": [ + 126, + 624, + 507, + 641 + ], + "score": 1.0, + "content": "effective neural extractive summarization: What works and what’s next. In Proceedings of", + "type": "text", + "cross_page": true + } + ], + "index": 46 + }, + { + "bbox": [ + 127, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 127, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "the 57th Annual Meeting of the Association for Computational Linguistics, pages 1049–1058,", + "type": "text", + "cross_page": true + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 648, + 409, + 662 + ], + "spans": [ + { + "bbox": [ + 126, + 648, + 409, + 662 + ], + "score": 1.0, + "content": "Florence, Italy, July 2019. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 662, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 676 + ], + "score": 1.0, + "content": "[80] Qingyu Zhou, Nan Yang, Furu Wei, Shaohan Huang, Ming Zhou, and Tiejun Zhao. Neural", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 673, + 506, + 687 + ], + "spans": [ + { + "bbox": [ + 127, + 673, + 506, + 687 + ], + "score": 1.0, + "content": "document summarization by jointly learning to score and select sentences. In Proceedings of the", + "type": "text", + "cross_page": true + } + ], + "index": 50 + }, + { + "bbox": [ + 126, + 682, + 507, + 700 + ], + "spans": [ + { + "bbox": [ + 126, + 682, + 507, + 700 + ], + "score": 1.0, + "content": "56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),", + "type": "text", + "cross_page": true + } + ], + "index": 51 + }, + { + "bbox": [ + 125, + 695, + 503, + 710 + ], + "spans": [ + { + "bbox": [ + 125, + 695, + 503, + 710 + ], + "score": 1.0, + "content": "pages 654–663, Melbourne, Australia, July 2018. Association for Computational Linguistics.", + "type": "text", + "cross_page": true + } + ], + "index": 52 + } + ], + "index": 26.5, + "bbox_fs": [ + 104, + 70, + 508, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 34, + 507, + 729 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 507, + 87 + ], + "score": 1.0, + "content": "[15] David Freedman, Robert Pisani, and Roger Purves. Statistics (international student edition).", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 127, + 84, + 406, + 96 + ], + "spans": [ + { + "bbox": [ + 127, + 84, + 406, + 96 + ], + "score": 1.0, + "content": "Pisani, R. Purves, 4th edn. WW Norton & Company, New York, 2007.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 97, + 507, + 111 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 507, + 111 + ], + "score": 1.0, + "content": "[16] Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. Bottom-up abstractive summarization.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 125, + 107, + 508, + 124 + ], + "spans": [ + { + "bbox": [ + 125, + 107, + 508, + 124 + ], + "score": 1.0, + "content": "In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 118, + 230, + 134 + ], + "spans": [ + { + "bbox": [ + 126, + 118, + 230, + 134 + ], + "score": 1.0, + "content": "pages 4098–4109, 2018.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 134, + 507, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 507, + 148 + ], + "score": 1.0, + "content": "[17] Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. Bottom-up abstractive summarization.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 126, + 144, + 507, + 160 + ], + "spans": [ + { + "bbox": [ + 126, + 144, + 507, + 160 + ], + "score": 1.0, + "content": "In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 126, + 156, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 126, + 156, + 506, + 169 + ], + "score": 1.0, + "content": "pages 4098–4109, Brussels, Belgium, October-November 2018. Association for Computational", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 165, + 180, + 182 + ], + "spans": [ + { + "bbox": [ + 126, + 165, + 180, + 182 + ], + "score": 1.0, + "content": "Linguistics.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 506, + 195 + ], + "score": 1.0, + "content": "[18] Max Grusky, Mor Naaman, and Yoav Artzi. Newsroom: A dataset of 1.3 million summaries", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 191, + 506, + 206 + ], + "spans": [ + { + "bbox": [ + 127, + 191, + 506, + 206 + ], + "score": 1.0, + "content": "with diverse extractive strategies. In Proceedings of the 2018 Conference of the North American", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 204, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 127, + 204, + 506, + 216 + ], + "score": 1.0, + "content": "Chapter of the Association for Computational Linguistics: Human Language Technologies,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 127, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "Volume 1 (Long Papers), pages 708–719, New Orleans, Louisiana, June 2018. Association for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 126, + 223, + 241, + 240 + ], + "spans": [ + { + "bbox": [ + 126, + 223, + 241, + 240 + ], + "score": 1.0, + "content": "Computational Linguistics.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 238, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 506, + 253 + ], + "score": 1.0, + "content": "[19] Max Grusky, Mor Naaman, and Yoav Artzi. Newsroom: A dataset of 1.3 million summaries", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 249, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 127, + 249, + 506, + 264 + ], + "score": 1.0, + "content": "with diverse extractive strategies. In Proceedings of the 2018 Conference of the North American", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 260, + 507, + 275 + ], + "spans": [ + { + "bbox": [ + 126, + 260, + 507, + 275 + ], + "score": 1.0, + "content": "Chapter of the Association for Computational Linguistics: Human Language Technologies,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 127, + 271, + 360, + 286 + ], + "spans": [ + { + "bbox": [ + 127, + 271, + 360, + 286 + ], + "score": 1.0, + "content": "Volume 1 (Long Papers), volume 1, pages 708–719, 2018.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 284, + 507, + 302 + ], + "spans": [ + { + "bbox": [ + 104, + 284, + 507, + 302 + ], + "score": 1.0, + "content": "[20] K. Hermann, Tomás Kociský, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suley-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 125, + 297, + 461, + 310 + ], + "spans": [ + { + "bbox": [ + 125, + 297, + 461, + 310 + ], + "score": 1.0, + "content": "man, and P. Blunsom. Teaching machines to read and comprehend. In NIPS, 2015.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "score": 1.0, + "content": "[21] Tsutomu Hirao, Manabu Okumura, Norihito Yasuda, and Hideki Isozaki. Supervised auto-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 127, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 127, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "matic evaluation for summarization with voted regression model. Information Processing &", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 125, + 333, + 284, + 347 + ], + "spans": [ + { + "bbox": [ + 125, + 333, + 284, + 347 + ], + "score": 1.0, + "content": "Management, 43(6):1521–1535, 2007.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 348, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 506, + 362 + ], + "score": 1.0, + "content": "[22] J. Edward Hu, Abhinav Singh, Nils Holzenberger, Matt Post, and Benjamin Van Durme. Large-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 127, + 358, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 127, + 358, + 483, + 373 + ], + "score": 1.0, + "content": "scale, diverse, paraphrastic bitexts via sampling and clustering. In Proceedings of the", + "type": "text" + }, + { + "bbox": [ + 483, + 359, + 505, + 370 + ], + "score": 0.33, + "content": "2 3 r d", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 125, + 367, + 507, + 386 + ], + "spans": [ + { + "bbox": [ + 125, + 367, + 507, + 386 + ], + "score": 1.0, + "content": "Conference on Computational Natural Language Learning (CoNLL), pages 44–54, Hong Kong,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 127, + 381, + 401, + 393 + ], + "spans": [ + { + "bbox": [ + 127, + 381, + 401, + 393 + ], + "score": 1.0, + "content": "China, November 2019. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 394, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 506, + 409 + ], + "score": 1.0, + "content": "[23] Sarthak Jain, Madeleine van Zuylen, Hannaneh Hajishirzi, and Iz Beltagy. Scirex: A challenge", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 126, + 406, + 495, + 419 + ], + "spans": [ + { + "bbox": [ + 126, + 406, + 495, + 419 + ], + "score": 1.0, + "content": "dataset for document-level information extraction. arXiv preprint arXiv:2005.00512, 2020.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "[24] Zhengbao Jiang, Frank F. Xu, Jun Araki, and Graham Neubig. How can we know what language", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 126, + 431, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 126, + 431, + 506, + 446 + ], + "score": 1.0, + "content": "models know? Transactions of the Association for Computational Linguistics, 8:423–438,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 126, + 441, + 155, + 456 + ], + "spans": [ + { + "bbox": [ + 126, + 441, + 155, + 456 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 456, + 507, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 507, + 471 + ], + "score": 1.0, + "content": "[25] Karen Sparck Jones. A statistical interpretation of term specificity and its application in retrieval.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 126, + 467, + 261, + 482 + ], + "spans": [ + { + "bbox": [ + 126, + 467, + 261, + 482 + ], + "score": 1.0, + "content": "Journal of documentation, 1972.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 482, + 507, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 507, + 496 + ], + "score": 1.0, + "content": "[26] Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, ˘", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 125, + 491, + 507, + 509 + ], + "spans": [ + { + "bbox": [ + 125, + 491, + 507, + 509 + ], + "score": 1.0, + "content": "Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 504, + 291, + 516 + ], + "spans": [ + { + "bbox": [ + 126, + 504, + 291, + 516 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2004.04906, 2020.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 516, + 473, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 473, + 533 + ], + "score": 1.0, + "content": "[27] M. G. Kendall. A new measure of rank correlation. Biometrika, 30(1/2):81–93, 1938.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 531, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 547 + ], + "score": 1.0, + "content": "[28] Philipp Koehn. Statistical significance tests for machine translation evaluation. In Proceedings", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 542, + 507, + 559 + ], + "spans": [ + { + "bbox": [ + 126, + 542, + 507, + 559 + ], + "score": 1.0, + "content": "of the 2004 Conference on Empirical Methods in Natural Language Processing, pages 388–395,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 127, + 555, + 421, + 569 + ], + "spans": [ + { + "bbox": [ + 127, + 555, + 421, + 569 + ], + "score": 1.0, + "content": "Barcelona, Spain, July 2004. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 568, + 461, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 461, + 582 + ], + "score": 1.0, + "content": "[29] Philipp Koehn. Statistical machine translation. Cambridge University Press, 2009.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 583, + 506, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 506, + 597 + ], + "score": 1.0, + "content": "[30] Wojciech Kryscinski, Bryan McCann, Caiming Xiong, and Richard Socher. Evaluating the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 594, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 127, + 594, + 506, + 609 + ], + "score": 1.0, + "content": "factual consistency of abstractive text summarization. In Proceedings of the 2020 Conference", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 127, + 605, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 127, + 605, + 506, + 619 + ], + "score": 1.0, + "content": "on Empirical Methods in Natural Language Processing (EMNLP), pages 9332–9346, Online,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 125, + 615, + 372, + 630 + ], + "spans": [ + { + "bbox": [ + 125, + 615, + 372, + 630 + ], + "score": 1.0, + "content": "November 2020. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 630, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 506, + 645 + ], + "score": 1.0, + "content": "[31] Teven Le Scao and Alexander Rush. How many data points is a prompt worth? In Proceedings", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 127, + 642, + 505, + 654 + ], + "spans": [ + { + "bbox": [ + 127, + 642, + 505, + 654 + ], + "score": 1.0, + "content": "of the 2021 Conference of the North American Chapter of the Association for Computational", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 127, + 652, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 127, + 652, + 506, + 667 + ], + "score": 1.0, + "content": "Linguistics: Human Language Technologies, pages 2627–2636, Online, June 2021. Association", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 125, + 662, + 254, + 677 + ], + "spans": [ + { + "bbox": [ + 125, + 662, + 254, + 677 + ], + "score": 1.0, + "content": "for Computational Linguistics.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 677, + 507, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 507, + 692 + ], + "score": 1.0, + "content": "[32] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed,", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 127, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 127, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Omer Levy, Ves Stoyanov, and Luke Zettlemoyer. Bart: Denoising sequence-to-sequence", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 127, + 699, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 127, + 699, + 506, + 714 + ], + "score": 1.0, + "content": "pre-training for natural language generation, translation, and comprehension. arXiv preprint", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 127, + 711, + 230, + 723 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 230, + 723 + ], + "score": 1.0, + "content": "arXiv:1910.13461, 2019.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 26.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 34, + 507, + 729 + ], + "lines": [], + "index": 26.5, + "bbox_fs": [ + 104, + 72, + 508, + 723 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 36, + 507, + 726 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "[33] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 126, + 83, + 507, + 98 + ], + "spans": [ + { + "bbox": [ + 126, + 83, + 507, + 98 + ], + "score": 1.0, + "content": "Levy, Veselin Stoyanov, and Luke Zettlemoyer. BART: Denoising sequence-to-sequence pre-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 127, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 127, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "training for natural language generation, translation, and comprehension. In Proceedings of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 105, + 506, + 120 + ], + "spans": [ + { + "bbox": [ + 127, + 105, + 506, + 120 + ], + "score": 1.0, + "content": "the 58th Annual Meeting of the Association for Computational Linguistics, pages 7871–7880,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 127, + 115, + 380, + 129 + ], + "spans": [ + { + "bbox": [ + 127, + 115, + 380, + 129 + ], + "score": 1.0, + "content": "Online, July 2020. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 130, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 506, + 145 + ], + "score": 1.0, + "content": "[34] Xiang Lisa Li and Percy Liang. Prefix-tuning: Optimizing continuous prompts for generation.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 126, + 141, + 251, + 155 + ], + "spans": [ + { + "bbox": [ + 126, + 141, + 251, + 155 + ], + "score": 1.0, + "content": "ArXiv, abs/2101.00190, 2021.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 107, + 154, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 107, + 154, + 506, + 169 + ], + "score": 1.0, + "content": "[35] Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. Text Summarization", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 127, + 167, + 213, + 179 + ], + "spans": [ + { + "bbox": [ + 127, + 167, + 213, + 179 + ], + "score": 1.0, + "content": "Branches Out, 2004.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 180, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 180, + 506, + 195 + ], + "score": 1.0, + "content": "[36] Chin-Yew Lin and Eduard Hovy. Automatic evaluation of summaries using n-gram co-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 191, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 127, + 191, + 505, + 205 + ], + "score": 1.0, + "content": "occurrence statistics. In Proceedings of the 2003 Human Language Technology Conference of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 202, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 127, + 202, + 506, + 217 + ], + "score": 1.0, + "content": "the North American Chapter of the Association for Computational Linguistics, pages 150–157,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 212, + 155, + 227 + ], + "spans": [ + { + "bbox": [ + 126, + 212, + 155, + 227 + ], + "score": 1.0, + "content": "2003.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 227, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 506, + 241 + ], + "score": 1.0, + "content": "[37] Ying Lin, Heng Ji, Fei Huang, and Lingfei Wu. A joint neural model for information extrac-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 127, + 238, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 127, + 238, + 506, + 252 + ], + "score": 1.0, + "content": "tion with global features. In Proceedings of the 58th Annual Meeting of the Association for", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 248, + 339, + 263 + ], + "spans": [ + { + "bbox": [ + 127, + 248, + 339, + 263 + ], + "score": 1.0, + "content": "Computational Linguistics, pages 7999–8009, 2020.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 506, + 277 + ], + "score": 1.0, + "content": "[38] Pengfei Liu, Jinlan Fu, Yang Xiao, Weizhe Yuan, Shuaicheng Chang, Junqi Dai, Yixin Liu,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 127, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 127, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "Zihuiwen Ye, and Graham Neubig. Explainaboard: An explainable leaderboard for nlp. arXiv", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 125, + 285, + 268, + 298 + ], + "spans": [ + { + "bbox": [ + 125, + 285, + 268, + 298 + ], + "score": 1.0, + "content": "preprint arXiv:2104.06387, 2021.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 298, + 507, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 507, + 314 + ], + "score": 1.0, + "content": "[39] Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 126, + 309, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 126, + 309, + 506, + 325 + ], + "score": 1.0, + "content": "Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 125, + 321, + 201, + 334 + ], + "spans": [ + { + "bbox": [ + 125, + 321, + 201, + 334 + ], + "score": 1.0, + "content": "processing, 2021.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 334, + 506, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 506, + 350 + ], + "score": 1.0, + "content": "[40] Yang Liu and Mirella Lapata. Text summarization with pretrained encoders. arXiv preprint", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 126, + 345, + 232, + 360 + ], + "spans": [ + { + "bbox": [ + 126, + 345, + 232, + 360 + ], + "score": 1.0, + "content": "arXiv:1908.08345, 2019.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 359, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 104, + 359, + 506, + 375 + ], + "score": 1.0, + "content": "[41] Yang Liu and Mirella Lapata. Text summarization with pretrained encoders. In Proceedings", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 127, + 371, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 127, + 371, + 505, + 385 + ], + "score": 1.0, + "content": "of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 126, + 381, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 126, + 381, + 506, + 397 + ], + "score": 1.0, + "content": "International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 127, + 394, + 502, + 406 + ], + "spans": [ + { + "bbox": [ + 127, + 394, + 502, + 406 + ], + "score": 1.0, + "content": "3730–3740, Hong Kong, China, November 2019. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 406, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 506, + 421 + ], + "score": 1.0, + "content": "[42] Yixin Liu and Pengfei Liu. Simcls: A simple framework for contrastive learning of abstractive", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 126, + 418, + 356, + 432 + ], + "spans": [ + { + "bbox": [ + 126, + 418, + 356, + 432 + ], + "score": 1.0, + "content": "summarization. arXiv preprint arXiv:2106.01890, 2021.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 432, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 445 + ], + "score": 1.0, + "content": "[43] Qingsong Ma, Ondˇrej Bojar, and Yvette Graham. Results of the WMT18 metrics shared", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 127, + 443, + 506, + 457 + ], + "spans": [ + { + "bbox": [ + 127, + 443, + 506, + 457 + ], + "score": 1.0, + "content": "task: Both characters and embeddings achieve good performance. In Proceedings of the Third", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 454, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 127, + 454, + 506, + 469 + ], + "score": 1.0, + "content": "Conference on Machine Translation: Shared Task Papers, pages 671–688, Belgium, Brussels,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 127, + 465, + 362, + 477 + ], + "spans": [ + { + "bbox": [ + 127, + 465, + 362, + 477 + ], + "score": 1.0, + "content": "October 2018. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 478, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 104, + 478, + 506, + 493 + ], + "score": 1.0, + "content": "[44] Qingsong Ma, Johnny Wei, Ondˇrej Bojar, and Yvette Graham. Results of the WMT19 metrics", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 127, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 127, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "shared task: Segment-level and strong MT systems pose big challenges. In Proceedings of", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 501, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 127, + 501, + 506, + 515 + ], + "score": 1.0, + "content": "the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 512, + 452, + 526 + ], + "spans": [ + { + "bbox": [ + 127, + 512, + 452, + 526 + ], + "score": 1.0, + "content": "62–90, Florence, Italy, August 2019. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 107, + 525, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 107, + 525, + 506, + 540 + ], + "score": 1.0, + "content": "[45] François Mairesse, Milica Gasic, Filip Jurcicek, Simon Keizer, Blaise Thomson, Kai Yu, and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 127, + 537, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 127, + 537, + 506, + 551 + ], + "score": 1.0, + "content": "Steve Young. Phrase-based statistical language generation using graphical models and active", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 127, + 548, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 127, + 548, + 506, + 562 + ], + "score": 1.0, + "content": "learning. In Proceedings of the 48th Annual Meeting of the Association for Computational", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 126, + 559, + 277, + 572 + ], + "spans": [ + { + "bbox": [ + 126, + 559, + 277, + 572 + ], + "score": 1.0, + "content": "Linguistics, pages 1552–1561, 2010.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 572, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 506, + 587 + ], + "score": 1.0, + "content": "[46] Nitika Mathur, Timothy Baldwin, and Trevor Cohn. Tangled up in BLEU: Reevaluating the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 583, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 127, + 583, + 506, + 598 + ], + "score": 1.0, + "content": "evaluation of automatic machine translation evaluation metrics. In Proceedings of the 58th", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 126, + 595, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 126, + 595, + 506, + 609 + ], + "score": 1.0, + "content": "Annual Meeting of the Association for Computational Linguistics, pages 4984–4997, Online,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 127, + 605, + 348, + 620 + ], + "spans": [ + { + "bbox": [ + 127, + 605, + 348, + 620 + ], + "score": 1.0, + "content": "July 2020. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 107, + 621, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 107, + 621, + 505, + 633 + ], + "score": 1.0, + "content": "[47] Shashi Narayan, Shay B Cohen, and Mirella Lapata. Don’t give me the details, just the summary!", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 127, + 631, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 127, + 631, + 506, + 645 + ], + "score": 1.0, + "content": "topic-aware convolutional neural networks for extreme summarization. In Proceedings of the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 640, + 507, + 656 + ], + "spans": [ + { + "bbox": [ + 126, + 640, + 507, + 656 + ], + "score": 1.0, + "content": "2018 Conference on Empirical Methods in Natural Language Processing, pages 1797–1807,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 127, + 653, + 155, + 665 + ], + "spans": [ + { + "bbox": [ + 127, + 653, + 155, + 665 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 666, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 506, + 680 + ], + "score": 1.0, + "content": "[48] Shashi Narayan, Shay B. Cohen, and Mirella Lapata. Ranking sentences for extractive sum-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 127, + 677, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 127, + 677, + 506, + 692 + ], + "score": 1.0, + "content": "marization with reinforcement learning. In Proceedings of the 2018 Conference of the North", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 126, + 690, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 126, + 690, + 506, + 702 + ], + "score": 1.0, + "content": "American Chapter of the Association for Computational Linguistics: Human Language Tech-", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 127, + 699, + 507, + 714 + ], + "spans": [ + { + "bbox": [ + 127, + 699, + 507, + 714 + ], + "score": 1.0, + "content": "nologies, Volume 1 (Long Papers), pages 1747–1759, New Orleans, Louisiana, June 2018.", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 127, + 712, + 302, + 723 + ], + "spans": [ + { + "bbox": [ + 127, + 712, + 302, + 723 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 27 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 36, + 507, + 726 + ], + "lines": [], + "index": 27, + "bbox_fs": [ + 104, + 72, + 507, + 723 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 68, + 507, + 727 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "[49] Ani Nenkova and Rebecca Passonneau. Evaluating content selection in summarization: The", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 126, + 84, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 126, + 84, + 506, + 96 + ], + "score": 1.0, + "content": "pyramid method. In Proceedings of the Human Language Technology Conference of the North", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 125, + 93, + 507, + 109 + ], + "spans": [ + { + "bbox": [ + 125, + 93, + 507, + 109 + ], + "score": 1.0, + "content": "American Chapter of the Association for Computational Linguistics: HLT-NAACL 2004, pages", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 106, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 127, + 106, + 505, + 118 + ], + "score": 1.0, + "content": "145–152, Boston, Massachusetts, USA, May 2 - May 7 2004. Association for Computational", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 117, + 179, + 130 + ], + "spans": [ + { + "bbox": [ + 126, + 117, + 179, + 130 + ], + "score": 1.0, + "content": "Linguistics.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "score": 1.0, + "content": "[50] Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 126, + 142, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 126, + 142, + 506, + 156 + ], + "score": 1.0, + "content": "evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 127, + 154, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 127, + 154, + 506, + 165 + ], + "score": 1.0, + "content": "for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA, July 2002.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 127, + 164, + 303, + 176 + ], + "spans": [ + { + "bbox": [ + 127, + 164, + 303, + 176 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 178, + 506, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 506, + 192 + ], + "score": 1.0, + "content": "[51] Maxime Peyrard, Teresa Botschen, and Iryna Gurevych. Learning to score system summaries", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 189, + 506, + 203 + ], + "spans": [ + { + "bbox": [ + 127, + 189, + 506, + 203 + ], + "score": 1.0, + "content": "for better content selection evaluation. In Proceedings of the Workshop on New Frontiers", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 200, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 126, + 200, + 506, + 214 + ], + "score": 1.0, + "content": "in Summarization, pages 74–84, Copenhagen, Denmark, September 2017. Association for", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 210, + 240, + 225 + ], + "spans": [ + { + "bbox": [ + 126, + 210, + 240, + 225 + ], + "score": 1.0, + "content": "Computational Linguistics.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 224, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 104, + 224, + 506, + 240 + ], + "score": 1.0, + "content": "[52] Maja Popovic. chrF: character n-gram F-score for automatic MT evaluation. In ´ Proceedings", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 126, + 235, + 507, + 251 + ], + "spans": [ + { + "bbox": [ + 126, + 235, + 507, + 251 + ], + "score": 1.0, + "content": "of the Tenth Workshop on Statistical Machine Translation, pages 392–395, Lisbon, Portugal,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 247, + 373, + 261 + ], + "spans": [ + { + "bbox": [ + 127, + 247, + 373, + 261 + ], + "score": 1.0, + "content": "September 2015. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 262, + 507, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 507, + 276 + ], + "score": 1.0, + "content": "[53] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 126, + 273, + 459, + 287 + ], + "spans": [ + { + "bbox": [ + 126, + 273, + 459, + 287 + ], + "score": 1.0, + "content": "Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 288, + 507, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 507, + 301 + ], + "score": 1.0, + "content": "[54] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 128, + 298, + 506, + 312 + ], + "spans": [ + { + "bbox": [ + 128, + 298, + 506, + 312 + ], + "score": 1.0, + "content": "Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 127, + 309, + 473, + 323 + ], + "spans": [ + { + "bbox": [ + 127, + 309, + 473, + 323 + ], + "score": 1.0, + "content": "text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67, 2020.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 324, + 507, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 507, + 338 + ], + "score": 1.0, + "content": "[55] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 128, + 335, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 128, + 335, + 505, + 349 + ], + "score": 1.0, + "content": "Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 126, + 345, + 412, + 358 + ], + "spans": [ + { + "bbox": [ + 126, + 345, + 412, + 358 + ], + "score": 1.0, + "content": "text-to-text transformer. J. Mach. Learn. Res., 21:140:1–140:67, 2020.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "[56] Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. COMET: A neural framework", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 127, + 371, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 127, + 371, + 506, + 385 + ], + "score": 1.0, + "content": "for MT evaluation. In Proceedings of the 2020 Conference on Empirical Methods in Natural", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 126, + 382, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 126, + 382, + 505, + 396 + ], + "score": 1.0, + "content": "Language Processing (EMNLP), pages 2685–2702, Online, November 2020. Association for", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 126, + 392, + 240, + 407 + ], + "spans": [ + { + "bbox": [ + 126, + 392, + 240, + 407 + ], + "score": 1.0, + "content": "Computational Linguistics.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 407, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 506, + 423 + ], + "score": 1.0, + "content": "[57] Laria Reynolds and Kyle McDonell. Prompt programming for large language models: Beyond", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 128, + 419, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 128, + 419, + 505, + 431 + ], + "score": 1.0, + "content": "the few-shot paradigm. In Extended Abstracts of the 2021 CHI Conference on Human Factors", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 429, + 291, + 444 + ], + "spans": [ + { + "bbox": [ + 125, + 429, + 291, + 444 + ], + "score": 1.0, + "content": "in Computing Systems, pages 1–7, 2021.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 444, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 458 + ], + "score": 1.0, + "content": "[58] Timo Schick and Hinrich Schütze. Exploiting cloze questions for few-shot text classification", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 455, + 418, + 469 + ], + "spans": [ + { + "bbox": [ + 127, + 455, + 418, + 469 + ], + "score": 1.0, + "content": "and natural language inference. arXiv preprint arXiv:2001.07676, 2020.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 469, + 507, + 486 + ], + "spans": [ + { + "bbox": [ + 104, + 469, + 507, + 486 + ], + "score": 1.0, + "content": "[59] Timo Schick and Hinrich Schütze. Few-shot text generation with pattern-exploiting training.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 127, + 480, + 290, + 494 + ], + "spans": [ + { + "bbox": [ + 127, + 480, + 290, + 494 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2012.11926, 2020.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 495, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 506, + 509 + ], + "score": 1.0, + "content": "[60] Timo Schick and Hinrich Schütze. It’s not just size that matters: Small language models are", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 507, + 384, + 520 + ], + "spans": [ + { + "bbox": [ + 126, + 507, + 384, + 520 + ], + "score": 1.0, + "content": "also few-shot learners. arXiv preprint arXiv:2009.07118, 2020.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 521, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 535 + ], + "score": 1.0, + "content": "[61] Abigail See, Peter J. Liu, and Christopher D. Manning. Get to the point: Summarization with", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 127, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 127, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "pointer-generator networks. In Proceedings of the 55th Annual Meeting of the Association for", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 127, + 543, + 507, + 557 + ], + "spans": [ + { + "bbox": [ + 127, + 543, + 507, + 557 + ], + "score": 1.0, + "content": "Computational Linguistics (Volume 1: Long Papers), pages 1073–1083, Vancouver, Canada,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 127, + 555, + 346, + 566 + ], + "spans": [ + { + "bbox": [ + 127, + 555, + 346, + 566 + ], + "score": 1.0, + "content": "July 2017. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 569, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 569, + 505, + 581 + ], + "score": 1.0, + "content": "[62] Thibault Sellam, Dipanjan Das, and Ankur Parikh. BLEURT: Learning robust metrics for text", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 127, + 579, + 506, + 593 + ], + "spans": [ + { + "bbox": [ + 127, + 579, + 506, + 593 + ], + "score": 1.0, + "content": "generation. In Proceedings of the 58th Annual Meeting of the Association for Computational", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 126, + 590, + 502, + 604 + ], + "spans": [ + { + "bbox": [ + 126, + 590, + 502, + 604 + ], + "score": 1.0, + "content": "Linguistics, pages 7881–7892, Online, July 2020. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 605, + 507, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 507, + 618 + ], + "score": 1.0, + "content": "[63] Taylor Shin, Yasaman Razeghi, Robert L Logan IV, Eric Wallace, and Sameer Singh. Auto-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 126, + 616, + 507, + 630 + ], + "spans": [ + { + "bbox": [ + 126, + 616, + 507, + 630 + ], + "score": 1.0, + "content": "prompt: Eliciting knowledge from language models with automatically generated prompts.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 126, + 626, + 291, + 641 + ], + "spans": [ + { + "bbox": [ + 126, + 626, + 291, + 641 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2010.15980, 2020.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 641, + 507, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 507, + 655 + ], + "score": 1.0, + "content": "[64] Miloš Stanojevic and Khalil Sima’an. BEER: BEtter evaluation as ranking. In ´ Proceedings of", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 127, + 653, + 507, + 666 + ], + "spans": [ + { + "bbox": [ + 127, + 653, + 507, + 666 + ], + "score": 1.0, + "content": "the Ninth Workshop on Statistical Machine Translation, pages 414–419, Baltimore, Maryland,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 126, + 662, + 374, + 677 + ], + "spans": [ + { + "bbox": [ + 126, + 662, + 374, + 677 + ], + "score": 1.0, + "content": "USA, June 2014. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "[65] Brian Thompson and Matt Post. Automatic machine translation evaluation in many languages", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 127, + 689, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 127, + 689, + 506, + 702 + ], + "score": 1.0, + "content": "via zero-shot paraphrasing. In Proceedings of the 2020 Conference on Empirical Methods in", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 126, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 126, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "Natural Language Processing (EMNLP), pages 90–121, Online, November 2020. Association", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 126, + 709, + 254, + 724 + ], + "spans": [ + { + "bbox": [ + 126, + 709, + 254, + 724 + ], + "score": 1.0, + "content": "for Computational Linguistics.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 26.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 68, + 507, + 727 + ], + "lines": [], + "index": 26.5, + "bbox_fs": [ + 104, + 72, + 507, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 41, + 507, + 712 + ], + "lines": [ + { + "bbox": [ + 104, + 70, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 104, + 70, + 507, + 87 + ], + "score": 1.0, + "content": "[66] Alex Wang, Kyunghyun Cho, and M. Lewis. Asking and answering questions to evaluate the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 126, + 83, + 326, + 97 + ], + "spans": [ + { + "bbox": [ + 126, + 83, + 326, + 97 + ], + "score": 1.0, + "content": "factual consistency of summaries. In ACL, 2020.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 97, + 508, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 508, + 113 + ], + "score": 1.0, + "content": "[67] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel Bowman.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 126, + 106, + 508, + 124 + ], + "spans": [ + { + "bbox": [ + 126, + 106, + 508, + 124 + ], + "score": 1.0, + "content": "GLUE: A multi-task benchmark and analysis platform for natural language understanding.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 125, + 118, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 125, + 118, + 506, + 135 + ], + "score": 1.0, + "content": "In Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 126, + 131, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 126, + 131, + 506, + 145 + ], + "score": 1.0, + "content": "Neural Networks for NLP, pages 353–355, Brussels, Belgium, November 2018. Association for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 127, + 142, + 240, + 156 + ], + "spans": [ + { + "bbox": [ + 127, + 142, + 240, + 156 + ], + "score": 1.0, + "content": "Computational Linguistics.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 156, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 506, + 171 + ], + "score": 1.0, + "content": "[68] Danqing Wang, Pengfei Liu, Yining Zheng, Xipeng Qiu, and Xuanjing Huang. Heterogeneous", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 167, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 126, + 167, + 506, + 182 + ], + "score": 1.0, + "content": "graph neural networks for extractive document summarization. In Proceedings of the 58th", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 126, + 178, + 506, + 192 + ], + "spans": [ + { + "bbox": [ + 126, + 178, + 506, + 192 + ], + "score": 1.0, + "content": "Annual Meeting of the Association for Computational Linguistics, pages 6209–6219, Online,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 125, + 188, + 348, + 204 + ], + "spans": [ + { + "bbox": [ + 125, + 188, + 348, + 204 + ], + "score": 1.0, + "content": "July 2020. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 202, + 507, + 219 + ], + "spans": [ + { + "bbox": [ + 107, + 202, + 507, + 219 + ], + "score": 1.0, + "content": "[69] Tsung-Hsien Wen, Milica Gasic, Nikola Mrksic, Pei-Hao Su, David Vandyke, and Steve Young.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 215, + 508, + 231 + ], + "spans": [ + { + "bbox": [ + 126, + 215, + 508, + 231 + ], + "score": 1.0, + "content": "Semantically conditioned lstm-based natural language generation for spoken dialogue systems.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 127, + 226, + 291, + 240 + ], + "spans": [ + { + "bbox": [ + 127, + 226, + 291, + 240 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1508.01745, 2015.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 239, + 506, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 506, + 256 + ], + "score": 1.0, + "content": "[70] Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 251, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 127, + 251, + 506, + 265 + ], + "score": 1.0, + "content": "Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 127, + 262, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 127, + 262, + 505, + 276 + ], + "score": 1.0, + "content": "Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 126, + 273, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 126, + 273, + 506, + 288 + ], + "score": 1.0, + "content": "Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. Transformers: State-of-the-art", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 126, + 285, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 126, + 285, + 506, + 299 + ], + "score": 1.0, + "content": "natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 126, + 296, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 126, + 296, + 506, + 309 + ], + "score": 1.0, + "content": "Natural Language Processing: System Demonstrations, pages 38–45, Online, October 2020.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 126, + 307, + 303, + 321 + ], + "spans": [ + { + "bbox": [ + 126, + 307, + 303, + 321 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "score": 1.0, + "content": "[71] Weizhe Yuan, Pengfei Liu, and Graham Neubig. Can we automate scientific reviewing? arXiv", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 126, + 333, + 266, + 345 + ], + "spans": [ + { + "bbox": [ + 126, + 333, + 266, + 345 + ], + "score": 1.0, + "content": "preprint arXiv:2102.00176, 2021.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 347, + 454, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 454, + 361 + ], + "score": 1.0, + "content": "[72] Jerrold H Zar. Spearman rank correlation. Encyclopedia of Biostatistics, 7, 2005.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 360, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 104, + 360, + 506, + 378 + ], + "score": 1.0, + "content": "[73] Haoyu Zhang, Jingjing Cai, Jianjun Xu, and Ji Wang. Pretraining-based natural language", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 126, + 372, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 126, + 372, + 506, + 388 + ], + "score": 1.0, + "content": "generation for text summarization. In Proceedings of the 23rd Conference on Computational", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 127, + 384, + 507, + 398 + ], + "spans": [ + { + "bbox": [ + 127, + 384, + 507, + 398 + ], + "score": 1.0, + "content": "Natural Language Learning (CoNLL), pages 789–797, Hong Kong, China, November 2019.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 128, + 396, + 302, + 407 + ], + "spans": [ + { + "bbox": [ + 128, + 396, + 302, + 407 + ], + "score": 1.0, + "content": "Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 410, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 506, + 424 + ], + "score": 1.0, + "content": "[74] Jingqing Zhang, Yao Zhao, Mohammad Saleh, and Peter Liu. Pegasus: Pre-training with", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 126, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 126, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "extracted gap-sentences for abstractive summarization. In International Conference on Machine", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 430, + 312, + 446 + ], + "spans": [ + { + "bbox": [ + 125, + 430, + 312, + 446 + ], + "score": 1.0, + "content": "Learning, pages 11328–11339. PMLR, 2020.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 447, + 507, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 507, + 461 + ], + "score": 1.0, + "content": "[75] Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. Bertscore:", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 457, + 507, + 471 + ], + "spans": [ + { + "bbox": [ + 127, + 457, + 507, + 471 + ], + "score": 1.0, + "content": "Evaluating text generation with bert. In International Conference on Learning Representations,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 126, + 467, + 154, + 482 + ], + "spans": [ + { + "bbox": [ + 126, + 467, + 154, + 482 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 483, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 506, + 497 + ], + "score": 1.0, + "content": "[76] Wei Zhao, Maxime Peyrard, Fei Liu, Yang Gao, Christian M. Meyer, and Steffen Eger. Mover-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 127, + 494, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 127, + 494, + 506, + 507 + ], + "score": 1.0, + "content": "Score: Text generation evaluating with contextualized embeddings and earth mover distance. In", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 504, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 126, + 504, + 506, + 519 + ], + "score": 1.0, + "content": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 516, + 507, + 530 + ], + "spans": [ + { + "bbox": [ + 127, + 516, + 507, + 530 + ], + "score": 1.0, + "content": "and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP),", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 125, + 526, + 509, + 542 + ], + "spans": [ + { + "bbox": [ + 125, + 526, + 509, + 542 + ], + "score": 1.0, + "content": "pages 563–578, Hong Kong, China, November 2019. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 540, + 507, + 557 + ], + "spans": [ + { + "bbox": [ + 104, + 540, + 507, + 557 + ], + "score": 1.0, + "content": "[77] Ming Zhong, Pengfei Liu, Yiran Chen, Danqing Wang, Xipeng Qiu, and Xuanjing Huang.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 126, + 552, + 467, + 567 + ], + "spans": [ + { + "bbox": [ + 126, + 552, + 467, + 567 + ], + "score": 1.0, + "content": "Extractive summarization as text matching. arXiv preprint arXiv:2004.08795, 2020.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 566, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 104, + 566, + 506, + 583 + ], + "score": 1.0, + "content": "[78] Ming Zhong, Pengfei Liu, Yiran Chen, Danqing Wang, Xipeng Qiu, and Xuanjing Huang.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 127, + 578, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 127, + 578, + 506, + 592 + ], + "score": 1.0, + "content": "Extractive summarization as text matching. In Proceedings of the 58th Annual Meeting of the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 126, + 589, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 126, + 589, + 506, + 603 + ], + "score": 1.0, + "content": "Association for Computational Linguistics, pages 6197–6208, Online, July 2020. Association", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 126, + 599, + 254, + 614 + ], + "spans": [ + { + "bbox": [ + 126, + 599, + 254, + 614 + ], + "score": 1.0, + "content": "for Computational Linguistics.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 614, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 104, + 614, + 506, + 629 + ], + "score": 1.0, + "content": "[79] Ming Zhong, Pengfei Liu, Danqing Wang, Xipeng Qiu, and Xuanjing Huang. Searching for", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 126, + 624, + 507, + 641 + ], + "spans": [ + { + "bbox": [ + 126, + 624, + 507, + 641 + ], + "score": 1.0, + "content": "effective neural extractive summarization: What works and what’s next. In Proceedings of", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 127, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 127, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "the 57th Annual Meeting of the Association for Computational Linguistics, pages 1049–1058,", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 648, + 409, + 662 + ], + "spans": [ + { + "bbox": [ + 126, + 648, + 409, + 662 + ], + "score": 1.0, + "content": "Florence, Italy, July 2019. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 662, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 676 + ], + "score": 1.0, + "content": "[80] Qingyu Zhou, Nan Yang, Furu Wei, Shaohan Huang, Ming Zhou, and Tiejun Zhao. Neural", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 127, + 673, + 506, + 687 + ], + "spans": [ + { + "bbox": [ + 127, + 673, + 506, + 687 + ], + "score": 1.0, + "content": "document summarization by jointly learning to score and select sentences. In Proceedings of the", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 126, + 682, + 507, + 700 + ], + "spans": [ + { + "bbox": [ + 126, + 682, + 507, + 700 + ], + "score": 1.0, + "content": "56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 125, + 695, + 503, + 710 + ], + "spans": [ + { + "bbox": [ + 125, + 695, + 503, + 710 + ], + "score": 1.0, + "content": "pages 654–663, Melbourne, Australia, July 2018. Association for Computational Linguistics.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 26 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 41, + 507, + 712 + ], + "lines": [], + "index": 26, + "bbox_fs": [ + 104, + 70, + 509, + 710 + ], + "lines_deleted": true + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_model.json b/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_model.json new file mode 100644 index 0000000000000000000000000000000000000000..64eaae3bf5713af07783730dfae2db6cfcde4fae --- /dev/null +++ b/parse/train/5Ya8PbvpZ9/5Ya8PbvpZ9_model.json @@ -0,0 +1,16552 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 397, + 737, + 1305, + 737, + 1305, + 1344, + 397, + 1344 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1465, + 1405, + 1465, + 1405, + 1741, + 298, + 1741 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1754, + 1403, + 1754, + 1403, + 1936, + 298, + 1936 + ], + "score": 0.975 + }, + { + "category_id": 0, + "poly": [ + 373, + 272, + 1328, + 272, + 1328, + 382, + 373, + 382 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 694, + 499, + 1005, + 499, + 1005, + 592, + 694, + 592 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 332, + 499, + 644, + 499, + 644, + 591, + 332, + 591 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 1052, + 498, + 1367, + 498, + 1367, + 592, + 1052, + 592 + ], + "score": 0.941 + }, + { + "category_id": 0, + "poly": [ + 299, + 1397, + 530, + 1397, + 530, + 1436, + 299, + 1436 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 787, + 670, + 912, + 670, + 912, + 706, + 787, + 706 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 297, + 2033, + 1070, + 2033, + 1070, + 2062, + 297, + 2062 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 330, + 1957, + 580, + 1957, + 580, + 1984, + 330, + 1984 + ], + "score": 0.884 + }, + { + "category_id": 15, + "poly": [ + 696.0, + 270.0, + 1012.0, + 270.0, + 1012.0, + 325.0, + 696.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 323.0, + 1331.0, + 323.0, + 1331.0, + 385.0, + 368.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1394.0, + 536.0, + 1394.0, + 536.0, + 1442.0, + 292.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 667.0, + 920.0, + 667.0, + 920.0, + 711.0, + 782.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2028.0, + 1074.0, + 2028.0, + 1074.0, + 2066.0, + 295.0, + 2066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1949.0, + 586.0, + 1949.0, + 586.0, + 1991.0, + 327.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 737.0, + 1308.0, + 737.0, + 1308.0, + 772.0, + 394.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 768.0, + 1306.0, + 768.0, + 1306.0, + 804.0, + 394.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 797.0, + 1307.0, + 797.0, + 1307.0, + 833.0, + 393.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 829.0, + 1306.0, + 829.0, + 1306.0, + 864.0, + 393.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 860.0, + 1306.0, + 860.0, + 1306.0, + 895.0, + 393.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 892.0, + 1305.0, + 892.0, + 1305.0, + 922.0, + 395.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 917.0, + 1306.0, + 917.0, + 1306.0, + 955.0, + 393.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 951.0, + 1305.0, + 951.0, + 1305.0, + 984.0, + 395.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 982.0, + 1305.0, + 982.0, + 1305.0, + 1014.0, + 394.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1011.0, + 1305.0, + 1011.0, + 1305.0, + 1044.0, + 394.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1038.0, + 1309.0, + 1038.0, + 1309.0, + 1078.0, + 392.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1070.0, + 1310.0, + 1070.0, + 1310.0, + 1107.0, + 393.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1102.0, + 1310.0, + 1102.0, + 1310.0, + 1137.0, + 393.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1132.0, + 1307.0, + 1132.0, + 1307.0, + 1168.0, + 393.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1164.0, + 1305.0, + 1164.0, + 1305.0, + 1196.0, + 394.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1192.0, + 1309.0, + 1192.0, + 1309.0, + 1227.0, + 393.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1222.0, + 1309.0, + 1222.0, + 1309.0, + 1257.0, + 394.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1254.0, + 1305.0, + 1254.0, + 1305.0, + 1287.0, + 398.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1284.0, + 1305.0, + 1284.0, + 1305.0, + 1316.0, + 395.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1314.0, + 1066.0, + 1314.0, + 1066.0, + 1347.0, + 394.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1465.0, + 1407.0, + 1465.0, + 1407.0, + 1500.0, + 296.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1496.0, + 1408.0, + 1496.0, + 1408.0, + 1532.0, + 293.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1529.0, + 1405.0, + 1529.0, + 1405.0, + 1561.0, + 296.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1556.0, + 1408.0, + 1556.0, + 1408.0, + 1593.0, + 295.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1587.0, + 1407.0, + 1587.0, + 1407.0, + 1624.0, + 293.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1616.0, + 1403.0, + 1616.0, + 1403.0, + 1653.0, + 295.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1650.0, + 1403.0, + 1650.0, + 1403.0, + 1683.0, + 296.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1679.0, + 1406.0, + 1679.0, + 1406.0, + 1713.0, + 293.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1709.0, + 1210.0, + 1709.0, + 1210.0, + 1744.0, + 293.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1756.0, + 1405.0, + 1756.0, + 1405.0, + 1787.0, + 296.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1786.0, + 1405.0, + 1786.0, + 1405.0, + 1821.0, + 294.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1816.0, + 1403.0, + 1816.0, + 1403.0, + 1848.0, + 296.0, + 1848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1845.0, + 1408.0, + 1845.0, + 1408.0, + 1881.0, + 292.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1876.0, + 1405.0, + 1876.0, + 1405.0, + 1910.0, + 293.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1909.0, + 714.0, + 1909.0, + 714.0, + 1939.0, + 294.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 495.0, + 951.0, + 495.0, + 951.0, + 536.0, + 750.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 526.0, + 1009.0, + 526.0, + 1009.0, + 568.0, + 690.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 562.0, + 1003.0, + 562.0, + 1003.0, + 595.0, + 696.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 496.0, + 571.0, + 496.0, + 571.0, + 532.0, + 409.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 526.0, + 648.0, + 526.0, + 648.0, + 567.0, + 330.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 563.0, + 641.0, + 563.0, + 641.0, + 594.0, + 337.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 499.0, + 1298.0, + 499.0, + 1298.0, + 532.0, + 1132.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 526.0, + 1369.0, + 526.0, + 1369.0, + 567.0, + 1051.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 561.0, + 1347.0, + 561.0, + 1347.0, + 594.0, + 1071.0, + 594.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 203, + 1405, + 203, + 1405, + 659, + 297, + 659 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 671, + 1405, + 671, + 1405, + 887, + 298, + 887 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1050, + 1403, + 1050, + 1403, + 1173, + 297, + 1173 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1855, + 1403, + 1855, + 1403, + 2008, + 299, + 2008 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 300, + 1263, + 1405, + 1263, + 1405, + 1354, + 300, + 1354 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 361, + 1375, + 1404, + 1375, + 1404, + 1834, + 361, + 1834 + ], + "score": 0.969 + }, + { + "category_id": 0, + "poly": [ + 299, + 993, + 619, + 993, + 619, + 1026, + 299, + 1026 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 299, + 1206, + 770, + 1206, + 770, + 1239, + 299, + 1239 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 299, + 926, + 542, + 926, + 542, + 965, + 299, + 965 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.721 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.125 + }, + { + "category_id": 13, + "poly": [ + 861, + 1113, + 1058, + 1113, + 1058, + 1143, + 861, + 1143 + ], + "score": 0.92, + "latex": "( s = s _ { 1 } , \\cdots , s _ { n } )" + }, + { + "category_id": 13, + "poly": [ + 299, + 1111, + 514, + 1111, + 514, + 1142, + 299, + 1142 + ], + "score": 0.9, + "latex": "( \\boldsymbol { h } = h _ { 1 } , \\cdots , h _ { m } )" + }, + { + "category_id": 13, + "poly": [ + 595, + 1144, + 772, + 1144, + 772, + 1173, + 595, + 1173 + ], + "score": 0.89, + "latex": "( r = r _ { 1 } , \\cdots , r _ { l } )" + }, + { + "category_id": 13, + "poly": [ + 620, + 795, + 662, + 795, + 662, + 824, + 620, + 824 + ], + "score": 0.84, + "latex": "3 \\%" + }, + { + "category_id": 13, + "poly": [ + 808, + 508, + 859, + 508, + 859, + 537, + 808, + 537 + ], + "score": 0.82, + "latex": "\\ S 3 . 2" + }, + { + "category_id": 13, + "poly": [ + 780, + 477, + 812, + 477, + 812, + 507, + 780, + 507 + ], + "score": 0.6, + "latex": "\\ S 4" + }, + { + "category_id": 13, + "poly": [ + 935, + 1325, + 966, + 1325, + 966, + 1354, + 935, + 1354 + ], + "score": 0.55, + "latex": "\\ S 4" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 991.0, + 622.0, + 991.0, + 622.0, + 1028.0, + 292.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1206.0, + 770.0, + 1206.0, + 770.0, + 1242.0, + 295.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 921.0, + 547.0, + 921.0, + 547.0, + 973.0, + 290.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 839.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 201.0, + 1405.0, + 201.0, + 1405.0, + 239.0, + 292.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 232.0, + 1408.0, + 232.0, + 1408.0, + 270.0, + 291.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 262.0, + 1406.0, + 262.0, + 1406.0, + 301.0, + 294.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 293.0, + 1406.0, + 293.0, + 1406.0, + 332.0, + 294.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 323.0, + 1406.0, + 323.0, + 1406.0, + 361.0, + 294.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 353.0, + 1406.0, + 353.0, + 1406.0, + 390.0, + 294.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 384.0, + 1405.0, + 384.0, + 1405.0, + 419.0, + 295.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 415.0, + 1405.0, + 415.0, + 1405.0, + 451.0, + 294.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 442.0, + 1406.0, + 442.0, + 1406.0, + 482.0, + 291.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 476.0, + 779.0, + 476.0, + 779.0, + 510.0, + 292.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 476.0, + 1405.0, + 476.0, + 1405.0, + 510.0, + 813.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 507.0, + 807.0, + 507.0, + 807.0, + 541.0, + 295.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 507.0, + 1406.0, + 507.0, + 1406.0, + 541.0, + 860.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 536.0, + 1406.0, + 536.0, + 1406.0, + 571.0, + 295.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 565.0, + 1406.0, + 565.0, + 1406.0, + 604.0, + 294.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 595.0, + 1406.0, + 595.0, + 1406.0, + 636.0, + 292.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 626.0, + 1106.0, + 626.0, + 1106.0, + 663.0, + 294.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 673.0, + 1408.0, + 673.0, + 1408.0, + 707.0, + 296.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 700.0, + 1405.0, + 700.0, + 1405.0, + 741.0, + 295.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 732.0, + 1405.0, + 732.0, + 1405.0, + 769.0, + 293.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 760.0, + 1407.0, + 760.0, + 1407.0, + 801.0, + 292.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 794.0, + 619.0, + 794.0, + 619.0, + 827.0, + 293.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 794.0, + 1408.0, + 794.0, + 1408.0, + 827.0, + 663.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 825.0, + 1403.0, + 825.0, + 1403.0, + 856.0, + 296.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 856.0, + 1136.0, + 856.0, + 1136.0, + 890.0, + 296.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1051.0, + 1405.0, + 1051.0, + 1405.0, + 1084.0, + 295.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1080.0, + 1407.0, + 1080.0, + 1407.0, + 1116.0, + 293.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1110.0, + 298.0, + 1110.0, + 298.0, + 1146.0, + 293.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1110.0, + 860.0, + 1110.0, + 860.0, + 1146.0, + 515.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1110.0, + 1404.0, + 1110.0, + 1404.0, + 1146.0, + 1059.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1143.0, + 594.0, + 1143.0, + 594.0, + 1175.0, + 295.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1143.0, + 1167.0, + 1143.0, + 1167.0, + 1175.0, + 773.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1855.0, + 1405.0, + 1855.0, + 1405.0, + 1888.0, + 295.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1883.0, + 1404.0, + 1883.0, + 1404.0, + 1920.0, + 294.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1916.0, + 1404.0, + 1916.0, + 1404.0, + 1949.0, + 295.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1945.0, + 1405.0, + 1945.0, + 1405.0, + 1982.0, + 295.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1975.0, + 716.0, + 1975.0, + 716.0, + 2012.0, + 295.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1262.0, + 1404.0, + 1262.0, + 1404.0, + 1299.0, + 294.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1294.0, + 1405.0, + 1294.0, + 1405.0, + 1328.0, + 295.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1322.0, + 934.0, + 1322.0, + 934.0, + 1357.0, + 294.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 1322.0, + 987.0, + 1322.0, + 987.0, + 1357.0, + 967.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1372.0, + 1404.0, + 1372.0, + 1404.0, + 1414.0, + 360.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1408.0, + 585.0, + 1408.0, + 585.0, + 1441.0, + 394.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1434.0, + 1405.0, + 1434.0, + 1405.0, + 1474.0, + 355.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1467.0, + 503.0, + 1467.0, + 503.0, + 1501.0, + 393.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1496.0, + 1406.0, + 1496.0, + 1406.0, + 1535.0, + 356.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1529.0, + 1405.0, + 1529.0, + 1405.0, + 1565.0, + 393.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1557.0, + 680.0, + 1557.0, + 680.0, + 1593.0, + 393.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1589.0, + 1405.0, + 1589.0, + 1405.0, + 1622.0, + 362.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1617.0, + 729.0, + 1617.0, + 729.0, + 1653.0, + 393.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1646.0, + 1404.0, + 1646.0, + 1404.0, + 1686.0, + 356.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1680.0, + 623.0, + 1680.0, + 623.0, + 1713.0, + 394.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1708.0, + 1405.0, + 1708.0, + 1405.0, + 1747.0, + 365.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1740.0, + 866.0, + 1740.0, + 866.0, + 1776.0, + 393.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1769.0, + 1405.0, + 1769.0, + 1405.0, + 1806.0, + 357.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1801.0, + 983.0, + 1801.0, + 983.0, + 1837.0, + 393.0, + 1837.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1358, + 1404, + 1358, + 1404, + 1662, + 297, + 1662 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 858, + 1404, + 858, + 1404, + 1071, + 298, + 1071 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1854, + 1403, + 1854, + 1403, + 2007, + 299, + 2007 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 1191, + 1404, + 1191, + 1404, + 1343, + 298, + 1343 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 301, + 751, + 1402, + 751, + 1402, + 843, + 301, + 843 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 299, + 1085, + 1401, + 1085, + 1401, + 1178, + 299, + 1178 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 297, + 579, + 1400, + 579, + 1400, + 643, + 297, + 643 + ], + "score": 0.951 + }, + { + "category_id": 4, + "poly": [ + 289, + 447, + 1403, + 447, + 1403, + 510, + 289, + 510 + ], + "score": 0.933 + }, + { + "category_id": 0, + "poly": [ + 298, + 1792, + 863, + 1792, + 863, + 1826, + 298, + 1826 + ], + "score": 0.927 + }, + { + "category_id": 0, + "poly": [ + 300, + 690, + 710, + 690, + 710, + 722, + 300, + 722 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 298, + 1716, + 519, + 1716, + 519, + 1754, + 298, + 1754 + ], + "score": 0.907 + }, + { + "category_id": 3, + "poly": [ + 335, + 219, + 1355, + 219, + 1355, + 431, + 335, + 431 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.62 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.38 + }, + { + "category_id": 3, + "poly": [ + 333, + 2, + 1358, + 2, + 1358, + 430, + 333, + 430 + ], + "score": 0.126 + }, + { + "category_id": 13, + "poly": [ + 1182, + 1145, + 1216, + 1145, + 1216, + 1175, + 1182, + 1175 + ], + "score": 0.86, + "latex": "{ \\bar { S } } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1008, + 453, + 1036, + 453, + 1036, + 481, + 1008, + 481 + ], + "score": 0.84, + "latex": "r _ { j }" + }, + { + "category_id": 13, + "poly": [ + 929, + 449, + 958, + 449, + 958, + 478, + 929, + 478 + ], + "score": 0.79, + "latex": "h _ { i }" + }, + { + "category_id": 13, + "poly": [ + 894, + 452, + 919, + 452, + 919, + 478, + 894, + 478 + ], + "score": 0.74, + "latex": "s _ { i }" + }, + { + "category_id": 13, + "poly": [ + 562, + 582, + 592, + 582, + 592, + 611, + 562, + 611 + ], + "score": 0.66, + "latex": "\\ S 4" + }, + { + "category_id": 13, + "poly": [ + 350, + 231, + 367, + 231, + 367, + 247, + 350, + 247 + ], + "score": 0.31, + "latex": "r _ { 1 }" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 444.0, + 893.0, + 444.0, + 893.0, + 482.0, + 293.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 444.0, + 928.0, + 444.0, + 928.0, + 482.0, + 920.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 444.0, + 1007.0, + 444.0, + 1007.0, + 482.0, + 959.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 444.0, + 1405.0, + 444.0, + 1405.0, + 482.0, + 1037.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 477.0, + 625.0, + 477.0, + 625.0, + 513.0, + 296.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1790.0, + 867.0, + 1790.0, + 867.0, + 1831.0, + 291.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 689.0, + 715.0, + 689.0, + 715.0, + 726.0, + 293.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1715.0, + 525.0, + 1715.0, + 525.0, + 1763.0, + 290.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 224.0, + 374.0, + 224.0, + 374.0, + 248.0, + 368.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 224.0, + 415.0, + 224.0, + 415.0, + 251.0, + 381.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 225.0, + 451.0, + 225.0, + 451.0, + 249.0, + 423.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 223.0, + 1188.0, + 223.0, + 1188.0, + 249.0, + 1124.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 357.0, + 441.0, + 357.0, + 441.0, + 397.0, + 364.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 362.0, + 732.0, + 362.0, + 732.0, + 394.0, + 658.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 365.0, + 804.0, + 365.0, + 804.0, + 390.0, + 780.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 365.0, + 884.0, + 365.0, + 884.0, + 391.0, + 858.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 363.0, + 980.0, + 363.0, + 980.0, + 393.0, + 949.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 363.0, + 1040.0, + 363.0, + 1040.0, + 391.0, + 1007.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 364.0, + 1075.0, + 364.0, + 1075.0, + 392.0, + 1042.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 366.0, + 1147.0, + 366.0, + 1147.0, + 386.0, + 1125.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 369.0, + 1222.0, + 369.0, + 1222.0, + 382.0, + 1207.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 393.0, + 474.0, + 393.0, + 474.0, + 437.0, + 330.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 396.0, + 697.0, + 396.0, + 697.0, + 435.0, + 538.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 394.0, + 995.0, + 394.0, + 995.0, + 437.0, + 862.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 398.0, + 1317.0, + 398.0, + 1317.0, + 433.0, + 1160.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 278.0, + 1192.0, + 278.0, + 1192.0, + 295.0, + 1159.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.25, + 368.5, + 519.25, + 368.5, + 519.25, + 385.0, + 511.25, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 366.0, + 569.0, + 366.0, + 569.0, + 389.0, + 542.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 365.5, + 603.0, + 365.5, + 603.0, + 388.0, + 581.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 363.0, + 842.0, + 363.0, + 842.0, + 390.5, + 820.0, + 390.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.75, + 363.0, + 938.75, + 363.0, + 938.75, + 392.0, + 914.75, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.75, + 365.5, + 1183.75, + 365.5, + 1183.75, + 383.5, + 1166.75, + 383.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 860.0, + 2058.0, + 860.0, + 2089.0, + 839.0, + 2089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 224.0, + 373.0, + 224.0, + 373.0, + 249.0, + 368.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 225.0, + 414.0, + 225.0, + 414.0, + 249.0, + 385.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 229.0, + 1138.0, + 229.0, + 1138.0, + 240.0, + 1129.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 230.0, + 1178.0, + 230.0, + 1178.0, + 240.0, + 1167.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 358.0, + 441.0, + 358.0, + 441.0, + 398.0, + 364.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 367.0, + 527.0, + 367.0, + 527.0, + 387.0, + 504.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 367.0, + 570.0, + 367.0, + 570.0, + 390.0, + 543.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 366.0, + 606.0, + 366.0, + 606.0, + 392.0, + 582.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 364.0, + 694.0, + 364.0, + 694.0, + 393.0, + 660.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 364.0, + 729.0, + 364.0, + 729.0, + 394.0, + 697.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 366.0, + 804.0, + 366.0, + 804.0, + 389.0, + 780.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 364.0, + 883.0, + 364.0, + 883.0, + 392.0, + 858.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 363.0, + 942.0, + 363.0, + 942.0, + 393.0, + 912.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 364.0, + 980.0, + 364.0, + 980.0, + 393.0, + 950.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 363.0, + 1037.0, + 363.0, + 1037.0, + 392.0, + 1007.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 363.0, + 1076.0, + 363.0, + 1076.0, + 392.0, + 1046.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 373.0, + 1183.0, + 373.0, + 1183.0, + 383.0, + 1171.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 394.0, + 474.0, + 394.0, + 474.0, + 437.0, + 329.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 395.0, + 698.0, + 395.0, + 698.0, + 436.0, + 537.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 392.0, + 996.0, + 392.0, + 996.0, + 438.0, + 860.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 396.0, + 1319.0, + 396.0, + 1319.0, + 434.0, + 1158.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 222.5, + 449.0, + 222.5, + 449.0, + 249.5, + 424.0, + 249.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.75, + 363.5, + 843.75, + 363.5, + 843.75, + 390.0, + 816.75, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1356.0, + 1406.0, + 1356.0, + 1406.0, + 1393.0, + 294.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1390.0, + 1404.0, + 1390.0, + 1404.0, + 1422.0, + 296.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1418.0, + 1407.0, + 1418.0, + 1407.0, + 1454.0, + 291.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1449.0, + 1406.0, + 1449.0, + 1406.0, + 1482.0, + 292.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1478.0, + 1406.0, + 1478.0, + 1406.0, + 1514.0, + 294.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1511.0, + 1405.0, + 1511.0, + 1405.0, + 1543.0, + 295.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1542.0, + 1404.0, + 1542.0, + 1404.0, + 1573.0, + 295.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1569.0, + 1407.0, + 1569.0, + 1407.0, + 1606.0, + 291.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1601.0, + 1405.0, + 1601.0, + 1405.0, + 1636.0, + 294.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1632.0, + 933.0, + 1632.0, + 933.0, + 1664.0, + 295.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 857.0, + 1409.0, + 857.0, + 1409.0, + 895.0, + 293.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 889.0, + 1409.0, + 889.0, + 1409.0, + 923.0, + 293.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 916.0, + 1408.0, + 916.0, + 1408.0, + 956.0, + 292.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 949.0, + 1405.0, + 949.0, + 1405.0, + 983.0, + 294.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 980.0, + 1406.0, + 980.0, + 1406.0, + 1015.0, + 294.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1009.0, + 1406.0, + 1009.0, + 1406.0, + 1045.0, + 293.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1039.0, + 1240.0, + 1039.0, + 1240.0, + 1077.0, + 294.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1853.0, + 1405.0, + 1853.0, + 1405.0, + 1890.0, + 294.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1880.0, + 1407.0, + 1880.0, + 1407.0, + 1923.0, + 293.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1950.0, + 293.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1946.0, + 1405.0, + 1946.0, + 1405.0, + 1979.0, + 297.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2015.0, + 293.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1188.0, + 1404.0, + 1188.0, + 1404.0, + 1226.0, + 293.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1218.0, + 1408.0, + 1218.0, + 1408.0, + 1258.0, + 293.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1250.0, + 1405.0, + 1250.0, + 1405.0, + 1288.0, + 293.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1281.0, + 1404.0, + 1281.0, + 1404.0, + 1317.0, + 293.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1313.0, + 1230.0, + 1313.0, + 1230.0, + 1347.0, + 293.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 747.0, + 1407.0, + 747.0, + 1407.0, + 789.0, + 295.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 783.0, + 1406.0, + 783.0, + 1406.0, + 816.0, + 296.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 810.0, + 751.0, + 810.0, + 751.0, + 846.0, + 295.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1084.0, + 1407.0, + 1084.0, + 1407.0, + 1122.0, + 294.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1116.0, + 1405.0, + 1116.0, + 1405.0, + 1150.0, + 295.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1144.0, + 1181.0, + 1144.0, + 1181.0, + 1181.0, + 294.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 1144.0, + 1406.0, + 1144.0, + 1406.0, + 1181.0, + 1217.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 574.0, + 561.0, + 574.0, + 561.0, + 620.0, + 293.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 574.0, + 1404.0, + 574.0, + 1404.0, + 620.0, + 593.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 611.0, + 1110.0, + 611.0, + 1110.0, + 646.0, + 293.0, + 646.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 601, + 1404, + 601, + 1404, + 725, + 297, + 725 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1642, + 1404, + 1642, + 1404, + 1765, + 298, + 1765 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 1777, + 1403, + 1777, + 1403, + 1901, + 298, + 1901 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 301, + 1465, + 1400, + 1465, + 1400, + 1560, + 301, + 1560 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 300, + 950, + 1406, + 950, + 1406, + 1376, + 300, + 1376 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 670, + 373, + 1029, + 373, + 1029, + 456, + 670, + 456 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 298, + 278, + 1402, + 278, + 1402, + 370, + 298, + 370 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 298, + 833, + 1405, + 833, + 1405, + 928, + 298, + 928 + ], + "score": 0.956 + }, + { + "category_id": 2, + "poly": [ + 297, + 1922, + 1405, + 1922, + 1405, + 2007, + 297, + 2007 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 607, + 739, + 1094, + 739, + 1094, + 823, + 607, + 823 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 294, + 202, + 1405, + 202, + 1405, + 266, + 294, + 266 + ], + "score": 0.949 + }, + { + "category_id": 0, + "poly": [ + 297, + 546, + 505, + 546, + 505, + 579, + 297, + 579 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 296, + 479, + 1400, + 479, + 1400, + 514, + 296, + 514 + ], + "score": 0.926 + }, + { + "category_id": 0, + "poly": [ + 298, + 1410, + 612, + 1410, + 612, + 1443, + 298, + 1443 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 298, + 1589, + 476, + 1589, + 476, + 1622, + 298, + 1622 + ], + "score": 0.908 + }, + { + "category_id": 9, + "poly": [ + 1366, + 396, + 1400, + 396, + 1400, + 426, + 1366, + 426 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1366, + 763, + 1400, + 763, + 1400, + 793, + 1366, + 793 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.783 + }, + { + "category_id": 14, + "poly": [ + 670, + 367, + 1028, + 367, + 1028, + 457, + 670, + 457 + ], + "score": 0.95, + "latex": "p ( \\mathbf { y } | \\mathbf { x } , \\theta ) = \\prod _ { t = 1 } ^ { m } p ( \\mathbf { y } _ { t } | \\mathbf { y } _ { < t } , \\mathbf { x } , \\theta )" + }, + { + "category_id": 14, + "poly": [ + 604, + 735, + 1094, + 735, + 1094, + 825, + 604, + 825 + ], + "score": 0.95, + "latex": "\\mathbf { B A R T S C O R E } = \\sum _ { t = 1 } ^ { m } \\omega _ { t } \\log p ( \\mathbf { y } _ { t } | \\mathbf { y } _ { < t } , \\mathbf { x } , \\theta )" + }, + { + "category_id": 13, + "poly": [ + 1032, + 1071, + 1137, + 1071, + 1137, + 1105, + 1032, + 1105 + ], + "score": 0.94, + "latex": "p ( \\boldsymbol { h } | \\boldsymbol { r } , \\boldsymbol { \\theta } )" + }, + { + "category_id": 13, + "poly": [ + 797, + 310, + 1015, + 310, + 1015, + 344, + 797, + 344 + ], + "score": 0.93, + "latex": "\\mathbf { y } = \\{ y _ { 1 } , \\dots , y _ { m } \\}" + }, + { + "category_id": 13, + "poly": [ + 971, + 950, + 1075, + 950, + 1075, + 984, + 971, + 984 + ], + "score": 0.93, + "latex": "p ( \\boldsymbol { h } | \\boldsymbol { s } , \\boldsymbol { \\theta } )" + }, + { + "category_id": 13, + "poly": [ + 1022, + 1162, + 1127, + 1162, + 1127, + 1196, + 1022, + 1196 + ], + "score": 0.93, + "latex": "p ( \\pmb { r } | \\pmb { h } , \\theta )" + }, + { + "category_id": 13, + "poly": [ + 898, + 1839, + 1243, + 1839, + 1243, + 1873, + 898, + 1873 + ], + "score": 0.93, + "latex": "\\mathbf { y } ^ { \\prime } = \\{ z _ { 1 } , \\cdot \\cdot \\cdot , z _ { l } , y _ { 1 } , \\cdot \\cdot \\cdot , y _ { m } \\}" + }, + { + "category_id": 13, + "poly": [ + 1191, + 278, + 1401, + 278, + 1401, + 313, + 1191, + 313 + ], + "score": 0.92, + "latex": "\\mathbf { x } = \\{ x _ { 1 } , \\cdots , x _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 600, + 1778, + 793, + 1778, + 793, + 1811, + 600, + 1811 + ], + "score": 0.9, + "latex": "\\mathbf { z } = \\{ z _ { 1 } , \\cdots , z _ { l } \\}" + }, + { + "category_id": 13, + "poly": [ + 438, + 1809, + 784, + 1809, + 784, + 1842, + 438, + 1842 + ], + "score": 0.9, + "latex": "\\mathbf { x } ^ { \\prime } = \\{ x _ { 1 } , \\cdot \\cdot \\cdot , x _ { n } , \\dot { z } _ { 1 } , \\cdot \\cdot \\cdot , z _ { l } \\}" + }, + { + "category_id": 13, + "poly": [ + 440, + 1072, + 526, + 1072, + 526, + 1102, + 440, + 1102 + ], + "score": 0.87, + "latex": "( r h )" + }, + { + "category_id": 13, + "poly": [ + 471, + 952, + 553, + 952, + 553, + 980, + 471, + 980 + ], + "score": 0.87, + "latex": "s \\to h" + }, + { + "category_id": 13, + "poly": [ + 927, + 1011, + 980, + 1011, + 980, + 1041, + 927, + 1041 + ], + "score": 0.87, + "latex": "\\ S 2 . 2" + }, + { + "category_id": 13, + "poly": [ + 411, + 1163, + 496, + 1163, + 496, + 1193, + 411, + 1193 + ], + "score": 0.86, + "latex": "( h \\to r" + }, + { + "category_id": 13, + "poly": [ + 425, + 1284, + 510, + 1284, + 510, + 1314, + 425, + 1314 + ], + "score": 0.85, + "latex": "( r h" + }, + { + "category_id": 13, + "poly": [ + 864, + 1224, + 917, + 1224, + 917, + 1253, + 864, + 1253 + ], + "score": 0.82, + "latex": "\\ S 2 . 2" + }, + { + "category_id": 13, + "poly": [ + 325, + 1284, + 352, + 1284, + 352, + 1312, + 325, + 1312 + ], + "score": 0.78, + "latex": "\\mathcal { F }" + }, + { + "category_id": 13, + "poly": [ + 745, + 281, + 761, + 281, + 761, + 307, + 745, + 307 + ], + "score": 0.78, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 448, + 1344, + 503, + 1344, + 503, + 1374, + 448, + 1374 + ], + "score": 0.76, + "latex": "\\ S 2 . 2 \\AA ," + }, + { + "category_id": 13, + "poly": [ + 1092, + 286, + 1112, + 286, + 1112, + 308, + 1092, + 308 + ], + "score": 0.75, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 681, + 315, + 709, + 315, + 709, + 338, + 681, + 338 + ], + "score": 0.75, + "latex": "m" + }, + { + "category_id": 13, + "poly": [ + 1113, + 1498, + 1131, + 1498, + 1131, + 1524, + 1113, + 1524 + ], + "score": 0.74, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 818, + 639, + 838, + 639, + 838, + 661, + 818, + 661 + ], + "score": 0.56, + "latex": "\\mathbf { x }" + }, + { + "category_id": 13, + "poly": [ + 652, + 345, + 673, + 345, + 673, + 367, + 652, + 367 + ], + "score": 0.55, + "latex": "\\mathbf { x }" + }, + { + "category_id": 13, + "poly": [ + 506, + 1781, + 518, + 1781, + 518, + 1805, + 506, + 1805 + ], + "score": 0.53, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 456, + 344, + 477, + 344, + 477, + 372, + 456, + 372 + ], + "score": 0.47, + "latex": "\\mathbf { y }" + }, + { + "category_id": 13, + "poly": [ + 1088, + 1471, + 1110, + 1471, + 1110, + 1498, + 1088, + 1498 + ], + "score": 0.39, + "latex": "\\mathbf { y }" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1472, + 1040, + 1472, + 1040, + 1495, + 1018, + 1495 + ], + "score": 0.38, + "latex": "\\mathbf { x }" + }, + { + "category_id": 13, + "poly": [ + 1306, + 1041, + 1374, + 1041, + 1374, + 1072, + 1306, + 1072 + ], + "score": 0.32, + "latex": "( \\ S 2 . 2 )" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1915.0, + 1410.0, + 1915.0, + 1410.0, + 1959.0, + 326.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1948.0, + 1406.0, + 1948.0, + 1406.0, + 1983.0, + 294.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1976.0, + 1210.0, + 1976.0, + 1210.0, + 2011.0, + 294.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 539.0, + 508.0, + 539.0, + 508.0, + 587.0, + 292.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1408.0, + 615.0, + 1408.0, + 615.0, + 1445.0, + 293.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1584.0, + 479.0, + 1584.0, + 479.0, + 1632.0, + 292.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 602.0, + 1405.0, + 602.0, + 1405.0, + 639.0, + 294.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 633.0, + 817.0, + 633.0, + 817.0, + 669.0, + 294.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 633.0, + 1406.0, + 633.0, + 1406.0, + 669.0, + 839.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 659.0, + 1405.0, + 659.0, + 1405.0, + 702.0, + 292.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 692.0, + 955.0, + 692.0, + 955.0, + 727.0, + 292.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1641.0, + 1406.0, + 1641.0, + 1406.0, + 1678.0, + 292.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1672.0, + 1406.0, + 1672.0, + 1406.0, + 1708.0, + 293.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1702.0, + 1405.0, + 1702.0, + 1405.0, + 1738.0, + 294.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1735.0, + 1117.0, + 1735.0, + 1117.0, + 1764.0, + 296.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1776.0, + 505.0, + 1776.0, + 505.0, + 1812.0, + 296.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1776.0, + 599.0, + 1776.0, + 599.0, + 1812.0, + 519.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1776.0, + 1405.0, + 1776.0, + 1405.0, + 1812.0, + 794.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1806.0, + 437.0, + 1806.0, + 437.0, + 1846.0, + 292.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1806.0, + 1409.0, + 1806.0, + 1409.0, + 1846.0, + 785.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1838.0, + 897.0, + 1838.0, + 897.0, + 1877.0, + 293.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 1838.0, + 1405.0, + 1838.0, + 1405.0, + 1877.0, + 1244.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1868.0, + 735.0, + 1868.0, + 735.0, + 1903.0, + 295.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1465.0, + 1017.0, + 1465.0, + 1017.0, + 1503.0, + 295.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1465.0, + 1087.0, + 1465.0, + 1087.0, + 1503.0, + 1041.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1465.0, + 1404.0, + 1465.0, + 1404.0, + 1503.0, + 1111.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1495.0, + 1112.0, + 1495.0, + 1112.0, + 1532.0, + 292.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1495.0, + 1406.0, + 1495.0, + 1406.0, + 1532.0, + 1132.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1526.0, + 1280.0, + 1526.0, + 1280.0, + 1562.0, + 293.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 948.0, + 470.0, + 948.0, + 470.0, + 985.0, + 299.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 948.0, + 970.0, + 948.0, + 970.0, + 985.0, + 554.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 948.0, + 1406.0, + 948.0, + 1406.0, + 985.0, + 1076.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 980.0, + 1406.0, + 980.0, + 1406.0, + 1015.0, + 322.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1010.0, + 926.0, + 1010.0, + 926.0, + 1044.0, + 319.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1010.0, + 1406.0, + 1010.0, + 1406.0, + 1044.0, + 981.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1040.0, + 1305.0, + 1040.0, + 1305.0, + 1076.0, + 320.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1040.0, + 1384.0, + 1040.0, + 1384.0, + 1076.0, + 1375.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1069.0, + 439.0, + 1069.0, + 439.0, + 1107.0, + 303.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1069.0, + 1031.0, + 1069.0, + 1031.0, + 1107.0, + 527.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 1069.0, + 1406.0, + 1069.0, + 1406.0, + 1107.0, + 1138.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1101.0, + 1404.0, + 1101.0, + 1404.0, + 1136.0, + 320.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1132.0, + 635.0, + 1132.0, + 635.0, + 1164.0, + 316.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1160.0, + 410.0, + 1160.0, + 410.0, + 1200.0, + 302.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1160.0, + 1021.0, + 1160.0, + 1021.0, + 1200.0, + 497.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1160.0, + 1406.0, + 1160.0, + 1406.0, + 1200.0, + 1128.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1192.0, + 1404.0, + 1192.0, + 1404.0, + 1227.0, + 320.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1221.0, + 863.0, + 1221.0, + 863.0, + 1258.0, + 320.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1221.0, + 1406.0, + 1221.0, + 1406.0, + 1258.0, + 918.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1253.0, + 1384.0, + 1253.0, + 1384.0, + 1288.0, + 319.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1281.0, + 324.0, + 1281.0, + 324.0, + 1317.0, + 303.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1281.0, + 424.0, + 1281.0, + 424.0, + 1317.0, + 353.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1281.0, + 1406.0, + 1281.0, + 1406.0, + 1317.0, + 511.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1311.0, + 1405.0, + 1311.0, + 1405.0, + 1350.0, + 320.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1341.0, + 447.0, + 1341.0, + 447.0, + 1379.0, + 320.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1341.0, + 1008.0, + 1341.0, + 1008.0, + 1379.0, + 504.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 276.0, + 744.0, + 276.0, + 744.0, + 316.0, + 293.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 276.0, + 1091.0, + 276.0, + 1091.0, + 316.0, + 762.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 276.0, + 1190.0, + 276.0, + 1190.0, + 316.0, + 1113.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 308.0, + 680.0, + 308.0, + 680.0, + 344.0, + 293.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 308.0, + 796.0, + 308.0, + 796.0, + 344.0, + 710.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 308.0, + 1407.0, + 308.0, + 1407.0, + 344.0, + 1016.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 340.0, + 455.0, + 340.0, + 455.0, + 372.0, + 293.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 340.0, + 651.0, + 340.0, + 651.0, + 372.0, + 478.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 340.0, + 805.0, + 340.0, + 805.0, + 372.0, + 674.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 833.0, + 1406.0, + 833.0, + 1406.0, + 871.0, + 295.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 865.0, + 1405.0, + 865.0, + 1405.0, + 901.0, + 292.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 895.0, + 1388.0, + 895.0, + 1388.0, + 932.0, + 292.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 201.0, + 1406.0, + 201.0, + 1406.0, + 240.0, + 294.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 233.0, + 1408.0, + 233.0, + 1408.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 477.0, + 1404.0, + 477.0, + 1404.0, + 519.0, + 292.0, + 519.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 255, + 1405, + 255, + 1405, + 500, + 298, + 500 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 902, + 1484, + 1413, + 1484, + 1413, + 1909, + 902, + 1909 + ], + "score": 0.98, + "html": "
TasksDatasetsEval. Perspectives
SUMREALSUMCov
SummEvalCOH FAC FLU INFO
NeR18COH FLU REL INFO
Rank19 QAGS-C QAGS-XFAC
DE FI GU KK IT RU ZHADE FLU
MT D2TBAGEL SFHOT SFRESINFO
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1101, + 1405, + 1101, + 1405, + 1253, + 298, + 1253 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 859, + 1405, + 859, + 1405, + 1011, + 298, + 1011 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1646, + 881, + 1646, + 881, + 1921, + 298, + 1921 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1449, + 881, + 1449, + 881, + 1632, + 298, + 1632 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1343, + 880, + 1343, + 880, + 1435, + 298, + 1435 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 299, + 616, + 1402, + 616, + 1402, + 709, + 299, + 709 + ], + "score": 0.961 + }, + { + "category_id": 6, + "poly": [ + 900, + 1344, + 1404, + 1344, + 1404, + 1468, + 900, + 1468 + ], + "score": 0.947 + }, + { + "category_id": 2, + "poly": [ + 294, + 1949, + 1405, + 1949, + 1405, + 2007, + 294, + 2007 + ], + "score": 0.934 + }, + { + "category_id": 0, + "poly": [ + 299, + 202, + 585, + 202, + 585, + 235, + 299, + 235 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 300, + 1047, + 742, + 1047, + 742, + 1079, + 300, + 1079 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 299, + 748, + 631, + 748, + 631, + 780, + 299, + 780 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 298, + 545, + 519, + 545, + 519, + 583, + 298, + 583 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 299, + 805, + 611, + 805, + 611, + 838, + 299, + 838 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 299, + 1289, + 485, + 1289, + 485, + 1321, + 299, + 1321 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.768 + }, + { + "category_id": 13, + "poly": [ + 376, + 467, + 434, + 467, + 434, + 497, + 376, + 497 + ], + "score": 0.89, + "latex": "5 e ^ { - 5 }" + }, + { + "category_id": 13, + "poly": [ + 864, + 860, + 916, + 860, + 916, + 890, + 864, + 890 + ], + "score": 0.85, + "latex": "\\ S 2 . 3" + }, + { + "category_id": 13, + "poly": [ + 1222, + 1436, + 1274, + 1436, + 1274, + 1466, + 1222, + 1466 + ], + "score": 0.85, + "latex": "\\ S 2 . 2" + }, + { + "category_id": 13, + "poly": [ + 818, + 1978, + 865, + 1978, + 865, + 2006, + 818, + 2006 + ], + "score": 0.82, + "latex": "\\ S 2 . 2" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 1345.0, + 1404.0, + 1345.0, + 1404.0, + 1376.0, + 900.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 1375.0, + 1404.0, + 1375.0, + 1404.0, + 1406.0, + 900.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1406.0, + 1403.0, + 1406.0, + 1403.0, + 1437.0, + 898.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1438.0, + 1221.0, + 1438.0, + 1221.0, + 1467.0, + 896.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 1438.0, + 1284.0, + 1438.0, + 1284.0, + 1467.0, + 1275.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1943.0, + 1405.0, + 1943.0, + 1405.0, + 1983.0, + 330.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1978.0, + 817.0, + 1978.0, + 817.0, + 2009.0, + 297.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1978.0, + 873.0, + 1978.0, + 873.0, + 2009.0, + 866.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 200.0, + 589.0, + 200.0, + 589.0, + 239.0, + 294.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1047.0, + 745.0, + 1047.0, + 745.0, + 1083.0, + 295.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 748.0, + 634.0, + 748.0, + 634.0, + 784.0, + 294.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 538.0, + 525.0, + 538.0, + 525.0, + 594.0, + 290.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 803.0, + 614.0, + 803.0, + 614.0, + 842.0, + 295.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1286.0, + 488.0, + 1286.0, + 488.0, + 1326.0, + 294.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 254.0, + 1405.0, + 254.0, + 1405.0, + 290.0, + 293.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 285.0, + 1407.0, + 285.0, + 1407.0, + 321.0, + 293.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 315.0, + 1405.0, + 315.0, + 1405.0, + 351.0, + 293.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 348.0, + 1405.0, + 348.0, + 1405.0, + 381.0, + 296.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 375.0, + 1406.0, + 375.0, + 1406.0, + 415.0, + 292.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 407.0, + 1406.0, + 407.0, + 1406.0, + 441.0, + 295.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 435.0, + 1405.0, + 435.0, + 1405.0, + 475.0, + 292.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 464.0, + 375.0, + 464.0, + 375.0, + 503.0, + 291.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 464.0, + 1239.0, + 464.0, + 1239.0, + 503.0, + 435.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1102.0, + 1403.0, + 1102.0, + 1403.0, + 1135.0, + 297.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1131.0, + 1402.0, + 1131.0, + 1402.0, + 1164.0, + 297.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1161.0, + 1408.0, + 1161.0, + 1408.0, + 1199.0, + 293.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1193.0, + 1405.0, + 1193.0, + 1405.0, + 1226.0, + 295.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1222.0, + 1308.0, + 1222.0, + 1308.0, + 1257.0, + 293.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 856.0, + 863.0, + 856.0, + 863.0, + 896.0, + 295.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 856.0, + 1405.0, + 856.0, + 1405.0, + 896.0, + 917.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 886.0, + 1407.0, + 886.0, + 1407.0, + 926.0, + 293.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 917.0, + 1408.0, + 917.0, + 1408.0, + 957.0, + 292.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 947.0, + 1405.0, + 947.0, + 1405.0, + 985.0, + 293.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 980.0, + 874.0, + 980.0, + 874.0, + 1015.0, + 293.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1646.0, + 885.0, + 1646.0, + 885.0, + 1678.0, + 296.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1677.0, + 884.0, + 1677.0, + 884.0, + 1709.0, + 296.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1708.0, + 883.0, + 1708.0, + 883.0, + 1740.0, + 295.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1737.0, + 884.0, + 1737.0, + 884.0, + 1770.0, + 294.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1769.0, + 880.0, + 1769.0, + 880.0, + 1799.0, + 295.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1797.0, + 880.0, + 1797.0, + 880.0, + 1833.0, + 293.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1828.0, + 880.0, + 1828.0, + 880.0, + 1861.0, + 295.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1859.0, + 883.0, + 1859.0, + 883.0, + 1893.0, + 296.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1889.0, + 880.0, + 1889.0, + 880.0, + 1923.0, + 296.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1448.0, + 884.0, + 1448.0, + 884.0, + 1481.0, + 296.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1482.0, + 885.0, + 1482.0, + 885.0, + 1509.0, + 294.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1511.0, + 883.0, + 1511.0, + 883.0, + 1541.0, + 296.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1538.0, + 881.0, + 1538.0, + 881.0, + 1573.0, + 294.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1571.0, + 884.0, + 1571.0, + 884.0, + 1604.0, + 294.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1602.0, + 588.0, + 1602.0, + 588.0, + 1633.0, + 297.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1344.0, + 881.0, + 1344.0, + 881.0, + 1375.0, + 297.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1370.0, + 885.0, + 1370.0, + 885.0, + 1409.0, + 294.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1403.0, + 860.0, + 1403.0, + 860.0, + 1439.0, + 295.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 613.0, + 1404.0, + 613.0, + 1404.0, + 653.0, + 292.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 648.0, + 1406.0, + 648.0, + 1406.0, + 682.0, + 295.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 678.0, + 934.0, + 678.0, + 934.0, + 711.0, + 294.0, + 711.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1365, + 1404, + 1365, + 1404, + 1640, + 297, + 1640 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1230, + 1402, + 1230, + 1402, + 1351, + 299, + 1351 + ], + "score": 0.978 + }, + { + "category_id": 5, + "poly": [ + 322, + 1014, + 1375, + 1014, + 1375, + 1134, + 322, + 1134 + ], + "score": 0.977, + "html": "
UsageNumberSeedExample
s→h70in summaryin short，inword， atosum up
h←r34inother wordstorephrase it，that istosay，i.e.
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1653, + 1405, + 1653, + 1405, + 1928, + 297, + 1928 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 757, + 1403, + 757, + 1403, + 910, + 298, + 910 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 462, + 1403, + 462, + 1403, + 615, + 298, + 615 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 295, + 1403, + 295, + 1403, + 448, + 298, + 448 + ], + "score": 0.975 + }, + { + "category_id": 2, + "poly": [ + 329, + 1947, + 1178, + 1947, + 1178, + 2007, + 329, + 2007 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 295, + 202, + 1401, + 202, + 1401, + 265, + 295, + 265 + ], + "score": 0.946 + }, + { + "category_id": 0, + "poly": [ + 299, + 704, + 563, + 704, + 563, + 738, + 299, + 738 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 298, + 648, + 432, + 648, + 432, + 681, + 298, + 681 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 298, + 1177, + 480, + 1177, + 480, + 1210, + 298, + 1210 + ], + "score": 0.908 + }, + { + "category_id": 6, + "poly": [ + 299, + 934, + 1397, + 934, + 1397, + 997, + 299, + 997 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.803 + }, + { + "category_id": 13, + "poly": [ + 549, + 1397, + 636, + 1397, + 636, + 1425, + 549, + 1425 + ], + "score": 0.9, + "latex": "s h" + }, + { + "category_id": 13, + "poly": [ + 962, + 1397, + 1049, + 1397, + 1049, + 1425, + 962, + 1425 + ], + "score": 0.9, + "latex": "h r" + }, + { + "category_id": 13, + "poly": [ + 468, + 1777, + 556, + 1777, + 556, + 1806, + 468, + 1806 + ], + "score": 0.86, + "latex": "( h \\to r" + }, + { + "category_id": 13, + "poly": [ + 1188, + 1837, + 1281, + 1837, + 1281, + 1866, + 1188, + 1866 + ], + "score": 0.85, + "latex": "s h" + }, + { + "category_id": 13, + "poly": [ + 530, + 1716, + 583, + 1716, + 583, + 1746, + 530, + 1746 + ], + "score": 0.82, + "latex": "\\ S 2 . 2 \\AA" + }, + { + "category_id": 13, + "poly": [ + 327, + 1685, + 387, + 1685, + 387, + 1716, + 327, + 1716 + ], + "score": 0.77, + "latex": "\\ S 3 . 2 )" + }, + { + "category_id": 13, + "poly": [ + 779, + 787, + 840, + 787, + 840, + 819, + 779, + 819 + ], + "score": 0.58, + "latex": "[ 6 3 ] ^ { 4 }" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1939.0, + 1181.0, + 1939.0, + 1181.0, + 1983.0, + 329.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1970.0, + 1180.0, + 1970.0, + 1180.0, + 2014.0, + 328.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 699.0, + 566.0, + 699.0, + 566.0, + 746.0, + 293.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 641.0, + 438.0, + 641.0, + 438.0, + 691.0, + 292.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1171.0, + 482.0, + 1171.0, + 482.0, + 1219.0, + 293.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 930.0, + 1403.0, + 930.0, + 1403.0, + 972.0, + 293.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 964.0, + 797.0, + 964.0, + 797.0, + 1000.0, + 295.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2091.0, + 840.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1367.0, + 1404.0, + 1367.0, + 1404.0, + 1400.0, + 296.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1396.0, + 548.0, + 1396.0, + 548.0, + 1433.0, + 292.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1396.0, + 961.0, + 1396.0, + 961.0, + 1433.0, + 637.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 1396.0, + 1405.0, + 1396.0, + 1405.0, + 1433.0, + 1050.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1426.0, + 1405.0, + 1426.0, + 1405.0, + 1462.0, + 294.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1457.0, + 1404.0, + 1457.0, + 1404.0, + 1493.0, + 294.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1487.0, + 1406.0, + 1487.0, + 1406.0, + 1523.0, + 292.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1517.0, + 1406.0, + 1517.0, + 1406.0, + 1553.0, + 295.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1547.0, + 1407.0, + 1547.0, + 1407.0, + 1583.0, + 294.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1575.0, + 1406.0, + 1575.0, + 1406.0, + 1614.0, + 294.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1607.0, + 595.0, + 1607.0, + 595.0, + 1645.0, + 294.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1231.0, + 1404.0, + 1231.0, + 1404.0, + 1263.0, + 295.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1262.0, + 1406.0, + 1262.0, + 1406.0, + 1294.0, + 295.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1292.0, + 1406.0, + 1292.0, + 1406.0, + 1324.0, + 297.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1319.0, + 1249.0, + 1319.0, + 1249.0, + 1357.0, + 292.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1651.0, + 1405.0, + 1651.0, + 1405.0, + 1690.0, + 294.0, + 1690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1682.0, + 326.0, + 1682.0, + 326.0, + 1720.0, + 292.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 1682.0, + 1406.0, + 1682.0, + 1406.0, + 1720.0, + 388.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1712.0, + 529.0, + 1712.0, + 529.0, + 1751.0, + 294.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1712.0, + 1406.0, + 1712.0, + 1406.0, + 1751.0, + 584.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1745.0, + 1406.0, + 1745.0, + 1406.0, + 1781.0, + 294.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1774.0, + 467.0, + 1774.0, + 467.0, + 1813.0, + 294.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1774.0, + 1405.0, + 1774.0, + 1405.0, + 1813.0, + 557.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1804.0, + 1403.0, + 1804.0, + 1403.0, + 1840.0, + 295.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1836.0, + 1187.0, + 1836.0, + 1187.0, + 1873.0, + 295.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1836.0, + 1405.0, + 1836.0, + 1405.0, + 1873.0, + 1282.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1867.0, + 1403.0, + 1867.0, + 1403.0, + 1900.0, + 295.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1896.0, + 1328.0, + 1896.0, + 1328.0, + 1932.0, + 294.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 756.0, + 1407.0, + 756.0, + 1407.0, + 793.0, + 294.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 786.0, + 778.0, + 786.0, + 778.0, + 823.0, + 294.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 786.0, + 1405.0, + 786.0, + 1405.0, + 823.0, + 841.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 816.0, + 1407.0, + 816.0, + 1407.0, + 852.0, + 293.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 844.0, + 1407.0, + 844.0, + 1407.0, + 885.0, + 292.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 878.0, + 1010.0, + 878.0, + 1010.0, + 914.0, + 293.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 459.0, + 1404.0, + 459.0, + 1404.0, + 498.0, + 293.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 490.0, + 1405.0, + 490.0, + 1405.0, + 528.0, + 293.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 523.0, + 1405.0, + 523.0, + 1405.0, + 560.0, + 294.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 554.0, + 1404.0, + 554.0, + 1404.0, + 587.0, + 296.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 582.0, + 1315.0, + 582.0, + 1315.0, + 622.0, + 292.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 294.0, + 1407.0, + 294.0, + 1407.0, + 331.0, + 294.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 323.0, + 1407.0, + 323.0, + 1407.0, + 362.0, + 292.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 354.0, + 1407.0, + 354.0, + 1407.0, + 393.0, + 293.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 385.0, + 1405.0, + 385.0, + 1405.0, + 422.0, + 293.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 418.0, + 682.0, + 418.0, + 682.0, + 451.0, + 294.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 202.0, + 1405.0, + 202.0, + 1405.0, + 239.0, + 294.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 232.0, + 1281.0, + 232.0, + 1281.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 303, + 919, + 1397, + 919, + 1397, + 1318, + 303, + 1318 + ], + "score": 0.985, + "html": "
REALSummSummEvalNeR18Avg.
CovCOHFACFLUINFOCOHFLUINFOREL
ROUGE-10.4980.1670.1600.1150.3260.0950.1040.1300.1470.194
ROUGE-20.4230.1840.1870.1590.2900.0260.0480.0790.0910.165
ROUGE-L0.4880.1280.1150.1050.3110.0640.0720.0890.1060.164
BERTScore0.4400.2840.1100.1930.3120.1470.1700.1310.1630.217
MoverScore0.3720.1590.1570.1290.3180.1610.1200.1880.1950.200
PRISM0.4110.2490.3450.2540.2120.5730.5320.5610.5530.410
BARTSCORE0.4410.322†0.3110.2480.2640.679†0.670t0.646†0.604t0.465
+ CNN0.4750.448‡0.382†0.356t0.356t0.653†0.640t0.616†0.5670.499
+ CNN+Para0.4710.424†0.401‡0.378t0.3130.657t0.652t0.614†0.5620.497
+Ω+Prompt0.4880.4070.3780.338f0.368f0.701t0.679t0.686t0.6200.518
" + }, + { + "category_id": 5, + "poly": [ + 313, + 305, + 1385, + 305, + 1385, + 762, + 313, + 762 + ], + "score": 0.983, + "html": "
de-enfi-en gu-enkk-enlt-enru-enzh-enAvg.
SUPERVISED METHODS
BLEURT0.1740.3740.3130.3720.3880.2200.4360.325
COMET0.2190.3690.3160.3780.4050.2260.4620.339
UNSUPERVISED METHODS
BLEU0.0540.2360.1940.2760.2490.1150.3210.206
CHRF0.1230.2920.2400.3230.3040.1770.3710.261
PRISM0.1990.3660.3200.3620.3820.2200.4340.326
BERTScore0.1900.3540.2920.3510.3810.2210.4300.317
BARTSCORE0.1560.3350.2730.3240.3220.1670.3890.281
+CNN0.1900.3650.3000.3480.3840.2080.4250.317
+ CNN+Para0.205t0.370t0.3160.378t0.386†0.219_0.442t0.331
+ CNN +Para +Prompt0.2380.3740.3180.376t0.386+0.2190.4470.337
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1732, + 1404, + 1732, + 1404, + 2008, + 297, + 2008 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1395, + 1403, + 1395, + 1403, + 1549, + 297, + 1549 + ], + "score": 0.975 + }, + { + "category_id": 6, + "poly": [ + 298, + 195, + 1404, + 195, + 1404, + 291, + 298, + 291 + ], + "score": 0.958 + }, + { + "category_id": 6, + "poly": [ + 299, + 809, + 1404, + 809, + 1404, + 904, + 299, + 904 + ], + "score": 0.939 + }, + { + "category_id": 0, + "poly": [ + 300, + 1671, + 630, + 1671, + 630, + 1704, + 300, + 1704 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 300, + 1606, + 617, + 1606, + 617, + 1639, + 300, + 1639 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.703 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2084, + 841, + 2084 + ], + "score": 0.158 + }, + { + "category_id": 13, + "poly": [ + 302, + 1458, + 414, + 1458, + 414, + 1488, + 302, + 1488 + ], + "score": 0.86, + "latex": "( p < 0 . 0 5 )" + }, + { + "category_id": 13, + "poly": [ + 1349, + 1458, + 1367, + 1458, + 1367, + 1489, + 1349, + 1489 + ], + "score": 0.81, + "latex": "\\ddagger" + }, + { + "category_id": 13, + "poly": [ + 995, + 1428, + 1012, + 1428, + 1012, + 1458, + 995, + 1458 + ], + "score": 0.8, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 1203, + 1857, + 1288, + 1857, + 1288, + 1885, + 1203, + 1885 + ], + "score": 0.66, + "latex": "{ \\tt d e } \\mathrm { - } \\in \\mathrm { n }" + }, + { + "category_id": 13, + "poly": [ + 1188, + 842, + 1228, + 842, + 1228, + 872, + 1188, + 872 + ], + "score": 0.64, + "latex": "( \\Omega )" + }, + { + "category_id": 13, + "poly": [ + 1381, + 1400, + 1407, + 1400, + 1407, + 1425, + 1381, + 1425 + ], + "score": 0.59, + "latex": "<" + }, + { + "category_id": 13, + "poly": [ + 1057, + 1766, + 1203, + 1766, + 1203, + 1795, + 1057, + 1795 + ], + "score": 0.4, + "latex": "\\left( \\mathrm { C N N + P a r a } \\right)" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 195.0, + 1404.0, + 195.0, + 1404.0, + 230.0, + 294.0, + 230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 227.0, + 1404.0, + 227.0, + 1404.0, + 261.0, + 294.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 256.0, + 1387.0, + 256.0, + 1387.0, + 293.0, + 293.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 808.0, + 1406.0, + 808.0, + 1406.0, + 846.0, + 293.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 842.0, + 1187.0, + 842.0, + 1187.0, + 876.0, + 295.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 842.0, + 1409.0, + 842.0, + 1409.0, + 876.0, + 1229.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 871.0, + 1080.0, + 871.0, + 1080.0, + 905.0, + 294.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1670.0, + 632.0, + 1670.0, + 632.0, + 1706.0, + 294.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1602.0, + 621.0, + 1602.0, + 621.0, + 1646.0, + 293.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2092.0, + 840.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1734.0, + 1404.0, + 1734.0, + 1404.0, + 1767.0, + 296.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1762.0, + 1056.0, + 1762.0, + 1056.0, + 1800.0, + 292.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 1762.0, + 1404.0, + 1762.0, + 1404.0, + 1800.0, + 1204.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1792.0, + 1407.0, + 1792.0, + 1407.0, + 1832.0, + 294.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1819.0, + 1408.0, + 1819.0, + 1408.0, + 1862.0, + 292.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1853.0, + 1202.0, + 1853.0, + 1202.0, + 1894.0, + 294.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 1853.0, + 1406.0, + 1853.0, + 1406.0, + 1894.0, + 1289.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1883.0, + 1407.0, + 1883.0, + 1407.0, + 1922.0, + 292.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1952.0, + 292.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1985.0, + 294.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1977.0, + 933.0, + 1977.0, + 933.0, + 2010.0, + 295.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1395.0, + 1380.0, + 1395.0, + 1380.0, + 1432.0, + 295.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1423.0, + 994.0, + 1423.0, + 994.0, + 1463.0, + 292.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1423.0, + 1405.0, + 1423.0, + 1405.0, + 1463.0, + 1013.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1456.0, + 301.0, + 1456.0, + 301.0, + 1492.0, + 292.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1456.0, + 1348.0, + 1456.0, + 1348.0, + 1492.0, + 415.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1368.0, + 1456.0, + 1405.0, + 1456.0, + 1405.0, + 1492.0, + 1368.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1487.0, + 1405.0, + 1487.0, + 1405.0, + 1524.0, + 293.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1517.0, + 397.0, + 1517.0, + 397.0, + 1552.0, + 294.0, + 1552.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 845, + 1360, + 1402, + 1360, + 1402, + 1747, + 845, + 1747 + ], + "score": 0.982, + "html": "
BAGELSFRESSFHOTAvg.
ROUGE-10.2340.1150.1180.156
ROUGE-20.1990.1160.0880.134
ROUGE-L0.1890.1030.1100.134
BERTScore0.2890.1560.1350.193
MoverScore0.2840.1530.1720.203
PRISM0.3050.1550.1960.219
BARTSCORE0.2470.164†0.1580.190
+ CNN0.3030.191†0.1900.228
+ CNN+Para0.330t0.185t0.211†0.242
+Ω+Prompt0.3360.238t0.235t0.270
" + }, + { + "category_id": 5, + "poly": [ + 887, + 690, + 1400, + 690, + 1400, + 1187, + 887, + 1187 + ], + "score": 0.982, + "html": "
Rank19 Q-CNNQ-XSUM
Acc.Pearson
ROUGE-10.5680.338-0.008
ROUGE-20.6300.4590.097
ROUGE-L0.5870.3570.024
BERTScore0.7130.5760.024
MoverScore0.7130.4140.054
PRISM0.7800.4790.025
FactCC [30] QAGS [66]0.700 0.7211 0.5451 0.175
Human [14] BARTSCORE0.83911 0.009
+CNN0.684 0.836‡0.661† 0.735±0.184‡
+ CNN+Para0.680t0.074
0.788
+CNN +Prompt70.7960.719f0.094
" + }, + { + "category_id": 1, + "poly": [ + 298, + 694, + 864, + 694, + 864, + 1118, + 298, + 1118 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 269, + 1405, + 269, + 1405, + 545, + 297, + 545 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1255, + 823, + 1255, + 823, + 1743, + 299, + 1743 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1884, + 1405, + 1884, + 1405, + 2007, + 299, + 2007 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 557, + 865, + 557, + 865, + 679, + 297, + 679 + ], + "score": 0.968 + }, + { + "category_id": 6, + "poly": [ + 885, + 582, + 1403, + 582, + 1403, + 673, + 885, + 673 + ], + "score": 0.929 + }, + { + "category_id": 0, + "poly": [ + 300, + 203, + 625, + 203, + 625, + 235, + 300, + 235 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 299, + 1188, + 526, + 1188, + 526, + 1219, + 299, + 1219 + ], + "score": 0.897 + }, + { + "category_id": 0, + "poly": [ + 299, + 1814, + 462, + 1814, + 462, + 1846, + 299, + 1846 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 858, + 2062, + 858, + 2084, + 841, + 2084 + ], + "score": 0.805 + }, + { + "category_id": 6, + "poly": [ + 845, + 1250, + 1403, + 1250, + 1403, + 1343, + 845, + 1343 + ], + "score": 0.782 + }, + { + "category_id": 1, + "poly": [ + 845, + 1250, + 1403, + 1250, + 1403, + 1343, + 845, + 1343 + ], + "score": 0.179 + }, + { + "category_id": 6, + "poly": [ + 884, + 582, + 1404, + 582, + 1404, + 673, + 884, + 673 + ], + "score": 0.101 + }, + { + "category_id": 13, + "poly": [ + 517, + 1650, + 591, + 1650, + 591, + 1681, + 517, + 1681 + ], + "score": 0.86, + "latex": "\\ S 4 . 3 . 2" + }, + { + "category_id": 13, + "poly": [ + 594, + 483, + 664, + 483, + 664, + 512, + 594, + 512 + ], + "score": 0.84, + "latex": "( > 0 . 6" + }, + { + "category_id": 13, + "poly": [ + 292, + 818, + 368, + 818, + 368, + 845, + 292, + 845 + ], + "score": 0.49, + "latex": "+ \\thinspace C N N" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 577.0, + 1409.0, + 577.0, + 1409.0, + 615.0, + 881.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 611.0, + 1404.0, + 611.0, + 1404.0, + 644.0, + 883.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 642.0, + 1206.0, + 642.0, + 1206.0, + 674.0, + 883.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 201.0, + 628.0, + 201.0, + 628.0, + 238.0, + 295.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1186.0, + 531.0, + 1186.0, + 531.0, + 1222.0, + 295.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1810.0, + 468.0, + 1810.0, + 468.0, + 1851.0, + 294.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2089.0, + 839.0, + 2089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1248.0, + 1404.0, + 1248.0, + 1404.0, + 1283.0, + 843.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1280.0, + 1407.0, + 1280.0, + 1407.0, + 1316.0, + 845.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1311.0, + 1044.0, + 1311.0, + 1044.0, + 1343.0, + 844.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 578.0, + 1409.0, + 578.0, + 1409.0, + 615.0, + 881.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 611.0, + 1405.0, + 611.0, + 1405.0, + 644.0, + 883.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 642.0, + 1206.0, + 642.0, + 1206.0, + 674.0, + 883.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 693.0, + 864.0, + 693.0, + 864.0, + 728.0, + 296.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 726.0, + 866.0, + 726.0, + 866.0, + 758.0, + 297.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 756.0, + 864.0, + 756.0, + 864.0, + 787.0, + 295.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 786.0, + 863.0, + 786.0, + 863.0, + 815.0, + 298.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 816.0, + 863.0, + 816.0, + 863.0, + 847.0, + 369.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 847.0, + 865.0, + 847.0, + 865.0, + 878.0, + 296.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 877.0, + 865.0, + 877.0, + 865.0, + 907.0, + 295.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 907.0, + 865.0, + 907.0, + 865.0, + 942.0, + 295.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 938.0, + 865.0, + 938.0, + 865.0, + 968.0, + 295.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 967.0, + 865.0, + 967.0, + 865.0, + 998.0, + 295.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 997.0, + 866.0, + 997.0, + 866.0, + 1031.0, + 295.0, + 1031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1028.0, + 868.0, + 1028.0, + 868.0, + 1063.0, + 295.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1060.0, + 867.0, + 1060.0, + 867.0, + 1091.0, + 297.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1089.0, + 832.0, + 1089.0, + 832.0, + 1121.0, + 296.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 269.0, + 1405.0, + 269.0, + 1405.0, + 303.0, + 294.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 295.0, + 1406.0, + 295.0, + 1406.0, + 338.0, + 291.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 330.0, + 1406.0, + 330.0, + 1406.0, + 367.0, + 294.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 357.0, + 1408.0, + 357.0, + 1408.0, + 399.0, + 292.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 385.0, + 1406.0, + 385.0, + 1406.0, + 431.0, + 291.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 419.0, + 1405.0, + 419.0, + 1405.0, + 458.0, + 294.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 450.0, + 1405.0, + 450.0, + 1405.0, + 488.0, + 292.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 479.0, + 593.0, + 479.0, + 593.0, + 518.0, + 292.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 479.0, + 1405.0, + 479.0, + 1405.0, + 518.0, + 665.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 511.0, + 1407.0, + 511.0, + 1407.0, + 548.0, + 294.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1255.0, + 825.0, + 1255.0, + 825.0, + 1290.0, + 296.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1286.0, + 826.0, + 1286.0, + 826.0, + 1317.0, + 295.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1315.0, + 826.0, + 1315.0, + 826.0, + 1350.0, + 296.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1346.0, + 824.0, + 1346.0, + 824.0, + 1378.0, + 296.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1378.0, + 826.0, + 1378.0, + 826.0, + 1409.0, + 296.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1407.0, + 825.0, + 1407.0, + 825.0, + 1442.0, + 295.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1437.0, + 825.0, + 1437.0, + 825.0, + 1471.0, + 295.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1467.0, + 825.0, + 1467.0, + 825.0, + 1501.0, + 295.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1499.0, + 824.0, + 1499.0, + 824.0, + 1528.0, + 296.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1528.0, + 826.0, + 1528.0, + 826.0, + 1562.0, + 295.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1557.0, + 827.0, + 1557.0, + 827.0, + 1595.0, + 295.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1589.0, + 825.0, + 1589.0, + 825.0, + 1624.0, + 295.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1616.0, + 826.0, + 1616.0, + 826.0, + 1653.0, + 294.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1648.0, + 516.0, + 1648.0, + 516.0, + 1684.0, + 295.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1648.0, + 825.0, + 1648.0, + 825.0, + 1684.0, + 592.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1680.0, + 825.0, + 1680.0, + 825.0, + 1713.0, + 295.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1711.0, + 425.0, + 1711.0, + 425.0, + 1747.0, + 294.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1884.0, + 1405.0, + 1884.0, + 1405.0, + 1920.0, + 294.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1914.0, + 1410.0, + 1914.0, + 1410.0, + 1950.0, + 292.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1946.0, + 1410.0, + 1946.0, + 1410.0, + 1982.0, + 293.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1975.0, + 940.0, + 1975.0, + 940.0, + 2009.0, + 294.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 557.0, + 868.0, + 557.0, + 868.0, + 590.0, + 295.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 587.0, + 865.0, + 587.0, + 865.0, + 620.0, + 295.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 619.0, + 869.0, + 619.0, + 869.0, + 653.0, + 294.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 647.0, + 759.0, + 647.0, + 759.0, + 682.0, + 295.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1248.0, + 1404.0, + 1248.0, + 1404.0, + 1283.0, + 843.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1280.0, + 1407.0, + 1280.0, + 1407.0, + 1316.0, + 845.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1311.0, + 1044.0, + 1311.0, + 1044.0, + 1343.0, + 844.0, + 1343.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1344, + 1404, + 1344, + 1404, + 1679, + 298, + 1679 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 821, + 1404, + 821, + 1404, + 1004, + 298, + 1004 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1041, + 1405, + 1041, + 1405, + 1255, + 298, + 1255 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1767, + 1403, + 1767, + 1403, + 1920, + 298, + 1920 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 304, + 215, + 1393, + 215, + 1393, + 499, + 304, + 499 + ], + "score": 0.966 + }, + { + "category_id": 4, + "poly": [ + 298, + 515, + 1404, + 515, + 1404, + 609, + 298, + 609 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 297, + 724, + 1402, + 724, + 1402, + 785, + 297, + 785 + ], + "score": 0.951 + }, + { + "category_id": 2, + "poly": [ + 297, + 1950, + 1404, + 1950, + 1404, + 2008, + 297, + 2008 + ], + "score": 0.943 + }, + { + "category_id": 0, + "poly": [ + 299, + 671, + 645, + 671, + 645, + 703, + 299, + 703 + ], + "score": 0.922 + }, + { + "category_id": 0, + "poly": [ + 299, + 1289, + 582, + 1289, + 582, + 1322, + 299, + 1322 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 299, + 1713, + 542, + 1713, + 542, + 1746, + 299, + 1746 + ], + "score": 0.908 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2084, + 840, + 2084 + ], + "score": 0.81 + }, + { + "category_id": 13, + "poly": [ + 402, + 1587, + 474, + 1587, + 474, + 1617, + 402, + 1617 + ], + "score": 0.85, + "latex": "\\ S 4 . 3 . 2" + }, + { + "category_id": 13, + "poly": [ + 1318, + 1648, + 1390, + 1648, + 1390, + 1679, + 1318, + 1679 + ], + "score": 0.81, + "latex": "\\ S 4 . 3 . 2" + }, + { + "category_id": 13, + "poly": [ + 666, + 854, + 685, + 854, + 685, + 880, + 666, + 880 + ], + "score": 0.81, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1057, + 854, + 1075, + 854, + 1075, + 881, + 1057, + 881 + ], + "score": 0.79, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1328, + 915, + 1347, + 915, + 1347, + 941, + 1328, + 941 + ], + "score": 0.78, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 412, + 1979, + 489, + 1979, + 489, + 2005, + 412, + 2005 + ], + "score": 0.52, + "latex": "\\operatorname { k k } - \\ e \\mathrm { e n }" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 205.0, + 382.0, + 205.0, + 382.0, + 241.0, + 326.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 257.0, + 382.0, + 257.0, + 382.0, + 312.0, + 306.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 229.0, + 618.0, + 229.0, + 618.0, + 311.0, + 561.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 232.0, + 722.0, + 232.0, + 722.0, + 316.0, + 644.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 222.0, + 996.0, + 222.0, + 996.0, + 255.0, + 758.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 221.0, + 1134.0, + 221.0, + 1134.0, + 390.0, + 1057.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 295.0, + 379.0, + 295.0, + 379.0, + 400.0, + 301.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 301.0, + 618.0, + 301.0, + 618.0, + 331.0, + 559.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 303.0, + 680.0, + 303.0, + 680.0, + 409.0, + 639.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 296.0, + 1130.0, + 296.0, + 1130.0, + 353.0, + 1090.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 327.0, + 718.0, + 327.0, + 718.0, + 362.0, + 666.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 365.0, + 379.0, + 365.0, + 379.0, + 398.0, + 336.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 375.0, + 720.0, + 375.0, + 720.0, + 407.0, + 676.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 375.0, + 1127.0, + 375.0, + 1127.0, + 412.0, + 1092.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 418.0, + 571.0, + 418.0, + 571.0, + 466.0, + 327.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 440.0, + 619.0, + 440.0, + 619.0, + 457.0, + 603.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 419.0, + 719.0, + 419.0, + 719.0, + 450.0, + 664.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 435.0, + 956.0, + 435.0, + 956.0, + 466.0, + 733.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 436.0, + 1033.0, + 436.0, + 1033.0, + 465.0, + 967.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 433.0, + 1206.0, + 433.0, + 1206.0, + 466.0, + 1146.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 432.0, + 1289.0, + 432.0, + 1289.0, + 468.0, + 1232.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 431.0, + 1372.0, + 431.0, + 1372.0, + 468.0, + 1316.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 465.0, + 567.0, + 465.0, + 567.0, + 503.0, + 378.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 463.0, + 953.0, + 463.0, + 953.0, + 501.0, + 735.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 466.0, + 1358.0, + 466.0, + 1358.0, + 501.0, + 1085.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.25, + 268.5, + 444.25, + 268.5, + 444.25, + 289.0, + 402.25, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 292.0, + 585.0, + 292.0, + 585.0, + 302.0, + 568.0, + 302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.75, + 338.5, + 1114.75, + 338.5, + 1114.75, + 376.5, + 1101.75, + 376.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 515.0, + 1405.0, + 515.0, + 1405.0, + 552.0, + 294.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 545.0, + 1404.0, + 545.0, + 1404.0, + 581.0, + 293.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 577.0, + 1184.0, + 577.0, + 1184.0, + 614.0, + 293.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1944.0, + 1404.0, + 1944.0, + 1404.0, + 1983.0, + 329.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1975.0, + 411.0, + 1975.0, + 411.0, + 2010.0, + 294.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1975.0, + 1166.0, + 1975.0, + 1166.0, + 2010.0, + 490.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 667.0, + 647.0, + 667.0, + 647.0, + 710.0, + 294.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1284.0, + 585.0, + 1284.0, + 585.0, + 1330.0, + 293.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1711.0, + 546.0, + 1711.0, + 546.0, + 1750.0, + 295.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1342.0, + 1404.0, + 1342.0, + 1404.0, + 1379.0, + 293.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1373.0, + 1406.0, + 1373.0, + 1406.0, + 1410.0, + 292.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1405.0, + 1406.0, + 1405.0, + 1406.0, + 1440.0, + 294.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1436.0, + 1404.0, + 1436.0, + 1404.0, + 1471.0, + 294.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1463.0, + 1405.0, + 1463.0, + 1405.0, + 1504.0, + 292.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1497.0, + 1405.0, + 1497.0, + 1405.0, + 1531.0, + 293.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1524.0, + 1405.0, + 1524.0, + 1405.0, + 1561.0, + 293.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1557.0, + 1404.0, + 1557.0, + 1404.0, + 1592.0, + 294.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1587.0, + 401.0, + 1587.0, + 401.0, + 1621.0, + 294.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1587.0, + 1406.0, + 1587.0, + 1406.0, + 1621.0, + 475.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1618.0, + 1404.0, + 1618.0, + 1404.0, + 1652.0, + 294.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1647.0, + 1317.0, + 1647.0, + 1317.0, + 1682.0, + 294.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1391.0, + 1647.0, + 1401.0, + 1647.0, + 1401.0, + 1682.0, + 1391.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 822.0, + 1405.0, + 822.0, + 1405.0, + 857.0, + 296.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 852.0, + 665.0, + 852.0, + 665.0, + 887.0, + 292.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 852.0, + 1056.0, + 852.0, + 1056.0, + 887.0, + 686.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 852.0, + 1407.0, + 852.0, + 1407.0, + 887.0, + 1076.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 883.0, + 1406.0, + 883.0, + 1406.0, + 919.0, + 294.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 913.0, + 1327.0, + 913.0, + 1327.0, + 947.0, + 292.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 913.0, + 1405.0, + 913.0, + 1405.0, + 947.0, + 1348.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 942.0, + 1409.0, + 942.0, + 1409.0, + 979.0, + 293.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 973.0, + 1121.0, + 973.0, + 1121.0, + 1007.0, + 295.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1040.0, + 1407.0, + 1040.0, + 1407.0, + 1074.0, + 295.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1068.0, + 1408.0, + 1068.0, + 1408.0, + 1108.0, + 292.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1100.0, + 1408.0, + 1100.0, + 1408.0, + 1138.0, + 293.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1131.0, + 1406.0, + 1131.0, + 1406.0, + 1166.0, + 293.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1160.0, + 1406.0, + 1160.0, + 1406.0, + 1197.0, + 293.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1193.0, + 1405.0, + 1193.0, + 1405.0, + 1228.0, + 293.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1222.0, + 421.0, + 1222.0, + 421.0, + 1259.0, + 292.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1770.0, + 1404.0, + 1770.0, + 1404.0, + 1799.0, + 298.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1797.0, + 1405.0, + 1797.0, + 1405.0, + 1833.0, + 294.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1828.0, + 1404.0, + 1828.0, + 1404.0, + 1863.0, + 294.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1859.0, + 1404.0, + 1859.0, + 1404.0, + 1891.0, + 293.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1888.0, + 1405.0, + 1888.0, + 1405.0, + 1923.0, + 294.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 723.0, + 1404.0, + 723.0, + 1404.0, + 758.0, + 295.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 753.0, + 656.0, + 753.0, + 656.0, + 787.0, + 295.0, + 787.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1762, + 1406, + 1762, + 1406, + 2007, + 298, + 2007 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1193, + 1404, + 1193, + 1404, + 1344, + 298, + 1344 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1359, + 1404, + 1359, + 1404, + 1511, + 298, + 1511 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1526, + 1405, + 1526, + 1405, + 1649, + 298, + 1649 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 924, + 1406, + 924, + 1406, + 1079, + 298, + 1079 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 299, + 788, + 1405, + 788, + 1405, + 912, + 299, + 912 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 298, + 660, + 1408, + 660, + 1408, + 755, + 298, + 755 + ], + "score": 0.961 + }, + { + "category_id": 4, + "poly": [ + 297, + 411, + 1405, + 411, + 1405, + 504, + 297, + 504 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 296, + 565, + 1402, + 565, + 1402, + 629, + 296, + 629 + ], + "score": 0.932 + }, + { + "category_id": 0, + "poly": [ + 297, + 1121, + 850, + 1121, + 850, + 1160, + 297, + 1160 + ], + "score": 0.931 + }, + { + "category_id": 3, + "poly": [ + 304, + 192, + 1407, + 192, + 1407, + 399, + 304, + 399 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 300, + 1693, + 561, + 1693, + 561, + 1731, + 300, + 1731 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.848 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 410.0, + 1403.0, + 410.0, + 1403.0, + 448.0, + 295.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 443.0, + 1405.0, + 443.0, + 1405.0, + 476.0, + 292.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 472.0, + 1108.0, + 472.0, + 1108.0, + 509.0, + 292.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1120.0, + 854.0, + 1120.0, + 854.0, + 1168.0, + 292.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 210.0, + 361.0, + 210.0, + 361.0, + 233.0, + 337.0, + 233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 215.0, + 332.0, + 215.0, + 332.0, + 356.0, + 303.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 258.0, + 358.0, + 258.0, + 358.0, + 275.0, + 344.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 349.0, + 352.0, + 349.0, + 352.0, + 360.0, + 342.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 371.0, + 1383.0, + 371.0, + 1383.0, + 401.0, + 384.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1689.0, + 565.0, + 1689.0, + 565.0, + 1737.0, + 295.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 871.0, + 2057.0, + 871.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1761.0, + 1408.0, + 1761.0, + 1408.0, + 1799.0, + 293.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1793.0, + 1406.0, + 1793.0, + 1406.0, + 1826.0, + 293.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1823.0, + 1408.0, + 1823.0, + 1408.0, + 1857.0, + 293.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1853.0, + 1406.0, + 1853.0, + 1406.0, + 1889.0, + 293.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1884.0, + 1403.0, + 1884.0, + 1403.0, + 1918.0, + 295.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1912.0, + 1406.0, + 1912.0, + 1406.0, + 1950.0, + 292.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1946.0, + 1409.0, + 1946.0, + 1409.0, + 1979.0, + 295.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1974.0, + 447.0, + 1974.0, + 447.0, + 2009.0, + 295.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1191.0, + 1408.0, + 1191.0, + 1408.0, + 1229.0, + 293.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1221.0, + 1404.0, + 1221.0, + 1404.0, + 1259.0, + 293.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1253.0, + 1406.0, + 1253.0, + 1406.0, + 1288.0, + 292.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1283.0, + 1405.0, + 1283.0, + 1405.0, + 1319.0, + 294.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1313.0, + 769.0, + 1313.0, + 769.0, + 1347.0, + 295.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1360.0, + 1404.0, + 1360.0, + 1404.0, + 1393.0, + 296.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1390.0, + 1401.0, + 1390.0, + 1401.0, + 1423.0, + 296.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1418.0, + 1405.0, + 1418.0, + 1405.0, + 1456.0, + 293.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1448.0, + 1404.0, + 1448.0, + 1404.0, + 1486.0, + 295.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1481.0, + 599.0, + 1481.0, + 599.0, + 1515.0, + 295.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1527.0, + 1405.0, + 1527.0, + 1405.0, + 1560.0, + 297.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1556.0, + 1408.0, + 1556.0, + 1408.0, + 1591.0, + 295.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1590.0, + 1407.0, + 1590.0, + 1407.0, + 1622.0, + 295.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1619.0, + 1367.0, + 1619.0, + 1367.0, + 1651.0, + 295.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 922.0, + 1409.0, + 922.0, + 1409.0, + 962.0, + 293.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 954.0, + 1408.0, + 954.0, + 1408.0, + 994.0, + 293.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 984.0, + 1407.0, + 984.0, + 1407.0, + 1025.0, + 292.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1016.0, + 1406.0, + 1016.0, + 1406.0, + 1052.0, + 293.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1048.0, + 447.0, + 1048.0, + 447.0, + 1083.0, + 296.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 786.0, + 1410.0, + 786.0, + 1410.0, + 826.0, + 293.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 820.0, + 1409.0, + 820.0, + 1409.0, + 852.0, + 294.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 849.0, + 1410.0, + 849.0, + 1410.0, + 885.0, + 294.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 880.0, + 459.0, + 880.0, + 459.0, + 914.0, + 294.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 661.0, + 1408.0, + 661.0, + 1408.0, + 695.0, + 295.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 693.0, + 1408.0, + 693.0, + 1408.0, + 727.0, + 295.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 721.0, + 1053.0, + 721.0, + 1053.0, + 758.0, + 293.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 565.0, + 1406.0, + 565.0, + 1406.0, + 601.0, + 295.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 594.0, + 483.0, + 594.0, + 483.0, + 632.0, + 294.0, + 632.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 0, + "poly": [ + 299, + 200, + 455, + 200, + 455, + 235, + 299, + 235 + ], + "score": 0.893 + }, + { + "category_id": 1, + "poly": [ + 291, + 196, + 1410, + 196, + 1410, + 2009, + 291, + 2009 + ], + "score": 0.854 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 863, + 2061, + 863, + 2086, + 836, + 2086 + ], + "score": 0.61 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 863, + 2061, + 863, + 2086, + 836, + 2086 + ], + "score": 0.483 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 196.0, + 460.0, + 196.0, + 460.0, + 241.0, + 295.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2098.0, + 832.0, + 2098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2098.0, + 832.0, + 2098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 195.0, + 462.0, + 195.0, + 462.0, + 242.0, + 294.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 254.0, + 1410.0, + 254.0, + 1410.0, + 299.0, + 303.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 285.0, + 1410.0, + 285.0, + 1410.0, + 329.0, + 351.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 317.0, + 646.0, + 317.0, + 646.0, + 353.0, + 351.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 355.0, + 1408.0, + 355.0, + 1408.0, + 401.0, + 303.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 385.0, + 1408.0, + 385.0, + 1408.0, + 431.0, + 349.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 419.0, + 1406.0, + 419.0, + 1406.0, + 457.0, + 353.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 447.0, + 1410.0, + 447.0, + 1410.0, + 489.0, + 351.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 475.0, + 1195.0, + 475.0, + 1195.0, + 519.0, + 351.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 521.0, + 1406.0, + 521.0, + 1406.0, + 558.0, + 307.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 553.0, + 1406.0, + 553.0, + 1406.0, + 590.0, + 353.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 582.0, + 1408.0, + 582.0, + 1408.0, + 620.0, + 355.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 612.0, + 1346.0, + 612.0, + 1346.0, + 650.0, + 355.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 650.0, + 1410.0, + 650.0, + 1410.0, + 692.0, + 303.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 684.0, + 1408.0, + 684.0, + 1408.0, + 722.0, + 353.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 714.0, + 1410.0, + 714.0, + 1410.0, + 752.0, + 353.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 742.0, + 1035.0, + 742.0, + 1035.0, + 782.0, + 351.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 786.0, + 1406.0, + 786.0, + 1406.0, + 824.0, + 307.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 816.0, + 1410.0, + 816.0, + 1410.0, + 853.0, + 353.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 843.0, + 1410.0, + 843.0, + 1410.0, + 885.0, + 349.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 877.0, + 842.0, + 877.0, + 842.0, + 915.0, + 353.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 913.0, + 1410.0, + 913.0, + 1410.0, + 959.0, + 303.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 943.0, + 1408.0, + 943.0, + 1408.0, + 991.0, + 349.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 979.0, + 931.0, + 979.0, + 931.0, + 1017.0, + 353.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1017.0, + 1406.0, + 1017.0, + 1406.0, + 1061.0, + 305.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1051.0, + 861.0, + 1051.0, + 861.0, + 1089.0, + 353.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1091.0, + 1408.0, + 1091.0, + 1408.0, + 1128.0, + 307.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1118.0, + 1408.0, + 1118.0, + 1408.0, + 1162.0, + 351.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1152.0, + 1404.0, + 1152.0, + 1404.0, + 1190.0, + 353.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1182.0, + 1249.0, + 1182.0, + 1249.0, + 1220.0, + 353.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1218.0, + 1408.0, + 1218.0, + 1408.0, + 1264.0, + 303.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1254.0, + 1406.0, + 1254.0, + 1406.0, + 1292.0, + 353.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1288.0, + 1404.0, + 1288.0, + 1404.0, + 1320.0, + 355.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1312.0, + 1308.0, + 1312.0, + 1308.0, + 1354.0, + 347.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1356.0, + 1408.0, + 1356.0, + 1408.0, + 1393.0, + 293.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1385.0, + 1408.0, + 1385.0, + 1408.0, + 1423.0, + 353.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1415.0, + 1406.0, + 1415.0, + 1406.0, + 1453.0, + 353.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1445.0, + 1406.0, + 1445.0, + 1406.0, + 1483.0, + 353.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1477.0, + 1408.0, + 1477.0, + 1408.0, + 1515.0, + 355.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1507.0, + 1364.0, + 1507.0, + 1364.0, + 1545.0, + 351.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1551.0, + 1406.0, + 1551.0, + 1406.0, + 1583.0, + 295.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1581.0, + 1402.0, + 1581.0, + 1402.0, + 1613.0, + 357.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1607.0, + 1410.0, + 1607.0, + 1410.0, + 1648.0, + 349.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1637.0, + 1133.0, + 1637.0, + 1133.0, + 1676.0, + 351.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1680.0, + 1404.0, + 1680.0, + 1404.0, + 1718.0, + 293.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1708.0, + 1408.0, + 1708.0, + 1408.0, + 1752.0, + 351.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1738.0, + 1408.0, + 1738.0, + 1408.0, + 1780.0, + 349.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1770.0, + 486.0, + 1770.0, + 486.0, + 1808.0, + 351.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1810.0, + 1410.0, + 1810.0, + 1410.0, + 1852.0, + 289.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1840.0, + 1412.0, + 1840.0, + 1412.0, + 1882.0, + 349.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1884.0, + 1410.0, + 1884.0, + 1410.0, + 1922.0, + 293.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1916.0, + 1408.0, + 1916.0, + 1408.0, + 1953.0, + 353.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1945.0, + 1406.0, + 1945.0, + 1406.0, + 1983.0, + 353.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1975.0, + 943.0, + 1975.0, + 943.0, + 2013.0, + 355.0, + 2013.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.791 + }, + { + "category_id": 1, + "poly": [ + 293, + 97, + 1411, + 97, + 1411, + 2027, + 293, + 2027 + ], + "score": 0.601 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.182 + }, + { + "category_id": 13, + "poly": [ + 1344, + 999, + 1403, + 999, + 1403, + 1028, + 1344, + 1028 + ], + "score": 0.33, + "latex": "2 3 r d" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 201.0, + 1410.0, + 201.0, + 1410.0, + 242.0, + 294.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 235.0, + 1129.0, + 235.0, + 1129.0, + 269.0, + 355.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 271.0, + 1410.0, + 271.0, + 1410.0, + 311.0, + 294.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 299.0, + 1412.0, + 299.0, + 1412.0, + 347.0, + 349.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 373.0, + 1410.0, + 373.0, + 1410.0, + 413.0, + 294.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 400.0, + 1410.0, + 400.0, + 1410.0, + 445.0, + 351.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 434.0, + 1406.0, + 434.0, + 1406.0, + 472.0, + 351.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 461.0, + 501.0, + 461.0, + 501.0, + 506.0, + 352.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 504.0, + 1408.0, + 504.0, + 1408.0, + 542.0, + 292.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 533.0, + 1408.0, + 533.0, + 1408.0, + 574.0, + 355.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 567.0, + 1406.0, + 567.0, + 1406.0, + 601.0, + 355.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 595.0, + 1408.0, + 595.0, + 1408.0, + 635.0, + 353.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 622.0, + 670.0, + 622.0, + 670.0, + 667.0, + 351.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 662.0, + 1406.0, + 662.0, + 1406.0, + 703.0, + 292.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 694.0, + 1408.0, + 694.0, + 1408.0, + 734.0, + 353.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 724.0, + 1410.0, + 724.0, + 1410.0, + 766.0, + 351.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 755.0, + 1002.0, + 755.0, + 1002.0, + 796.0, + 353.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 791.0, + 1410.0, + 791.0, + 1410.0, + 840.0, + 290.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 825.0, + 1283.0, + 825.0, + 1283.0, + 863.0, + 349.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 865.0, + 1408.0, + 865.0, + 1408.0, + 906.0, + 294.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 899.0, + 1404.0, + 899.0, + 1404.0, + 933.0, + 355.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 927.0, + 791.0, + 927.0, + 791.0, + 965.0, + 349.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 967.0, + 1408.0, + 967.0, + 1408.0, + 1007.0, + 296.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 996.0, + 1343.0, + 996.0, + 1343.0, + 1037.0, + 353.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 996.0, + 1408.0, + 996.0, + 1408.0, + 1037.0, + 1404.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1022.0, + 1410.0, + 1022.0, + 1410.0, + 1073.0, + 349.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1060.0, + 1114.0, + 1060.0, + 1114.0, + 1094.0, + 355.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1096.0, + 1406.0, + 1096.0, + 1406.0, + 1138.0, + 294.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1128.0, + 1376.0, + 1128.0, + 1376.0, + 1166.0, + 351.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1168.0, + 1406.0, + 1168.0, + 1406.0, + 1208.0, + 296.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1199.0, + 1408.0, + 1199.0, + 1408.0, + 1240.0, + 351.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1227.0, + 433.0, + 1227.0, + 433.0, + 1269.0, + 351.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1269.0, + 1410.0, + 1269.0, + 1410.0, + 1309.0, + 296.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1299.0, + 725.0, + 1299.0, + 725.0, + 1339.0, + 351.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1339.0, + 1410.0, + 1339.0, + 1410.0, + 1379.0, + 294.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1364.0, + 1410.0, + 1364.0, + 1410.0, + 1415.0, + 349.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1400.0, + 809.0, + 1400.0, + 809.0, + 1436.0, + 351.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1436.0, + 1315.0, + 1436.0, + 1315.0, + 1481.0, + 292.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1476.0, + 1406.0, + 1476.0, + 1406.0, + 1521.0, + 294.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1508.0, + 1410.0, + 1508.0, + 1410.0, + 1555.0, + 351.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1542.0, + 1171.0, + 1542.0, + 1171.0, + 1582.0, + 353.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1580.0, + 1281.0, + 1580.0, + 1281.0, + 1618.0, + 294.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1620.0, + 1408.0, + 1620.0, + 1408.0, + 1660.0, + 294.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1652.0, + 1408.0, + 1652.0, + 1408.0, + 1692.0, + 353.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1682.0, + 1408.0, + 1682.0, + 1408.0, + 1722.0, + 353.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1709.0, + 1036.0, + 1709.0, + 1036.0, + 1751.0, + 349.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1751.0, + 1406.0, + 1751.0, + 1406.0, + 1792.0, + 294.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1785.0, + 1404.0, + 1785.0, + 1404.0, + 1819.0, + 355.0, + 1819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1813.0, + 1406.0, + 1813.0, + 1406.0, + 1853.0, + 353.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1840.0, + 708.0, + 1840.0, + 708.0, + 1883.0, + 349.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1410.0, + 1882.0, + 1410.0, + 1923.0, + 294.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1912.0, + 1408.0, + 1912.0, + 1408.0, + 1952.0, + 353.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1944.0, + 1408.0, + 1944.0, + 1408.0, + 1984.0, + 353.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1975.0, + 641.0, + 1975.0, + 641.0, + 2009.0, + 353.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.25, + 330.5, + 639.25, + 330.5, + 639.25, + 372.5, + 350.25, + 372.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.776 + }, + { + "category_id": 1, + "poly": [ + 293, + 100, + 1409, + 100, + 1409, + 2018, + 293, + 2018 + ], + "score": 0.712 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.189 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 201.0, + 1408.0, + 201.0, + 1408.0, + 241.0, + 294.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 231.0, + 1410.0, + 231.0, + 1410.0, + 273.0, + 351.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 262.0, + 1406.0, + 262.0, + 1406.0, + 302.0, + 353.0, + 302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 294.0, + 1408.0, + 294.0, + 1408.0, + 334.0, + 353.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 321.0, + 1056.0, + 321.0, + 1056.0, + 361.0, + 353.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 363.0, + 1408.0, + 363.0, + 1408.0, + 403.0, + 294.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 393.0, + 699.0, + 393.0, + 699.0, + 433.0, + 351.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 430.0, + 1406.0, + 430.0, + 1406.0, + 470.0, + 298.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 464.0, + 593.0, + 464.0, + 593.0, + 498.0, + 355.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 502.0, + 1408.0, + 502.0, + 1408.0, + 542.0, + 296.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 531.0, + 1404.0, + 531.0, + 1404.0, + 571.0, + 355.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 563.0, + 1408.0, + 563.0, + 1408.0, + 603.0, + 353.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 590.0, + 433.0, + 590.0, + 433.0, + 632.0, + 351.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 632.0, + 1408.0, + 632.0, + 1408.0, + 672.0, + 294.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 662.0, + 1406.0, + 662.0, + 1406.0, + 702.0, + 353.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 691.0, + 944.0, + 691.0, + 944.0, + 731.0, + 355.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 731.0, + 1408.0, + 731.0, + 1408.0, + 771.0, + 294.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 765.0, + 1404.0, + 765.0, + 1404.0, + 798.0, + 355.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 792.0, + 745.0, + 792.0, + 745.0, + 830.0, + 348.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 828.0, + 1410.0, + 828.0, + 1410.0, + 874.0, + 292.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 861.0, + 1406.0, + 861.0, + 1406.0, + 903.0, + 351.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 893.0, + 559.0, + 893.0, + 559.0, + 930.0, + 348.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 929.0, + 1406.0, + 929.0, + 1406.0, + 973.0, + 292.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 960.0, + 646.0, + 960.0, + 646.0, + 1000.0, + 351.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 998.0, + 1408.0, + 998.0, + 1408.0, + 1042.0, + 289.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1032.0, + 1404.0, + 1032.0, + 1404.0, + 1072.0, + 353.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1059.0, + 1406.0, + 1059.0, + 1406.0, + 1103.0, + 351.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1095.0, + 1396.0, + 1095.0, + 1396.0, + 1128.0, + 355.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1130.0, + 1406.0, + 1130.0, + 1406.0, + 1170.0, + 294.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1162.0, + 990.0, + 1162.0, + 990.0, + 1200.0, + 351.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1200.0, + 1406.0, + 1200.0, + 1406.0, + 1238.0, + 292.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1231.0, + 1406.0, + 1231.0, + 1406.0, + 1271.0, + 353.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1263.0, + 1406.0, + 1263.0, + 1406.0, + 1303.0, + 353.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1292.0, + 1007.0, + 1292.0, + 1007.0, + 1326.0, + 355.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1328.0, + 1408.0, + 1328.0, + 1408.0, + 1372.0, + 289.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1362.0, + 1406.0, + 1362.0, + 1406.0, + 1402.0, + 353.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1393.0, + 1406.0, + 1393.0, + 1406.0, + 1433.0, + 353.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1423.0, + 1256.0, + 1423.0, + 1256.0, + 1463.0, + 353.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1460.0, + 1406.0, + 1460.0, + 1406.0, + 1500.0, + 298.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1492.0, + 1406.0, + 1492.0, + 1406.0, + 1532.0, + 353.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1524.0, + 1406.0, + 1524.0, + 1406.0, + 1563.0, + 353.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1553.0, + 771.0, + 1553.0, + 771.0, + 1591.0, + 351.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1591.0, + 1406.0, + 1591.0, + 1406.0, + 1631.0, + 296.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1622.0, + 1406.0, + 1622.0, + 1406.0, + 1662.0, + 353.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1654.0, + 1408.0, + 1654.0, + 1408.0, + 1694.0, + 351.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1683.0, + 967.0, + 1683.0, + 967.0, + 1723.0, + 353.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1725.0, + 1404.0, + 1725.0, + 1404.0, + 1759.0, + 298.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1753.0, + 1406.0, + 1753.0, + 1406.0, + 1793.0, + 355.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1780.0, + 1410.0, + 1780.0, + 1410.0, + 1824.0, + 351.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1814.0, + 431.0, + 1814.0, + 431.0, + 1849.0, + 353.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1851.0, + 1408.0, + 1851.0, + 1408.0, + 1891.0, + 296.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1883.0, + 1406.0, + 1883.0, + 1406.0, + 1923.0, + 353.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1917.0, + 1406.0, + 1917.0, + 1406.0, + 1950.0, + 351.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1944.0, + 1410.0, + 1944.0, + 1410.0, + 1984.0, + 353.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1978.0, + 840.0, + 1978.0, + 840.0, + 2011.0, + 355.0, + 2011.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.836 + }, + { + "category_id": 1, + "poly": [ + 294, + 190, + 1409, + 190, + 1409, + 2021, + 294, + 2021 + ], + "score": 0.622 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 871.0, + 2057.0, + 871.0, + 2095.0, + 832.0, + 2095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 202.0, + 1407.0, + 202.0, + 1407.0, + 241.0, + 294.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 235.0, + 1407.0, + 235.0, + 1407.0, + 269.0, + 350.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 259.0, + 1409.0, + 259.0, + 1409.0, + 305.0, + 348.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 297.0, + 1403.0, + 297.0, + 1403.0, + 329.0, + 354.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 325.0, + 498.0, + 325.0, + 498.0, + 363.0, + 352.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 365.0, + 1407.0, + 365.0, + 1407.0, + 404.0, + 294.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 395.0, + 1407.0, + 395.0, + 1407.0, + 434.0, + 352.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 428.0, + 1407.0, + 428.0, + 1407.0, + 460.0, + 354.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 458.0, + 842.0, + 458.0, + 842.0, + 490.0, + 354.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 496.0, + 1407.0, + 496.0, + 1407.0, + 534.0, + 294.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 526.0, + 1407.0, + 526.0, + 1407.0, + 564.0, + 354.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 558.0, + 1407.0, + 558.0, + 1407.0, + 597.0, + 352.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 584.0, + 668.0, + 584.0, + 668.0, + 625.0, + 350.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 623.0, + 1407.0, + 623.0, + 1407.0, + 669.0, + 290.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 655.0, + 1411.0, + 655.0, + 1411.0, + 699.0, + 352.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 687.0, + 1037.0, + 687.0, + 1037.0, + 725.0, + 354.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 729.0, + 1409.0, + 729.0, + 1409.0, + 768.0, + 294.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 760.0, + 1275.0, + 760.0, + 1275.0, + 798.0, + 352.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 800.0, + 1409.0, + 800.0, + 1409.0, + 838.0, + 294.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 830.0, + 1407.0, + 830.0, + 1407.0, + 868.0, + 356.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 860.0, + 1315.0, + 860.0, + 1315.0, + 898.0, + 354.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 900.0, + 1409.0, + 900.0, + 1409.0, + 939.0, + 294.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 933.0, + 1405.0, + 933.0, + 1405.0, + 971.0, + 356.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 961.0, + 1145.0, + 961.0, + 1145.0, + 997.0, + 352.0, + 997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1003.0, + 1407.0, + 1003.0, + 1407.0, + 1041.0, + 296.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1033.0, + 1407.0, + 1033.0, + 1407.0, + 1071.0, + 354.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1063.0, + 1405.0, + 1063.0, + 1405.0, + 1101.0, + 352.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1091.0, + 668.0, + 1091.0, + 668.0, + 1132.0, + 350.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1132.0, + 1407.0, + 1132.0, + 1407.0, + 1176.0, + 294.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1166.0, + 1403.0, + 1166.0, + 1403.0, + 1198.0, + 356.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1192.0, + 811.0, + 1192.0, + 811.0, + 1234.0, + 348.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1236.0, + 1405.0, + 1236.0, + 1405.0, + 1274.0, + 294.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1264.0, + 1163.0, + 1264.0, + 1163.0, + 1303.0, + 354.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1303.0, + 1409.0, + 1303.0, + 1409.0, + 1351.0, + 290.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1335.0, + 808.0, + 1335.0, + 808.0, + 1373.0, + 354.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1377.0, + 1407.0, + 1377.0, + 1407.0, + 1415.0, + 294.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1409.0, + 1069.0, + 1409.0, + 1069.0, + 1447.0, + 352.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1449.0, + 1407.0, + 1449.0, + 1407.0, + 1488.0, + 294.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1480.0, + 1405.0, + 1480.0, + 1405.0, + 1518.0, + 354.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1510.0, + 1409.0, + 1510.0, + 1409.0, + 1548.0, + 354.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1542.0, + 963.0, + 1542.0, + 963.0, + 1574.0, + 354.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1582.0, + 1405.0, + 1582.0, + 1405.0, + 1614.0, + 296.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1610.0, + 1407.0, + 1610.0, + 1407.0, + 1649.0, + 354.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1641.0, + 1397.0, + 1641.0, + 1397.0, + 1679.0, + 352.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1681.0, + 1409.0, + 1681.0, + 1409.0, + 1719.0, + 296.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1713.0, + 1409.0, + 1713.0, + 1409.0, + 1751.0, + 352.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1739.0, + 809.0, + 1739.0, + 809.0, + 1781.0, + 352.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1783.0, + 1409.0, + 1783.0, + 1409.0, + 1822.0, + 294.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1814.0, + 1409.0, + 1814.0, + 1409.0, + 1852.0, + 354.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1840.0, + 1039.0, + 1840.0, + 1039.0, + 1882.0, + 352.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1884.0, + 1407.0, + 1884.0, + 1407.0, + 1922.0, + 294.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1914.0, + 1407.0, + 1914.0, + 1407.0, + 1952.0, + 354.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1944.0, + 1407.0, + 1944.0, + 1407.0, + 1982.0, + 352.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1972.0, + 706.0, + 1972.0, + 706.0, + 2013.0, + 350.0, + 2013.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.717 + }, + { + "category_id": 1, + "poly": [ + 292, + 116, + 1410, + 116, + 1410, + 1979, + 292, + 1979 + ], + "score": 0.686 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.346 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 197.0, + 1410.0, + 197.0, + 1410.0, + 244.0, + 290.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 232.0, + 907.0, + 232.0, + 907.0, + 270.0, + 352.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 270.0, + 1412.0, + 270.0, + 1412.0, + 315.0, + 292.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 297.0, + 1412.0, + 297.0, + 1412.0, + 346.0, + 350.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 330.0, + 1408.0, + 330.0, + 1408.0, + 377.0, + 348.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 365.0, + 1408.0, + 365.0, + 1408.0, + 403.0, + 352.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 395.0, + 669.0, + 395.0, + 669.0, + 434.0, + 354.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 436.0, + 1406.0, + 436.0, + 1406.0, + 475.0, + 294.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 465.0, + 1408.0, + 465.0, + 1408.0, + 506.0, + 350.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 497.0, + 1406.0, + 497.0, + 1406.0, + 536.0, + 352.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 524.0, + 967.0, + 524.0, + 967.0, + 567.0, + 348.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 563.0, + 1410.0, + 563.0, + 1410.0, + 610.0, + 298.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 598.0, + 1412.0, + 598.0, + 1412.0, + 643.0, + 352.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 630.0, + 809.0, + 630.0, + 809.0, + 669.0, + 354.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 665.0, + 1408.0, + 665.0, + 1408.0, + 712.0, + 296.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 698.0, + 1406.0, + 698.0, + 1406.0, + 737.0, + 354.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 729.0, + 1404.0, + 729.0, + 1404.0, + 767.0, + 354.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 759.0, + 1408.0, + 759.0, + 1408.0, + 800.0, + 350.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 792.0, + 1408.0, + 792.0, + 1408.0, + 831.0, + 352.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 823.0, + 1408.0, + 823.0, + 1408.0, + 861.0, + 352.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 853.0, + 843.0, + 853.0, + 843.0, + 892.0, + 352.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 890.0, + 1406.0, + 890.0, + 1406.0, + 935.0, + 292.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 925.0, + 741.0, + 925.0, + 741.0, + 961.0, + 350.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 966.0, + 1262.0, + 966.0, + 1262.0, + 1005.0, + 294.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1002.0, + 1408.0, + 1002.0, + 1408.0, + 1050.0, + 290.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1035.0, + 1408.0, + 1035.0, + 1408.0, + 1078.0, + 350.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1068.0, + 1410.0, + 1068.0, + 1410.0, + 1107.0, + 354.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1101.0, + 841.0, + 1101.0, + 841.0, + 1133.0, + 356.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1140.0, + 1406.0, + 1140.0, + 1406.0, + 1178.0, + 294.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1170.0, + 1406.0, + 1170.0, + 1406.0, + 1207.0, + 352.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1197.0, + 869.0, + 1197.0, + 869.0, + 1240.0, + 348.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1242.0, + 1410.0, + 1242.0, + 1410.0, + 1281.0, + 294.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1272.0, + 1410.0, + 1272.0, + 1410.0, + 1311.0, + 354.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1299.0, + 430.0, + 1299.0, + 430.0, + 1340.0, + 352.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1344.0, + 1408.0, + 1344.0, + 1408.0, + 1383.0, + 294.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1373.0, + 1406.0, + 1373.0, + 1406.0, + 1411.0, + 354.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1401.0, + 1406.0, + 1401.0, + 1406.0, + 1444.0, + 350.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1434.0, + 1410.0, + 1434.0, + 1410.0, + 1473.0, + 354.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1463.0, + 1414.0, + 1463.0, + 1414.0, + 1506.0, + 348.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1501.0, + 1410.0, + 1501.0, + 1410.0, + 1549.0, + 290.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1534.0, + 1298.0, + 1534.0, + 1298.0, + 1577.0, + 350.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1573.0, + 1408.0, + 1573.0, + 1408.0, + 1620.0, + 290.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1608.0, + 1406.0, + 1608.0, + 1406.0, + 1647.0, + 354.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1638.0, + 1406.0, + 1638.0, + 1406.0, + 1677.0, + 350.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1665.0, + 707.0, + 1665.0, + 707.0, + 1706.0, + 350.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1706.0, + 1408.0, + 1706.0, + 1408.0, + 1749.0, + 290.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1735.0, + 1410.0, + 1735.0, + 1410.0, + 1782.0, + 350.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1769.0, + 1406.0, + 1769.0, + 1406.0, + 1808.0, + 354.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1800.0, + 1137.0, + 1800.0, + 1137.0, + 1839.0, + 352.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1841.0, + 1406.0, + 1841.0, + 1406.0, + 1880.0, + 294.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1872.0, + 1406.0, + 1872.0, + 1406.0, + 1910.0, + 354.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1896.0, + 1410.0, + 1896.0, + 1410.0, + 1945.0, + 350.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1931.0, + 1398.0, + 1931.0, + 1398.0, + 1974.0, + 348.0, + 1974.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/6YEQUn0QICG/images/03cfaf28c16a6c81076be28c9e3d3ea9f7e19c2f436436b3651b685a8ac7f6f1.jpg b/parse/train/6YEQUn0QICG/images/03cfaf28c16a6c81076be28c9e3d3ea9f7e19c2f436436b3651b685a8ac7f6f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b43d1779969b4371d85bfa1d7d152ff23d1ea151 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/03cfaf28c16a6c81076be28c9e3d3ea9f7e19c2f436436b3651b685a8ac7f6f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a96845069b9f4ff5289348f1e5e3821c91569d1405f425bf8cd5a33d9fbb7c03 +size 15726 diff --git a/parse/train/6YEQUn0QICG/images/0ebeb0c9d52d66cd2d7d1e388a22e633f984eaffaaa2af85dafd22cc5b1ca11a.jpg b/parse/train/6YEQUn0QICG/images/0ebeb0c9d52d66cd2d7d1e388a22e633f984eaffaaa2af85dafd22cc5b1ca11a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..77d1a92a2ac3d4377bbd39b2e34fb3ea36ad2eea --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/0ebeb0c9d52d66cd2d7d1e388a22e633f984eaffaaa2af85dafd22cc5b1ca11a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f794e480cc20faec9954a865599b26670f3d1f3ea0694f4e636a1f60a266c45 +size 22888 diff --git a/parse/train/6YEQUn0QICG/images/12374984512f0120a266c28c530fb940caf11f9814989378f3db97717b4d3f77.jpg b/parse/train/6YEQUn0QICG/images/12374984512f0120a266c28c530fb940caf11f9814989378f3db97717b4d3f77.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8fa7a84eb9255413e72224067ff5854d04463df0 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/12374984512f0120a266c28c530fb940caf11f9814989378f3db97717b4d3f77.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd5eb0eec4f2473dce40fa804fd8c807bf860df9bfaae21592f2a72ba2479986 +size 23631 diff --git a/parse/train/6YEQUn0QICG/images/16ff2ba7407495bace8c955a54164a1f63bda76bf8210709284d897843a70504.jpg b/parse/train/6YEQUn0QICG/images/16ff2ba7407495bace8c955a54164a1f63bda76bf8210709284d897843a70504.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3de242346df76540e86692036a8a12a9c45cfcfd --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/16ff2ba7407495bace8c955a54164a1f63bda76bf8210709284d897843a70504.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc4b08c317ffc7f8011d83712c518ce06860adb3ed1336095a3a828135b5785d +size 24470 diff --git a/parse/train/6YEQUn0QICG/images/17a9ef49d564a047bf5db1ce46f8ec5696661b3ba76dd82b931ba1640cd365d4.jpg b/parse/train/6YEQUn0QICG/images/17a9ef49d564a047bf5db1ce46f8ec5696661b3ba76dd82b931ba1640cd365d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..83ed5b40dbd272b93bea4070f7129f32fcd2f59c --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/17a9ef49d564a047bf5db1ce46f8ec5696661b3ba76dd82b931ba1640cd365d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:796c28fa380286e2b4bbbb12b2c9c7574d5b4a66e0aec63ef6a54b9dac713ac9 +size 6639 diff --git a/parse/train/6YEQUn0QICG/images/196f0e3e1fbe5e265f1ae776b173c4a2ad9f4f7bfe9a669a4b053fcf2519c56a.jpg b/parse/train/6YEQUn0QICG/images/196f0e3e1fbe5e265f1ae776b173c4a2ad9f4f7bfe9a669a4b053fcf2519c56a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..949dcad02a12505ec2a88c19ddf95a5447c87698 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/196f0e3e1fbe5e265f1ae776b173c4a2ad9f4f7bfe9a669a4b053fcf2519c56a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5369cbaa7766b4fdd696e72533e09389893371fe7de17ee993013d46fcec62bc +size 130448 diff --git a/parse/train/6YEQUn0QICG/images/1c49cd4a1352be09d7a1290da665ec569cbec5edb890303af31a816cfa95079b.jpg b/parse/train/6YEQUn0QICG/images/1c49cd4a1352be09d7a1290da665ec569cbec5edb890303af31a816cfa95079b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0fe9dcc52d5afb166b365065913504d0c1f1ac72 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/1c49cd4a1352be09d7a1290da665ec569cbec5edb890303af31a816cfa95079b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84cd580efae60e9b2493e749ae1bb6efe4406a0cd8c3daa77647ffb2bd51f1ab +size 73183 diff --git a/parse/train/6YEQUn0QICG/images/1e2840200a60301b07e3c5b5b4a806db2d4ae5b04e98ee86a534faade5c820e0.jpg b/parse/train/6YEQUn0QICG/images/1e2840200a60301b07e3c5b5b4a806db2d4ae5b04e98ee86a534faade5c820e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e3abe62dd827528a303c4d6a2b75772bf1a71eaf --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/1e2840200a60301b07e3c5b5b4a806db2d4ae5b04e98ee86a534faade5c820e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:14012bf7aa0f588ea20ab88be77ee1dc6fffe9a1442b29041d6e9a6241d4c4e9 +size 52049 diff --git a/parse/train/6YEQUn0QICG/images/1e44ce48e3e0ee9de712a1269355b4872cd79606cb44f04314b4244c3c06b432.jpg b/parse/train/6YEQUn0QICG/images/1e44ce48e3e0ee9de712a1269355b4872cd79606cb44f04314b4244c3c06b432.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5d8aa61f8416bc402c71dad0494ec60a14161fdb --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/1e44ce48e3e0ee9de712a1269355b4872cd79606cb44f04314b4244c3c06b432.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e013faf7771b115c812d83de72d5bc5d72d2439963e8d333be9202cbc64ecea +size 4111 diff --git a/parse/train/6YEQUn0QICG/images/2041880eb73898a1126028ce47dd82cf4feed3f9f69db2f9392f17505694e616.jpg b/parse/train/6YEQUn0QICG/images/2041880eb73898a1126028ce47dd82cf4feed3f9f69db2f9392f17505694e616.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eea6dfba95417cbede8a2973aca7acc186cb3c9a --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/2041880eb73898a1126028ce47dd82cf4feed3f9f69db2f9392f17505694e616.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cffd1ff771f9aa9f711dcf2a722fd1d592032c8604b7798a13c534e154d6f72 +size 13735 diff --git a/parse/train/6YEQUn0QICG/images/37bf373cc0ce6c43044fb24349c30838212d2604e29a059195c0d90b218609f8.jpg b/parse/train/6YEQUn0QICG/images/37bf373cc0ce6c43044fb24349c30838212d2604e29a059195c0d90b218609f8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1d1a3f85c410309506ce6742ab1d8be00e74d67f --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/37bf373cc0ce6c43044fb24349c30838212d2604e29a059195c0d90b218609f8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f827e10404e6f8f7a37e021855f9bf2ab3fb8b11cec5c1d56e403b54bd5bdce3 +size 4595 diff --git a/parse/train/6YEQUn0QICG/images/38f977efc05dbe6b767563413356c6a50cf918f0febe3b2b47c0223723c1dfcd.jpg b/parse/train/6YEQUn0QICG/images/38f977efc05dbe6b767563413356c6a50cf918f0febe3b2b47c0223723c1dfcd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..29e2dad25beaf5834016bcd2490d69d8b9363fa5 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/38f977efc05dbe6b767563413356c6a50cf918f0febe3b2b47c0223723c1dfcd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ceefaf01a844090ac7ce44a1fc659cfb12b34151bd4380ca40c67b044c37cc7 +size 50727 diff --git a/parse/train/6YEQUn0QICG/images/3a625261941d3cdd95e90410f2a1988eb70923008fe596735e44e700a55fc4ba.jpg b/parse/train/6YEQUn0QICG/images/3a625261941d3cdd95e90410f2a1988eb70923008fe596735e44e700a55fc4ba.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1d9f10d32a964356f8ece7afcad67aa0673e0f84 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/3a625261941d3cdd95e90410f2a1988eb70923008fe596735e44e700a55fc4ba.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c42c509161a60a538f904062bd0e136e0be5e27ac08389a047c0c70a04e2ade0 +size 31493 diff --git a/parse/train/6YEQUn0QICG/images/3cb2cdabb35d5e52e3308eee1550d60e963503b41ad7093f8f4e6bbb51705352.jpg b/parse/train/6YEQUn0QICG/images/3cb2cdabb35d5e52e3308eee1550d60e963503b41ad7093f8f4e6bbb51705352.jpg new file mode 100644 index 0000000000000000000000000000000000000000..12d74f85b8c13923c1b5974f2fb69ad6085693d9 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/3cb2cdabb35d5e52e3308eee1550d60e963503b41ad7093f8f4e6bbb51705352.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0d35edff85fe6ff0e4bb6ed09bfd99485c80a10efb1bfe7113976e4b13515f0 +size 18018 diff --git a/parse/train/6YEQUn0QICG/images/3d4fe8ab0dd277b457dcd52fb36a4c6e6fec2c77b067e775581c4409dca89a46.jpg b/parse/train/6YEQUn0QICG/images/3d4fe8ab0dd277b457dcd52fb36a4c6e6fec2c77b067e775581c4409dca89a46.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f20480fda3caaac236d74ab3f95f47f112af2f89 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/3d4fe8ab0dd277b457dcd52fb36a4c6e6fec2c77b067e775581c4409dca89a46.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:014c640cde10ee49e4268aa7e9c65ff8d4971435c89f9a5a706f592f078a1682 +size 52224 diff --git a/parse/train/6YEQUn0QICG/images/3da4dd1e3d585fcd139bfa6130668a4ed359338445d8915bd3d2e3e3cea18dfa.jpg b/parse/train/6YEQUn0QICG/images/3da4dd1e3d585fcd139bfa6130668a4ed359338445d8915bd3d2e3e3cea18dfa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6333be6273ae9ef9489b05daf3edbf0bf0ce0cf1 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/3da4dd1e3d585fcd139bfa6130668a4ed359338445d8915bd3d2e3e3cea18dfa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e7455d3d0b49d786c7f68340513735cceede72aa37a2fb6059ff8d7cde3de3b +size 11221 diff --git a/parse/train/6YEQUn0QICG/images/42adba1d27911cac0168a460d92bea14f1a24a1905e5e221b0233fc636cf4d86.jpg b/parse/train/6YEQUn0QICG/images/42adba1d27911cac0168a460d92bea14f1a24a1905e5e221b0233fc636cf4d86.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7481888023ccc01b8a877366a1586de5bff32769 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/42adba1d27911cac0168a460d92bea14f1a24a1905e5e221b0233fc636cf4d86.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d575b13a855fbde05c9de22b85a444586955585ee1c3878598807923a48a5c70 +size 5611 diff --git a/parse/train/6YEQUn0QICG/images/450bfce125adec3429ba426fb339d739e456643c2bfb9e6603f4d0c065903a8c.jpg b/parse/train/6YEQUn0QICG/images/450bfce125adec3429ba426fb339d739e456643c2bfb9e6603f4d0c065903a8c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d36477c76e3df30fd78c469fe1d4bb8ff8c436e2 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/450bfce125adec3429ba426fb339d739e456643c2bfb9e6603f4d0c065903a8c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bbea0e9010e819a497c32bd8ceb91ac9750b056e3da274dcb0953e8f32972332 +size 8169 diff --git a/parse/train/6YEQUn0QICG/images/4704d2e63ad9c9b6813f1f6f6384541715c3881346459df4d24594991619589d.jpg b/parse/train/6YEQUn0QICG/images/4704d2e63ad9c9b6813f1f6f6384541715c3881346459df4d24594991619589d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c86a616baa21c6fd682c28598ac0fe246824baad --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/4704d2e63ad9c9b6813f1f6f6384541715c3881346459df4d24594991619589d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e11b0fc430b16fb89e5ad5d5b18c05e03a9b5e54cb82cc4ff67ce94719de7656 +size 5687 diff --git a/parse/train/6YEQUn0QICG/images/50afbaadf11043239ce45934848c71334ab45f5170b7c9140f8ef9040b89222e.jpg b/parse/train/6YEQUn0QICG/images/50afbaadf11043239ce45934848c71334ab45f5170b7c9140f8ef9040b89222e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..33e120d055fffa1e81875eebe93e1d2711c523f5 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/50afbaadf11043239ce45934848c71334ab45f5170b7c9140f8ef9040b89222e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bbe7d4ebffc4682d58a0e38355e3b7731be12866b08ad8c789cc8783092fbe96 +size 121130 diff --git a/parse/train/6YEQUn0QICG/images/592b93681ae621869581c4fd331f63c28a592ef7454a342c5c746587ca312960.jpg b/parse/train/6YEQUn0QICG/images/592b93681ae621869581c4fd331f63c28a592ef7454a342c5c746587ca312960.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d82bfe40ee800b925cf9ab086f5327c2696479eb --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/592b93681ae621869581c4fd331f63c28a592ef7454a342c5c746587ca312960.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad6316002e2440bdf8b1ccbc75f3629d9f3a08f9e55c41282ab0a5bb3ae01c62 +size 4103 diff --git a/parse/train/6YEQUn0QICG/images/594a7c670b193ac730a3ef0c708ab4dad345be10505b6f4606980977a5109d39.jpg b/parse/train/6YEQUn0QICG/images/594a7c670b193ac730a3ef0c708ab4dad345be10505b6f4606980977a5109d39.jpg new file mode 100644 index 0000000000000000000000000000000000000000..30b3928a3a5009860e1d7081f36a6e77572d9946 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/594a7c670b193ac730a3ef0c708ab4dad345be10505b6f4606980977a5109d39.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e5e267d6c224b2d8c689a43086c6beab6223974317c2c7e45a7f782ff4f8353 +size 30675 diff --git a/parse/train/6YEQUn0QICG/images/62420797776691eb600624283304fc2d3cea09d683088c4f0fb1d74b5fed52e9.jpg b/parse/train/6YEQUn0QICG/images/62420797776691eb600624283304fc2d3cea09d683088c4f0fb1d74b5fed52e9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b8de99537761391dd5cea252fae5d062f1510bc6 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/62420797776691eb600624283304fc2d3cea09d683088c4f0fb1d74b5fed52e9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72657735559d81b93124d78ad6642455122c700baae14040819484457278bafd +size 64444 diff --git a/parse/train/6YEQUn0QICG/images/65ac6893bdeb06868dcdecfe5716400b6060cc2ce9508f2466348cea415e00d8.jpg b/parse/train/6YEQUn0QICG/images/65ac6893bdeb06868dcdecfe5716400b6060cc2ce9508f2466348cea415e00d8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b7943ae37f787eb7a936031198eb1d317fb2a0cb --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/65ac6893bdeb06868dcdecfe5716400b6060cc2ce9508f2466348cea415e00d8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf05cd2a1f4262dda662cd93954425ecb368daeb51fa2fc9a51fd03d54c09da7 +size 7517 diff --git a/parse/train/6YEQUn0QICG/images/67382127c3f90e789edb85c8213efdbc77e02c3b30b239547f486ae44400fbfd.jpg b/parse/train/6YEQUn0QICG/images/67382127c3f90e789edb85c8213efdbc77e02c3b30b239547f486ae44400fbfd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93ee25f8b53f63e41d84518d1642f344ef61d00c --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/67382127c3f90e789edb85c8213efdbc77e02c3b30b239547f486ae44400fbfd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b068aba3edd978a0f82c6538dad872560f78ddfc8e92e3da8b77fb417759a27 +size 212489 diff --git a/parse/train/6YEQUn0QICG/images/6e99ced75e8b47eecde9e96d64eff11f2a1f59f39117c620094161accbacf4e7.jpg b/parse/train/6YEQUn0QICG/images/6e99ced75e8b47eecde9e96d64eff11f2a1f59f39117c620094161accbacf4e7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f2816594a858737ad870eb1a9b7e5e472c44d3b8 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/6e99ced75e8b47eecde9e96d64eff11f2a1f59f39117c620094161accbacf4e7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe990cc301976e181b1463fc9d4f01d1fbf60033cf82c911fbb3f5c72604b627 +size 33892 diff --git a/parse/train/6YEQUn0QICG/images/76722d4bdb0b4be51d8cd8e96e47b86980f1e7bffd7e9a52a36feb9f16a95d95.jpg b/parse/train/6YEQUn0QICG/images/76722d4bdb0b4be51d8cd8e96e47b86980f1e7bffd7e9a52a36feb9f16a95d95.jpg new file mode 100644 index 0000000000000000000000000000000000000000..862183ae043e26bd564d782f44bc7678303137a4 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/76722d4bdb0b4be51d8cd8e96e47b86980f1e7bffd7e9a52a36feb9f16a95d95.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b8db0db1c00ba6c17888ca83a0b92c5da87db1eee462c55e016dfb705b3082a +size 12227 diff --git a/parse/train/6YEQUn0QICG/images/7f2e394665e0dae34d2dd5668dd167cd80f46f657c0708330454942f164c74d2.jpg b/parse/train/6YEQUn0QICG/images/7f2e394665e0dae34d2dd5668dd167cd80f46f657c0708330454942f164c74d2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..618288b91b3b5e5e17822d8de0211e83e0deaf71 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/7f2e394665e0dae34d2dd5668dd167cd80f46f657c0708330454942f164c74d2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1779f2bba6ee2ef4e834956a5c2749ba469536d1bdd541c7869c7346be17bddd +size 78765 diff --git a/parse/train/6YEQUn0QICG/images/837cf66a339aa1233ba0a9a4118e63f4b06695b5a87f45e162d2ac9efd91e4fe.jpg b/parse/train/6YEQUn0QICG/images/837cf66a339aa1233ba0a9a4118e63f4b06695b5a87f45e162d2ac9efd91e4fe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9a2da56b79c6b22890087e433824c1e321c7bbbb --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/837cf66a339aa1233ba0a9a4118e63f4b06695b5a87f45e162d2ac9efd91e4fe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad7c93977ed22fce75ca2cfc888f4908a8b3d5830a3832837faf7ca7b991f2d6 +size 29773 diff --git a/parse/train/6YEQUn0QICG/images/864b5caa4ce593142b7a99d8d067ff3cd3a4e7bb964686dfba97d079dfb68d90.jpg b/parse/train/6YEQUn0QICG/images/864b5caa4ce593142b7a99d8d067ff3cd3a4e7bb964686dfba97d079dfb68d90.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a7f1a71aa1dce033637a12a0a2ef1afefaffff6 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/864b5caa4ce593142b7a99d8d067ff3cd3a4e7bb964686dfba97d079dfb68d90.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26ca2e3ae77dfdf04e09a49319368e7bbc024074fe72a3af5dd53cfb4cbe7426 +size 13952 diff --git a/parse/train/6YEQUn0QICG/images/8c355270dca8f569d576772bf622cc1f7dd0ef6d217595b8a4d203e3a5c70fd3.jpg b/parse/train/6YEQUn0QICG/images/8c355270dca8f569d576772bf622cc1f7dd0ef6d217595b8a4d203e3a5c70fd3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0c35e748c990d6fae78f26025ed1dbdca2762bb5 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/8c355270dca8f569d576772bf622cc1f7dd0ef6d217595b8a4d203e3a5c70fd3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6b6cd94ef30080dc16ab01e9090245d36ca81e8e7f88ecb300d11e2f42b95be +size 7583 diff --git a/parse/train/6YEQUn0QICG/images/8d7c7d6c88c4050284340d7dcaeea9f2d4425955c2a65e22c1e97f87fe84f7e5.jpg b/parse/train/6YEQUn0QICG/images/8d7c7d6c88c4050284340d7dcaeea9f2d4425955c2a65e22c1e97f87fe84f7e5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..da296f408c3bead2bc003d1a985f1d7abc547a11 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/8d7c7d6c88c4050284340d7dcaeea9f2d4425955c2a65e22c1e97f87fe84f7e5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22ad660ce9f29203ae87ec923c4a5d6abad2ae91ffb07d2463a41f0af6622b49 +size 32363 diff --git a/parse/train/6YEQUn0QICG/images/8e2a2fdf81a8efaa896c5c6775847ee949dc6ed4cbc13d8ab7d8e9051736773b.jpg b/parse/train/6YEQUn0QICG/images/8e2a2fdf81a8efaa896c5c6775847ee949dc6ed4cbc13d8ab7d8e9051736773b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6dcdf6fcc4b60859722b3654e8ae7432b3fb2063 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/8e2a2fdf81a8efaa896c5c6775847ee949dc6ed4cbc13d8ab7d8e9051736773b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:646bef92a582163365f3afea6e3e40b94a4e018025a995689fe3b1b1bb4fc9d6 +size 34078 diff --git a/parse/train/6YEQUn0QICG/images/9092214067a0f430a7de5f89cc805954beb11139b563366e7a32f0fae63cf12a.jpg b/parse/train/6YEQUn0QICG/images/9092214067a0f430a7de5f89cc805954beb11139b563366e7a32f0fae63cf12a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4fd7fba7830b75a7fdc2e3e35de4abb8f4b70ebd --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/9092214067a0f430a7de5f89cc805954beb11139b563366e7a32f0fae63cf12a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5b2747b10ceb6901f2bdcb8ced8efee84eaf7a554c3e2f194f2ec216416e1e8 +size 30660 diff --git a/parse/train/6YEQUn0QICG/images/95f881ad2ea770f8eff5d3510f8d601a42dbc51af6b6e252c4bdad5e958c734b.jpg b/parse/train/6YEQUn0QICG/images/95f881ad2ea770f8eff5d3510f8d601a42dbc51af6b6e252c4bdad5e958c734b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e438360429eef848058cb07995b4b86d2a29d377 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/95f881ad2ea770f8eff5d3510f8d601a42dbc51af6b6e252c4bdad5e958c734b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f6abbd833e01b1cae6f26530ca0c67c476983868868b0f51841dedec5dbdf5e +size 17794 diff --git a/parse/train/6YEQUn0QICG/images/97eac8b4d8f28d29fdd8a00e11e739dc403181d1e75c030f77cd8de50fbeec3c.jpg b/parse/train/6YEQUn0QICG/images/97eac8b4d8f28d29fdd8a00e11e739dc403181d1e75c030f77cd8de50fbeec3c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..46d33afe7a8b29b9f4e2828d504fd3747430b483 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/97eac8b4d8f28d29fdd8a00e11e739dc403181d1e75c030f77cd8de50fbeec3c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b772c1f14c57a1c0421254dffda7c062ca1361e87cee1d41c6700f30e134f533 +size 46384 diff --git a/parse/train/6YEQUn0QICG/images/ae95917e3702dcabb53411a932cba7f69f42d75dd387a3f6048fe8c6a1cd2b57.jpg b/parse/train/6YEQUn0QICG/images/ae95917e3702dcabb53411a932cba7f69f42d75dd387a3f6048fe8c6a1cd2b57.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a9220784cafcf3eb14111f722af4f6ccacb36087 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/ae95917e3702dcabb53411a932cba7f69f42d75dd387a3f6048fe8c6a1cd2b57.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71d1b160b4c358477f88c6b82a46f0b06f01e0a435a767ee6e9cae4cfd095b59 +size 59227 diff --git a/parse/train/6YEQUn0QICG/images/afffa84bcb7822b25a768f3855507e618e3b37388e89de1a3235dda8444e22ad.jpg b/parse/train/6YEQUn0QICG/images/afffa84bcb7822b25a768f3855507e618e3b37388e89de1a3235dda8444e22ad.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5253e693194bca710c7990966c713a54f88a688c --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/afffa84bcb7822b25a768f3855507e618e3b37388e89de1a3235dda8444e22ad.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:999989dbc2198e21145838793bdb196669344136ae5eccc226db9a8b574461b7 +size 26953 diff --git a/parse/train/6YEQUn0QICG/images/b32247955ee01e7e112a07a28ac329e581fcd0e128198c86fc1317a3f38e07d4.jpg b/parse/train/6YEQUn0QICG/images/b32247955ee01e7e112a07a28ac329e581fcd0e128198c86fc1317a3f38e07d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89a56a8b98391ee63ffe3035f9cb106da05c4173 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/b32247955ee01e7e112a07a28ac329e581fcd0e128198c86fc1317a3f38e07d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95620692f82d352db853ded324a6b5ac2b3817edcdd2e595a031722c9cb5e2fe +size 5207 diff --git a/parse/train/6YEQUn0QICG/images/b3343cfec2b698a0bdd144bee549d8b5f236f353b87985e5af3b05942e106f38.jpg b/parse/train/6YEQUn0QICG/images/b3343cfec2b698a0bdd144bee549d8b5f236f353b87985e5af3b05942e106f38.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ac88b3f04585a329f7378addcc995203fc9a690e --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/b3343cfec2b698a0bdd144bee549d8b5f236f353b87985e5af3b05942e106f38.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5715fd3ab00557ae0dd58966d6d302f4ff47718dd0b6c1750a02c644e2768bbe +size 17081 diff --git a/parse/train/6YEQUn0QICG/images/b57b6746cae300287423d1e9406c319d55b772985782445dd06b3b8d05f73605.jpg b/parse/train/6YEQUn0QICG/images/b57b6746cae300287423d1e9406c319d55b772985782445dd06b3b8d05f73605.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b8fec47804a53fc27c392048669ce52a75588e61 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/b57b6746cae300287423d1e9406c319d55b772985782445dd06b3b8d05f73605.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:586b9263d565ec231b4af79046d5d1db6ec071d41c64e24c2072f8fae5cd38f8 +size 30592 diff --git a/parse/train/6YEQUn0QICG/images/ca4b0cbbed428d4c4d78a89ad10306e889af9761f60870bec124a73930f50c80.jpg b/parse/train/6YEQUn0QICG/images/ca4b0cbbed428d4c4d78a89ad10306e889af9761f60870bec124a73930f50c80.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4e937d73c75c188bc7c6aa0b65385b2dc5d4c75c --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/ca4b0cbbed428d4c4d78a89ad10306e889af9761f60870bec124a73930f50c80.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02c9d4e4bb32d594249d8de9bc58da59f2f95c9f4aecd3623622a7dba111cd6f +size 2879 diff --git a/parse/train/6YEQUn0QICG/images/ce9352a4e9719f8d6d8c60d1ffcfb951d822e718b958b45fd612caec856c50ff.jpg b/parse/train/6YEQUn0QICG/images/ce9352a4e9719f8d6d8c60d1ffcfb951d822e718b958b45fd612caec856c50ff.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bbeaa1572f809a7ea335e2ccd863ff9cdcf1cc5c --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/ce9352a4e9719f8d6d8c60d1ffcfb951d822e718b958b45fd612caec856c50ff.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb4a5fe796e583752f420e5e83164ab3ca10f85964edab7db566d78b0e08c7cd +size 4158 diff --git a/parse/train/6YEQUn0QICG/images/d687299802ced41f36eca848a835b397cf4dc0886d9e774b815d04bc823ab281.jpg b/parse/train/6YEQUn0QICG/images/d687299802ced41f36eca848a835b397cf4dc0886d9e774b815d04bc823ab281.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6fc6b9d1e4b94eca295483e7e88c0fcf01d3c590 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/d687299802ced41f36eca848a835b397cf4dc0886d9e774b815d04bc823ab281.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01c3ded185728829cd8738af81bd0f88e7ce04e1188e81de6c79589922c94e18 +size 11189 diff --git a/parse/train/6YEQUn0QICG/images/de2139ee056dc3049f3ee388cb07af214c48fff81a3a80db929b6239791599e4.jpg b/parse/train/6YEQUn0QICG/images/de2139ee056dc3049f3ee388cb07af214c48fff81a3a80db929b6239791599e4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..22fd462511e2ef8d741d0ac7902844fdbdaf05cd --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/de2139ee056dc3049f3ee388cb07af214c48fff81a3a80db929b6239791599e4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c32276b775e6dd29df8f00dbb6af7c621bb3a4268945df548c4bdb63b5ebc7d +size 17155 diff --git a/parse/train/6YEQUn0QICG/images/e09dac46b1290aabfd16ec089c81813a885135d3a47390fdd1398379d01d55e3.jpg b/parse/train/6YEQUn0QICG/images/e09dac46b1290aabfd16ec089c81813a885135d3a47390fdd1398379d01d55e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c38328074127e679d4a03aee839a6e7ca5f67616 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/e09dac46b1290aabfd16ec089c81813a885135d3a47390fdd1398379d01d55e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19a4f3d5cc60a23fac88731b78a4a9ec3914e41cf579405452f60448e5a9782a +size 7612 diff --git a/parse/train/6YEQUn0QICG/images/e3548fafe73d4dfa7ac338e9784ba49eb51eab6be80d8c89ec31b936d763f849.jpg b/parse/train/6YEQUn0QICG/images/e3548fafe73d4dfa7ac338e9784ba49eb51eab6be80d8c89ec31b936d763f849.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dec2e688c16442905b3927370a33d982740642a7 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/e3548fafe73d4dfa7ac338e9784ba49eb51eab6be80d8c89ec31b936d763f849.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b67871de713b74a4afea9c8776720771e5ff8986f4a6cf9e7be06c62f768f0c1 +size 25047 diff --git a/parse/train/6YEQUn0QICG/images/f89aea762d88a47eca254b6cb254866dddb524145b70f4efe4d2948575263b61.jpg b/parse/train/6YEQUn0QICG/images/f89aea762d88a47eca254b6cb254866dddb524145b70f4efe4d2948575263b61.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cfb6ac9c90114c140c43c04fcc82961ba54addf2 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/f89aea762d88a47eca254b6cb254866dddb524145b70f4efe4d2948575263b61.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90cff1d2ea504f3fef5489b864d88db5f68e16bf0f4afd359f92b4341fb4b9d4 +size 3619 diff --git a/parse/train/6YEQUn0QICG/images/f8a72763d5a177c72f73e7b09726efca9fb803d2939d33b3260b7cb14a9cec0e.jpg b/parse/train/6YEQUn0QICG/images/f8a72763d5a177c72f73e7b09726efca9fb803d2939d33b3260b7cb14a9cec0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9bc5fa5c537bc5e4281fad72a8a1c43d49c8a949 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/f8a72763d5a177c72f73e7b09726efca9fb803d2939d33b3260b7cb14a9cec0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d201141b8f9f9e50a41e18509048dd1a5cc82b6eeb604bd0142e0fdaa38da7c7 +size 3831 diff --git a/parse/train/6YEQUn0QICG/images/fcb6e3a2e5e9f5e8652f972a2f18bf8ec01e07ea99bc7809579e0dbd4729caf0.jpg b/parse/train/6YEQUn0QICG/images/fcb6e3a2e5e9f5e8652f972a2f18bf8ec01e07ea99bc7809579e0dbd4729caf0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5ec1f2f2a1bfd0ee323d7dde256680faa94d0cf --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/fcb6e3a2e5e9f5e8652f972a2f18bf8ec01e07ea99bc7809579e0dbd4729caf0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:764489f7e3a9688e36d3565f2895ede295196a0413ac8cacc65c9fd7aba781d4 +size 6902 diff --git a/parse/train/6YEQUn0QICG/images/ffd4143cbe6a7365db699025cb628091fc8f1aeac7067fd59311043d243b4bcc.jpg b/parse/train/6YEQUn0QICG/images/ffd4143cbe6a7365db699025cb628091fc8f1aeac7067fd59311043d243b4bcc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..af8e0b10077e2f4b7c3cb437871771a946157049 --- /dev/null +++ b/parse/train/6YEQUn0QICG/images/ffd4143cbe6a7365db699025cb628091fc8f1aeac7067fd59311043d243b4bcc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da21b72859b0ac5994e3529483623cedd30ee64b2a24a493d0c3da68e6853aed +size 36974 diff --git a/parse/train/74RmfBweB60/images/07b2e5847f06f569d996bfff2cef68ba4e7e99ebccd9cf274448fbddd63318e4.jpg b/parse/train/74RmfBweB60/images/07b2e5847f06f569d996bfff2cef68ba4e7e99ebccd9cf274448fbddd63318e4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1d21bb58d2c84ab951398ec481342141745b2a4a --- /dev/null +++ b/parse/train/74RmfBweB60/images/07b2e5847f06f569d996bfff2cef68ba4e7e99ebccd9cf274448fbddd63318e4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c20080d41fa049fad18a7b54d634234191521f3575e6100cc19d4f0cae0a808e +size 9426 diff --git a/parse/train/74RmfBweB60/images/0dfa08fcfd9718653016b28f0cfebe97143b66b36a3bdfdbe1308aa1e3e265a4.jpg b/parse/train/74RmfBweB60/images/0dfa08fcfd9718653016b28f0cfebe97143b66b36a3bdfdbe1308aa1e3e265a4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5d535dfa1c9da63af42de9f18211678f0809295 --- /dev/null +++ b/parse/train/74RmfBweB60/images/0dfa08fcfd9718653016b28f0cfebe97143b66b36a3bdfdbe1308aa1e3e265a4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e55bf66f32e7386127de78fd38cacf17c80db25c59258c08ca3baf25fdb5ac1 +size 2679 diff --git a/parse/train/74RmfBweB60/images/1da1f00199904a4b6c97f5e9c76f9a3df834798c6291155ebe93e90d726b2886.jpg b/parse/train/74RmfBweB60/images/1da1f00199904a4b6c97f5e9c76f9a3df834798c6291155ebe93e90d726b2886.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c91859cf5ad1f17dae043c08bd9d93817ecde1b4 --- /dev/null +++ b/parse/train/74RmfBweB60/images/1da1f00199904a4b6c97f5e9c76f9a3df834798c6291155ebe93e90d726b2886.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ffccba050bc83608c8582d65485efeda3f91bea73a140967266cb82dbe62670f +size 10442 diff --git a/parse/train/74RmfBweB60/images/2059d579c93d1602d1e72439a83460122b2d3c51c9a3214e4c20249b2456a9c9.jpg b/parse/train/74RmfBweB60/images/2059d579c93d1602d1e72439a83460122b2d3c51c9a3214e4c20249b2456a9c9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..06eab842feed8b22c294a94fe568486294bd576d --- /dev/null +++ b/parse/train/74RmfBweB60/images/2059d579c93d1602d1e72439a83460122b2d3c51c9a3214e4c20249b2456a9c9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e090c4e03f5babad0024aba83539e4873d1cdc6f62536cec95fb4558c3d8a985 +size 12769 diff --git a/parse/train/74RmfBweB60/images/213e7eea9f5e11eb13db31a851e4ce7321f6ff73c1f9234656441c96d93ddf5f.jpg b/parse/train/74RmfBweB60/images/213e7eea9f5e11eb13db31a851e4ce7321f6ff73c1f9234656441c96d93ddf5f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..962df9334d5a6d7caade07ef4dc3030c0c65a69e --- /dev/null +++ b/parse/train/74RmfBweB60/images/213e7eea9f5e11eb13db31a851e4ce7321f6ff73c1f9234656441c96d93ddf5f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:799a4a8ffd50ef42bd68798167c280b19833e5ab063fb13696ec1dc4641b10cc +size 9619 diff --git a/parse/train/74RmfBweB60/images/2682d9ed9998363f783340595232da07def6ae1b62593530707c1b063a745524.jpg b/parse/train/74RmfBweB60/images/2682d9ed9998363f783340595232da07def6ae1b62593530707c1b063a745524.jpg new file mode 100644 index 0000000000000000000000000000000000000000..094e009572f71e415b379c0bb4d1bcb4ebb8b42b --- /dev/null +++ b/parse/train/74RmfBweB60/images/2682d9ed9998363f783340595232da07def6ae1b62593530707c1b063a745524.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e896d05f6bafefe77b5cc8992d4a022c340583b1f241954e1816ba69ab5d47e +size 9111 diff --git a/parse/train/74RmfBweB60/images/2b46e9efc74fa0a534222968a3d20091a3a422fea976cf78d559559513d2e0d6.jpg b/parse/train/74RmfBweB60/images/2b46e9efc74fa0a534222968a3d20091a3a422fea976cf78d559559513d2e0d6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c74050b21a5b2e486e1ae27ccb5b0618006f05c0 --- /dev/null +++ b/parse/train/74RmfBweB60/images/2b46e9efc74fa0a534222968a3d20091a3a422fea976cf78d559559513d2e0d6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:180be7adc47ba23697112f27e0d89bdf13d7b96a700489bd58b0de2e17e70d6d +size 11497 diff --git a/parse/train/74RmfBweB60/images/3702b093fedebbe58ca32ee1f3510f7779f94f8298da5387a3f661e7ff847f92.jpg b/parse/train/74RmfBweB60/images/3702b093fedebbe58ca32ee1f3510f7779f94f8298da5387a3f661e7ff847f92.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e97d06605239ba74d165e7dd55f9c1af60d951d --- /dev/null +++ b/parse/train/74RmfBweB60/images/3702b093fedebbe58ca32ee1f3510f7779f94f8298da5387a3f661e7ff847f92.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b434c9163de1b94b6a2a5da2ed7446180695b82cdb2b08887c16755c43297c9b +size 17173 diff --git a/parse/train/74RmfBweB60/images/3ae3b2a7eb130747659835faa8d01046b129d51408a7d9c08d0dd4aa00a0a848.jpg b/parse/train/74RmfBweB60/images/3ae3b2a7eb130747659835faa8d01046b129d51408a7d9c08d0dd4aa00a0a848.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9715c683de2da21b538066979442f7a2be2dd129 --- /dev/null +++ b/parse/train/74RmfBweB60/images/3ae3b2a7eb130747659835faa8d01046b129d51408a7d9c08d0dd4aa00a0a848.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4558d6b799fbe93d3d0f5695649ca5efba258da0adcb05bafe844e238f66e683 +size 9789 diff --git a/parse/train/74RmfBweB60/images/46d9ad77b55d1cc670745026c47b3d350c5530aabb855c6c7b775b9a5c3422f4.jpg b/parse/train/74RmfBweB60/images/46d9ad77b55d1cc670745026c47b3d350c5530aabb855c6c7b775b9a5c3422f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b0171c72dc63f05ede486dd28d30208377455fe1 --- /dev/null +++ b/parse/train/74RmfBweB60/images/46d9ad77b55d1cc670745026c47b3d350c5530aabb855c6c7b775b9a5c3422f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0bca96fc5b2f9766e497bd2924f9ce3859b0f5b18fb25fab85f36c808180d295 +size 9615 diff --git a/parse/train/74RmfBweB60/images/4e4a6167f1bf0473e8d7a714c8b874f38c7ea875b79a1295462ea5a711766939.jpg b/parse/train/74RmfBweB60/images/4e4a6167f1bf0473e8d7a714c8b874f38c7ea875b79a1295462ea5a711766939.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e53e1b5576c5cca5294bc089261b3ce4fd515874 --- /dev/null +++ b/parse/train/74RmfBweB60/images/4e4a6167f1bf0473e8d7a714c8b874f38c7ea875b79a1295462ea5a711766939.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ace9edf2cc83216ffba647a7743fa63f6a296734989e759e1fb6b88058ed6d0a +size 59228 diff --git a/parse/train/74RmfBweB60/images/53f7ab40917a8ede5122762d6391a76581e4da919bf92190bc43075b2c8c3060.jpg b/parse/train/74RmfBweB60/images/53f7ab40917a8ede5122762d6391a76581e4da919bf92190bc43075b2c8c3060.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e4fc85db1ac02843eba8b73d11755ce2ed862c87 --- /dev/null +++ b/parse/train/74RmfBweB60/images/53f7ab40917a8ede5122762d6391a76581e4da919bf92190bc43075b2c8c3060.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a020095a1ea5e5b48affe2c0587daca4eedb603069bd8768afe462eb1cd898e +size 5997 diff --git a/parse/train/74RmfBweB60/images/57dd0f2231a644892d0d759e737298493ccd260916dd0c48818739b8fbeb7aef.jpg b/parse/train/74RmfBweB60/images/57dd0f2231a644892d0d759e737298493ccd260916dd0c48818739b8fbeb7aef.jpg new file mode 100644 index 0000000000000000000000000000000000000000..deeb2e96560455de762b8a5a9fb083e97a220609 --- /dev/null +++ b/parse/train/74RmfBweB60/images/57dd0f2231a644892d0d759e737298493ccd260916dd0c48818739b8fbeb7aef.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f02566df19844d3ece596e33145259d621587c3de60d2e299098e5451537464 +size 14206 diff --git a/parse/train/74RmfBweB60/images/71555ba46498e3920949431f5517f0cfdf53213e02f54c9ada168880b5c7fd49.jpg b/parse/train/74RmfBweB60/images/71555ba46498e3920949431f5517f0cfdf53213e02f54c9ada168880b5c7fd49.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9a4823a251718da21019f5835bae59395382c4eb --- /dev/null +++ b/parse/train/74RmfBweB60/images/71555ba46498e3920949431f5517f0cfdf53213e02f54c9ada168880b5c7fd49.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b77c6f44f27932b9d5dcf199623604a6ead46c3b2031eb299a71eb33a7ad4e72 +size 17791 diff --git a/parse/train/74RmfBweB60/images/7cd63fa4d3a31b21549b1c3cba203d1cfde83dc305268984dcc577e9d1c629f3.jpg b/parse/train/74RmfBweB60/images/7cd63fa4d3a31b21549b1c3cba203d1cfde83dc305268984dcc577e9d1c629f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d3c5f23d66c3a906748593fd1a600f913a2f6407 --- /dev/null +++ b/parse/train/74RmfBweB60/images/7cd63fa4d3a31b21549b1c3cba203d1cfde83dc305268984dcc577e9d1c629f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3ee7b5c51ad5126ba55f6b1e8363723c507393d8dbfe920921dc421040f55496 +size 48022 diff --git a/parse/train/74RmfBweB60/images/82747ffa22810c00af6604f8c9488a585bb5fc2f6fc901ae3a034f3572b482f0.jpg b/parse/train/74RmfBweB60/images/82747ffa22810c00af6604f8c9488a585bb5fc2f6fc901ae3a034f3572b482f0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..443e258c6754a5f7bd420e6791990e20801a03f2 --- /dev/null +++ b/parse/train/74RmfBweB60/images/82747ffa22810c00af6604f8c9488a585bb5fc2f6fc901ae3a034f3572b482f0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15a5a9880429a8fd3aaeb5e6f0d550aa579963b46039c0dc8ce560c377970480 +size 9369 diff --git a/parse/train/74RmfBweB60/images/9c9ca3a937c7498ac455080e8128e2e20c04105f9efdeb52eea00b6054ed9608.jpg b/parse/train/74RmfBweB60/images/9c9ca3a937c7498ac455080e8128e2e20c04105f9efdeb52eea00b6054ed9608.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7177fb188c7c5d9208fc3031c2f65787ce8628d0 --- /dev/null +++ b/parse/train/74RmfBweB60/images/9c9ca3a937c7498ac455080e8128e2e20c04105f9efdeb52eea00b6054ed9608.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3cccd0160d08de00c3af175917ffa04154f6d72ade4cf93b241299ce4a4414c3 +size 19647 diff --git a/parse/train/74RmfBweB60/images/9dc3554546cd0d7e8714d2a8742b8e7b54e42d716313e7ff2b236bd90a374b64.jpg b/parse/train/74RmfBweB60/images/9dc3554546cd0d7e8714d2a8742b8e7b54e42d716313e7ff2b236bd90a374b64.jpg new file mode 100644 index 0000000000000000000000000000000000000000..42c63f50c052b3ef54827132310b3cb0bca89922 --- /dev/null +++ b/parse/train/74RmfBweB60/images/9dc3554546cd0d7e8714d2a8742b8e7b54e42d716313e7ff2b236bd90a374b64.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1469c14fab9a82a6e47c31d8aedd5a2c828abaada6cfdcab0300d76809bbc879 +size 8753 diff --git a/parse/train/74RmfBweB60/images/a1e0157530e8a4c8bc9d0803138f0eba224ba8f93f9a46dce23f9f9876f0d1d5.jpg b/parse/train/74RmfBweB60/images/a1e0157530e8a4c8bc9d0803138f0eba224ba8f93f9a46dce23f9f9876f0d1d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d66663a54764255e69e99cd4e0beaea1224aebf5 --- /dev/null +++ b/parse/train/74RmfBweB60/images/a1e0157530e8a4c8bc9d0803138f0eba224ba8f93f9a46dce23f9f9876f0d1d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3963bb6577e3e85cbb327d355383bbae23d22d5fce882e12b5864d3084ad8dfa +size 13767 diff --git a/parse/train/74RmfBweB60/images/a84cc1579eae155c735ed137faef7ebb0f81235573bb5194136ab35bb171c683.jpg b/parse/train/74RmfBweB60/images/a84cc1579eae155c735ed137faef7ebb0f81235573bb5194136ab35bb171c683.jpg new file mode 100644 index 0000000000000000000000000000000000000000..338a09481bcb78f4e517f08c6f47dccb1b73b1c2 --- /dev/null +++ b/parse/train/74RmfBweB60/images/a84cc1579eae155c735ed137faef7ebb0f81235573bb5194136ab35bb171c683.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4428c5dbce2658c5f1996ae94567c35cee43b8f77acdf4b46e64fda1a4211c5 +size 5102 diff --git a/parse/train/74RmfBweB60/images/b4d50912a83264c73c0021a71bd120be521f76f0af08da4df50bf194bda49736.jpg b/parse/train/74RmfBweB60/images/b4d50912a83264c73c0021a71bd120be521f76f0af08da4df50bf194bda49736.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b8aabd51ded4c39b981615e6038511bdccc7b8bd --- /dev/null +++ b/parse/train/74RmfBweB60/images/b4d50912a83264c73c0021a71bd120be521f76f0af08da4df50bf194bda49736.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:debd2fd66c7d268aad7eb374e0b1649ed780d75c68ba7efa6e78f0a1c3cc1dc7 +size 7455 diff --git a/parse/train/74RmfBweB60/images/be090bb26a96971777947c1ec6bf114494d258ce86d67ce09fbd7459d4b4d822.jpg b/parse/train/74RmfBweB60/images/be090bb26a96971777947c1ec6bf114494d258ce86d67ce09fbd7459d4b4d822.jpg new file mode 100644 index 0000000000000000000000000000000000000000..73c77fbc01720ecf5fb9f0b9efc2f69618e8eaf3 --- /dev/null +++ b/parse/train/74RmfBweB60/images/be090bb26a96971777947c1ec6bf114494d258ce86d67ce09fbd7459d4b4d822.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04940e8d4e80d999e9b0093c1617a0a9f42200c8303d3a763d721e37b6b2ad5e +size 4791 diff --git a/parse/train/74RmfBweB60/images/be159e0ed6417306b53ac4f96ca81f6ad60044ec1cf69fdcd8c0e3698c23ba8a.jpg b/parse/train/74RmfBweB60/images/be159e0ed6417306b53ac4f96ca81f6ad60044ec1cf69fdcd8c0e3698c23ba8a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0b886d809e1af3eb669a8b15a2fa3bcacc20fd08 --- /dev/null +++ b/parse/train/74RmfBweB60/images/be159e0ed6417306b53ac4f96ca81f6ad60044ec1cf69fdcd8c0e3698c23ba8a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d1b700863a688ffba41856253db24c8f0d061b03708ac79205675eb1927b63e +size 10928 diff --git a/parse/train/74RmfBweB60/images/cd4673555690bc9e7c2c134b940cbf6ce609f112d808752990bda7448b1a8d47.jpg b/parse/train/74RmfBweB60/images/cd4673555690bc9e7c2c134b940cbf6ce609f112d808752990bda7448b1a8d47.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4e74b21630c234d131419e454c4a57f26c7b01a4 --- /dev/null +++ b/parse/train/74RmfBweB60/images/cd4673555690bc9e7c2c134b940cbf6ce609f112d808752990bda7448b1a8d47.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:54c1a0b5e615b0dca20177a728104063616aca2de804259d1f28df69e394842f +size 10858 diff --git a/parse/train/74RmfBweB60/images/d2ba603658bc41dc17088723550a675658f813d3ce0cc45bf0c47f8124f2dac5.jpg b/parse/train/74RmfBweB60/images/d2ba603658bc41dc17088723550a675658f813d3ce0cc45bf0c47f8124f2dac5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d4d175e0539806994a455a9e54c558a90f9e891 --- /dev/null +++ b/parse/train/74RmfBweB60/images/d2ba603658bc41dc17088723550a675658f813d3ce0cc45bf0c47f8124f2dac5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8589d7ca026b2e3821eaf42bcb5247141d5b7caf7dc5ce12878a5d050e68fe03 +size 7731 diff --git a/parse/train/74RmfBweB60/images/ddd811cf563572a33890a12c24fdb862f3bb0d98bf6a0999a7e50461d2e1a650.jpg b/parse/train/74RmfBweB60/images/ddd811cf563572a33890a12c24fdb862f3bb0d98bf6a0999a7e50461d2e1a650.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e2dcbc0b1fea100d2ad4045985ea9f17c8d71f4a --- /dev/null +++ b/parse/train/74RmfBweB60/images/ddd811cf563572a33890a12c24fdb862f3bb0d98bf6a0999a7e50461d2e1a650.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7a742ca3d97ca8b2b2399c863cdf089cb6fcb51817d97496c7caf6a7c05fda4 +size 6762 diff --git a/parse/train/74RmfBweB60/images/f3de3e826aa012c6e58901a8e3b22025d2112634453e74fd01df633f9fe9af64.jpg b/parse/train/74RmfBweB60/images/f3de3e826aa012c6e58901a8e3b22025d2112634453e74fd01df633f9fe9af64.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3e1e0b479e291ae3f6d13fd0bdff40544ebc02c7 --- /dev/null +++ b/parse/train/74RmfBweB60/images/f3de3e826aa012c6e58901a8e3b22025d2112634453e74fd01df633f9fe9af64.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b98352d8de6284e43316ee76e09af6ff649acfa554479dedd7c65e5c57078704 +size 6376 diff --git a/parse/train/7wCBOfJ8hJM/images/1877c35898e19a3fe5549972edb284899fbbb1b9ac44dfa14b08a27116ff8eb7.jpg b/parse/train/7wCBOfJ8hJM/images/1877c35898e19a3fe5549972edb284899fbbb1b9ac44dfa14b08a27116ff8eb7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b72a4b24ca7b05304305f66fd12f07a28da78fd6 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/1877c35898e19a3fe5549972edb284899fbbb1b9ac44dfa14b08a27116ff8eb7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b455794ae70440d8255d68dd713b7be95d89ed6bc63af12ca6bd52e36b7de3e8 +size 127668 diff --git a/parse/train/7wCBOfJ8hJM/images/20a1ccf1cf30c46d7bddacd08ee4088dcfc1c8141981161f775b1c03074811e6.jpg b/parse/train/7wCBOfJ8hJM/images/20a1ccf1cf30c46d7bddacd08ee4088dcfc1c8141981161f775b1c03074811e6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..84289d50bc1e51f54aa9e552b09af7aee558ce76 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/20a1ccf1cf30c46d7bddacd08ee4088dcfc1c8141981161f775b1c03074811e6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:12223af2d4a8a4ae97cc7827389ed652fc683c1965b00722b838836f466bc9f5 +size 65721 diff --git a/parse/train/7wCBOfJ8hJM/images/339e767a8fc160f109859447be16d050e9c88d0aaa6bfa252036c719ad0df8d7.jpg b/parse/train/7wCBOfJ8hJM/images/339e767a8fc160f109859447be16d050e9c88d0aaa6bfa252036c719ad0df8d7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..431616d4f851f9e656086d0e9631473dcfd4095f --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/339e767a8fc160f109859447be16d050e9c88d0aaa6bfa252036c719ad0df8d7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13ae18b78d078ec21165010d82e0f53a5fe048cc0c720649fbdd3dbb8fe321ed +size 7562 diff --git a/parse/train/7wCBOfJ8hJM/images/4a894a653b5bc86d1ace69b7d9355010f59c2f95ab0e272cc0d4147ac07474f3.jpg b/parse/train/7wCBOfJ8hJM/images/4a894a653b5bc86d1ace69b7d9355010f59c2f95ab0e272cc0d4147ac07474f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..26d84caea983d4e727e49562339b0a02c0ce5cb8 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/4a894a653b5bc86d1ace69b7d9355010f59c2f95ab0e272cc0d4147ac07474f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc420a1623cbb05e0636238bdc7307652f812baa1c5f2f04d8c28427fa2f349d +size 26387 diff --git a/parse/train/7wCBOfJ8hJM/images/4b913a52b71a1038797f492c6259f1bfa60cc26de6e5b0d577f8997d355de1e7.jpg b/parse/train/7wCBOfJ8hJM/images/4b913a52b71a1038797f492c6259f1bfa60cc26de6e5b0d577f8997d355de1e7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d66817f457d8e4fd87550a0dec321cb1cc9fbd83 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/4b913a52b71a1038797f492c6259f1bfa60cc26de6e5b0d577f8997d355de1e7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2685d9224ea22da9cf320755eaed52de16a0de0bf8c6581340d7bdf606e7290d +size 139951 diff --git a/parse/train/7wCBOfJ8hJM/images/72d8a278cbb129abd9187f3c258b785f290f66f7123b95dfbe6d12c289ab8319.jpg b/parse/train/7wCBOfJ8hJM/images/72d8a278cbb129abd9187f3c258b785f290f66f7123b95dfbe6d12c289ab8319.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5c0b2dad32812ecaad8193f80ebef9ee618f2b5e --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/72d8a278cbb129abd9187f3c258b785f290f66f7123b95dfbe6d12c289ab8319.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ba4e78aaba32c9b3324e19cea4a4bbfabb80869ec04ad7f9bf824b0a358c794b +size 95487 diff --git a/parse/train/7wCBOfJ8hJM/images/84aad7135ccff9fe33818b0517a5def1faf3b3778bc796d4a7c37898510181cc.jpg b/parse/train/7wCBOfJ8hJM/images/84aad7135ccff9fe33818b0517a5def1faf3b3778bc796d4a7c37898510181cc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..11df41d5f26920f52111033c28fcb3fbf8f50988 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/84aad7135ccff9fe33818b0517a5def1faf3b3778bc796d4a7c37898510181cc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:376e38ae9754440401f7ecc8542eaef9ce53ca8646cd824a976b6e8542c7c14b +size 9682 diff --git a/parse/train/7wCBOfJ8hJM/images/9777eb0ba0977a96cd11b5a3eb397a62eb248e42d00cdaf275c011e3d748301e.jpg b/parse/train/7wCBOfJ8hJM/images/9777eb0ba0977a96cd11b5a3eb397a62eb248e42d00cdaf275c011e3d748301e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4869ee06fa24e2fdc9ddc00118abbb4012356d26 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/9777eb0ba0977a96cd11b5a3eb397a62eb248e42d00cdaf275c011e3d748301e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a0d8fccff402ed02b677ac25c5104453a5fc423fbc4b42ddcb67e0085f5c8900 +size 119203 diff --git a/parse/train/7wCBOfJ8hJM/images/a4458de4cbe1120f80319e2c06caa93f42845ef43970f6b0419636d0edbf21f4.jpg b/parse/train/7wCBOfJ8hJM/images/a4458de4cbe1120f80319e2c06caa93f42845ef43970f6b0419636d0edbf21f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d32933df96a5fdbfc49ad69100ec009476925ae2 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/a4458de4cbe1120f80319e2c06caa93f42845ef43970f6b0419636d0edbf21f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37bda39470f8137f0bb664cab6229efe3fb8178f6fc514386b343ff0f515a4b0 +size 54715 diff --git a/parse/train/7wCBOfJ8hJM/images/b0c064f07c0b3808598b641df9f86b4812b89dc790b092c99f30b19cefd8ffe4.jpg b/parse/train/7wCBOfJ8hJM/images/b0c064f07c0b3808598b641df9f86b4812b89dc790b092c99f30b19cefd8ffe4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f25a2465f2bc9d9ca6741925a3a2b71438ec5a49 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/b0c064f07c0b3808598b641df9f86b4812b89dc790b092c99f30b19cefd8ffe4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a4aba0614b3025d1dca4195f77682ceb4e3e6728cf1e4fbd1203dc8d62323e5 +size 185631 diff --git a/parse/train/7wCBOfJ8hJM/images/b5694d2a21ee6fdec4ab2d43b8b2ec473fb204c592a569633c1b625bead70a7b.jpg b/parse/train/7wCBOfJ8hJM/images/b5694d2a21ee6fdec4ab2d43b8b2ec473fb204c592a569633c1b625bead70a7b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..39c876ea333e1387391f8bf169cd7490c31bfb2a --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/b5694d2a21ee6fdec4ab2d43b8b2ec473fb204c592a569633c1b625bead70a7b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd99c2d5d66003569dd9dd6d49ab4221fa2c5f4421b642bdba2b9d9ba70b2c43 +size 55490 diff --git a/parse/train/7wCBOfJ8hJM/images/f5357ba624ce5a00f05a13ec397eee0f47d08d1ea75dc036874a5aa5e4611522.jpg b/parse/train/7wCBOfJ8hJM/images/f5357ba624ce5a00f05a13ec397eee0f47d08d1ea75dc036874a5aa5e4611522.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a0492eb679096d8bf2e30794dee9ec1c05218b33 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/f5357ba624ce5a00f05a13ec397eee0f47d08d1ea75dc036874a5aa5e4611522.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9224bbf2ecc5dacbeae5c100336551ea4e2f926f8d579366145f4961654c48e5 +size 6271 diff --git a/parse/train/7wCBOfJ8hJM/images/f76678d9e136b82f35983e81e76fe6881d314a201492125ddfa207c61811a6d0.jpg b/parse/train/7wCBOfJ8hJM/images/f76678d9e136b82f35983e81e76fe6881d314a201492125ddfa207c61811a6d0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..de095a90027b585ed9d8e732c16a4073a1801895 --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/f76678d9e136b82f35983e81e76fe6881d314a201492125ddfa207c61811a6d0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e05d3e2b655c71986d9460b675b755cd55f5c4346dea1f57ffacef5b191f4325 +size 81495 diff --git a/parse/train/7wCBOfJ8hJM/images/ffebda3037765255e36913287c15f19f093247f9555a28236f8315bb31082fba.jpg b/parse/train/7wCBOfJ8hJM/images/ffebda3037765255e36913287c15f19f093247f9555a28236f8315bb31082fba.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1cb59fbd9409ed4446b4db7b26b0320e16bc8f2a --- /dev/null +++ b/parse/train/7wCBOfJ8hJM/images/ffebda3037765255e36913287c15f19f093247f9555a28236f8315bb31082fba.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:366f3981385b67f2e681a8a1a69e75ec79e4efb759c238dbab75bd86d797cdcc +size 22801 diff --git a/parse/train/A7pvvrlv68/images/17af84fd9f604ba2f997624ee6c90f575596714ce9e4803e6164c51b50f0351f.jpg b/parse/train/A7pvvrlv68/images/17af84fd9f604ba2f997624ee6c90f575596714ce9e4803e6164c51b50f0351f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..06640516b825915828ac219e9d78788e73e5d5dc --- /dev/null +++ b/parse/train/A7pvvrlv68/images/17af84fd9f604ba2f997624ee6c90f575596714ce9e4803e6164c51b50f0351f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc11858b1181f685cb1a385f437743253d6b4387d05f6db98f978db931881c5c +size 6218 diff --git a/parse/train/A7pvvrlv68/images/436b1c7ee6aa211b3ef781943de1fd6602816335ca173c3ebbe3c25afe9a1439.jpg b/parse/train/A7pvvrlv68/images/436b1c7ee6aa211b3ef781943de1fd6602816335ca173c3ebbe3c25afe9a1439.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b0bdd05375825afab8007fdb05c0f0b594a20583 --- /dev/null +++ b/parse/train/A7pvvrlv68/images/436b1c7ee6aa211b3ef781943de1fd6602816335ca173c3ebbe3c25afe9a1439.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3acc469697fac40b8483bdc2743d37981e708e1f4b68359581f107e370dbfc33 +size 5534 diff --git a/parse/train/A7pvvrlv68/images/536d2591f15be89f8ff1bf26d90cf75e55d832a5c4d603a591c841bcb51445e5.jpg b/parse/train/A7pvvrlv68/images/536d2591f15be89f8ff1bf26d90cf75e55d832a5c4d603a591c841bcb51445e5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e83944a3a610bc46970a8738a14b338b4dad018c --- /dev/null +++ b/parse/train/A7pvvrlv68/images/536d2591f15be89f8ff1bf26d90cf75e55d832a5c4d603a591c841bcb51445e5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f34bb87c0cea4a70a794ab0e37f5d1bd64de9f548e1350d79be1201dee5a3cb7 +size 13715 diff --git a/parse/train/A7pvvrlv68/images/57a8b36e4658a261dfb590691de231137f36c566012ab250bf1024a961ec2b90.jpg b/parse/train/A7pvvrlv68/images/57a8b36e4658a261dfb590691de231137f36c566012ab250bf1024a961ec2b90.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5c81265fad2573b221e2ad631a4ed0f71243116f --- /dev/null +++ b/parse/train/A7pvvrlv68/images/57a8b36e4658a261dfb590691de231137f36c566012ab250bf1024a961ec2b90.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:349b83060efe8f06e3e3f84a978216f5975cef8ce30006c306d4bcf1c2607efa +size 9271 diff --git a/parse/train/A7pvvrlv68/images/6af06c98f410630170c917e8cab50477e1aa6be60e3cfa4697976649ab5238e6.jpg b/parse/train/A7pvvrlv68/images/6af06c98f410630170c917e8cab50477e1aa6be60e3cfa4697976649ab5238e6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf894b540f956643601df37d749d55dbc284c88a --- /dev/null +++ b/parse/train/A7pvvrlv68/images/6af06c98f410630170c917e8cab50477e1aa6be60e3cfa4697976649ab5238e6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f964eaef46ca1caa7abedf7606b514a9819ee87c9b4ddde82bfc3dedeb36cd05 +size 57523 diff --git a/parse/train/A7pvvrlv68/images/6ba0f65fb5a621c0d95553507d10f2e27c01940d6043cdf28397d76590409b64.jpg b/parse/train/A7pvvrlv68/images/6ba0f65fb5a621c0d95553507d10f2e27c01940d6043cdf28397d76590409b64.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d8259aebf78a057fb8b21d9dd5ca1d14ffa6b718 --- /dev/null +++ b/parse/train/A7pvvrlv68/images/6ba0f65fb5a621c0d95553507d10f2e27c01940d6043cdf28397d76590409b64.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6067b7fb243c5c144afee4ecc969d714fda6739101fd3ceb68ea3ffc18525f10 +size 5806 diff --git a/parse/train/A7pvvrlv68/images/871673288d167bcbd541079e5997b278cca429d8ca64fd48b785eb05c1db1a12.jpg b/parse/train/A7pvvrlv68/images/871673288d167bcbd541079e5997b278cca429d8ca64fd48b785eb05c1db1a12.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a73d39d5da274f597293c8f60a994a2269347c25 --- /dev/null +++ b/parse/train/A7pvvrlv68/images/871673288d167bcbd541079e5997b278cca429d8ca64fd48b785eb05c1db1a12.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2fe44f357eb2024e64e1ac8758f03f8438a3af912571410069071c2ebf7134c +size 53255 diff --git a/parse/train/A7pvvrlv68/images/956f97dc2146646d797257b5483737aca4c5f1c6a8ae6460f9c3c6c2a9f3f493.jpg b/parse/train/A7pvvrlv68/images/956f97dc2146646d797257b5483737aca4c5f1c6a8ae6460f9c3c6c2a9f3f493.jpg new file mode 100644 index 0000000000000000000000000000000000000000..54cbe12bb561b90671e31875519aff69dbc57b4f --- /dev/null +++ b/parse/train/A7pvvrlv68/images/956f97dc2146646d797257b5483737aca4c5f1c6a8ae6460f9c3c6c2a9f3f493.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5b632b3862890afb096134ee15bbabadd89a188d29c8c228fdaf3536716ff37 +size 13017 diff --git a/parse/train/A7pvvrlv68/images/a468f1bb1946757538c10a82e4cba5957f0190cedcd03a0cdcef5e345becbb5b.jpg b/parse/train/A7pvvrlv68/images/a468f1bb1946757538c10a82e4cba5957f0190cedcd03a0cdcef5e345becbb5b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5b12c93ff9186de6df66953d0c7c4d90184664f --- /dev/null +++ b/parse/train/A7pvvrlv68/images/a468f1bb1946757538c10a82e4cba5957f0190cedcd03a0cdcef5e345becbb5b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9751350bf47c9e1891bc1922c0664894cf4188daf9faec3fbf5e7097f26e1bcd +size 14826 diff --git a/parse/train/A7pvvrlv68/images/bbf0b59144e8d6846c1fc593ed08bd606e371673e597d9ebfbeb2703730f9ed1.jpg b/parse/train/A7pvvrlv68/images/bbf0b59144e8d6846c1fc593ed08bd606e371673e597d9ebfbeb2703730f9ed1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9763716b331108a8e9136578019e3309137fd02d --- /dev/null +++ b/parse/train/A7pvvrlv68/images/bbf0b59144e8d6846c1fc593ed08bd606e371673e597d9ebfbeb2703730f9ed1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b03ae2113f4c9c54b4e8a61cfe8962fafe38ffed9fae00a8b869ed5c48d55593 +size 14903 diff --git a/parse/train/B1e9Y2NYvS/B1e9Y2NYvS.md b/parse/train/B1e9Y2NYvS/B1e9Y2NYvS.md new file mode 100644 index 0000000000000000000000000000000000000000..b38b948b2fbf913941118ad1055a74b633aa1b62 --- /dev/null +++ b/parse/train/B1e9Y2NYvS/B1e9Y2NYvS.md @@ -0,0 +1,318 @@ +# ON ROBUSTNESS OF NEURAL ORDINARY DIFFERENTIAL EQUATIONS + +Hanshu YAN\*, Jiawei DU\*, Vincent Y. F. TAN & Jiashi FENG + +Department of Electrical and Computer Engineering +National University of Singapore +{hanshu.yan, dujiawei}@u.nus.edu, {vtan, elefjia}@nus.edu.sg + +# ABSTRACT + +Neural ordinary differential equations (ODEs) have been attracting increasing attention in various research domains recently. There have been some works studying optimization issues and approximation capabilities of neural ODEs, but their robustness is still yet unclear. In this work, we fill this important gap by exploring robustness properties of neural ODEs both empirically and theoretically. We first present an empirical study on the robustness of the neural ODE-based networks (ODENets) by exposing them to inputs with various types of perturbations and subsequently investigating the changes of the corresponding outputs. In contrast to conventional convolutional neural networks (CNNs), we find that the ODENets are more robust against both random Gaussian perturbations and adversarial attack examples. We then provide an insightful understanding of this phenomenon by exploiting a certain desirable property of the flow of a continuous-time ODE, namely that integral curves are non-intersecting. Our work suggests that, due to their intrinsic robustness, it is promising to use neural ODEs as a basic block for building robust deep network models. To further enhance the robustness of vanilla neural ODEs, we propose the time-invariant steady neural ODE (TisODE), which regularizes the flow on perturbed data via the time-invariant property and the imposition of a steady-state constraint. We show that the TisODE method outperforms vanilla neural ODEs and also can work in conjunction with other state-of-the-art architectural methods to build more robust deep networks. + +# 1 INTRODUCTION + +Neural ordinary differential equations (Chen et al., 2018) form a family of models that approximate nonlinear mappings by using continuous-time ODEs. Due to their desirable properties, such as invertibility and parameter efficiency, neural ODEs have attracted increasing attention recently (Dupont et al., 2019; Liu et al., 2019). For example, Grathwohl et al. (2018) proposed a neural ODE-based generative model—the FFJORD—to solve inverse problems; Quaglino et al. (2019) used a higher-order approximation of the states in a neural ODE, and proposed the SNet to accelerate computation. Along with the wider deployment of neural ODEs, robustness issues come to the fore. However, the robustness of neural ODEs is still yet unclear. In particular, it is unclear how robust neural ODEs are in comparison to the widely-used CNNs. Robustness properties of CNNs have been studied extensively. In this work, we present the first systematic study on exploring the robustness properties of neural ODEs. + +To do so, we consider the task of image classification. We expect that results would be similar for other machine learning tasks such as regression. Neural ODEs are dimension-preserving mappings, but a classification model transforms a high-dimensional input—such as an image—into an output whose dimension is equal to the number of classes. Thus, we consider the neural ODE-based classification network (ODENet) whose architecture is shown in Figure 1. An ODENet consists of three components: the feature extractor (FE) consists of convolutional layers which maps an input datum to a multi-channel feature map, a neural ODE that serves as the nonlinear representation mapping (RM), and the fully-connected classifier (FCC) that generates a prediction vector based on the output of the RM. + +The robustness of a classification model can be evaluated through the lens of its performance on perturbed images. To comprehensively investigate the robustness of neural ODEs, we perturb original images with commonly-used perturbations, namely, random Gaussian noise (Szegedy et al., 2013) and harmful adversarial examples (Goodfellow et al., 2014; Madry et al., 2017). We conduct experiments in two common settings—training the model only on authentic non-perturbed images and training the model on authentic images as well as the Gaussian perturbed ones. We observe that ODENets are more robust compared to CNN models against all types of perturbations in both settings. We then provide an insightful understanding of such intriguing robustness of neural ODEs by exploiting a certain property of the flow (Dupont et al., 2019), namely that integral curves that start at distinct initial states are nonintersecting. The flow of a continuous-time ODE is defined as the family of solutions/paths traversed by the state, starting from different initial points, and an integral curve is a specific solution for a given initial point. The non-intersecting property indicates that an integral curve starting from some point is constrained by the integral curves starting from that point’s neighborhood. Thus, in an ODENet, if a correctly classified datum is slightly perturbed, the integral curve associated to its perturbed version would not change too much from the original one. Consequently, the perturbed datum could still be correctly classified. Thus, there exists intrinsic robustness regularization in ODENets, which is absent from CNNs. + +![](images/6e9bd948731c27f7167c50dec38ca4fa802294fd3834ad917ec0b4023c814c57.jpg) +Figure 1: The architecture of an ODENet. The neural ODE block serves as a dimension-preserving nonlinear mapping. + +Motivated by this property of the neural ODE flow, we attempt to explore a more robust neural ODE architecture by introducing stronger regularization on the flow. We thus propose a Time-Invariant Steady neural ODE (TisODE). The TisODE removes the time dependence of the dynamics in an ODE and imposes a steady-state constraint on the integral curves. Removing the time dependence of the derivative results in the time-invariant property of the ODE. To wit, given a solution ${ \bf z } _ { 1 } ( t )$ , another solution $\widetilde { \mathbf { z } } _ { 1 } ( t )$ , with an initial state $\tilde { { \bf z } } _ { 1 } ( 0 ) \bar { { \bf \phi } } = { \bf \bar { z } } _ { 1 } ( T ^ { \prime } )$ for some $T ^ { \prime } > 0$ , can be regarded as the $- T ^ { \prime }$ - shift version of ${ \bf z } _ { 1 } ( t )$ . Such a time-invariant property would make bounding the difference between output states convenient. To elaborate, let the output of a neural ODE correspond to states at time $T > 0$ . By the time-invariant property, the difference between outputs, $\| \widetilde { \mathbf z } _ { 1 } ( \bar { T } ) - \mathbf z _ { 1 } ( T ) \|$ , equals to $\| { \bf z } _ { 1 } ( T + T ^ { \prime } ) - { \bf z } _ { 1 } ( T ) \|$ . To control this distance, a steady-state regularization term is introduced to the overall objective to constrain the change of a state after time exceeds $T$ . With the time-invariant property and the steady-state term, we show that TisODE even is more robust. We do so by evaluating the robustness of TisODE-based classifiers against various types of perturbations and observe that such models are more robust than vanilla ODE-based models. + +In addition, some other effective architectural solutions have also been recently proposed to improve the robustness of CNNs. For example, Xie et al. (2017) randomly resizes or pads zeros into test images to destroy the specific structure of adversarial perturbations. Besides, the model proposed by Xie et al. (2019) contains feature denoising filters to remove the feature-level patterns of adversarial examples. We conduct experiments to show that our proposed TisODE can work seamlessly and in conjunction with these methods to further boost the robustness of deep models. Thus, the proposed TisODE can be used as a generally applicable and effective component for improving the robustness of deep models. + +In summary, our contributions are as follows. Firstly, we are the first to provide a systematic empirical study on the robustness of neural ODEs and find that the neural ODE-based models are more robust compared to conventional CNN models. This finding inspires new applications of neural ODEs in improving robustness of deep models, a problem that concerns many deep learning theorists and practitioners alike. Secondly, we propose the TisODE method, which is simple yet effective in significantly boosting the robustness of neural ODEs. Moreover, the proposed TisODE can also be used in conjunction with other state-of-the-art robust architectures. Thus, TisODE can serve as a drop-in module to improve the robustness of deep models effectively. + +# 2 PRELIMINARIES ON NEURAL ODE + +It has been shown that a residual block (He et al., 2016) can be interpreted as the discrete approximation of an ODE by setting the discretization step to be one. When the discretization step approaches zero, it yields a family of neural networks, which are called neural ODEs (Chen et al., 2018). Formally, in a neural ODE, the relation between input and output is characterized by the following set of equations: + +$$ +\frac { \mathrm { d } { \mathbf z } ( t ) } { \mathrm { d } t } = f _ { \boldsymbol \theta } ( { \mathbf z } ( t ) , t ) , \quad { \mathbf z } ( 0 ) = { \mathbf z } _ { \mathrm { i n } } , \quad { \mathbf z } _ { \mathrm { o u t } } = { \mathbf z } ( T ) , +$$ + +where $f _ { \theta } : \mathbb { R } ^ { d } \times [ 0 , \infty ) \mathbb { R } ^ { d }$ denotes the trainable layers that are parameterized by weights $\theta$ and $\mathbf { z } : [ 0 , \infty ) \mathbb { R } ^ { d }$ represents the $d$ -dimensional state of the neural ODE. We assume that $f _ { \theta }$ is continuous in $t$ and globally Lipschitz continuous in $\mathbf { z }$ . In this case, the input $\mathbf { z } _ { \mathrm { i n } }$ of the neural ODE corresponds to the state at $t = 0$ , and the output $\mathbf { z } _ { \mathrm { o u t } }$ is associated to the state at some $T \in ( 0 , \infty )$ . Because $f _ { \theta }$ governs how the state changes with respect to time $t$ , we also use $f _ { \theta }$ to denote the dynamics of the neural ODE. + +Given input $\mathbf { z } _ { \mathrm { i n } }$ , the output $\mathbf { z } _ { \mathrm { o u t } }$ can be computed by solving the ODE in (1). If $T$ is fixed, the output $\mathbf { z } _ { \mathrm { o u t } }$ only depends on the input $\mathbf { z } _ { \mathrm { i n } }$ and the dynamics $f _ { \theta }$ , which also corresponds to the weighted layers in the neural ODE. Therefore, the neural ODE can be represented as the $d$ -dimensional function $\phi _ { T } ( \cdot , \cdot )$ of the input $\mathbf { z } _ { \mathrm { i n } }$ and the dynamics $f _ { \theta }$ , i.e., + +$$ +\mathbf { z } _ { \mathrm { o u t } } = \mathbf { z } ( T ) = \mathbf { z } ( 0 ) + \int _ { 0 } ^ { T } f _ { \theta } ( \mathbf { z } ( t ) , t ) \mathrm { d } t = \phi _ { T } ( \mathbf { z } _ { \mathrm { i n } } , f _ { \theta } ) . +$$ + +The terminal time $T$ of the output state ${ \mathbf z } ( T )$ is set to be 1 in practice. Several methods have been proposed for training neural ODEs, such as the adjoint sensitivity method (Chen et al., 2018), SNet (Quaglino et al., 2019), and the auto-differentiation technique (Paszke et al., 2017). In this work, we use the most straightforward technique, i.e., updating the weights $\theta$ with the autodifferentiation technique in the PyTorch framework. + +# 3 AN EMPIRICAL STUDY ON THE ROBUSTNESS OF ODENETS + +Robustness of deep models has gained increased attention, as it is imperative that deep models employed in critical applications, such as healthcare, are robust. The robustness of a model is measured by the sensitivity of the prediction with respect to small perturbations on the inputs. In this study, we consider three commonly-used perturbation schemes, namely random Gaussian perturbations, FGSM (Goodfellow et al., 2014) adversarial examples, and PGD (Madry et al., 2017) adversarial examples. These perturbation schemes reflect noise and adversarial robustness properties of the investigated models respectively. We evaluate the robustness via the classification accuracies on perturbed images, in which the original non-perturbed versions of these images are all correctly classified. + +For a fair comparison with conventional CNN models, we made sure that the number of parameters of an ODENet is close to that of its counterpart CNN model. Specifically, the ODENet shares the same network architecture with the CNN model for the FE and FCC parts. The only difference is that, for the RM part, the input of the ODE-based RM is concatenated with one more channel which represents the time $t$ , while the RM in a CNN model has a skip connection and serves as a residual block. During the training phase, all the hyperparameters are kept the same, including training epochs, learning rate schedules, and weight decay coefficients. Each model is trained three times with different random seeds, and we report the average performance (classification accuracy) together with the standard deviation. + +# 3.1 EXPERIMENTAL SETTINGS + +Dataset: We conduct experiments to compare the robustness of ODENets with CNN models on three datasets, i.e., the MNIST (LeCun et al., 1998), the SVHN (Netzer et al., 2011), and a subset of the ImageNet datset (Deng et al., 2009). We call the subset ImgNet10 since it is collected from 10 synsets of ImageNet: dog, bird, car, fish, monkey, turtle, lizard, bridge, cow, and crab. We selected 3,000 training images and 300 test images from each synset and resized all images to $1 2 8 \times 1 2 8$ . + +Architectures: On the MNIST dataset, both the ODENet and the CNN model consists of four convolutional layers and one fully-connected layer. The total number of parameters of the two models is around $1 4 0 \mathrm { k }$ . On the SVHN dataset, the networks are similar to those for the MNIST; we only changed the input channels of the first convolutional layer to three. On the ImgNet10 dataset, there are nine convolutional layers and one fully-connected layer for both the ODENet and the CNN model. The numbers of parameters is approximately $2 8 0 \mathrm { k }$ . In practice, the neural ODE can be solved with different numerical solvers such as the Euler method and the Runge-Kutta methods (Chen et al., 2018). Here, we use the easily-implemented Euler method in the experiments. To balance the computation and the continuity of the flow, we solve the ODE initial value problem in equation (1) by the Euler method with step size 0.1. Our implementation builds on the open-source neural ODE codes. 1 Details on the network architectures are included in the Appendix. + +Training: The experiments are conducted using two settings on each dataset—training models only with original non-perturbed images and training models on original images together with their perturbed versions. In both settings, we added a weight decay term into the training objective to regularize the norm of the weights, since this can help control the model’s representation capacity and improve the robustness of a neural network (Sokolic´ et al., 2017). In the second setting, images perturbed with random Gaussian noise are used to fine-tune the models, because augmenting the dataset with small perturbations can possibly improve the robustness of models and synthesizing Gaussian noise does not incur excessive computation time. + +# 3.2 ROBUSTNESS OF ODENETS TRAINED ONLY ON NON-PERTURBED IMAGES + +The first question we are interested in is how robust ODENets are against perturbations if the model is only trained on original non-perturbed images. We train CNNs and ODEnets to perform classification on three datasets and set the weight decay parameters for all models to be 0.0005. We make sure that both the well-trained ODENets and CNN models have satisfactory performances on original non-perturbed images, i.e., around $9 9 . 5 \%$ for MNIST, $9 5 . 0 \%$ for the SVHN, and $8 0 . 0 \%$ for ImgNet10. + +Since Gaussian noise is ubiquitous in modeling image degradation, we first evaluated the robustness of the models in the presence of zero-mean random Gaussian perturbations. It has also been shown that a deep model is vulnerable to harmful adversarial examples, such as the FGSM (Goodfellow et al., 2014). We are also interested in how robust ODENets are in the presence of adversarial examples. The standard deviation $\sigma$ of Gaussian noise and the $l _ { \infty }$ -norm $\epsilon$ of the FGSM attack for each dataset are shown in Table 1. + +Table 1: Robustness comparison of different models. We report their mean classification accuracies $( \% )$ and standard deviations (mean $\pm$ std) on perturbed images from the MNIST, the SVHN, and the ImgNet10 datasets. Two types of perturbations are used—zero-mean Gaussian noise and FGSM adversarial attack. The results show that ODENets are much more robust in comparison to CNN models. + +
Gaussian noiseAdversarial attack
MNISTσ=50σ=75σ=100FGSM-0.15FGSM-0.3FGSM-0.5
CNN98.1±0.785.8±4.356.4±5.663.4±2.324.0±8.98.3±3.2
ODENet98.7±0.690.6±5.473.2±8.683.5±0.942.1±2.414.3±2.1
SVHNg=15g=25σ=35FGSM-3/255FGSM-5/255FGSM-8/255
CNN90.0±1.276.3±2.760.9±3.929.2±2.913.7±1.95.4±1.5
ODENet95.7±0.788.1±1.578.2±2.158.2±2.343.0±1.330.9±1.4
ImgNet10σ=10σ=15σ = 25FGSM-5/255FGSM-8/255FGSM-16/255
CNN80.1±1.863.3±2.040.8±2.728.5±0.518.1±0.79.4±1.2
ODENet81.9±2.067.5±2.048.7±2.636.2±1.027.2±1.114.4±1.7
+ +From the results in Table 1, we observe that the ODENets demonstrate superior robustness compared to CNNs for all types of perturbations. On the MNIST dataset, in the presence of Gaussian perturbations with a large $\sigma$ of 100, the ODENet produces much higher accuracy on perturbed images compared to the CNN model ( $7 3 . 2 \%$ vs. $5 6 . 4 \%$ ). For the FGSM-0.3 adversarial examples, the accuracy of ONEnet is around twice as high as that of the CNN model. On the SVHN dataset, ODENets significantly outperform CNN models, e.g., for the FGSM-5/255 examples, the accuracy of the ODENet is $4 3 . 0 \%$ , which is much higher than that of the CNN model $( 1 3 . 7 \% )$ . On the ImgNet10, for both cases of $\sigma = 2 5$ and FGSM-8/255, ODENet outperforms CNNs by a large margin of around $9 \%$ . + +# 3.3 ROBUSTNESS OF ODENETS TRAINED ON ORIGINAL IMAGES TOGETHER WITH GAUSSIAN PERTURBATIONS + +Training a model on original images together with their perturbed versions can improve the robustness of the model. As mentioned previously, Gaussian noise is commonly assumed to be present in real-world images. Synthesizing Gaussian noise is also fast and easy. Thus, we add random Gaussian noise into the original images to generate their perturbed versions. ODENets and CNN models are both trained on original images together with their perturbed versions. The standard deviation of the added Gaussian noise is randomly chosen from $\bar { \{ 5 0 , 7 5 , 1 0 0 \} }$ on the MNIST dataset, $\{ 1 5 , 2 5 , 3 5 \}$ on the SVHN dataset, and $\{ 1 0 , 1 5 , 2 5 \}$ on the ImgNet10. All other hyperparameters are kept the same as above. + +Table 2: Robustness comparison of different models. We report their mean classification accuracies $( \% )$ and standard deviations (mean $\pm$ std) on perturbed images from the MNIST, the SVHN, and the ImgNet10 datsets. Three types of perturbations are used—zero-mean Gaussian noise, FGSM adversarial attack, and PGD adversarial attack. The results show that ODENets are more robust compared to CNN models. + +
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
ImgNet10σ= 25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
+ +The robustness of the models is evaluated under Gaussian perturbations, FGSM adversarial examples, and PGD (Madry et al., 2017) adversarial examples. The latter is a stronger attacker compared to the FGSM. The $l _ { \infty }$ -norm $\epsilon$ of the PGD attack for each dataset is shown in Table 2. Based on the results, we observe that ODENets consistently outperform CNN models on both two datasets. On the MNIST dataset, the ODENet outperforms the CNN against all types of perturbations. In particular, for the PGD-0.2 adversarial examples, the accuracy of the ODENet $( 6 4 . 7 \% )$ is much higher than that of the CNN $( 3 2 . 9 \% )$ . Besides, for the PGD-0.3 attack, the CNN is completely misled by the adversarial examples, but the ODENet can still classify perturbed images with an accuracy of $1 3 . 0 \%$ . On the SVHN dataset, ODENets also show superior robustness in comparison to CNN models. For all the adversarial examples, ODENets outperform CNN models by a margin of at least 10 percentage points. On the ImgNet10 dataset, the ODENet also performs better than CNN models against all forms of adversarial examples. + +# 3.4 INSIGHTS ON THE ROBUSTNESS OF ODENETS + +From the results in Sections 3.2 and 3.3, we find ODENets are more robust compared to CNN models. Here, we attempt to provide an intuitive understanding of the robustness of the neural ODE. In an ODENet, given some datum, the FE extracts an informative feature map from the datum. The neural ODE, serving as the RM, takes as input the feature map and performs a nonlinear mapping. In practice, we use the weight decay technique during training which regularizes the norm of weights in the FE part, so that the change of feature map in terms of a small perturbation on the input can be controlled. We aim to show that, in the neural ODE, a small change on the feature map will not lead to a large deviation from the original output associated with the feature map. + +Theorem 1 (ODE integral curves do not intersect (Coddington & Levinson, 1955; Younes, 2010; Dupont et al., 2019)). Let ${ \bf z } _ { 1 } ( t )$ and ${ \bf z } _ { 2 } ( t )$ be two solutions of the ODE in (1) with different initial conditions, i.e. ${ \bf z } _ { 1 } ( 0 ) \neq { \bf z } _ { 2 } ( 0 )$ . In (1), $f _ { \theta }$ is continuous in t and globally Lipschitz continuous in z. Then, it holds that ${ \bf z } _ { 1 } ( t ) \neq { \bf z } _ { 2 } ( t )$ for all $t \in [ 0 , \infty )$ . + +To illustrate this theorem, considering a simple 1- dimensional system in which the state is a scalar. As shown in Figure 2, equation (1) has a solution $z _ { 1 } ( t )$ starting from $\bar { A _ { 1 } } = ( 0 , \bar { z _ { 1 } } ( 0 ) )$ , where $z _ { 1 } ( 0 )$ is the feature of some datum. Equation (1) also has another two solutions $z _ { 2 } ( t )$ and $z _ { 3 } ( t )$ , whose starting points $A _ { 2 } = ( 0 , z _ { 2 } ( 0 ) )$ and $A _ { 3 } ~ = ~ ( 0 , z _ { 3 } ( 0 ) )$ , both of which are close to $A _ { 1 }$ . Suppose $A _ { 1 }$ is between $A _ { 2 }$ and $A _ { 3 }$ . By Theorem 1, we know that the integral curve $z _ { 1 } ( t )$ is always sandwiched between the integral curves $z _ { 2 } ( t )$ and $z _ { 3 } ( t )$ . + +Now, let $\epsilon < \operatorname* { m i n } \{ | z _ { 2 } ( 0 ) - z _ { 1 } ( 0 ) | , | z _ { 3 } ( 0 ) - z _ { 1 } ( 0 ) | \} .$ . Consider a solution $\widetilde { z } _ { 1 } ( t )$ of equation (1). The integral curve $\widetilde { z } _ { 1 } ( t )$ starts from a point $\widetilde { A } _ { 1 } = ( 0 , \widetilde { z } _ { 1 } ( 0 ) )$ . The point $\widetilde { A } _ { 1 }$ is !in the $\epsilon$ -neighborhood of $A _ { 1 }$ with $| \widetilde z _ { 1 } ( 0 ) - z _ { 1 } ( 0 ) | < \epsilon .$ . By Theorem 1, we know that $| \tilde { z } _ { 1 } ( T ) - z _ { 1 } ( T ) | \leq | z _ { 3 } ( T ) -$ $z _ { 2 } ( T ) |$ !. In other words, if any perturbation smaller than $\epsilon$ is added to the scalar $z _ { 1 } ( 0 )$ in $A _ { 1 }$ , the deviation from the original output $z _ { 1 } ( T )$ is bounded by the distance between $z _ { 2 } ( T )$ and $z _ { 3 } ( T )$ . In contrast, in a CNN model, there is no such bound on the deviation from the original output. Thus, we opine that due to this non-intersecting property, ODENets are intrinsically robust. + +![](images/901235eac73c549d6bdb2646bd2f84cf3a82d0581b3c2860c0361d6a9e2870e9.jpg) +Figure 2: No integral curves intersect. The integral curve starting from $\widetilde { A _ { 1 } }$ is always sandwiched between two integral curves starting from $A _ { 1 }$ and $A _ { 3 }$ . + +# 4 TISODE: BOOSTING THE ROBUSTNESS OF NEURAL ODES + +In the previous section, we presented an empirical study on the robustness of ODENets and observed that ODENets are more robust compared to CNN models. In this section, we explore how to boost the robustness of the vanilla neural ODE model further. This motivates the proposal of time-invariant steady neural ODEs (TisODEs). + +# 4.1 TIME-INVARIANT STEADY NEURAL ODES + +From the discussion in Section 3.4, the key to improving the robustness of neural ODEs is to control the difference between neighboring integral curves. By Grownall’s inequality (Howard, 1998) (see Theorem 2 in the Appendix), we know that the difference between two terminal states is bounded by the difference between initial states multiplied by the exponential of the dynamics’ Lipschitz constant. However, it is very difficult to bound the Lipschitz constant of the dynamics directly. Alternatively, we propose to achieve the goal of controlling the output deviation by following two steps: (i) removing the time dependence of the dynamics and (ii) imposing a certain steady-state constraint. + +![](images/b7cae751cc80ff17df6fe622bceb23becc612933d297ca2ae9a8307363b3a95d.jpg) +Figure 3: An illustration of the timeinvariant property of ODEs. We can see that the curve $\widetilde { \mathbf { z } } _ { 1 } ( t )$ is exactly the horizontal translation of ${ \bf z } _ { 1 } ( t )$ on the interval $[ T ^ { \prime } , \infty )$ . + +In the neural ODE characterized by equation (1), the dynamics $f _ { \theta } ( \mathbf { z } ( t ) , t )$ depends on both the state ${ \bf z } ( t )$ at time $t$ and the time $t$ itself. In contrast, if the neural ODE is modified to be time-invariant, the time dependence of the dynamics is removed. Consequently, the dynamics depends only on the state z. So, we can rewrite the dynamics function as $f _ { \boldsymbol { \theta } } ( \mathbf { z } )$ , and the neural ODE is characterized as + +$$ +\left\{ \begin{array} { l l } { \displaystyle \frac { \mathrm { d } \mathbf { z } ( t ) } { \mathrm { d } t } = f _ { \boldsymbol { \theta } } ( \mathbf { z } ( t ) ) ; } \\ { \mathbf { z } ( 0 ) = \mathbf { z } _ { \mathrm { i n } } ; } \\ { \mathbf { z } _ { \mathrm { o u t } } = \mathbf { z } ( T ) . } \end{array} \right. +$$ + +Let ${ \bf z } _ { 1 } ( t )$ be a solution of (2) on $[ 0 , \infty )$ and $\epsilon > 0$ be a small positive value. We define the set $\mathbb { M } _ { 1 } = \{ ( \mathbf { z } _ { 1 } ( t ) , t ) | t \in [ 0 , T ]$ , " $\| \mathbf { z } _ { 1 } ( t ) - \mathbf { z } _ { 1 } ( 0 ) \| \leq \epsilon \}$ . This set contains all points on the curve of ${ \bf z } _ { 1 } ( t )$ during $[ 0 , T ]$ that are also inside the $\epsilon$ -neighborhood of ${ \bf z } _ { 1 } ( 0 )$ . For some element $( { \bf z } _ { 1 } ( T ^ { \prime } ) , T ^ { \prime } ) \in \mathbb { M } _ { 1 }$ , let $\widetilde { \mathbf { z } } _ { 1 } ( t )$ be the solution of (2) which starts from $\widetilde { \mathbf z } _ { 1 } ( 0 ) = \mathbf z _ { 1 } ( T ^ { \prime } )$ . Then we have + +$$ +\widetilde { { \mathbf z } } _ { 1 } ( t ) = { \mathbf z } _ { 1 } ( t + T ^ { \prime } ) +$$ + +for all $t$ in $[ 0 , \infty )$ . The property shown in equation (3) is known as the time-invariant property. It indicates that the integral curve $\widetilde { \mathbf { z } } _ { 1 } ( t )$ is the $- T ^ { \prime }$ shift of ${ \bf z } _ { 1 } ( t )$ (Figure 3). + +We can regard $\widetilde { \mathbf { z } } _ { 1 } ( 0 )$ as a slightly perturbed version of ${ \bf z } _ { 1 } ( 0 )$ , and we are interested in how large the difference between $\widetilde { \mathbf z } _ { 1 } ( T )$ and ${ \bf z } _ { 1 } ( T )$ is. In a robust model, the difference should be small. By equation (3), we have $\| \mathbf { \tilde { z } } _ { 1 } ( T ) - \mathbf { z } _ { 1 } ( T ) \| = \| \mathbf { z } _ { 1 } ( T + T ^ { \prime } ) - \mathbf { z } _ { 1 } ( T ) \|$ . Since $T ^ { \prime } \in [ 0 , T ]$ , the difference between ${ \bf z } _ { 1 } ( T )$ and $\widetilde { \mathbf z } _ { 1 } ( T )$ can be bounded as follows, + +$$ +\| \widetilde { \mathbf z } _ { 1 } ( T ) - \mathbf z _ { 1 } ( T ) \| = \left\| \int _ { T } ^ { T + T ^ { \prime } } f _ { \theta } ( \mathbf z _ { 1 } ( t ) ) \mathrm { d } t \right\| \leq \left\| \int _ { T } ^ { T + T ^ { \prime } } | f _ { \theta } ( \mathbf z _ { 1 } ( t ) ) | \mathrm { d } t \right\| \leq \left\| \int _ { T } ^ { 2 T } | f _ { \theta } ( \mathbf z _ { 1 } ( t ) ) | \mathrm { d } t \right\| , +$$ + +where all norms are $\ell _ { 2 }$ norms and $| f _ { \theta } |$ denotes the element-wise absolute operation of a vectorvalued function $f _ { \theta }$ . That is to say, the difference between $\widetilde { \mathbf z } _ { 1 } ( T )$ and ${ \bf z } _ { 1 } ( T )$ can be bounded by only using the information of the curve ${ \bf z } _ { 1 } ( t )$ . For any $t ^ { \prime } \in [ 0 , T ]$ and element $( { \bf z } _ { 1 } ( t ^ { \prime } ) , t ^ { \prime } ) \in \mathbb { M } _ { 1 }$ , consider the integral curve that starts from ${ \bf z } _ { 1 } ( t ^ { \prime } )$ . The difference between the output state of this curve and ${ \bf z } _ { 1 } ( T )$ satisfies inequality (4). + +Therefore, we propose to add an additional term $L _ { \mathrm { s s } }$ to the loss function when training the timeinvariant neural ODE: + +$$ +L _ { \mathrm { s s } } = \sum _ { i = 1 } ^ { N } \left\| \int _ { T } ^ { 2 T } | f _ { \theta } ( \mathbf { z } _ { i } ( t ) ) | \mathrm { d } t \right\| , +$$ + +where $N$ is the number of samples in the training set and ${ \bf z } _ { i } ( t )$ is the solution whose initial state equals to the feature of the $i ^ { \mathrm { t h } }$ sample. The regularization term $L _ { \mathrm { s s } }$ is termed as the steady-state loss. This terminology “steady state” is borrowed from the dynamical systems literature. In a stable dynamical system, the states stabilize around a fixed point, known as the steady-state, as time tends to infinity. If we can ensure that $L _ { \mathrm { s s } }$ is small, for each sample, the outputs of all the points in $\mathbb { M } _ { i }$ will stabilize around ${ \bf z } _ { i } ( T )$ . Consequently, the model is robust. This modification of the neural ODE is dubbed Time-invariant steady neural $O D E$ . + +# 4.2 EVALUATING ROBUSTNESS OF TISODE-BASED CLASSIFIERS + +Here, we conduct experiments to evaluate the robustness of our proposed TisODE, and compare TisODE-based models with the vanilla ODENets. We train all models with original non-perturbed images together with their Gaussian perturbed versions. The regularization parameter for the steadystate loss $L _ { \mathrm { s s } }$ is set to be 0.1. All other hyperparameters are exactly the same as those in Section 3.3. + +From the results in Table 3, we can see that our proposed TisODE-based models are clearly more robust compared to vanilla ODENets. On the MNIST dataset, when combating FGSM-0.3 attacks, the TisODE-based models outperform vanilla ODENets by more than 4 percentage points. For the FGSM-0.5 adversarial examples, the accuracy of the TisODE-based model is 6 percentage points better. On the SVHN dataset, the TisODE-based models perform better in terms of all forms of adversarial examples. On the ImgNet10 dataset, the TisODE-based models also outperform vanilla ODE-based models on all types of perturbations. In the presence of FGSM and PGD-5/255 examples, the accuracies are enhanced by more than 2 percentage points. + +Table 3: Classification accuracy (mean $\pm$ std in $\%$ ) on perturbed images from MNIST, SVHN and ImgNet10. To evaluate the robustness of classifiers, we use three types of perturbations, namely zero-mean Gaussian noise with standard deviation $\sigma$ , FGSM attack and PGD attack. From the results, the proposed TisODE effectively improve the robustness of the vanilla neural ODE. + +
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
TisODE99.6±0.075.7±1.426.5±3.867.4±1.513.2±1.0
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
TisODE94.9±0.151.6±1.238.2±1.952.0±0.928.2±0.3
ImgNet10g=25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
TisODE92.8±0.444.3±0.731.4±1.131.1±1.214.5±1.1
+ +4.3 TISODE - A GENERALLY APPLICABLE DROP-IN TECHNIQUE FOR IMPROVING THE ROBUSTNESS OF DEEP NETWORKS + +In view of the excellent robustness of the TisODE, we claim that the proposed TisODE can be used as a general drop-in module for improving the robustness of deep networks. We support this claim by showing the TisODE can work in conjunction with other state-of-the-art techniques and further boost the models’ robustness. These techniques include the feature denoising (FDn) method (Xie et al., 2019) and the input randomization (IR) method (Xie et al., 2017). We conduct experiments on the MNIST and SVHN datasets. All models are trained with original non-perturbed images together with their Gaussian perturbed versions. We show that models using the FDn/IRd technique becomes much more robust when equipped with the TisODE. In the FDn experiments, the dot-product nonlocal denoising layer (Xie et al., 2019) is added to the head of the fully-connected classifier. + +Table 4: Classification accuracy (mean $\pm$ std in $\%$ ) on perturbed images from MNIST and SVHN. We evaluate against three types of perturbations, namely zero-mean Gaussian noise with standard deviation $\sigma$ , FGSM attack and PGD attack. From the results, upon the CNNs modified with FDn and IRd, using TisODE can further improve the robustness. + +
Gaussian noiseAdversarial attack
MNISTσ= 100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
CNN-FDn TisODE-FDn99.0±0.174.0±4.132.6±5.358.9±4.08.2±2.6
CNN-IRd99.4±0.0 95.3±0.980.6±2.3 78.1±2.240.4±5.7 36.7±2.172.6±2.4 79.6±1.928.2±3.6 55.5±2.9
TisODE-IRd97.6±0.186.8±2.349.1±0.288.8±0.966.0±0.9
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN
90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
CNN-FDn92.4±0.143.8±1.431.5±3.040.0±2.619.6±3.4
TisODE-FDn95.2±0.157.8±1.748.2±2.053.4±2.932.3±1.0
CNN-IRd84.9±1.265.8±0.454.7±1.274.0±0.564.5±0.8
TisODE-IRd91.7±0.574.4±1.261.9±1.881.6±0.871.0±0.5
+ +From Table 4, we observe that both FDn and IRd can effectively improve the adversarial robustness of vanilla CNN models (CNN-FDn, CNN-IRd). Furthermore, combining our proposed TisODE with FDn or IRd (TisODE-FDn, TisODE-IRd), the adversarial robustness of the resultant model is significantly enhanced. For example, on the MNIST dataset, the additional use of our TisODE increases the accuracies on the PGD-0.3 examples by at least 10 percentage points for both FDn $8 . 2 \%$ to $2 8 . 2 \% )$ and IRd $5 5 . 5 \%$ to $6 6 . 0 \%$ ). However, on both MNIST and SVHN datasets, the IRd technique improves the robustness against adversarial examples, but its performance is worse on random Gaussian noise. With the help of the TisODE, the degradation in the robustness against random Gaussian noise can be effectively ameliorated. + +# 5 RELATED WORKS + +In this section, we briefly review related works on the neural ODE and works concerning improving the robustness of deep neural networks. + +Neural ODE: The neural ODE (Chen et al., 2018) method models the input and output as two states of a continuous-time dynamical system by approximating the dynamics of this system with trainable layers. Before the proposal of neural ODE, the idea of modeling nonlinear mappings using continuous-time dynamical systems was proposed in Weinan (2017). Lu et al. (2017) also showed that several popular network architectures could be interpreted as the discretization of a continuoustime ODE. For example, the ResNet (He et al., 2016) and PolyNet (Zhang et al., 2017) are associated with the Euler scheme and the FractalNet (Larsson et al., 2016) is related to the Runge-Kutta scheme. In contrast to these discretization models, neural ODEs are endowed with an intrinsic invertibility property, which yields a family of invertible models for solving inverse problems (Ardizzone et al., 2018), such as the FFJORD (Grathwohl et al., 2018). + +Recently, many researchers have conducted studies on neural ODEs from the perspectives of optimization techniques, approximation capabilities, and generalization. Concerning the optimization of neural ODEs, the auto-differentiation techniques can effectively train ODENets, but the training procedure is computationally and memory inefficient. To address this problem, Chen et al. (2018) proposed to compute gradients using the adjoint sensitivity method (Pontryagin, 2018), in which there is no need to store any intermediate quantities of the forward pass. Also in Quaglino et al. (2019), the authors proposed the SNet which accelerates the neural ODEs by expressing their dynamics as truncated series of Legendre polynomials. Concerning the approximation capability, Dupont et al. (2019) pointed out the limitations in approximation capabilities of neural ODEs because of the preserving of input topology. The authors proposed an augmented neural ODE which increases the dimension of states by concatenating zeros so that complex mappings can be learned with simple flow. The most relevant work to ours concerns strategies to improve the generalization of neural ODEs. In Liu et al. (2019), the authors proposed the neural stochastic differential equation (SDE) by injecting random noise to the dynamics function and showed that the generalization and robustness of vanilla neural ODEs could be improved. However, our improvement on the neural ODEs is explored from a different perspective by introducing constraints on the flow. We empirically found that our proposal and the neural SDE can work in tandem to further boost the robustness of neural ODEs. + +Robust Improvement: A straightforward way of improving the robustness of a model is to smooth the loss surface by controlling the spectral norm of the Jacobian matrix of the loss function (Sokolic´ et al., 2017). In terms of adversarial examples (Carlini & Wagner, 2017; Chen et al., 2017), researchers have proposed adversarial training strategies (Madry et al., 2017; Elsayed et al., 2018; Trame\`r et al., 2017) in which the model is fine-tuned with adversarial examples generated in realtime. However, generating adversarial examples is not computationally efficient, and there exists a trade-off between the adversarial robustness and the performance on original non-perturbed images (Yan et al., 2018; Tsipras et al., 2018). In Wang et al. (2018a), the authors model the ResNet as a transport equation, in which the adversarial vulnerability can be interpreted as the irregularity of the decision boundary. Consequently, a diffusion term is introduced to enhance the robustness of the neural nets. Besides, there are also some works that propose novel architectural defense mechanisms against adversarial examples. For example, Xie et al. (2017) utilized random resizing and random padding to destroy the specific structure of adversarial perturbations; Wang et al. (2018b) and Wang et al. (2018c) improved the robustness of neural networks by replacing the output layers with novel interpolating functions; In Xie et al. (2019), the authors designed a feature denoising filter that can remove the perturbation’s pattern from feature maps. In this work, we explore the intrinsic robustness of a specific novel architecture (neural ODE), and show that the proposed TisODE can improve the robustness of deep networks and can also work in tandem with these state-of-the-art methods Xie et al. (2017; 2019) to achieve further improvements. + +# 6 CONCLUSION + +In this paper, we first empirically study the robustness of neural ODEs. Our studies reveal that neural ODE-based models are superior in terms of robustness compared to CNN models. We then explore how to further boost the robustness of vanilla neural ODEs and propose the TisODE. Finally, we show that the proposed TisODE outperforms the vanilla neural ODE and also can work in conjunction with other state-of-the-art techniques to further improve the robustness of deep networks. Thus, the TisODE method is an effective drop-in module for building robust deep models. + +# ACKNOWLEDGEMENT + +This work is funded by a Singapore National Research Foundation (NRF) Fellowship (R-263-000- D02-281). + +Jiashi Feng was partially supported by NUS IDS R-263-000-C67-646, ECRA R-263-000-C87-133, MOE Tier-II R-263-000-D17-112 and AI.SG R-263-000-D97-490 + +# REFERENCES + +Lynton Ardizzone, Jakob Kruse, Sebastian Wirkert, Daniel Rahner, Eric W Pellegrini, Ralf S Klessen, Lena Maier-Hein, Carsten Rother, and Ullrich Ko¨the. Analyzing inverse problems with invertible neural networks. arXiv preprint arXiv:1808.04730, 2018. + +Nicholas Carlini and David Wagner. Towards evaluating the robustness of neural networks. In 2017 IEEE Symposium on Security and Privacy (SP), pp. 39–57. IEEE, 2017. + +Pin-Yu Chen, Huan Zhang, Yash Sharma, Jinfeng Yi, and Cho-Jui Hsieh. Zoo: Zeroth order optimization based black-box attacks to deep neural networks without training substitute models. In Proceedings of the 10th ACM Workshop on Artificial Intelligence and Security, pp. 15–26. ACM, 2017. + +Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differential equations. In Advances in neural information processing systems, pp. 6571–6583, 2018. + +Earl A Coddington and Norman Levinson. Theory of ordinary differential equations. Tata McGrawHill Education, 1955. + +Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee, 2009. + +Emilien Dupont, Arnaud Doucet, and Yee Whye Teh. Augmented neural odes. arXiv preprint arXiv:1904.01681, 2019. + +Gamaleldin F Elsayed, Shreya Shankar, Brian Cheung, Nicolas Papernot, Alex Kurakin, Ian Goodfellow, and Jascha Sohl-Dickstein. Adversarial examples that fool both human and computer vision. arXiv preprint arXiv:1802.08195, 10, 2018. + +Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial examples. arXiv preprint arXiv:1412.6572, 2014. + +Will Grathwohl, Ricky TQ Chen, Jesse Betterncourt, Ilya Sutskever, and David Duvenaud. Ffjord: Free-form continuous dynamics for scalable reversible generative models. arXiv preprint arXiv:1810.01367, 2018. + +Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016. + +Ralph Howard. The gronwall inequality. lecture notes, 1998. + +Gustav Larsson, Michael Maire, and Gregory Shakhnarovich. Fractalnet: Ultra-deep neural networks without residuals. arXiv preprint arXiv:1605.07648, 2016. + +Yann LeCun, Le´on Bottou, Yoshua Bengio, Patrick Haffner, et al. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. + +Xuanqing Liu, Si Si, Qin Cao, Sanjiv Kumar, and Cho-Jui Hsieh. Neural sde: Stabilizing neural ode networks with stochastic noise. arXiv preprint arXiv:1906.02355, 2019. + +Yiping Lu, Aoxiao Zhong, Quanzheng Li, and Bin Dong. Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations. arXiv preprint arXiv:1710.10121, 2017. + +Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. Towards deep learning models resistant to adversarial attacks. arXiv preprint arXiv:1706.06083, 2017. + +Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, and Andrew Y Ng. Reading digits in natural images with unsupervised feature learning. 2011. + +Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017. + +Lev Semenovich Pontryagin. Mathematical theory of optimal processes. Routledge, 2018. + +Alessio Quaglino, Marco Gallieri, Jonathan Masci, and Jan Koutn´ık. Accelerating neural odes with spectral elements. arXiv preprint arXiv:1906.07038, 2019. + +Jure Sokolic´, Raja Giryes, Guillermo Sapiro, and Miguel RD Rodrigues. Robust large margin deep neural networks. IEEE Transactions on Signal Processing, 65(16):4265–4280, 2017. + +Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow, and Rob Fergus. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013. + +Florian Trame\`r, Alexey Kurakin, Nicolas Papernot, Ian Goodfellow, Dan Boneh, and Patrick McDaniel. Ensemble adversarial training: Attacks and defenses. arXiv preprint arXiv:1705.07204, 2017. + +Dimitris Tsipras, Shibani Santurkar, Logan Engstrom, Alexander Turner, and Aleksander Madry. Robustness may be at odds with accuracy. arXiv preprint arXiv:1805.12152, 2018. + +B. Wang, B. Yuan, Z. Shi, and S. Osher. ResNets Ensemble via the Feynman-Kac Formalism to Improve Natural and Robust Accuracies. arXiv e-prints, art. arXiv:1811.10745, Nov 2018a. + +Bao Wang, Alex T Lin, Zuoqiang Shi, Wei Zhu, Penghang Yin, Andrea L Bertozzi, and Stanley J Osher. Adversarial defense via data dependent activation function and total variation minimization. arXiv preprint arXiv:1809.08516, 2018b. + +Bao Wang, Xiyang Luo, Zhen Li, Wei Zhu, Zuoqiang Shi, and Stanley Osher. Deep neural nets with interpolating function as output activation. In Advances in Neural Information Processing Systems, pp. 743–753, 2018c. + +E Weinan. A proposal on machine learning via dynamical systems. Communications in Mathematics and Statistics, 5(1):1–11, 2017. + +Cihang Xie, Jianyu Wang, Zhishuai Zhang, Zhou Ren, and Alan Yuille. Mitigating adversarial effects through randomization. arXiv preprint arXiv:1711.01991, 2017. + +Cihang Xie, Yuxin Wu, Laurens van der Maaten, Alan L Yuille, and Kaiming He. Feature denoising for improving adversarial robustness. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 501–509, 2019. + +Ziang Yan, Yiwen Guo, and Changshui Zhang. Deep defense: Training dnns with improved adversarial robustness. In Advances in Neural Information Processing Systems, pp. 419–428, 2018. + +Laurent Younes. Shapes and diffeomorphisms, volume 171. Springer, 2010. + +Xingcheng Zhang, Zhizhong Li, Chen Change Loy, and Dahua Lin. Polynet: A pursuit of structural diversity in very deep networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 718–726, 2017. + +# 7 APPENDIX + +7.1 NETWORKS USED ON THE MNIST, THE SVHN, AND THE IMGNET10 DATASETS + +Table 5: The architectures of the ODENets on different datasets. + +
MNISTRepetitionLayer
FE×1×1Conv(1,64,3,1) +GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
+ +
SVHNRepetitionLayer
FE×1×1Conv(3,64,3,1) + GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
+ +
ImgNet10RepetitionLayer
FE×1Conv(3,32,5,2)+GroupNorm
×1MaxPooling(2)
×1BaiscBlock(32, 64,2)
×1MaxPooling(2)
RM×3BaiscBlock(64,64,1)
FCC×1AdaptiveAvgPool2d + Linear(64,10)
+ +In Table 5, the four arguments of the Conv layer represent the input channel, output channel, kernel size, and the stride. The two arguments of the Linear layer represents the input dimension and the output dimension of this fully-connected layer. In the network on the ImgNet10, the BasicBlock refers to the standard architecture in (He et al., 2016), the three arguments of the BasicBlock represent the input channel, output channel and the stride of the Conv layers inside the block. Note that we replace the BatchNorm layers in BasicBlocks as the GroupNorm to guarantee that the dynamics of each datum is independent of other data in the same mini-batch. + +# 7.2 THE CONSTRUCTION OF IMGNET10 DATASET + +Table 6: The corresponding indexes to each class in the original ImageNet dataset + +
ClassIndexing
dogn02090721, n02091032, n02088094
birdn01532829, n01558993, n01534433
carn02814533, n03930630, n03100240
fishn01484850,n01491361, n01494475
monkeyn02483708,n02484975, n02486261
turtlen01664065,n01665541, n01667114
lizardn01677366,n01682714,n01685808
bridgen03933933,n04366367,n04311004
cown02403003,n02408429,n02410509
crabn01980166,n01978455,n01981276
+ +# 7.3 GRONWALL’S INEQUALITY + +We formally state the Gronwall’s Inequality here, following the version in (Howard, 1998). + +Theorem 2. Let $U \subset \mathbb { R } ^ { d }$ be an open set. Let $f : U \times [ 0 , T ] \to \mathbb { R } ^ { d }$ be a continuous function and let $\mathbf { z } _ { 1 }$ , $\mathbf { z } _ { 2 }$ : $[ 0 , T ] \to U$ satisfy the initial value problems: + +$$ +\begin{array} { r l } & { \frac { \mathrm { d } { \mathbf z } _ { 1 } ( t ) } { \mathrm { d } t } = f ( { \mathbf z } _ { 1 } ( t ) , t ) , \quad { \mathbf z } _ { 1 } ( t ) = { \mathbf x } _ { 1 } } \\ & { \frac { \mathrm { d } { \mathbf z } _ { 2 } ( t ) } { \mathrm { d } t } = f ( { \mathbf z } _ { 2 } ( t ) , t ) , \quad { \mathbf z } _ { 2 } ( t ) = { \mathbf x } _ { 2 } } \end{array} +$$ + +Assume there is a constant $C \geq 0$ such that, for all $t \in [ 0 , T ]$ , + +$$ +\| f ( \mathbf { z } _ { 2 } ( t ) , t ) - f ( \mathbf { z } _ { 1 } ( t ) , t ) ) \| \leq C \| \mathbf { z } _ { 2 } ( t ) - \mathbf { z } _ { 1 } ( t ) \| +$$ + +Then, for any $t \in [ 0 , T ]$ , + +$$ +\begin{array} { r } { \| \mathbf { z } _ { 1 } ( t ) - \mathbf { z } _ { 2 } ( t ) \| \leq \| \mathbf { x } _ { 2 } - \mathbf { x } _ { 1 } \| \cdot e ^ { C t } . } \end{array} +$$ + +# 7.4 MORE EXPERIMENTAL RESULTS + +# 7.4.1 COMPARISON IN THE SETTING OF ADVERSARIAL TRAINING + +We implement the adversarial training of the models on the MNIST dataset, and the adversarial examples for training are generated in real-time via the FGSM method (epsilon $_ { 1 = 0 . 3 }$ ) during each epoch (Madry et al., 2017). The results of the adversarially trained models are shown in Table 7. We can observe that the neural ODE-based models are consistently more robust than CNN models. The proposed TisODE also outperforms the vanilla neural ODE. + +Table 7: Classification accuracy $( \% )$ on perturbed images from MNIST. To evaluate the robustness of classifiers, we use three types of perturbations, namely zero-mean Gaussian noise with standard deviation $\sigma$ , FGSM attack and PGD attack. + +
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.3
CNN58.098.421.15.3
ODENet84.299.136.012.3
TisODE87.999.166.578.9
+ +# 7.4.2 EXPERIMENTS ON THE CIFAR10 DATASET + +We conduct experiments on CIFAR10 to compare the robustness of CNN and neural ODE-based models. We train all the models only with original non-perturbed images and evaluate the robustness of models against random Gaussian noise and FGSM adversarial attacks. The results are shown in Table 8. We can observe that the ONENet is more robust than the CNN model in terms of both the random noise and the FGSM attack. Besides, our proposal, TisODE, can improve the robustness of the vanilla neural ODE. + +Table 8: Classification accuracy $( \% )$ on perturbed images from CIFAR10. To evaluate the robustness of classifiers, we use two types of perturbations, namely zero-mean Gaussian noise with standard deviation $\sigma$ and FGSM attack. + +
Gaussian noiseAdversarial attack
CIFAR10σ=15σ=20FGSM-8/255FGSM-10/255
CNN70.257.624.318.4
ODENet72.660.631.226.0
TisODE74.362.033.626.8
+ +Here, we control the number of parameters to be the same for all kinds of models. We use a small network, which consists of five convolutional layers and one linear layer. + +Table 9: The architecture of the ODENet on CIFAR10. + +
RepetitionLayer
FE×1Conv(3,16,3,1)+GroupNorm+ReLU
×1Conv(16,32,3,2) + GroupNorm+ReLU
×1Conv(32,64,3,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
+ +# 7.4.3 AN EXTENSION ON THE COMPARISON BETWEEN CNNS AND ODENETS + +Here, we compare CNN and neural ODE-based models by controlling both the number of parameters and the number of function evaluations. We conduct experiments on the MNIST dataset, and all the models are trained only with original non-perturbed images. + +For the neural ODE-based models, the time range is set from 0 to 1. We use the Euler method, and the step size is set to be 0.05. Thus the number of evaluations is $1 / 0 . 0 5 = 2 0 $ . For the CNN models (specifically ResNet), we repeatedly concatenate the residual block for 20 times, and these 20 blocks share the same weights. Our experiments show that, in this condition, the neural ODE-based models still outperform the CNN models (FGSM-0.15: $8 7 . 5 \%$ vs. $8 1 . 9 \%$ , FGSM-0.3: $5 3 . 4 \%$ vs. $4 9 . 7 \%$ , PGD-0.2: $1 1 . 8 \%$ vs. $4 . 8 \%$ ). \ No newline at end of file diff --git a/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_content_list.json b/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..457e6a1f71f302cd3fc4ef6950331064f84c1627 --- /dev/null +++ b/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_content_list.json @@ -0,0 +1,1703 @@ +[ + { + "type": "text", + "text": "ON ROBUSTNESS OF NEURAL ORDINARY DIFFERENTIAL EQUATIONS ", + "text_level": 1, + "bbox": [ + 176, + 98, + 820, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Hanshu YAN\\*, Jiawei DU\\*, Vincent Y. F. TAN & Jiashi FENG ", + "bbox": [ + 183, + 172, + 619, + 188 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Department of Electrical and Computer Engineering \nNational University of Singapore \n{hanshu.yan, dujiawei}@u.nus.edu, {vtan, elefjia}@nus.edu.sg ", + "bbox": [ + 184, + 189, + 764, + 229 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 266, + 544, + 281 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Neural ordinary differential equations (ODEs) have been attracting increasing attention in various research domains recently. There have been some works studying optimization issues and approximation capabilities of neural ODEs, but their robustness is still yet unclear. In this work, we fill this important gap by exploring robustness properties of neural ODEs both empirically and theoretically. We first present an empirical study on the robustness of the neural ODE-based networks (ODENets) by exposing them to inputs with various types of perturbations and subsequently investigating the changes of the corresponding outputs. In contrast to conventional convolutional neural networks (CNNs), we find that the ODENets are more robust against both random Gaussian perturbations and adversarial attack examples. We then provide an insightful understanding of this phenomenon by exploiting a certain desirable property of the flow of a continuous-time ODE, namely that integral curves are non-intersecting. Our work suggests that, due to their intrinsic robustness, it is promising to use neural ODEs as a basic block for building robust deep network models. To further enhance the robustness of vanilla neural ODEs, we propose the time-invariant steady neural ODE (TisODE), which regularizes the flow on perturbed data via the time-invariant property and the imposition of a steady-state constraint. We show that the TisODE method outperforms vanilla neural ODEs and also can work in conjunction with other state-of-the-art architectural methods to build more robust deep networks. ", + "bbox": [ + 233, + 297, + 764, + 575 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 606, + 336, + 622 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Neural ordinary differential equations (Chen et al., 2018) form a family of models that approximate nonlinear mappings by using continuous-time ODEs. Due to their desirable properties, such as invertibility and parameter efficiency, neural ODEs have attracted increasing attention recently (Dupont et al., 2019; Liu et al., 2019). For example, Grathwohl et al. (2018) proposed a neural ODE-based generative model—the FFJORD—to solve inverse problems; Quaglino et al. (2019) used a higher-order approximation of the states in a neural ODE, and proposed the SNet to accelerate computation. Along with the wider deployment of neural ODEs, robustness issues come to the fore. However, the robustness of neural ODEs is still yet unclear. In particular, it is unclear how robust neural ODEs are in comparison to the widely-used CNNs. Robustness properties of CNNs have been studied extensively. In this work, we present the first systematic study on exploring the robustness properties of neural ODEs. ", + "bbox": [ + 174, + 638, + 823, + 790 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "To do so, we consider the task of image classification. We expect that results would be similar for other machine learning tasks such as regression. Neural ODEs are dimension-preserving mappings, but a classification model transforms a high-dimensional input—such as an image—into an output whose dimension is equal to the number of classes. Thus, we consider the neural ODE-based classification network (ODENet) whose architecture is shown in Figure 1. An ODENet consists of three components: the feature extractor (FE) consists of convolutional layers which maps an input datum to a multi-channel feature map, a neural ODE that serves as the nonlinear representation mapping (RM), and the fully-connected classifier (FCC) that generates a prediction vector based on the output of the RM. ", + "bbox": [ + 174, + 799, + 825, + 924 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The robustness of a classification model can be evaluated through the lens of its performance on perturbed images. To comprehensively investigate the robustness of neural ODEs, we perturb original images with commonly-used perturbations, namely, random Gaussian noise (Szegedy et al., 2013) and harmful adversarial examples (Goodfellow et al., 2014; Madry et al., 2017). We conduct experiments in two common settings—training the model only on authentic non-perturbed images and training the model on authentic images as well as the Gaussian perturbed ones. We observe that ODENets are more robust compared to CNN models against all types of perturbations in both settings. We then provide an insightful understanding of such intriguing robustness of neural ODEs by exploiting a certain property of the flow (Dupont et al., 2019), namely that integral curves that start at distinct initial states are nonintersecting. The flow of a continuous-time ODE is defined as the family of solutions/paths traversed by the state, starting from different initial points, and an integral curve is a specific solution for a given initial point. The non-intersecting property indicates that an integral curve starting from some point is constrained by the integral curves starting from that point’s neighborhood. Thus, in an ODENet, if a correctly classified datum is slightly perturbed, the integral curve associated to its perturbed version would not change too much from the original one. Consequently, the perturbed datum could still be correctly classified. Thus, there exists intrinsic robustness regularization in ODENets, which is absent from CNNs. ", + "bbox": [ + 174, + 104, + 614, + 353 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/6e9bd948731c27f7167c50dec38ca4fa802294fd3834ad917ec0b4023c814c57.jpg", + "image_caption": [ + "Figure 1: The architecture of an ODENet. The neural ODE block serves as a dimension-preserving nonlinear mapping. " + ], + "image_footnote": [], + "bbox": [ + 658, + 92, + 790, + 265 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 353, + 825, + 421 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Motivated by this property of the neural ODE flow, we attempt to explore a more robust neural ODE architecture by introducing stronger regularization on the flow. We thus propose a Time-Invariant Steady neural ODE (TisODE). The TisODE removes the time dependence of the dynamics in an ODE and imposes a steady-state constraint on the integral curves. Removing the time dependence of the derivative results in the time-invariant property of the ODE. To wit, given a solution ${ \\bf z } _ { 1 } ( t )$ , another solution $\\widetilde { \\mathbf { z } } _ { 1 } ( t )$ , with an initial state $\\tilde { { \\bf z } } _ { 1 } ( 0 ) \\bar { { \\bf \\phi } } = { \\bf \\bar { z } } _ { 1 } ( T ^ { \\prime } )$ for some $T ^ { \\prime } > 0$ , can be regarded as the $- T ^ { \\prime }$ - shift version of ${ \\bf z } _ { 1 } ( t )$ . Such a time-invariant property would make bounding the difference between output states convenient. To elaborate, let the output of a neural ODE correspond to states at time $T > 0$ . By the time-invariant property, the difference between outputs, $\\| \\widetilde { \\mathbf z } _ { 1 } ( \\bar { T } ) - \\mathbf z _ { 1 } ( T ) \\|$ , equals to $\\| { \\bf z } _ { 1 } ( T + T ^ { \\prime } ) - { \\bf z } _ { 1 } ( T ) \\|$ . To control this distance, a steady-state regularization term is introduced to the overall objective to constrain the change of a state after time exceeds $T$ . With the time-invariant property and the steady-state term, we show that TisODE even is more robust. We do so by evaluating the robustness of TisODE-based classifiers against various types of perturbations and observe that such models are more robust than vanilla ODE-based models. ", + "bbox": [ + 174, + 429, + 825, + 623 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In addition, some other effective architectural solutions have also been recently proposed to improve the robustness of CNNs. For example, Xie et al. (2017) randomly resizes or pads zeros into test images to destroy the specific structure of adversarial perturbations. Besides, the model proposed by Xie et al. (2019) contains feature denoising filters to remove the feature-level patterns of adversarial examples. We conduct experiments to show that our proposed TisODE can work seamlessly and in conjunction with these methods to further boost the robustness of deep models. Thus, the proposed TisODE can be used as a generally applicable and effective component for improving the robustness of deep models. ", + "bbox": [ + 174, + 631, + 825, + 742 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In summary, our contributions are as follows. Firstly, we are the first to provide a systematic empirical study on the robustness of neural ODEs and find that the neural ODE-based models are more robust compared to conventional CNN models. This finding inspires new applications of neural ODEs in improving robustness of deep models, a problem that concerns many deep learning theorists and practitioners alike. Secondly, we propose the TisODE method, which is simple yet effective in significantly boosting the robustness of neural ODEs. Moreover, the proposed TisODE can also be used in conjunction with other state-of-the-art robust architectures. Thus, TisODE can serve as a drop-in module to improve the robustness of deep models effectively. ", + "bbox": [ + 173, + 748, + 825, + 861 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 PRELIMINARIES ON NEURAL ODE ", + "text_level": 1, + "bbox": [ + 174, + 102, + 491, + 118 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "It has been shown that a residual block (He et al., 2016) can be interpreted as the discrete approximation of an ODE by setting the discretization step to be one. When the discretization step approaches zero, it yields a family of neural networks, which are called neural ODEs (Chen et al., 2018). Formally, in a neural ODE, the relation between input and output is characterized by the following set of equations: ", + "bbox": [ + 173, + 132, + 825, + 202 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/97910f4443c6f9c371655e158097ca883dd18d4e639302bc0560ca17790de48a.jpg", + "text": "$$\n\\frac { \\mathrm { d } { \\mathbf z } ( t ) } { \\mathrm { d } t } = f _ { \\boldsymbol \\theta } ( { \\mathbf z } ( t ) , t ) , \\quad { \\mathbf z } ( 0 ) = { \\mathbf z } _ { \\mathrm { i n } } , \\quad { \\mathbf z } _ { \\mathrm { o u t } } = { \\mathbf z } ( T ) ,\n$$", + "text_format": "latex", + "bbox": [ + 328, + 200, + 668, + 232 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $f _ { \\theta } : \\mathbb { R } ^ { d } \\times [ 0 , \\infty ) \\mathbb { R } ^ { d }$ denotes the trainable layers that are parameterized by weights $\\theta$ and $\\mathbf { z } : [ 0 , \\infty ) \\mathbb { R } ^ { d }$ represents the $d$ -dimensional state of the neural ODE. We assume that $f _ { \\theta }$ is continuous in $t$ and globally Lipschitz continuous in $\\mathbf { z }$ . In this case, the input $\\mathbf { z } _ { \\mathrm { i n } }$ of the neural ODE corresponds to the state at $t = 0$ , and the output $\\mathbf { z } _ { \\mathrm { o u t } }$ is associated to the state at some $T \\in ( 0 , \\infty )$ . Because $f _ { \\theta }$ governs how the state changes with respect to time $t$ , we also use $f _ { \\theta }$ to denote the dynamics of the neural ODE. ", + "bbox": [ + 174, + 236, + 825, + 320 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Given input $\\mathbf { z } _ { \\mathrm { i n } }$ , the output $\\mathbf { z } _ { \\mathrm { o u t } }$ can be computed by solving the ODE in (1). If $T$ is fixed, the output $\\mathbf { z } _ { \\mathrm { o u t } }$ only depends on the input $\\mathbf { z } _ { \\mathrm { i n } }$ and the dynamics $f _ { \\theta }$ , which also corresponds to the weighted layers in the neural ODE. Therefore, the neural ODE can be represented as the $d$ -dimensional function $\\phi _ { T } ( \\cdot , \\cdot )$ of the input $\\mathbf { z } _ { \\mathrm { i n } }$ and the dynamics $f _ { \\theta }$ , i.e., ", + "bbox": [ + 174, + 327, + 825, + 383 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/ee04e9b71d2db1ef8c0ab434f849a920d8934ab30d56fbcaea2c2fdc1e2122ec.jpg", + "text": "$$\n\\mathbf { z } _ { \\mathrm { o u t } } = \\mathbf { z } ( T ) = \\mathbf { z } ( 0 ) + \\int _ { 0 } ^ { T } f _ { \\theta } ( \\mathbf { z } ( t ) , t ) \\mathrm { d } t = \\phi _ { T } ( \\mathbf { z } _ { \\mathrm { i n } } , f _ { \\theta } ) .\n$$", + "text_format": "latex", + "bbox": [ + 312, + 388, + 686, + 425 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The terminal time $T$ of the output state ${ \\mathbf z } ( T )$ is set to be 1 in practice. Several methods have been proposed for training neural ODEs, such as the adjoint sensitivity method (Chen et al., 2018), SNet (Quaglino et al., 2019), and the auto-differentiation technique (Paszke et al., 2017). In this work, we use the most straightforward technique, i.e., updating the weights $\\theta$ with the autodifferentiation technique in the PyTorch framework. ", + "bbox": [ + 174, + 431, + 825, + 502 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 AN EMPIRICAL STUDY ON THE ROBUSTNESS OF ODENETS ", + "text_level": 1, + "bbox": [ + 174, + 522, + 692, + 537 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Robustness of deep models has gained increased attention, as it is imperative that deep models employed in critical applications, such as healthcare, are robust. The robustness of a model is measured by the sensitivity of the prediction with respect to small perturbations on the inputs. In this study, we consider three commonly-used perturbation schemes, namely random Gaussian perturbations, FGSM (Goodfellow et al., 2014) adversarial examples, and PGD (Madry et al., 2017) adversarial examples. These perturbation schemes reflect noise and adversarial robustness properties of the investigated models respectively. We evaluate the robustness via the classification accuracies on perturbed images, in which the original non-perturbed versions of these images are all correctly classified. ", + "bbox": [ + 173, + 553, + 825, + 679 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "For a fair comparison with conventional CNN models, we made sure that the number of parameters of an ODENet is close to that of its counterpart CNN model. Specifically, the ODENet shares the same network architecture with the CNN model for the FE and FCC parts. The only difference is that, for the RM part, the input of the ODE-based RM is concatenated with one more channel which represents the time $t$ , while the RM in a CNN model has a skip connection and serves as a residual block. During the training phase, all the hyperparameters are kept the same, including training epochs, learning rate schedules, and weight decay coefficients. Each model is trained three times with different random seeds, and we report the average performance (classification accuracy) together with the standard deviation. ", + "bbox": [ + 174, + 685, + 825, + 810 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1 EXPERIMENTAL SETTINGS ", + "text_level": 1, + "bbox": [ + 176, + 828, + 397, + 842 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Dataset: We conduct experiments to compare the robustness of ODENets with CNN models on three datasets, i.e., the MNIST (LeCun et al., 1998), the SVHN (Netzer et al., 2011), and a subset of the ImageNet datset (Deng et al., 2009). We call the subset ImgNet10 since it is collected from 10 synsets of ImageNet: dog, bird, car, fish, monkey, turtle, lizard, bridge, cow, and crab. We selected 3,000 training images and 300 test images from each synset and resized all images to $1 2 8 \\times 1 2 8$ . ", + "bbox": [ + 174, + 854, + 825, + 924 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Architectures: On the MNIST dataset, both the ODENet and the CNN model consists of four convolutional layers and one fully-connected layer. The total number of parameters of the two models is around $1 4 0 \\mathrm { k }$ . On the SVHN dataset, the networks are similar to those for the MNIST; we only changed the input channels of the first convolutional layer to three. On the ImgNet10 dataset, there are nine convolutional layers and one fully-connected layer for both the ODENet and the CNN model. The numbers of parameters is approximately $2 8 0 \\mathrm { k }$ . In practice, the neural ODE can be solved with different numerical solvers such as the Euler method and the Runge-Kutta methods (Chen et al., 2018). Here, we use the easily-implemented Euler method in the experiments. To balance the computation and the continuity of the flow, we solve the ODE initial value problem in equation (1) by the Euler method with step size 0.1. Our implementation builds on the open-source neural ODE codes. 1 Details on the network architectures are included in the Appendix. ", + "bbox": [ + 174, + 103, + 825, + 256 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Training: The experiments are conducted using two settings on each dataset—training models only with original non-perturbed images and training models on original images together with their perturbed versions. In both settings, we added a weight decay term into the training objective to regularize the norm of the weights, since this can help control the model’s representation capacity and improve the robustness of a neural network (Sokolic´ et al., 2017). In the second setting, images perturbed with random Gaussian noise are used to fine-tune the models, because augmenting the dataset with small perturbations can possibly improve the robustness of models and synthesizing Gaussian noise does not incur excessive computation time. ", + "bbox": [ + 174, + 263, + 825, + 375 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2 ROBUSTNESS OF ODENETS TRAINED ONLY ON NON-PERTURBED IMAGES ", + "text_level": 1, + "bbox": [ + 178, + 393, + 722, + 407 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The first question we are interested in is how robust ODENets are against perturbations if the model is only trained on original non-perturbed images. We train CNNs and ODEnets to perform classification on three datasets and set the weight decay parameters for all models to be 0.0005. We make sure that both the well-trained ODENets and CNN models have satisfactory performances on original non-perturbed images, i.e., around $9 9 . 5 \\%$ for MNIST, $9 5 . 0 \\%$ for the SVHN, and $8 0 . 0 \\%$ for ImgNet10. ", + "bbox": [ + 174, + 420, + 825, + 503 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Since Gaussian noise is ubiquitous in modeling image degradation, we first evaluated the robustness of the models in the presence of zero-mean random Gaussian perturbations. It has also been shown that a deep model is vulnerable to harmful adversarial examples, such as the FGSM (Goodfellow et al., 2014). We are also interested in how robust ODENets are in the presence of adversarial examples. The standard deviation $\\sigma$ of Gaussian noise and the $l _ { \\infty }$ -norm $\\epsilon$ of the FGSM attack for each dataset are shown in Table 1. ", + "bbox": [ + 174, + 510, + 823, + 593 + ], + "page_idx": 3 + }, + { + "type": "table", + "img_path": "images/5c76587be27929f6a5c670ff5d42c94d3fd7b15e85b1409c548bdb5f3efde2a1.jpg", + "table_caption": [ + "Table 1: Robustness comparison of different models. We report their mean classification accuracies $( \\% )$ and standard deviations (mean $\\pm$ std) on perturbed images from the MNIST, the SVHN, and the ImgNet10 datasets. Two types of perturbations are used—zero-mean Gaussian noise and FGSM adversarial attack. The results show that ODENets are much more robust in comparison to CNN models. " + ], + "table_footnote": [], + "table_body": "
Gaussian noiseAdversarial attack
MNISTσ=50σ=75σ=100FGSM-0.15FGSM-0.3FGSM-0.5
CNN98.1±0.785.8±4.356.4±5.663.4±2.324.0±8.98.3±3.2
ODENet98.7±0.690.6±5.473.2±8.683.5±0.942.1±2.414.3±2.1
SVHNg=15g=25σ=35FGSM-3/255FGSM-5/255FGSM-8/255
CNN90.0±1.276.3±2.760.9±3.929.2±2.913.7±1.95.4±1.5
ODENet95.7±0.788.1±1.578.2±2.158.2±2.343.0±1.330.9±1.4
ImgNet10σ=10σ=15σ = 25FGSM-5/255FGSM-8/255FGSM-16/255
CNN80.1±1.863.3±2.040.8±2.728.5±0.518.1±0.79.4±1.2
ODENet81.9±2.067.5±2.048.7±2.636.2±1.027.2±1.114.4±1.7
", + "bbox": [ + 191, + 689, + 802, + 851 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "From the results in Table 1, we observe that the ODENets demonstrate superior robustness compared to CNNs for all types of perturbations. On the MNIST dataset, in the presence of Gaussian perturbations with a large $\\sigma$ of 100, the ODENet produces much higher accuracy on perturbed images compared to the CNN model ( $7 3 . 2 \\%$ vs. $5 6 . 4 \\%$ ). For the FGSM-0.3 adversarial examples, the accuracy of ONEnet is around twice as high as that of the CNN model. On the SVHN dataset, ODENets significantly outperform CNN models, e.g., for the FGSM-5/255 examples, the accuracy of the ODENet is $4 3 . 0 \\%$ , which is much higher than that of the CNN model $( 1 3 . 7 \\% )$ . On the ImgNet10, for both cases of $\\sigma = 2 5$ and FGSM-8/255, ODENet outperforms CNNs by a large margin of around $9 \\%$ . ", + "bbox": [ + 174, + 868, + 823, + 897 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 200 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.3 ROBUSTNESS OF ODENETS TRAINED ON ORIGINAL IMAGES TOGETHER WITH GAUSSIAN PERTURBATIONS ", + "text_level": 1, + "bbox": [ + 176, + 219, + 750, + 247 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Training a model on original images together with their perturbed versions can improve the robustness of the model. As mentioned previously, Gaussian noise is commonly assumed to be present in real-world images. Synthesizing Gaussian noise is also fast and easy. Thus, we add random Gaussian noise into the original images to generate their perturbed versions. ODENets and CNN models are both trained on original images together with their perturbed versions. The standard deviation of the added Gaussian noise is randomly chosen from $\\bar { \\{ 5 0 , 7 5 , 1 0 0 \\} }$ on the MNIST dataset, $\\{ 1 5 , 2 5 , 3 5 \\}$ on the SVHN dataset, and $\\{ 1 0 , 1 5 , 2 5 \\}$ on the ImgNet10. All other hyperparameters are kept the same as above. ", + "bbox": [ + 174, + 260, + 825, + 371 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/c863f23ea70d5ead2741c14baf42496133f48535a60953861fb5e41251daade3.jpg", + "table_caption": [ + "Table 2: Robustness comparison of different models. We report their mean classification accuracies $( \\% )$ and standard deviations (mean $\\pm$ std) on perturbed images from the MNIST, the SVHN, and the ImgNet10 datsets. Three types of perturbations are used—zero-mean Gaussian noise, FGSM adversarial attack, and PGD adversarial attack. The results show that ODENets are more robust compared to CNN models. " + ], + "table_footnote": [], + "table_body": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
ImgNet10σ= 25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
", + "bbox": [ + 222, + 468, + 772, + 626 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The robustness of the models is evaluated under Gaussian perturbations, FGSM adversarial examples, and PGD (Madry et al., 2017) adversarial examples. The latter is a stronger attacker compared to the FGSM. The $l _ { \\infty }$ -norm $\\epsilon$ of the PGD attack for each dataset is shown in Table 2. Based on the results, we observe that ODENets consistently outperform CNN models on both two datasets. On the MNIST dataset, the ODENet outperforms the CNN against all types of perturbations. In particular, for the PGD-0.2 adversarial examples, the accuracy of the ODENet $( 6 4 . 7 \\% )$ is much higher than that of the CNN $( 3 2 . 9 \\% )$ . Besides, for the PGD-0.3 attack, the CNN is completely misled by the adversarial examples, but the ODENet can still classify perturbed images with an accuracy of $1 3 . 0 \\%$ . On the SVHN dataset, ODENets also show superior robustness in comparison to CNN models. For all the adversarial examples, ODENets outperform CNN models by a margin of at least 10 percentage points. On the ImgNet10 dataset, the ODENet also performs better than CNN models against all forms of adversarial examples. ", + "bbox": [ + 173, + 641, + 825, + 808 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.4 INSIGHTS ON THE ROBUSTNESS OF ODENETS", + "text_level": 1, + "bbox": [ + 176, + 827, + 532, + 842 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "From the results in Sections 3.2 and 3.3, we find ODENets are more robust compared to CNN models. Here, we attempt to provide an intuitive understanding of the robustness of the neural ODE. In an ODENet, given some datum, the FE extracts an informative feature map from the datum. The neural ODE, serving as the RM, takes as input the feature map and performs a nonlinear mapping. In practice, we use the weight decay technique during training which regularizes the norm of weights in the FE part, so that the change of feature map in terms of a small perturbation on the input can be controlled. We aim to show that, in the neural ODE, a small change on the feature map will not lead to a large deviation from the original output associated with the feature map. ", + "bbox": [ + 174, + 853, + 823, + 924 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 103, + 823, + 146 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Theorem 1 (ODE integral curves do not intersect (Coddington & Levinson, 1955; Younes, 2010; Dupont et al., 2019)). Let ${ \\bf z } _ { 1 } ( t )$ and ${ \\bf z } _ { 2 } ( t )$ be two solutions of the ODE in (1) with different initial conditions, i.e. ${ \\bf z } _ { 1 } ( 0 ) \\neq { \\bf z } _ { 2 } ( 0 )$ . In (1), $f _ { \\theta }$ is continuous in t and globally Lipschitz continuous in z. Then, it holds that ${ \\bf z } _ { 1 } ( t ) \\neq { \\bf z } _ { 2 } ( t )$ for all $t \\in [ 0 , \\infty )$ . ", + "bbox": [ + 174, + 152, + 825, + 210 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "To illustrate this theorem, considering a simple 1- dimensional system in which the state is a scalar. As shown in Figure 2, equation (1) has a solution $z _ { 1 } ( t )$ starting from $\\bar { A _ { 1 } } = ( 0 , \\bar { z _ { 1 } } ( 0 ) )$ , where $z _ { 1 } ( 0 )$ is the feature of some datum. Equation (1) also has another two solutions $z _ { 2 } ( t )$ and $z _ { 3 } ( t )$ , whose starting points $A _ { 2 } = ( 0 , z _ { 2 } ( 0 ) )$ and $A _ { 3 } ~ = ~ ( 0 , z _ { 3 } ( 0 ) )$ , both of which are close to $A _ { 1 }$ . Suppose $A _ { 1 }$ is between $A _ { 2 }$ and $A _ { 3 }$ . By Theorem 1, we know that the integral curve $z _ { 1 } ( t )$ is always sandwiched between the integral curves $z _ { 2 } ( t )$ and $z _ { 3 } ( t )$ . ", + "bbox": [ + 174, + 223, + 549, + 363 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Now, let $\\epsilon < \\operatorname* { m i n } \\{ | z _ { 2 } ( 0 ) - z _ { 1 } ( 0 ) | , | z _ { 3 } ( 0 ) - z _ { 1 } ( 0 ) | \\} .$ . Consider a solution $\\widetilde { z } _ { 1 } ( t )$ of equation (1). The integral curve $\\widetilde { z } _ { 1 } ( t )$ starts from a point $\\widetilde { A } _ { 1 } = ( 0 , \\widetilde { z } _ { 1 } ( 0 ) )$ . The point $\\widetilde { A } _ { 1 }$ is !in the $\\epsilon$ -neighborhood of $A _ { 1 }$ with $| \\widetilde z _ { 1 } ( 0 ) - z _ { 1 } ( 0 ) | < \\epsilon .$ . By Theorem 1, we know that $| \\tilde { z } _ { 1 } ( T ) - z _ { 1 } ( T ) | \\leq | z _ { 3 } ( T ) -$ $z _ { 2 } ( T ) |$ !. In other words, if any perturbation smaller than $\\epsilon$ is added to the scalar $z _ { 1 } ( 0 )$ in $A _ { 1 }$ , the deviation from the original output $z _ { 1 } ( T )$ is bounded by the distance between $z _ { 2 } ( T )$ and $z _ { 3 } ( T )$ . In contrast, in a CNN model, there is no such bound on the deviation from the original output. Thus, we opine that due to this non-intersecting property, ODENets are intrinsically robust. ", + "bbox": [ + 174, + 369, + 549, + 443 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/901235eac73c549d6bdb2646bd2f84cf3a82d0581b3c2860c0361d6a9e2870e9.jpg", + "image_caption": [ + "Figure 2: No integral curves intersect. The integral curve starting from $\\widetilde { A _ { 1 } }$ is always sandwiched between two integral curves starting from $A _ { 1 }$ and $A _ { 3 }$ . " + ], + "image_footnote": [], + "bbox": [ + 568, + 219, + 799, + 352 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 443, + 825, + 498 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4 TISODE: BOOSTING THE ROBUSTNESS OF NEURAL ODES ", + "text_level": 1, + "bbox": [ + 174, + 525, + 686, + 540 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In the previous section, we presented an empirical study on the robustness of ODENets and observed that ODENets are more robust compared to CNN models. In this section, we explore how to boost the robustness of the vanilla neural ODE model further. This motivates the proposal of time-invariant steady neural ODEs (TisODEs). ", + "bbox": [ + 174, + 559, + 825, + 614 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.1 TIME-INVARIANT STEADY NEURAL ODES ", + "text_level": 1, + "bbox": [ + 173, + 638, + 504, + 652 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "From the discussion in Section 3.4, the key to improving the robustness of neural ODEs is to control the difference between neighboring integral curves. By Grownall’s inequality (Howard, 1998) (see Theorem 2 in the Appendix), we know that the difference between two terminal states is bounded by the difference between initial states multiplied by the exponential of the dynamics’ Lipschitz constant. However, it is very difficult to bound the Lipschitz constant of the dynamics directly. Alternatively, we propose to achieve the goal of controlling the output deviation by following two steps: (i) removing the time dependence of the dynamics and (ii) imposing a certain steady-state constraint. ", + "bbox": [ + 449, + 666, + 823, + 847 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/b7cae751cc80ff17df6fe622bceb23becc612933d297ca2ae9a8307363b3a95d.jpg", + "image_caption": [ + "Figure 3: An illustration of the timeinvariant property of ODEs. We can see that the curve $\\widetilde { \\mathbf { z } } _ { 1 } ( t )$ is exactly the horizontal translation of ${ \\bf z } _ { 1 } ( t )$ on the interval $[ T ^ { \\prime } , \\infty )$ . " + ], + "image_footnote": [], + "bbox": [ + 176, + 670, + 416, + 818 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In the neural ODE characterized by equation (1), the dynamics $f _ { \\theta } ( \\mathbf { z } ( t ) , t )$ depends on both the state ${ \\bf z } ( t )$ at time $t$ and the time $t$ itself. In contrast, if the neural ODE is modified to be time-invariant, the time dependence of the dynamics is removed. Consequently, the dynamics depends only on the state z. So, we can rewrite the dynamics function as $f _ { \\boldsymbol { \\theta } } ( \\mathbf { z } )$ , and the neural ODE is characterized as ", + "bbox": [ + 449, + 854, + 823, + 910 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 166, + 910, + 823, + 924 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 102, + 640, + 119 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/f46c4a1faca9cc61c86100722af41f1f7581c94ec9c4520e6d2130757a216c58.jpg", + "text": "$$\n\\left\\{ \\begin{array} { l l } { \\displaystyle \\frac { \\mathrm { d } \\mathbf { z } ( t ) } { \\mathrm { d } t } = f _ { \\boldsymbol { \\theta } } ( \\mathbf { z } ( t ) ) ; } \\\\ { \\mathbf { z } ( 0 ) = \\mathbf { z } _ { \\mathrm { i n } } ; } \\\\ { \\mathbf { z } _ { \\mathrm { o u t } } = \\mathbf { z } ( T ) . } \\end{array} \\right.\n$$", + "text_format": "latex", + "bbox": [ + 419, + 127, + 560, + 193 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Let ${ \\bf z } _ { 1 } ( t )$ be a solution of (2) on $[ 0 , \\infty )$ and $\\epsilon > 0$ be a small positive value. We define the set $\\mathbb { M } _ { 1 } = \\{ ( \\mathbf { z } _ { 1 } ( t ) , t ) | t \\in [ 0 , T ]$ , \" $\\| \\mathbf { z } _ { 1 } ( t ) - \\mathbf { z } _ { 1 } ( 0 ) \\| \\leq \\epsilon \\}$ . This set contains all points on the curve of ${ \\bf z } _ { 1 } ( t )$ during $[ 0 , T ]$ that are also inside the $\\epsilon$ -neighborhood of ${ \\bf z } _ { 1 } ( 0 )$ . For some element $( { \\bf z } _ { 1 } ( T ^ { \\prime } ) , T ^ { \\prime } ) \\in \\mathbb { M } _ { 1 }$ , let $\\widetilde { \\mathbf { z } } _ { 1 } ( t )$ be the solution of (2) which starts from $\\widetilde { \\mathbf z } _ { 1 } ( 0 ) = \\mathbf z _ { 1 } ( T ^ { \\prime } )$ . Then we have ", + "bbox": [ + 173, + 208, + 825, + 265 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/99ba3d941298bb987e3bce08ab9b157bb13e597a10070baaba0ae0a88f784c91.jpg", + "text": "$$\n\\widetilde { { \\mathbf z } } _ { 1 } ( t ) = { \\mathbf z } _ { 1 } ( t + T ^ { \\prime } )\n$$", + "text_format": "latex", + "bbox": [ + 433, + 272, + 563, + 290 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "for all $t$ in $[ 0 , \\infty )$ . The property shown in equation (3) is known as the time-invariant property. It indicates that the integral curve $\\widetilde { \\mathbf { z } } _ { 1 } ( t )$ is the $- T ^ { \\prime }$ shift of ${ \\bf z } _ { 1 } ( t )$ (Figure 3). ", + "bbox": [ + 174, + 299, + 826, + 329 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We can regard $\\widetilde { \\mathbf { z } } _ { 1 } ( 0 )$ as a slightly perturbed version of ${ \\bf z } _ { 1 } ( 0 )$ , and we are interested in how large the difference between $\\widetilde { \\mathbf z } _ { 1 } ( T )$ and ${ \\bf z } _ { 1 } ( T )$ is. In a robust model, the difference should be small. By equation (3), we have $\\| \\mathbf { \\tilde { z } } _ { 1 } ( T ) - \\mathbf { z } _ { 1 } ( T ) \\| = \\| \\mathbf { z } _ { 1 } ( T + T ^ { \\prime } ) - \\mathbf { z } _ { 1 } ( T ) \\|$ . Since $T ^ { \\prime } \\in [ 0 , T ]$ , the difference between ${ \\bf z } _ { 1 } ( T )$ and $\\widetilde { \\mathbf z } _ { 1 } ( T )$ can be bounded as follows, ", + "bbox": [ + 174, + 334, + 825, + 391 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/dda5a5162f628120cc1d9f8b14d9625d581af070a04ac8f089aaa2d9e4fe900a.jpg", + "text": "$$\n\\| \\widetilde { \\mathbf z } _ { 1 } ( T ) - \\mathbf z _ { 1 } ( T ) \\| = \\left\\| \\int _ { T } ^ { T + T ^ { \\prime } } f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) \\mathrm { d } t \\right\\| \\leq \\left\\| \\int _ { T } ^ { T + T ^ { \\prime } } | f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) | \\mathrm { d } t \\right\\| \\leq \\left\\| \\int _ { T } ^ { 2 T } | f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) | \\mathrm { d } t \\right\\| ,\n$$", + "text_format": "latex", + "bbox": [ + 186, + 398, + 808, + 443 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where all norms are $\\ell _ { 2 }$ norms and $| f _ { \\theta } |$ denotes the element-wise absolute operation of a vectorvalued function $f _ { \\theta }$ . That is to say, the difference between $\\widetilde { \\mathbf z } _ { 1 } ( T )$ and ${ \\bf z } _ { 1 } ( T )$ can be bounded by only using the information of the curve ${ \\bf z } _ { 1 } ( t )$ . For any $t ^ { \\prime } \\in [ 0 , T ]$ and element $( { \\bf z } _ { 1 } ( t ^ { \\prime } ) , t ^ { \\prime } ) \\in \\mathbb { M } _ { 1 }$ , consider the integral curve that starts from ${ \\bf z } _ { 1 } ( t ^ { \\prime } )$ . The difference between the output state of this curve and ${ \\bf z } _ { 1 } ( T )$ satisfies inequality (4). ", + "bbox": [ + 174, + 452, + 825, + 523 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Therefore, we propose to add an additional term $L _ { \\mathrm { s s } }$ to the loss function when training the timeinvariant neural ODE: ", + "bbox": [ + 174, + 529, + 821, + 556 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/dead5c3a609cf3ee30da32ebeb2ca3d5179c956d041557ad30986bda72525deb.jpg", + "text": "$$\nL _ { \\mathrm { s s } } = \\sum _ { i = 1 } ^ { N } \\left\\| \\int _ { T } ^ { 2 T } | f _ { \\theta } ( \\mathbf { z } _ { i } ( t ) ) | \\mathrm { d } t \\right\\| ,\n$$", + "text_format": "latex", + "bbox": [ + 387, + 556, + 609, + 601 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where $N$ is the number of samples in the training set and ${ \\bf z } _ { i } ( t )$ is the solution whose initial state equals to the feature of the $i ^ { \\mathrm { t h } }$ sample. The regularization term $L _ { \\mathrm { s s } }$ is termed as the steady-state loss. This terminology “steady state” is borrowed from the dynamical systems literature. In a stable dynamical system, the states stabilize around a fixed point, known as the steady-state, as time tends to infinity. If we can ensure that $L _ { \\mathrm { s s } }$ is small, for each sample, the outputs of all the points in $\\mathbb { M } _ { i }$ will stabilize around ${ \\bf z } _ { i } ( T )$ . Consequently, the model is robust. This modification of the neural ODE is dubbed Time-invariant steady neural $O D E$ . ", + "bbox": [ + 173, + 604, + 825, + 703 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.2 EVALUATING ROBUSTNESS OF TISODE-BASED CLASSIFIERS ", + "text_level": 1, + "bbox": [ + 176, + 722, + 632, + 737 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Here, we conduct experiments to evaluate the robustness of our proposed TisODE, and compare TisODE-based models with the vanilla ODENets. We train all models with original non-perturbed images together with their Gaussian perturbed versions. The regularization parameter for the steadystate loss $L _ { \\mathrm { s s } }$ is set to be 0.1. All other hyperparameters are exactly the same as those in Section 3.3. ", + "bbox": [ + 174, + 750, + 825, + 805 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "From the results in Table 3, we can see that our proposed TisODE-based models are clearly more robust compared to vanilla ODENets. On the MNIST dataset, when combating FGSM-0.3 attacks, the TisODE-based models outperform vanilla ODENets by more than 4 percentage points. For the FGSM-0.5 adversarial examples, the accuracy of the TisODE-based model is 6 percentage points better. On the SVHN dataset, the TisODE-based models perform better in terms of all forms of adversarial examples. On the ImgNet10 dataset, the TisODE-based models also outperform vanilla ODE-based models on all types of perturbations. In the presence of FGSM and PGD-5/255 examples, the accuracies are enhanced by more than 2 percentage points. ", + "bbox": [ + 173, + 811, + 825, + 924 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/564d88a3d2f0ff8bcf0003170d92128d6d885f175ce78aa9082a0b2fbed99adb.jpg", + "table_caption": [ + "Table 3: Classification accuracy (mean $\\pm$ std in $\\%$ ) on perturbed images from MNIST, SVHN and ImgNet10. To evaluate the robustness of classifiers, we use three types of perturbations, namely zero-mean Gaussian noise with standard deviation $\\sigma$ , FGSM attack and PGD attack. From the results, the proposed TisODE effectively improve the robustness of the vanilla neural ODE. " + ], + "table_footnote": [ + "4.3 TISODE - A GENERALLY APPLICABLE DROP-IN TECHNIQUE FOR IMPROVING THE ROBUSTNESS OF DEEP NETWORKS " + ], + "table_body": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
TisODE99.6±0.075.7±1.426.5±3.867.4±1.513.2±1.0
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
TisODE94.9±0.151.6±1.238.2±1.952.0±0.928.2±0.3
ImgNet10g=25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
TisODE92.8±0.444.3±0.731.4±1.131.1±1.214.5±1.1
", + "bbox": [ + 222, + 167, + 772, + 363 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In view of the excellent robustness of the TisODE, we claim that the proposed TisODE can be used as a general drop-in module for improving the robustness of deep networks. We support this claim by showing the TisODE can work in conjunction with other state-of-the-art techniques and further boost the models’ robustness. These techniques include the feature denoising (FDn) method (Xie et al., 2019) and the input randomization (IR) method (Xie et al., 2017). We conduct experiments on the MNIST and SVHN datasets. All models are trained with original non-perturbed images together with their Gaussian perturbed versions. We show that models using the FDn/IRd technique becomes much more robust when equipped with the TisODE. In the FDn experiments, the dot-product nonlocal denoising layer (Xie et al., 2019) is added to the head of the fully-connected classifier. ", + "bbox": [ + 173, + 416, + 825, + 542 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/b0ade5e612e54414b3aa94f7b78718bad3bcceec3355deacbdb5d82dc1e7092d.jpg", + "table_caption": [ + "Table 4: Classification accuracy (mean $\\pm$ std in $\\%$ ) on perturbed images from MNIST and SVHN. We evaluate against three types of perturbations, namely zero-mean Gaussian noise with standard deviation $\\sigma$ , FGSM attack and PGD attack. From the results, upon the CNNs modified with FDn and IRd, using TisODE can further improve the robustness. " + ], + "table_footnote": [], + "table_body": "
Gaussian noiseAdversarial attack
MNISTσ= 100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
CNN-FDn TisODE-FDn99.0±0.174.0±4.132.6±5.358.9±4.08.2±2.6
CNN-IRd99.4±0.0 95.3±0.980.6±2.3 78.1±2.240.4±5.7 36.7±2.172.6±2.4 79.6±1.928.2±3.6 55.5±2.9
TisODE-IRd97.6±0.186.8±2.349.1±0.288.8±0.966.0±0.9
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN
90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
CNN-FDn92.4±0.143.8±1.431.5±3.040.0±2.619.6±3.4
TisODE-FDn95.2±0.157.8±1.748.2±2.053.4±2.932.3±1.0
CNN-IRd84.9±1.265.8±0.454.7±1.274.0±0.564.5±0.8
TisODE-IRd91.7±0.574.4±1.261.9±1.881.6±0.871.0±0.5
", + "bbox": [ + 212, + 623, + 781, + 811 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "From Table 4, we observe that both FDn and IRd can effectively improve the adversarial robustness of vanilla CNN models (CNN-FDn, CNN-IRd). Furthermore, combining our proposed TisODE with FDn or IRd (TisODE-FDn, TisODE-IRd), the adversarial robustness of the resultant model is significantly enhanced. For example, on the MNIST dataset, the additional use of our TisODE increases the accuracies on the PGD-0.3 examples by at least 10 percentage points for both FDn $8 . 2 \\%$ to $2 8 . 2 \\% )$ and IRd $5 5 . 5 \\%$ to $6 6 . 0 \\%$ ). However, on both MNIST and SVHN datasets, the IRd technique improves the robustness against adversarial examples, but its performance is worse on random Gaussian noise. With the help of the TisODE, the degradation in the robustness against random Gaussian noise can be effectively ameliorated. ", + "bbox": [ + 173, + 825, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 103, + 823, + 132 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "5 RELATED WORKS ", + "text_level": 1, + "bbox": [ + 176, + 176, + 349, + 193 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In this section, we briefly review related works on the neural ODE and works concerning improving the robustness of deep neural networks. ", + "bbox": [ + 174, + 222, + 821, + 251 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Neural ODE: The neural ODE (Chen et al., 2018) method models the input and output as two states of a continuous-time dynamical system by approximating the dynamics of this system with trainable layers. Before the proposal of neural ODE, the idea of modeling nonlinear mappings using continuous-time dynamical systems was proposed in Weinan (2017). Lu et al. (2017) also showed that several popular network architectures could be interpreted as the discretization of a continuoustime ODE. For example, the ResNet (He et al., 2016) and PolyNet (Zhang et al., 2017) are associated with the Euler scheme and the FractalNet (Larsson et al., 2016) is related to the Runge-Kutta scheme. In contrast to these discretization models, neural ODEs are endowed with an intrinsic invertibility property, which yields a family of invertible models for solving inverse problems (Ardizzone et al., 2018), such as the FFJORD (Grathwohl et al., 2018). ", + "bbox": [ + 174, + 257, + 825, + 396 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Recently, many researchers have conducted studies on neural ODEs from the perspectives of optimization techniques, approximation capabilities, and generalization. Concerning the optimization of neural ODEs, the auto-differentiation techniques can effectively train ODENets, but the training procedure is computationally and memory inefficient. To address this problem, Chen et al. (2018) proposed to compute gradients using the adjoint sensitivity method (Pontryagin, 2018), in which there is no need to store any intermediate quantities of the forward pass. Also in Quaglino et al. (2019), the authors proposed the SNet which accelerates the neural ODEs by expressing their dynamics as truncated series of Legendre polynomials. Concerning the approximation capability, Dupont et al. (2019) pointed out the limitations in approximation capabilities of neural ODEs because of the preserving of input topology. The authors proposed an augmented neural ODE which increases the dimension of states by concatenating zeros so that complex mappings can be learned with simple flow. The most relevant work to ours concerns strategies to improve the generalization of neural ODEs. In Liu et al. (2019), the authors proposed the neural stochastic differential equation (SDE) by injecting random noise to the dynamics function and showed that the generalization and robustness of vanilla neural ODEs could be improved. However, our improvement on the neural ODEs is explored from a different perspective by introducing constraints on the flow. We empirically found that our proposal and the neural SDE can work in tandem to further boost the robustness of neural ODEs. ", + "bbox": [ + 174, + 405, + 825, + 652 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Robust Improvement: A straightforward way of improving the robustness of a model is to smooth the loss surface by controlling the spectral norm of the Jacobian matrix of the loss function (Sokolic´ et al., 2017). In terms of adversarial examples (Carlini & Wagner, 2017; Chen et al., 2017), researchers have proposed adversarial training strategies (Madry et al., 2017; Elsayed et al., 2018; Trame\\`r et al., 2017) in which the model is fine-tuned with adversarial examples generated in realtime. However, generating adversarial examples is not computationally efficient, and there exists a trade-off between the adversarial robustness and the performance on original non-perturbed images (Yan et al., 2018; Tsipras et al., 2018). In Wang et al. (2018a), the authors model the ResNet as a transport equation, in which the adversarial vulnerability can be interpreted as the irregularity of the decision boundary. Consequently, a diffusion term is introduced to enhance the robustness of the neural nets. Besides, there are also some works that propose novel architectural defense mechanisms against adversarial examples. For example, Xie et al. (2017) utilized random resizing and random padding to destroy the specific structure of adversarial perturbations; Wang et al. (2018b) and Wang et al. (2018c) improved the robustness of neural networks by replacing the output layers with novel interpolating functions; In Xie et al. (2019), the authors designed a feature denoising filter that can remove the perturbation’s pattern from feature maps. In this work, we explore the intrinsic robustness of a specific novel architecture (neural ODE), and show that the proposed TisODE can improve the robustness of deep networks and can also work in tandem with these state-of-the-art methods Xie et al. (2017; 2019) to achieve further improvements. ", + "bbox": [ + 174, + 659, + 825, + 924 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "6 CONCLUSION ", + "text_level": 1, + "bbox": [ + 174, + 102, + 318, + 117 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this paper, we first empirically study the robustness of neural ODEs. Our studies reveal that neural ODE-based models are superior in terms of robustness compared to CNN models. We then explore how to further boost the robustness of vanilla neural ODEs and propose the TisODE. Finally, we show that the proposed TisODE outperforms the vanilla neural ODE and also can work in conjunction with other state-of-the-art techniques to further improve the robustness of deep networks. Thus, the TisODE method is an effective drop-in module for building robust deep models. ", + "bbox": [ + 174, + 133, + 825, + 217 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "ACKNOWLEDGEMENT ", + "text_level": 1, + "bbox": [ + 176, + 238, + 357, + 253 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "This work is funded by a Singapore National Research Foundation (NRF) Fellowship (R-263-000- D02-281). ", + "bbox": [ + 169, + 268, + 821, + 296 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Jiashi Feng was partially supported by NUS IDS R-263-000-C67-646, ECRA R-263-000-C87-133, MOE Tier-II R-263-000-D17-112 and AI.SG R-263-000-D97-490 ", + "bbox": [ + 169, + 304, + 823, + 332 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 176, + 102, + 287, + 117 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Lynton Ardizzone, Jakob Kruse, Sebastian Wirkert, Daniel Rahner, Eric W Pellegrini, Ralf S Klessen, Lena Maier-Hein, Carsten Rother, and Ullrich Ko¨the. Analyzing inverse problems with invertible neural networks. arXiv preprint arXiv:1808.04730, 2018. ", + "bbox": [ + 176, + 126, + 823, + 169 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Nicholas Carlini and David Wagner. Towards evaluating the robustness of neural networks. In 2017 IEEE Symposium on Security and Privacy (SP), pp. 39–57. IEEE, 2017. ", + "bbox": [ + 171, + 179, + 823, + 208 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Pin-Yu Chen, Huan Zhang, Yash Sharma, Jinfeng Yi, and Cho-Jui Hsieh. Zoo: Zeroth order optimization based black-box attacks to deep neural networks without training substitute models. In Proceedings of the 10th ACM Workshop on Artificial Intelligence and Security, pp. 15–26. ACM, 2017. ", + "bbox": [ + 173, + 217, + 825, + 273 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differential equations. In Advances in neural information processing systems, pp. 6571–6583, 2018. ", + "bbox": [ + 171, + 284, + 820, + 313 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Earl A Coddington and Norman Levinson. Theory of ordinary differential equations. Tata McGrawHill Education, 1955. ", + "bbox": [ + 173, + 321, + 823, + 352 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee, 2009. ", + "bbox": [ + 173, + 359, + 825, + 404 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Emilien Dupont, Arnaud Doucet, and Yee Whye Teh. Augmented neural odes. arXiv preprint arXiv:1904.01681, 2019. ", + "bbox": [ + 169, + 412, + 825, + 443 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Gamaleldin F Elsayed, Shreya Shankar, Brian Cheung, Nicolas Papernot, Alex Kurakin, Ian Goodfellow, and Jascha Sohl-Dickstein. Adversarial examples that fool both human and computer vision. arXiv preprint arXiv:1802.08195, 10, 2018. ", + "bbox": [ + 173, + 452, + 823, + 496 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial examples. arXiv preprint arXiv:1412.6572, 2014. ", + "bbox": [ + 169, + 505, + 823, + 534 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Will Grathwohl, Ricky TQ Chen, Jesse Betterncourt, Ilya Sutskever, and David Duvenaud. Ffjord: Free-form continuous dynamics for scalable reversible generative models. arXiv preprint arXiv:1810.01367, 2018. ", + "bbox": [ + 174, + 542, + 825, + 587 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016. ", + "bbox": [ + 174, + 595, + 825, + 638 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ralph Howard. The gronwall inequality. lecture notes, 1998. ", + "bbox": [ + 173, + 648, + 575, + 664 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Gustav Larsson, Michael Maire, and Gregory Shakhnarovich. Fractalnet: Ultra-deep neural networks without residuals. arXiv preprint arXiv:1605.07648, 2016. ", + "bbox": [ + 174, + 672, + 823, + 703 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yann LeCun, Le´on Bottou, Yoshua Bengio, Patrick Haffner, et al. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. ", + "bbox": [ + 174, + 712, + 823, + 742 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Xuanqing Liu, Si Si, Qin Cao, Sanjiv Kumar, and Cho-Jui Hsieh. Neural sde: Stabilizing neural ode networks with stochastic noise. arXiv preprint arXiv:1906.02355, 2019. ", + "bbox": [ + 173, + 751, + 825, + 781 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yiping Lu, Aoxiao Zhong, Quanzheng Li, and Bin Dong. Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations. arXiv preprint arXiv:1710.10121, 2017. ", + "bbox": [ + 174, + 790, + 825, + 833 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. Towards deep learning models resistant to adversarial attacks. arXiv preprint arXiv:1706.06083, 2017. ", + "bbox": [ + 174, + 842, + 823, + 885 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, and Andrew Y Ng. Reading digits in natural images with unsupervised feature learning. 2011. ", + "bbox": [ + 176, + 895, + 821, + 924 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017. ", + "bbox": [ + 174, + 103, + 825, + 146 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Lev Semenovich Pontryagin. Mathematical theory of optimal processes. Routledge, 2018. ", + "bbox": [ + 169, + 155, + 767, + 171 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Alessio Quaglino, Marco Gallieri, Jonathan Masci, and Jan Koutn´ık. Accelerating neural odes with spectral elements. arXiv preprint arXiv:1906.07038, 2019. ", + "bbox": [ + 173, + 179, + 821, + 208 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jure Sokolic´, Raja Giryes, Guillermo Sapiro, and Miguel RD Rodrigues. Robust large margin deep neural networks. IEEE Transactions on Signal Processing, 65(16):4265–4280, 2017. ", + "bbox": [ + 173, + 215, + 823, + 246 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow, and Rob Fergus. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013. ", + "bbox": [ + 173, + 253, + 823, + 284 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Florian Trame\\`r, Alexey Kurakin, Nicolas Papernot, Ian Goodfellow, Dan Boneh, and Patrick McDaniel. Ensemble adversarial training: Attacks and defenses. arXiv preprint arXiv:1705.07204, 2017. ", + "bbox": [ + 176, + 291, + 823, + 334 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Dimitris Tsipras, Shibani Santurkar, Logan Engstrom, Alexander Turner, and Aleksander Madry. Robustness may be at odds with accuracy. arXiv preprint arXiv:1805.12152, 2018. ", + "bbox": [ + 171, + 343, + 823, + 372 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "B. Wang, B. Yuan, Z. Shi, and S. Osher. ResNets Ensemble via the Feynman-Kac Formalism to Improve Natural and Robust Accuracies. arXiv e-prints, art. arXiv:1811.10745, Nov 2018a. ", + "bbox": [ + 173, + 381, + 821, + 410 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Bao Wang, Alex T Lin, Zuoqiang Shi, Wei Zhu, Penghang Yin, Andrea L Bertozzi, and Stanley J Osher. Adversarial defense via data dependent activation function and total variation minimization. arXiv preprint arXiv:1809.08516, 2018b. ", + "bbox": [ + 174, + 419, + 823, + 462 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Bao Wang, Xiyang Luo, Zhen Li, Wei Zhu, Zuoqiang Shi, and Stanley Osher. Deep neural nets with interpolating function as output activation. In Advances in Neural Information Processing Systems, pp. 743–753, 2018c. ", + "bbox": [ + 173, + 469, + 823, + 513 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "E Weinan. A proposal on machine learning via dynamical systems. Communications in Mathematics and Statistics, 5(1):1–11, 2017. ", + "bbox": [ + 176, + 522, + 823, + 551 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Cihang Xie, Jianyu Wang, Zhishuai Zhang, Zhou Ren, and Alan Yuille. Mitigating adversarial effects through randomization. arXiv preprint arXiv:1711.01991, 2017. ", + "bbox": [ + 174, + 559, + 825, + 589 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Cihang Xie, Yuxin Wu, Laurens van der Maaten, Alan L Yuille, and Kaiming He. Feature denoising for improving adversarial robustness. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 501–509, 2019. ", + "bbox": [ + 173, + 597, + 823, + 641 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Ziang Yan, Yiwen Guo, and Changshui Zhang. Deep defense: Training dnns with improved adversarial robustness. In Advances in Neural Information Processing Systems, pp. 419–428, 2018. ", + "bbox": [ + 173, + 648, + 823, + 679 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Laurent Younes. Shapes and diffeomorphisms, volume 171. Springer, 2010. ", + "bbox": [ + 173, + 686, + 673, + 703 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Xingcheng Zhang, Zhizhong Li, Chen Change Loy, and Dahua Lin. Polynet: A pursuit of structural diversity in very deep networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 718–726, 2017. ", + "bbox": [ + 174, + 712, + 826, + 753 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "7 APPENDIX ", + "text_level": 1, + "bbox": [ + 174, + 102, + 294, + 117 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "7.1 NETWORKS USED ON THE MNIST, THE SVHN, AND THE IMGNET10 DATASETS ", + "bbox": [ + 168, + 136, + 766, + 151 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/d883b2f0037322e3b140127861d9b4cbcecf0aad825dcfc8ae8472d4037da268.jpg", + "table_caption": [ + "Table 5: The architectures of the ODENets on different datasets. " + ], + "table_footnote": [], + "table_body": "
MNISTRepetitionLayer
FE×1×1Conv(1,64,3,1) +GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "bbox": [ + 287, + 200, + 709, + 282 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/faf3961048327cb024192b9d8e2992e6f976673ff550b4ad755809e130397b17.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
SVHNRepetitionLayer
FE×1×1Conv(3,64,3,1) + GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "bbox": [ + 285, + 295, + 707, + 376 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/716551303bb6992cbdca9e029eee5f1d7cf0b6fc48418513eee4e672c4c78e95.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
ImgNet10RepetitionLayer
FE×1Conv(3,32,5,2)+GroupNorm
×1MaxPooling(2)
×1BaiscBlock(32, 64,2)
×1MaxPooling(2)
RM×3BaiscBlock(64,64,1)
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "bbox": [ + 285, + 388, + 709, + 496 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "In Table 5, the four arguments of the Conv layer represent the input channel, output channel, kernel size, and the stride. The two arguments of the Linear layer represents the input dimension and the output dimension of this fully-connected layer. In the network on the ImgNet10, the BasicBlock refers to the standard architecture in (He et al., 2016), the three arguments of the BasicBlock represent the input channel, output channel and the stride of the Conv layers inside the block. Note that we replace the BatchNorm layers in BasicBlocks as the GroupNorm to guarantee that the dynamics of each datum is independent of other data in the same mini-batch. ", + "bbox": [ + 173, + 515, + 825, + 613 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "7.2 THE CONSTRUCTION OF IMGNET10 DATASET ", + "text_level": 1, + "bbox": [ + 176, + 633, + 531, + 648 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/1a44f870fa6c81d744953453aa8fe7c73defdee5af67ea0f098fac0f2d07d1c9.jpg", + "table_caption": [ + "Table 6: The corresponding indexes to each class in the original ImageNet dataset " + ], + "table_footnote": [], + "table_body": "
ClassIndexing
dogn02090721, n02091032, n02088094
birdn01532829, n01558993, n01534433
carn02814533, n03930630, n03100240
fishn01484850,n01491361, n01494475
monkeyn02483708,n02484975, n02486261
turtlen01664065,n01665541, n01667114
lizardn01677366,n01682714,n01685808
bridgen03933933,n04366367,n04311004
cown02403003,n02408429,n02410509
crabn01980166,n01978455,n01981276
", + "bbox": [ + 333, + 695, + 660, + 842 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "7.3 GRONWALL’S INEQUALITY ", + "text_level": 1, + "bbox": [ + 174, + 878, + 401, + 893 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "We formally state the Gronwall’s Inequality here, following the version in (Howard, 1998). ", + "bbox": [ + 168, + 906, + 769, + 922 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Theorem 2. Let $U \\subset \\mathbb { R } ^ { d }$ be an open set. Let $f : U \\times [ 0 , T ] \\to \\mathbb { R } ^ { d }$ be a continuous function and let $\\mathbf { z } _ { 1 }$ , $\\mathbf { z } _ { 2 }$ : $[ 0 , T ] \\to U$ satisfy the initial value problems: ", + "bbox": [ + 169, + 102, + 825, + 132 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/7f6eeb32253c8cad90d543707eb2749cd6d45ec22143c726afbc82c5047bbd0d.jpg", + "text": "$$\n\\begin{array} { r l } & { \\frac { \\mathrm { d } { \\mathbf z } _ { 1 } ( t ) } { \\mathrm { d } t } = f ( { \\mathbf z } _ { 1 } ( t ) , t ) , \\quad { \\mathbf z } _ { 1 } ( t ) = { \\mathbf x } _ { 1 } } \\\\ & { \\frac { \\mathrm { d } { \\mathbf z } _ { 2 } ( t ) } { \\mathrm { d } t } = f ( { \\mathbf z } _ { 2 } ( t ) , t ) , \\quad { \\mathbf z } _ { 2 } ( t ) = { \\mathbf x } _ { 2 } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 379, + 142, + 619, + 207 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Assume there is a constant $C \\geq 0$ such that, for all $t \\in [ 0 , T ]$ , ", + "bbox": [ + 173, + 213, + 578, + 229 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/d44d70833a33bcbc6c66109b2ee246c42ecdacd3538edc2233bfcfea280df71a.jpg", + "text": "$$\n\\| f ( \\mathbf { z } _ { 2 } ( t ) , t ) - f ( \\mathbf { z } _ { 1 } ( t ) , t ) ) \\| \\leq C \\| \\mathbf { z } _ { 2 } ( t ) - \\mathbf { z } _ { 1 } ( t ) \\|\n$$", + "text_format": "latex", + "bbox": [ + 336, + 238, + 660, + 257 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Then, for any $t \\in [ 0 , T ]$ , ", + "bbox": [ + 174, + 265, + 333, + 281 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/9e1ae5bce98edee3c029a1c7316e4e9f396d1d4a5d6b60e76e8363662abe1da1.jpg", + "text": "$$\n\\begin{array} { r } { \\| \\mathbf { z } _ { 1 } ( t ) - \\mathbf { z } _ { 2 } ( t ) \\| \\leq \\| \\mathbf { x } _ { 2 } - \\mathbf { x } _ { 1 } \\| \\cdot e ^ { C t } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 379, + 284, + 619, + 303 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "7.4 MORE EXPERIMENTAL RESULTS ", + "text_level": 1, + "bbox": [ + 176, + 321, + 436, + 335 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "7.4.1 COMPARISON IN THE SETTING OF ADVERSARIAL TRAINING ", + "text_level": 1, + "bbox": [ + 174, + 348, + 642, + 363 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "We implement the adversarial training of the models on the MNIST dataset, and the adversarial examples for training are generated in real-time via the FGSM method (epsilon $_ { 1 = 0 . 3 }$ ) during each epoch (Madry et al., 2017). The results of the adversarially trained models are shown in Table 7. We can observe that the neural ODE-based models are consistently more robust than CNN models. The proposed TisODE also outperforms the vanilla neural ODE. ", + "bbox": [ + 173, + 373, + 826, + 444 + ], + "page_idx": 13 + }, + { + "type": "table", + "img_path": "images/55430c76602428adbc3f3d6582221d15213646ccc38819102c5ce51011e89650.jpg", + "table_caption": [ + "Table 7: Classification accuracy $( \\% )$ on perturbed images from MNIST. To evaluate the robustness of classifiers, we use three types of perturbations, namely zero-mean Gaussian noise with standard deviation $\\sigma$ , FGSM attack and PGD attack. " + ], + "table_footnote": [], + "table_body": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.3
CNN58.098.421.15.3
ODENet84.299.136.012.3
TisODE87.999.166.578.9
", + "bbox": [ + 290, + 515, + 702, + 592 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "7.4.2 EXPERIMENTS ON THE CIFAR10 DATASET ", + "text_level": 1, + "bbox": [ + 174, + 621, + 524, + 636 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "We conduct experiments on CIFAR10 to compare the robustness of CNN and neural ODE-based models. We train all the models only with original non-perturbed images and evaluate the robustness of models against random Gaussian noise and FGSM adversarial attacks. The results are shown in Table 8. We can observe that the ONENet is more robust than the CNN model in terms of both the random noise and the FGSM attack. Besides, our proposal, TisODE, can improve the robustness of the vanilla neural ODE. ", + "bbox": [ + 173, + 646, + 826, + 729 + ], + "page_idx": 13 + }, + { + "type": "table", + "img_path": "images/2bad224570b817a9b629ba9432e5070888e5d4466432b80c75d325e9d0218670.jpg", + "table_caption": [ + "Table 8: Classification accuracy $( \\% )$ on perturbed images from CIFAR10. To evaluate the robustness of classifiers, we use two types of perturbations, namely zero-mean Gaussian noise with standard deviation $\\sigma$ and FGSM attack. " + ], + "table_footnote": [], + "table_body": "
Gaussian noiseAdversarial attack
CIFAR10σ=15σ=20FGSM-8/255FGSM-10/255
CNN70.257.624.318.4
ODENet72.660.631.226.0
TisODE74.362.033.626.8
", + "bbox": [ + 297, + 797, + 696, + 876 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Here, we control the number of parameters to be the same for all kinds of models. We use a small network, which consists of five convolutional layers and one linear layer. ", + "bbox": [ + 173, + 895, + 821, + 924 + ], + "page_idx": 13 + }, + { + "type": "table", + "img_path": "images/43938662417564eb0ebd0d23ae05bc447cbb134bb8df69de1457887746166857.jpg", + "table_caption": [ + "Table 9: The architecture of the ODENet on CIFAR10. " + ], + "table_footnote": [], + "table_body": "
RepetitionLayer
FE×1Conv(3,16,3,1)+GroupNorm+ReLU
×1Conv(16,32,3,2) + GroupNorm+ReLU
×1Conv(32,64,3,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "bbox": [ + 302, + 127, + 692, + 223 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "7.4.3 AN EXTENSION ON THE COMPARISON BETWEEN CNNS AND ODENETS ", + "text_level": 1, + "bbox": [ + 173, + 248, + 722, + 263 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Here, we compare CNN and neural ODE-based models by controlling both the number of parameters and the number of function evaluations. We conduct experiments on the MNIST dataset, and all the models are trained only with original non-perturbed images. ", + "bbox": [ + 176, + 272, + 825, + 315 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "For the neural ODE-based models, the time range is set from 0 to 1. We use the Euler method, and the step size is set to be 0.05. Thus the number of evaluations is $1 / 0 . 0 5 = 2 0 $ . For the CNN models (specifically ResNet), we repeatedly concatenate the residual block for 20 times, and these 20 blocks share the same weights. Our experiments show that, in this condition, the neural ODE-based models still outperform the CNN models (FGSM-0.15: $8 7 . 5 \\%$ vs. $8 1 . 9 \\%$ , FGSM-0.3: $5 3 . 4 \\%$ vs. $4 9 . 7 \\%$ , PGD-0.2: $1 1 . 8 \\%$ vs. $4 . 8 \\%$ ). ", + "bbox": [ + 173, + 321, + 825, + 405 + ], + "page_idx": 14 + } +] \ No newline at end of file diff --git a/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_middle.json b/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..26ecbac3924dba66310aa0ced7a37cb6e81ab53c --- /dev/null +++ b/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_middle.json @@ -0,0 +1,40465 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 502, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 504, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 504, + 97 + ], + "score": 1.0, + "content": "ON ROBUSTNESS OF NEURAL ORDINARY DIFFEREN-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 235, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 235, + 119 + ], + "score": 1.0, + "content": "TIAL EQUATIONS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 137, + 379, + 149 + ], + "lines": [ + { + "bbox": [ + 111, + 137, + 380, + 150 + ], + "spans": [ + { + "bbox": [ + 111, + 137, + 380, + 150 + ], + "score": 1.0, + "content": "Hanshu YAN*, Jiawei DU*, Vincent Y. F. TAN & Jiashi FENG", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 113, + 150, + 468, + 182 + ], + "lines": [ + { + "bbox": [ + 111, + 147, + 324, + 162 + ], + "spans": [ + { + "bbox": [ + 111, + 147, + 324, + 162 + ], + "score": 1.0, + "content": "Department of Electrical and Computer Engineering", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 111, + 159, + 248, + 172 + ], + "spans": [ + { + "bbox": [ + 111, + 159, + 248, + 172 + ], + "score": 1.0, + "content": "National University of Singapore", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 110, + 167, + 471, + 186 + ], + "spans": [ + { + "bbox": [ + 110, + 167, + 471, + 186 + ], + "score": 1.0, + "content": "{hanshu.yan, dujiawei}@u.nus.edu, {vtan, elefjia}@nus.edu.sg", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 278, + 211, + 333, + 223 + ], + "lines": [ + { + "bbox": [ + 276, + 210, + 335, + 224 + ], + "spans": [ + { + "bbox": [ + 276, + 210, + 335, + 224 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 143, + 236, + 468, + 456 + ], + "lines": [ + { + "bbox": [ + 141, + 237, + 469, + 249 + ], + "spans": [ + { + "bbox": [ + 141, + 237, + 469, + 249 + ], + "score": 1.0, + "content": "Neural ordinary differential equations (ODEs) have been attracting increasing at-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 247, + 469, + 261 + ], + "spans": [ + { + "bbox": [ + 141, + 247, + 469, + 261 + ], + "score": 1.0, + "content": "tention in various research domains recently. There have been some works study-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 259, + 469, + 271 + ], + "spans": [ + { + "bbox": [ + 141, + 259, + 469, + 271 + ], + "score": 1.0, + "content": "ing optimization issues and approximation capabilities of neural ODEs, but their", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 269, + 470, + 283 + ], + "spans": [ + { + "bbox": [ + 141, + 269, + 470, + 283 + ], + "score": 1.0, + "content": "robustness is still yet unclear. In this work, we fill this important gap by exploring", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 281, + 470, + 293 + ], + "spans": [ + { + "bbox": [ + 141, + 281, + 470, + 293 + ], + "score": 1.0, + "content": "robustness properties of neural ODEs both empirically and theoretically. We first", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 292, + 470, + 303 + ], + "spans": [ + { + "bbox": [ + 141, + 292, + 470, + 303 + ], + "score": 1.0, + "content": "present an empirical study on the robustness of the neural ODE-based networks", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 303, + 470, + 316 + ], + "spans": [ + { + "bbox": [ + 141, + 303, + 470, + 316 + ], + "score": 1.0, + "content": "(ODENets) by exposing them to inputs with various types of perturbations and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 314, + 469, + 326 + ], + "spans": [ + { + "bbox": [ + 141, + 314, + 469, + 326 + ], + "score": 1.0, + "content": "subsequently investigating the changes of the corresponding outputs. In contrast", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 325, + 469, + 336 + ], + "spans": [ + { + "bbox": [ + 141, + 325, + 469, + 336 + ], + "score": 1.0, + "content": "to conventional convolutional neural networks (CNNs), we find that the ODENets", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 336, + 469, + 347 + ], + "spans": [ + { + "bbox": [ + 141, + 336, + 469, + 347 + ], + "score": 1.0, + "content": "are more robust against both random Gaussian perturbations and adversarial attack", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 347, + 470, + 359 + ], + "spans": [ + { + "bbox": [ + 141, + 347, + 470, + 359 + ], + "score": 1.0, + "content": "examples. We then provide an insightful understanding of this phenomenon by ex-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 357, + 469, + 370 + ], + "spans": [ + { + "bbox": [ + 141, + 357, + 469, + 370 + ], + "score": 1.0, + "content": "ploiting a certain desirable property of the flow of a continuous-time ODE, namely", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 368, + 469, + 381 + ], + "spans": [ + { + "bbox": [ + 141, + 368, + 469, + 381 + ], + "score": 1.0, + "content": "that integral curves are non-intersecting. Our work suggests that, due to their in-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 378, + 470, + 392 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 470, + 392 + ], + "score": 1.0, + "content": "trinsic robustness, it is promising to use neural ODEs as a basic block for building", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 390, + 470, + 402 + ], + "spans": [ + { + "bbox": [ + 141, + 390, + 470, + 402 + ], + "score": 1.0, + "content": "robust deep network models. To further enhance the robustness of vanilla neural", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 401, + 470, + 414 + ], + "spans": [ + { + "bbox": [ + 141, + 401, + 470, + 414 + ], + "score": 1.0, + "content": "ODEs, we propose the time-invariant steady neural ODE (TisODE), which regu-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 413, + 469, + 424 + ], + "spans": [ + { + "bbox": [ + 141, + 413, + 469, + 424 + ], + "score": 1.0, + "content": "larizes the flow on perturbed data via the time-invariant property and the imposi-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 423, + 469, + 435 + ], + "spans": [ + { + "bbox": [ + 141, + 423, + 469, + 435 + ], + "score": 1.0, + "content": "tion of a steady-state constraint. We show that the TisODE method outperforms", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 434, + 469, + 446 + ], + "spans": [ + { + "bbox": [ + 142, + 434, + 469, + 446 + ], + "score": 1.0, + "content": "vanilla neural ODEs and also can work in conjunction with other state-of-the-art", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 444, + 376, + 457 + ], + "spans": [ + { + "bbox": [ + 141, + 444, + 376, + 457 + ], + "score": 1.0, + "content": "architectural methods to build more robust deep networks.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 108, + 480, + 206, + 493 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 208, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 208, + 496 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 504, + 626 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "Neural ordinary differential equations (Chen et al., 2018) form a family of models that approxi-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 518, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 506, + 529 + ], + "score": 1.0, + "content": "mate nonlinear mappings by using continuous-time ODEs. Due to their desirable properties, such", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "as invertibility and parameter efficiency, neural ODEs have attracted increasing attention recently", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "(Dupont et al., 2019; Liu et al., 2019). For example, Grathwohl et al. (2018) proposed a neural", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 550, + 504, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 504, + 561 + ], + "score": 1.0, + "content": "ODE-based generative model—the FFJORD—to solve inverse problems; Quaglino et al. (2019)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "used a higher-order approximation of the states in a neural ODE, and proposed the SNet to acceler-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 573, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 505, + 584 + ], + "score": 1.0, + "content": "ate computation. Along with the wider deployment of neural ODEs, robustness issues come to the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "fore. However, the robustness of neural ODEs is still yet unclear. In particular, it is unclear how", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 595, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 605 + ], + "score": 1.0, + "content": "robust neural ODEs are in comparison to the widely-used CNNs. Robustness properties of CNNs", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "have been studied extensively. In this work, we present the first systematic study on exploring the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 616, + 261, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 261, + 627 + ], + "score": 1.0, + "content": "robustness properties of neural ODEs.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 633, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 632, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 506, + 646 + ], + "score": 1.0, + "content": "To do so, we consider the task of image classification. We expect that results would be similar for", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 642, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 658 + ], + "score": 1.0, + "content": "other machine learning tasks such as regression. Neural ODEs are dimension-preserving mappings,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "but a classification model transforms a high-dimensional input—such as an image—into an output", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "whose dimension is equal to the number of classes. Thus, we consider the neural ODE-based classi-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "fication network (ODENet) whose architecture is shown in Figure 1. An ODENet consists of three", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "components: the feature extractor (FE) consists of convolutional layers which maps an input datum", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "score": 1.0, + "content": "to a multi-channel feature map, a neural ODE that serves as the nonlinear representation mapping", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "(RM), and the fully-connected classifier (FCC) that generates a prediction vector based on the output", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 721, + 152, + 731 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 152, + 731 + ], + "score": 1.0, + "content": "of the RM.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 502, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 504, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 504, + 97 + ], + "score": 1.0, + "content": "ON ROBUSTNESS OF NEURAL ORDINARY DIFFEREN-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 235, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 235, + 119 + ], + "score": 1.0, + "content": "TIAL EQUATIONS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 137, + 379, + 149 + ], + "lines": [ + { + "bbox": [ + 111, + 137, + 380, + 150 + ], + "spans": [ + { + "bbox": [ + 111, + 137, + 380, + 150 + ], + "score": 1.0, + "content": "Hanshu YAN*, Jiawei DU*, Vincent Y. F. TAN & Jiashi FENG", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 111, + 137, + 380, + 150 + ] + }, + { + "type": "list", + "bbox": [ + 113, + 150, + 468, + 182 + ], + "lines": [ + { + "bbox": [ + 111, + 147, + 324, + 162 + ], + "spans": [ + { + "bbox": [ + 111, + 147, + 324, + 162 + ], + "score": 1.0, + "content": "Department of Electrical and Computer Engineering", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 159, + 248, + 172 + ], + "spans": [ + { + "bbox": [ + 111, + 159, + 248, + 172 + ], + "score": 1.0, + "content": "National University of Singapore", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 167, + 471, + 186 + ], + "spans": [ + { + "bbox": [ + 110, + 167, + 471, + 186 + ], + "score": 1.0, + "content": "{hanshu.yan, dujiawei}@u.nus.edu, {vtan, elefjia}@nus.edu.sg", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + } + ], + "index": 4, + "bbox_fs": [ + 110, + 147, + 471, + 186 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 211, + 333, + 223 + ], + "lines": [ + { + "bbox": [ + 276, + 210, + 335, + 224 + ], + "spans": [ + { + "bbox": [ + 276, + 210, + 335, + 224 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 143, + 236, + 468, + 456 + ], + "lines": [ + { + "bbox": [ + 141, + 237, + 469, + 249 + ], + "spans": [ + { + "bbox": [ + 141, + 237, + 469, + 249 + ], + "score": 1.0, + "content": "Neural ordinary differential equations (ODEs) have been attracting increasing at-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 247, + 469, + 261 + ], + "spans": [ + { + "bbox": [ + 141, + 247, + 469, + 261 + ], + "score": 1.0, + "content": "tention in various research domains recently. There have been some works study-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 259, + 469, + 271 + ], + "spans": [ + { + "bbox": [ + 141, + 259, + 469, + 271 + ], + "score": 1.0, + "content": "ing optimization issues and approximation capabilities of neural ODEs, but their", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 269, + 470, + 283 + ], + "spans": [ + { + "bbox": [ + 141, + 269, + 470, + 283 + ], + "score": 1.0, + "content": "robustness is still yet unclear. In this work, we fill this important gap by exploring", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 281, + 470, + 293 + ], + "spans": [ + { + "bbox": [ + 141, + 281, + 470, + 293 + ], + "score": 1.0, + "content": "robustness properties of neural ODEs both empirically and theoretically. We first", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 292, + 470, + 303 + ], + "spans": [ + { + "bbox": [ + 141, + 292, + 470, + 303 + ], + "score": 1.0, + "content": "present an empirical study on the robustness of the neural ODE-based networks", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 303, + 470, + 316 + ], + "spans": [ + { + "bbox": [ + 141, + 303, + 470, + 316 + ], + "score": 1.0, + "content": "(ODENets) by exposing them to inputs with various types of perturbations and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 314, + 469, + 326 + ], + "spans": [ + { + "bbox": [ + 141, + 314, + 469, + 326 + ], + "score": 1.0, + "content": "subsequently investigating the changes of the corresponding outputs. In contrast", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 325, + 469, + 336 + ], + "spans": [ + { + "bbox": [ + 141, + 325, + 469, + 336 + ], + "score": 1.0, + "content": "to conventional convolutional neural networks (CNNs), we find that the ODENets", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 336, + 469, + 347 + ], + "spans": [ + { + "bbox": [ + 141, + 336, + 469, + 347 + ], + "score": 1.0, + "content": "are more robust against both random Gaussian perturbations and adversarial attack", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 347, + 470, + 359 + ], + "spans": [ + { + "bbox": [ + 141, + 347, + 470, + 359 + ], + "score": 1.0, + "content": "examples. We then provide an insightful understanding of this phenomenon by ex-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 357, + 469, + 370 + ], + "spans": [ + { + "bbox": [ + 141, + 357, + 469, + 370 + ], + "score": 1.0, + "content": "ploiting a certain desirable property of the flow of a continuous-time ODE, namely", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 368, + 469, + 381 + ], + "spans": [ + { + "bbox": [ + 141, + 368, + 469, + 381 + ], + "score": 1.0, + "content": "that integral curves are non-intersecting. Our work suggests that, due to their in-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 378, + 470, + 392 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 470, + 392 + ], + "score": 1.0, + "content": "trinsic robustness, it is promising to use neural ODEs as a basic block for building", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 390, + 470, + 402 + ], + "spans": [ + { + "bbox": [ + 141, + 390, + 470, + 402 + ], + "score": 1.0, + "content": "robust deep network models. To further enhance the robustness of vanilla neural", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 401, + 470, + 414 + ], + "spans": [ + { + "bbox": [ + 141, + 401, + 470, + 414 + ], + "score": 1.0, + "content": "ODEs, we propose the time-invariant steady neural ODE (TisODE), which regu-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 413, + 469, + 424 + ], + "spans": [ + { + "bbox": [ + 141, + 413, + 469, + 424 + ], + "score": 1.0, + "content": "larizes the flow on perturbed data via the time-invariant property and the imposi-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 423, + 469, + 435 + ], + "spans": [ + { + "bbox": [ + 141, + 423, + 469, + 435 + ], + "score": 1.0, + "content": "tion of a steady-state constraint. We show that the TisODE method outperforms", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 434, + 469, + 446 + ], + "spans": [ + { + "bbox": [ + 142, + 434, + 469, + 446 + ], + "score": 1.0, + "content": "vanilla neural ODEs and also can work in conjunction with other state-of-the-art", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 444, + 376, + 457 + ], + "spans": [ + { + "bbox": [ + 141, + 444, + 376, + 457 + ], + "score": 1.0, + "content": "architectural methods to build more robust deep networks.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 16.5, + "bbox_fs": [ + 141, + 237, + 470, + 457 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 480, + 206, + 493 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 208, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 208, + 496 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 504, + 626 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "Neural ordinary differential equations (Chen et al., 2018) form a family of models that approxi-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 518, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 506, + 529 + ], + "score": 1.0, + "content": "mate nonlinear mappings by using continuous-time ODEs. Due to their desirable properties, such", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "as invertibility and parameter efficiency, neural ODEs have attracted increasing attention recently", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "(Dupont et al., 2019; Liu et al., 2019). For example, Grathwohl et al. (2018) proposed a neural", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 550, + 504, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 504, + 561 + ], + "score": 1.0, + "content": "ODE-based generative model—the FFJORD—to solve inverse problems; Quaglino et al. (2019)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "used a higher-order approximation of the states in a neural ODE, and proposed the SNet to acceler-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 573, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 505, + 584 + ], + "score": 1.0, + "content": "ate computation. Along with the wider deployment of neural ODEs, robustness issues come to the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "fore. However, the robustness of neural ODEs is still yet unclear. In particular, it is unclear how", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 595, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 605 + ], + "score": 1.0, + "content": "robust neural ODEs are in comparison to the widely-used CNNs. Robustness properties of CNNs", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "have been studied extensively. In this work, we present the first systematic study on exploring the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 616, + 261, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 261, + 627 + ], + "score": 1.0, + "content": "robustness properties of neural ODEs.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 506, + 506, + 627 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 633, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 632, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 506, + 646 + ], + "score": 1.0, + "content": "To do so, we consider the task of image classification. We expect that results would be similar for", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 642, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 658 + ], + "score": 1.0, + "content": "other machine learning tasks such as regression. Neural ODEs are dimension-preserving mappings,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "but a classification model transforms a high-dimensional input—such as an image—into an output", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "whose dimension is equal to the number of classes. Thus, we consider the neural ODE-based classi-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "fication network (ODENet) whose architecture is shown in Figure 1. An ODENet consists of three", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "components: the feature extractor (FE) consists of convolutional layers which maps an input datum", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "score": 1.0, + "content": "to a multi-channel feature map, a neural ODE that serves as the nonlinear representation mapping", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "(RM), and the fully-connected classifier (FCC) that generates a prediction vector based on the output", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 721, + 152, + 731 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 152, + 731 + ], + "score": 1.0, + "content": "of the RM.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 632, + 506, + 731 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 83, + 376, + 280 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 376, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 376, + 94 + ], + "score": 1.0, + "content": "The robustness of a classification model can be evaluated through", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 376, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 376, + 106 + ], + "score": 1.0, + "content": "the lens of its performance on perturbed images. To comprehen-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 376, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 376, + 118 + ], + "score": 1.0, + "content": "sively investigate the robustness of neural ODEs, we perturb orig-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 376, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 376, + 128 + ], + "score": 1.0, + "content": "inal images with commonly-used perturbations, namely, random", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 376, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 376, + 138 + ], + "score": 1.0, + "content": "Gaussian noise (Szegedy et al., 2013) and harmful adversarial ex-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 376, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 376, + 149 + ], + "score": 1.0, + "content": "amples (Goodfellow et al., 2014; Madry et al., 2017). We conduct", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 149, + 376, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 376, + 160 + ], + "score": 1.0, + "content": "experiments in two common settings—training the model only on", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 376, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 376, + 172 + ], + "score": 1.0, + "content": "authentic non-perturbed images and training the model on authen-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 376, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 376, + 182 + ], + "score": 1.0, + "content": "tic images as well as the Gaussian perturbed ones. We observe that", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 181, + 376, + 193 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 376, + 193 + ], + "score": 1.0, + "content": "ODENets are more robust compared to CNN models against all", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 192, + 376, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 376, + 204 + ], + "score": 1.0, + "content": "types of perturbations in both settings. We then provide an insight-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 203, + 376, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 376, + 216 + ], + "score": 1.0, + "content": "ful understanding of such intriguing robustness of neural ODEs", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 214, + 376, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 376, + 226 + ], + "score": 1.0, + "content": "by exploiting a certain property of the flow (Dupont et al., 2019),", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 225, + 376, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 376, + 236 + ], + "score": 1.0, + "content": "namely that integral curves that start at distinct initial states are non-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 236, + 376, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 376, + 248 + ], + "score": 1.0, + "content": "intersecting. The flow of a continuous-time ODE is defined as the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 247, + 376, + 259 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 376, + 259 + ], + "score": 1.0, + "content": "family of solutions/paths traversed by the state, starting from dif-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 258, + 376, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 376, + 270 + ], + "score": 1.0, + "content": "ferent initial points, and an integral curve is a specific solution for", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 269, + 376, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 376, + 281 + ], + "score": 1.0, + "content": "a given initial point. The non-intersecting property indicates that", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 8.5 + }, + { + "type": "image", + "bbox": [ + 403, + 73, + 484, + 210 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 403, + 73, + 484, + 210 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 403, + 73, + 484, + 210 + ], + "spans": [ + { + "bbox": [ + 403, + 73, + 484, + 210 + ], + "score": 0.95, + "type": "image", + "image_path": "6e9bd948731c27f7167c50dec38ca4fa802294fd3834ad917ec0b4023c814c57.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 403, + 73, + 484, + 210 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 383, + 217, + 504, + 272 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 383, + 216, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 383, + 216, + 505, + 228 + ], + "score": 1.0, + "content": "Figure 1: The architecture", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 383, + 228, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 383, + 228, + 505, + 239 + ], + "score": 1.0, + "content": "of an ODENet. The neu-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 383, + 239, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 383, + 239, + 505, + 250 + ], + "score": 1.0, + "content": "ral ODE block serves as a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 383, + 250, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 383, + 250, + 505, + 263 + ], + "score": 1.0, + "content": "dimension-preserving nonlin-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 382, + 261, + 438, + 275 + ], + "spans": [ + { + "bbox": [ + 382, + 261, + 438, + 275 + ], + "score": 1.0, + "content": "ear mapping.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 280, + 505, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 280, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 505, + 291 + ], + "score": 1.0, + "content": "an integral curve starting from some point is constrained by the integral curves starting from that", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 290, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 505, + 303 + ], + "score": 1.0, + "content": "point’s neighborhood. Thus, in an ODENet, if a correctly classified datum is slightly perturbed, the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 301, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 505, + 315 + ], + "score": 1.0, + "content": "integral curve associated to its perturbed version would not change too much from the original one.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "score": 1.0, + "content": "Consequently, the perturbed datum could still be correctly classified. Thus, there exists intrinsic", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 324, + 379, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 379, + 335 + ], + "score": 1.0, + "content": "robustness regularization in ODENets, which is absent from CNNs.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 505, + 494 + ], + "lines": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "Motivated by this property of the neural ODE flow, we attempt to explore a more robust neural ODE", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "score": 1.0, + "content": "architecture by introducing stronger regularization on the flow. We thus propose a Time-Invariant", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "Steady neural ODE (TisODE). The TisODE removes the time dependence of the dynamics in an", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 372, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 386 + ], + "score": 1.0, + "content": "ODE and imposes a steady-state constraint on the integral curves. Removing the time dependence", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 479, + 398 + ], + "score": 1.0, + "content": "of the derivative results in the time-invariant property of the ODE. To wit, given a solution", + "type": "text" + }, + { + "bbox": [ + 479, + 384, + 501, + 397 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 384, + 505, + 398 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 395, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 174, + 408 + ], + "score": 1.0, + "content": "another solution", + "type": "text" + }, + { + "bbox": [ + 174, + 395, + 197, + 407 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 396, + 282, + 408 + ], + "score": 1.0, + "content": ", with an initial state", + "type": "text" + }, + { + "bbox": [ + 283, + 395, + 349, + 407 + ], + "score": 0.94, + "content": "\\tilde { { \\bf z } } _ { 1 } ( 0 ) \\bar { { \\bf \\phi } } = { \\bf \\bar { z } } _ { 1 } ( T ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 396, + 390, + 408 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 390, + 396, + 422, + 406 + ], + "score": 0.92, + "content": "T ^ { \\prime } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 396, + 506, + 408 + ], + "score": 1.0, + "content": ", can be regarded as", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 405, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 121, + 420 + ], + "score": 1.0, + "content": "the", + "type": "text" + }, + { + "bbox": [ + 121, + 406, + 140, + 416 + ], + "score": 0.87, + "content": "- T ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 405, + 207, + 420 + ], + "score": 1.0, + "content": "- shift version of", + "type": "text" + }, + { + "bbox": [ + 207, + 406, + 228, + 418 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 405, + 506, + 420 + ], + "score": 1.0, + "content": ". Such a time-invariant property would make bounding the difference", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 418, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 429 + ], + "score": 1.0, + "content": "between output states convenient. To elaborate, let the output of a neural ODE correspond to states", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 137, + 441 + ], + "score": 1.0, + "content": "at time", + "type": "text" + }, + { + "bbox": [ + 137, + 428, + 166, + 438 + ], + "score": 0.89, + "content": "T > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 428, + 429, + 441 + ], + "score": 1.0, + "content": ". By the time-invariant property, the difference between outputs,", + "type": "text" + }, + { + "bbox": [ + 429, + 428, + 501, + 440 + ], + "score": 0.93, + "content": "\\| \\widetilde { \\mathbf z } _ { 1 } ( \\bar { T } ) - \\mathbf z _ { 1 } ( T ) \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 428, + 505, + 441 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 439, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 147, + 452 + ], + "score": 1.0, + "content": "equals to", + "type": "text" + }, + { + "bbox": [ + 147, + 439, + 245, + 451 + ], + "score": 0.93, + "content": "\\| { \\bf z } _ { 1 } ( T + T ^ { \\prime } ) - { \\bf z } _ { 1 } ( T ) \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 439, + 506, + 452 + ], + "score": 1.0, + "content": ". To control this distance, a steady-state regularization term is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 450, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 455, + 462 + ], + "score": 1.0, + "content": "introduced to the overall objective to constrain the change of a state after time exceeds", + "type": "text" + }, + { + "bbox": [ + 455, + 451, + 463, + 460 + ], + "score": 0.71, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 450, + 506, + 462 + ], + "score": 1.0, + "content": ". With the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "time-invariant property and the steady-state term, we show that TisODE even is more robust. We do", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "so by evaluating the robustness of TisODE-based classifiers against various types of perturbations", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 483, + 423, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 423, + 495 + ], + "score": 1.0, + "content": "and observe that such models are more robust than vanilla ODE-based models.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 505, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "score": 1.0, + "content": "In addition, some other effective architectural solutions have also been recently proposed to improve", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 511, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 505, + 522 + ], + "score": 1.0, + "content": "the robustness of CNNs. For example, Xie et al. (2017) randomly resizes or pads zeros into test", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "images to destroy the specific structure of adversarial perturbations. Besides, the model proposed by", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "Xie et al. (2019) contains feature denoising filters to remove the feature-level patterns of adversarial", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 544, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 505, + 556 + ], + "score": 1.0, + "content": "examples. We conduct experiments to show that our proposed TisODE can work seamlessly and in", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 555, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 567 + ], + "score": 1.0, + "content": "conjunction with these methods to further boost the robustness of deep models. Thus, the proposed", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 564, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 506, + 579 + ], + "score": 1.0, + "content": "TisODE can be used as a generally applicable and effective component for improving the robustness", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 576, + 172, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 172, + 590 + ], + "score": 1.0, + "content": "of deep models.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 46.5 + }, + { + "type": "text", + "bbox": [ + 106, + 593, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "In summary, our contributions are as follows. Firstly, we are the first to provide a systematic empir-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 605, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 616 + ], + "score": 1.0, + "content": "ical study on the robustness of neural ODEs and find that the neural ODE-based models are more", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "robust compared to conventional CNN models. This finding inspires new applications of neural", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 626, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 504, + 639 + ], + "score": 1.0, + "content": "ODEs in improving robustness of deep models, a problem that concerns many deep learning theo-", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "rists and practitioners alike. Secondly, we propose the TisODE method, which is simple yet effective", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "in significantly boosting the robustness of neural ODEs. Moreover, the proposed TisODE can also", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "score": 1.0, + "content": "be used in conjunction with other state-of-the-art robust architectures. Thus, TisODE can serve as a", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 105, + 670, + 385, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 385, + 684 + ], + "score": 1.0, + "content": "drop-in module to improve the robustness of deep models effectively.", + "type": "text" + } + ], + "index": 58 + } + ], + "index": 54.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 83, + 376, + 280 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 376, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 376, + 94 + ], + "score": 1.0, + "content": "The robustness of a classification model can be evaluated through", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 376, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 376, + 106 + ], + "score": 1.0, + "content": "the lens of its performance on perturbed images. To comprehen-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 376, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 376, + 118 + ], + "score": 1.0, + "content": "sively investigate the robustness of neural ODEs, we perturb orig-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 376, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 376, + 128 + ], + "score": 1.0, + "content": "inal images with commonly-used perturbations, namely, random", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 376, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 376, + 138 + ], + "score": 1.0, + "content": "Gaussian noise (Szegedy et al., 2013) and harmful adversarial ex-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 376, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 376, + 149 + ], + "score": 1.0, + "content": "amples (Goodfellow et al., 2014; Madry et al., 2017). We conduct", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 149, + 376, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 376, + 160 + ], + "score": 1.0, + "content": "experiments in two common settings—training the model only on", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 376, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 376, + 172 + ], + "score": 1.0, + "content": "authentic non-perturbed images and training the model on authen-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 376, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 376, + 182 + ], + "score": 1.0, + "content": "tic images as well as the Gaussian perturbed ones. We observe that", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 181, + 376, + 193 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 376, + 193 + ], + "score": 1.0, + "content": "ODENets are more robust compared to CNN models against all", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 192, + 376, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 376, + 204 + ], + "score": 1.0, + "content": "types of perturbations in both settings. We then provide an insight-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 203, + 376, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 376, + 216 + ], + "score": 1.0, + "content": "ful understanding of such intriguing robustness of neural ODEs", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 214, + 376, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 376, + 226 + ], + "score": 1.0, + "content": "by exploiting a certain property of the flow (Dupont et al., 2019),", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 225, + 376, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 376, + 236 + ], + "score": 1.0, + "content": "namely that integral curves that start at distinct initial states are non-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 236, + 376, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 376, + 248 + ], + "score": 1.0, + "content": "intersecting. The flow of a continuous-time ODE is defined as the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 247, + 376, + 259 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 376, + 259 + ], + "score": 1.0, + "content": "family of solutions/paths traversed by the state, starting from dif-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 258, + 376, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 376, + 270 + ], + "score": 1.0, + "content": "ferent initial points, and an integral curve is a specific solution for", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 269, + 376, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 376, + 281 + ], + "score": 1.0, + "content": "a given initial point. The non-intersecting property indicates that", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 280, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 505, + 291 + ], + "score": 1.0, + "content": "an integral curve starting from some point is constrained by the integral curves starting from that", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 290, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 505, + 303 + ], + "score": 1.0, + "content": "point’s neighborhood. Thus, in an ODENet, if a correctly classified datum is slightly perturbed, the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 301, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 505, + 315 + ], + "score": 1.0, + "content": "integral curve associated to its perturbed version would not change too much from the original one.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "score": 1.0, + "content": "Consequently, the perturbed datum could still be correctly classified. Thus, there exists intrinsic", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 324, + 379, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 379, + 335 + ], + "score": 1.0, + "content": "robustness regularization in ODENets, which is absent from CNNs.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 82, + 376, + 281 + ] + }, + { + "type": "image", + "bbox": [ + 403, + 73, + 484, + 210 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 403, + 73, + 484, + 210 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 403, + 73, + 484, + 210 + ], + "spans": [ + { + "bbox": [ + 403, + 73, + 484, + 210 + ], + "score": 0.95, + "type": "image", + "image_path": "6e9bd948731c27f7167c50dec38ca4fa802294fd3834ad917ec0b4023c814c57.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 403, + 73, + 484, + 210 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 383, + 217, + 504, + 272 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 383, + 216, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 383, + 216, + 505, + 228 + ], + "score": 1.0, + "content": "Figure 1: The architecture", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 383, + 228, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 383, + 228, + 505, + 239 + ], + "score": 1.0, + "content": "of an ODENet. The neu-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 383, + 239, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 383, + 239, + 505, + 250 + ], + "score": 1.0, + "content": "ral ODE block serves as a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 383, + 250, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 383, + 250, + 505, + 263 + ], + "score": 1.0, + "content": "dimension-preserving nonlin-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 382, + 261, + 438, + 275 + ], + "spans": [ + { + "bbox": [ + 382, + 261, + 438, + 275 + ], + "score": 1.0, + "content": "ear mapping.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 280, + 505, + 334 + ], + "lines": [], + "index": 26, + "bbox_fs": [ + 105, + 280, + 506, + 335 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 505, + 494 + ], + "lines": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "Motivated by this property of the neural ODE flow, we attempt to explore a more robust neural ODE", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "score": 1.0, + "content": "architecture by introducing stronger regularization on the flow. We thus propose a Time-Invariant", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "Steady neural ODE (TisODE). The TisODE removes the time dependence of the dynamics in an", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 372, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 386 + ], + "score": 1.0, + "content": "ODE and imposes a steady-state constraint on the integral curves. Removing the time dependence", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 479, + 398 + ], + "score": 1.0, + "content": "of the derivative results in the time-invariant property of the ODE. To wit, given a solution", + "type": "text" + }, + { + "bbox": [ + 479, + 384, + 501, + 397 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 384, + 505, + 398 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 395, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 174, + 408 + ], + "score": 1.0, + "content": "another solution", + "type": "text" + }, + { + "bbox": [ + 174, + 395, + 197, + 407 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 396, + 282, + 408 + ], + "score": 1.0, + "content": ", with an initial state", + "type": "text" + }, + { + "bbox": [ + 283, + 395, + 349, + 407 + ], + "score": 0.94, + "content": "\\tilde { { \\bf z } } _ { 1 } ( 0 ) \\bar { { \\bf \\phi } } = { \\bf \\bar { z } } _ { 1 } ( T ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 396, + 390, + 408 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 390, + 396, + 422, + 406 + ], + "score": 0.92, + "content": "T ^ { \\prime } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 396, + 506, + 408 + ], + "score": 1.0, + "content": ", can be regarded as", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 405, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 121, + 420 + ], + "score": 1.0, + "content": "the", + "type": "text" + }, + { + "bbox": [ + 121, + 406, + 140, + 416 + ], + "score": 0.87, + "content": "- T ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 405, + 207, + 420 + ], + "score": 1.0, + "content": "- shift version of", + "type": "text" + }, + { + "bbox": [ + 207, + 406, + 228, + 418 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 405, + 506, + 420 + ], + "score": 1.0, + "content": ". Such a time-invariant property would make bounding the difference", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 418, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 429 + ], + "score": 1.0, + "content": "between output states convenient. To elaborate, let the output of a neural ODE correspond to states", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 137, + 441 + ], + "score": 1.0, + "content": "at time", + "type": "text" + }, + { + "bbox": [ + 137, + 428, + 166, + 438 + ], + "score": 0.89, + "content": "T > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 428, + 429, + 441 + ], + "score": 1.0, + "content": ". By the time-invariant property, the difference between outputs,", + "type": "text" + }, + { + "bbox": [ + 429, + 428, + 501, + 440 + ], + "score": 0.93, + "content": "\\| \\widetilde { \\mathbf z } _ { 1 } ( \\bar { T } ) - \\mathbf z _ { 1 } ( T ) \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 428, + 505, + 441 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 439, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 147, + 452 + ], + "score": 1.0, + "content": "equals to", + "type": "text" + }, + { + "bbox": [ + 147, + 439, + 245, + 451 + ], + "score": 0.93, + "content": "\\| { \\bf z } _ { 1 } ( T + T ^ { \\prime } ) - { \\bf z } _ { 1 } ( T ) \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 439, + 506, + 452 + ], + "score": 1.0, + "content": ". To control this distance, a steady-state regularization term is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 450, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 455, + 462 + ], + "score": 1.0, + "content": "introduced to the overall objective to constrain the change of a state after time exceeds", + "type": "text" + }, + { + "bbox": [ + 455, + 451, + 463, + 460 + ], + "score": 0.71, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 450, + 506, + 462 + ], + "score": 1.0, + "content": ". With the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "time-invariant property and the steady-state term, we show that TisODE even is more robust. We do", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "so by evaluating the robustness of TisODE-based classifiers against various types of perturbations", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 483, + 423, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 423, + 495 + ], + "score": 1.0, + "content": "and observe that such models are more robust than vanilla ODE-based models.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 340, + 506, + 495 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 505, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "score": 1.0, + "content": "In addition, some other effective architectural solutions have also been recently proposed to improve", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 511, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 505, + 522 + ], + "score": 1.0, + "content": "the robustness of CNNs. For example, Xie et al. (2017) randomly resizes or pads zeros into test", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "images to destroy the specific structure of adversarial perturbations. Besides, the model proposed by", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "Xie et al. (2019) contains feature denoising filters to remove the feature-level patterns of adversarial", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 544, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 505, + 556 + ], + "score": 1.0, + "content": "examples. We conduct experiments to show that our proposed TisODE can work seamlessly and in", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 555, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 567 + ], + "score": 1.0, + "content": "conjunction with these methods to further boost the robustness of deep models. Thus, the proposed", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 564, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 506, + 579 + ], + "score": 1.0, + "content": "TisODE can be used as a generally applicable and effective component for improving the robustness", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 576, + 172, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 172, + 590 + ], + "score": 1.0, + "content": "of deep models.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 46.5, + "bbox_fs": [ + 105, + 499, + 506, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 593, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "In summary, our contributions are as follows. Firstly, we are the first to provide a systematic empir-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 605, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 616 + ], + "score": 1.0, + "content": "ical study on the robustness of neural ODEs and find that the neural ODE-based models are more", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "robust compared to conventional CNN models. This finding inspires new applications of neural", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 626, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 504, + 639 + ], + "score": 1.0, + "content": "ODEs in improving robustness of deep models, a problem that concerns many deep learning theo-", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "rists and practitioners alike. Secondly, we propose the TisODE method, which is simple yet effective", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "in significantly boosting the robustness of neural ODEs. Moreover, the proposed TisODE can also", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "score": 1.0, + "content": "be used in conjunction with other state-of-the-art robust architectures. Thus, TisODE can serve as a", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 105, + 670, + 385, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 385, + 684 + ], + "score": 1.0, + "content": "drop-in module to improve the robustness of deep models effectively.", + "type": "text" + } + ], + "index": 58 + } + ], + "index": 54.5, + "bbox_fs": [ + 105, + 594, + 506, + 684 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 301, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 302, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 302, + 96 + ], + "score": 1.0, + "content": "2 PRELIMINARIES ON NEURAL ODE", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 105, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "score": 1.0, + "content": "It has been shown that a residual block (He et al., 2016) can be interpreted as the discrete approxima-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "score": 1.0, + "content": "tion of an ODE by setting the discretization step to be one. When the discretization step approaches", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "score": 1.0, + "content": "zero, it yields a family of neural networks, which are called neural ODEs (Chen et al., 2018). For-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 506, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 506, + 152 + ], + "score": 1.0, + "content": "mally, in a neural ODE, the relation between input and output is characterized by the following set", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 150, + 161, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 150, + 161, + 162 + ], + "score": 1.0, + "content": "of equations:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 201, + 159, + 409, + 184 + ], + "lines": [ + { + "bbox": [ + 201, + 159, + 409, + 184 + ], + "spans": [ + { + "bbox": [ + 201, + 159, + 409, + 184 + ], + "score": 0.91, + "content": "\\frac { \\mathrm { d } { \\mathbf z } ( t ) } { \\mathrm { d } t } = f _ { \\boldsymbol \\theta } ( { \\mathbf z } ( t ) , t ) , \\quad { \\mathbf z } ( 0 ) = { \\mathbf z } _ { \\mathrm { i n } } , \\quad { \\mathbf z } _ { \\mathrm { o u t } } = { \\mathbf z } ( T ) ,", + "type": "interline_equation", + "image_path": "97910f4443c6f9c371655e158097ca883dd18d4e639302bc0560ca17790de48a.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 201, + 159, + 409, + 171.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 201, + 171.5, + 409, + 184.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 185, + 504, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 134, + 200 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 187, + 239, + 199 + ], + "score": 0.92, + "content": "f _ { \\theta } : \\mathbb { R } ^ { d } \\times [ 0 , \\infty ) \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 185, + 497, + 200 + ], + "score": 1.0, + "content": "denotes the trainable layers that are parameterized by weights", + "type": "text" + }, + { + "bbox": [ + 498, + 188, + 504, + 198 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 123, + 211 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 199, + 195, + 210 + ], + "score": 0.9, + "content": "\\mathbf { z } : [ 0 , \\infty ) \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 197, + 255, + 211 + ], + "score": 1.0, + "content": "represents the", + "type": "text" + }, + { + "bbox": [ + 256, + 199, + 262, + 208 + ], + "score": 0.82, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 197, + 483, + 211 + ], + "score": 1.0, + "content": "-dimensional state of the neural ODE. We assume that", + "type": "text" + }, + { + "bbox": [ + 484, + 199, + 494, + 210 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 197, + 506, + 211 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 209, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 163, + 221 + ], + "score": 1.0, + "content": "continuous in", + "type": "text" + }, + { + "bbox": [ + 163, + 211, + 168, + 219 + ], + "score": 0.79, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 209, + 316, + 221 + ], + "score": 1.0, + "content": "and globally Lipschitz continuous in", + "type": "text" + }, + { + "bbox": [ + 317, + 212, + 323, + 219 + ], + "score": 0.52, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 209, + 414, + 221 + ], + "score": 1.0, + "content": ". In this case, the input", + "type": "text" + }, + { + "bbox": [ + 415, + 211, + 428, + 221 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { \\mathrm { i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 209, + 506, + 221 + ], + "score": 1.0, + "content": "of the neural ODE", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 219, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 213, + 234 + ], + "score": 1.0, + "content": "corresponds to the state at", + "type": "text" + }, + { + "bbox": [ + 213, + 221, + 237, + 231 + ], + "score": 0.9, + "content": "t = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 219, + 300, + 234 + ], + "score": 1.0, + "content": ", and the output", + "type": "text" + }, + { + "bbox": [ + 300, + 222, + 319, + 232 + ], + "score": 0.89, + "content": "\\mathbf { z } _ { \\mathrm { o u t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 219, + 453, + 234 + ], + "score": 1.0, + "content": "is associated to the state at some", + "type": "text" + }, + { + "bbox": [ + 453, + 221, + 501, + 232 + ], + "score": 0.93, + "content": "T \\in ( 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 219, + 506, + 234 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 231, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 143, + 244 + ], + "score": 1.0, + "content": "Because", + "type": "text" + }, + { + "bbox": [ + 144, + 231, + 154, + 243 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 231, + 372, + 244 + ], + "score": 1.0, + "content": "governs how the state changes with respect to time", + "type": "text" + }, + { + "bbox": [ + 372, + 232, + 378, + 241 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 231, + 435, + 244 + ], + "score": 1.0, + "content": ", we also use", + "type": "text" + }, + { + "bbox": [ + 435, + 232, + 446, + 243 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 231, + 505, + 244 + ], + "score": 1.0, + "content": "to denote the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 243, + 225, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 225, + 253 + ], + "score": 1.0, + "content": "dynamics of the neural ODE.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 107, + 259, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 155, + 272 + ], + "score": 1.0, + "content": "Given input", + "type": "text" + }, + { + "bbox": [ + 155, + 261, + 168, + 270 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { \\mathrm { i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 259, + 214, + 272 + ], + "score": 1.0, + "content": ", the output", + "type": "text" + }, + { + "bbox": [ + 214, + 261, + 233, + 270 + ], + "score": 0.89, + "content": "\\mathbf { z } _ { \\mathrm { o u t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 259, + 420, + 272 + ], + "score": 1.0, + "content": "can be computed by solving the ODE in (1). If", + "type": "text" + }, + { + "bbox": [ + 420, + 260, + 428, + 269 + ], + "score": 0.83, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 259, + 506, + 272 + ], + "score": 1.0, + "content": "is fixed, the output", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 270, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 124, + 282 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { \\mathrm { o u t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 270, + 229, + 283 + ], + "score": 1.0, + "content": "only depends on the input", + "type": "text" + }, + { + "bbox": [ + 230, + 272, + 243, + 281 + ], + "score": 0.89, + "content": "\\mathbf { z } _ { \\mathrm { i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 270, + 315, + 283 + ], + "score": 1.0, + "content": "and the dynamics", + "type": "text" + }, + { + "bbox": [ + 315, + 271, + 325, + 282 + ], + "score": 0.88, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 270, + 505, + 283 + ], + "score": 1.0, + "content": ", which also corresponds to the weighted lay-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 281, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 409, + 293 + ], + "score": 1.0, + "content": "ers in the neural ODE. Therefore, the neural ODE can be represented as the", + "type": "text" + }, + { + "bbox": [ + 410, + 282, + 416, + 291 + ], + "score": 0.82, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 281, + 505, + 293 + ], + "score": 1.0, + "content": "-dimensional function", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 292, + 307, + 305 + ], + "spans": [ + { + "bbox": [ + 107, + 292, + 137, + 304 + ], + "score": 0.91, + "content": "\\phi _ { T } ( \\cdot , \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 292, + 187, + 305 + ], + "score": 1.0, + "content": "of the input", + "type": "text" + }, + { + "bbox": [ + 187, + 294, + 201, + 303 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { \\mathrm { i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 292, + 274, + 305 + ], + "score": 1.0, + "content": "and the dynamics", + "type": "text" + }, + { + "bbox": [ + 275, + 293, + 285, + 304 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 292, + 307, + 305 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 191, + 308, + 420, + 337 + ], + "lines": [ + { + "bbox": [ + 191, + 308, + 420, + 337 + ], + "spans": [ + { + "bbox": [ + 191, + 308, + 420, + 337 + ], + "score": 0.92, + "content": "\\mathbf { z } _ { \\mathrm { o u t } } = \\mathbf { z } ( T ) = \\mathbf { z } ( 0 ) + \\int _ { 0 } ^ { T } f _ { \\theta } ( \\mathbf { z } ( t ) , t ) \\mathrm { d } t = \\phi _ { T } ( \\mathbf { z } _ { \\mathrm { i n } } , f _ { \\theta } ) .", + "type": "interline_equation", + "image_path": "ee04e9b71d2db1ef8c0ab434f849a920d8934ab30d56fbcaea2c2fdc1e2122ec.jpg" + } + ] + } + ], + "index": 18.5, + "virtual_lines": [ + { + "bbox": [ + 191, + 308, + 420, + 322.5 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 191, + 322.5, + 420, + 337.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 342, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 185, + 354 + ], + "score": 1.0, + "content": "The terminal time", + "type": "text" + }, + { + "bbox": [ + 185, + 343, + 194, + 353 + ], + "score": 0.82, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 342, + 276, + 354 + ], + "score": 1.0, + "content": "of the output state", + "type": "text" + }, + { + "bbox": [ + 276, + 343, + 297, + 354 + ], + "score": 0.92, + "content": "{ \\mathbf z } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "is set to be 1 in practice. Several methods have", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 354, + 504, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 504, + 366 + ], + "score": 1.0, + "content": "been proposed for training neural ODEs, such as the adjoint sensitivity method (Chen et al., 2018),", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "score": 1.0, + "content": "SNet (Quaglino et al., 2019), and the auto-differentiation technique (Paszke et al., 2017). In", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 375, + 437, + 388 + ], + "score": 1.0, + "content": "this work, we use the most straightforward technique, i.e., updating the weights", + "type": "text" + }, + { + "bbox": [ + 437, + 376, + 443, + 385 + ], + "score": 0.78, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "with the auto-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 386, + 316, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 386, + 316, + 398 + ], + "score": 1.0, + "content": "differentiation technique in the PyTorch framework.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 107, + 414, + 424, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 413, + 426, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 426, + 428 + ], + "score": 1.0, + "content": "3 AN EMPIRICAL STUDY ON THE ROBUSTNESS OF ODENETS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 106, + 438, + 505, + 538 + ], + "lines": [ + { + "bbox": [ + 106, + 439, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 504, + 451 + ], + "score": 1.0, + "content": "Robustness of deep models has gained increased attention, as it is imperative that deep models em-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "ployed in critical applications, such as healthcare, are robust. The robustness of a model is measured", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 460, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 474 + ], + "score": 1.0, + "content": "by the sensitivity of the prediction with respect to small perturbations on the inputs. In this study,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 472, + 504, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 472, + 504, + 484 + ], + "score": 1.0, + "content": "we consider three commonly-used perturbation schemes, namely random Gaussian perturbations,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "FGSM (Goodfellow et al., 2014) adversarial examples, and PGD (Madry et al., 2017) adversarial", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 493, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 505, + 506 + ], + "score": 1.0, + "content": "examples. These perturbation schemes reflect noise and adversarial robustness properties of the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 505, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 505, + 517 + ], + "score": 1.0, + "content": "investigated models respectively. We evaluate the robustness via the classification accuracies on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 515, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 104, + 515, + 505, + 529 + ], + "score": 1.0, + "content": "perturbed images, in which the original non-perturbed versions of these images are all correctly", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 527, + 148, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 148, + 538 + ], + "score": 1.0, + "content": "classified.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 543, + 505, + 642 + ], + "lines": [ + { + "bbox": [ + 106, + 544, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 556 + ], + "score": 1.0, + "content": "For a fair comparison with conventional CNN models, we made sure that the number of parameters", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "score": 1.0, + "content": "of an ODENet is close to that of its counterpart CNN model. Specifically, the ODENet shares the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 565, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 578 + ], + "score": 1.0, + "content": "same network architecture with the CNN model for the FE and FCC parts. The only difference", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 577, + 504, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 504, + 588 + ], + "score": 1.0, + "content": "is that, for the RM part, the input of the ODE-based RM is concatenated with one more channel", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 588, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 214, + 599 + ], + "score": 1.0, + "content": "which represents the time", + "type": "text" + }, + { + "bbox": [ + 214, + 588, + 220, + 597 + ], + "score": 0.58, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 588, + 506, + 599 + ], + "score": 1.0, + "content": ", while the RM in a CNN model has a skip connection and serves as", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 597, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 505, + 612 + ], + "score": 1.0, + "content": "a residual block. During the training phase, all the hyperparameters are kept the same, including", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "score": 1.0, + "content": "training epochs, learning rate schedules, and weight decay coefficients. Each model is trained three", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 619, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 505, + 633 + ], + "score": 1.0, + "content": "times with different random seeds, and we report the average performance (classification accuracy)", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 632, + 254, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 254, + 644 + ], + "score": 1.0, + "content": "together with the standard deviation.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 39 + }, + { + "type": "title", + "bbox": [ + 108, + 656, + 243, + 667 + ], + "lines": [ + { + "bbox": [ + 106, + 656, + 244, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 244, + 668 + ], + "score": 1.0, + "content": "3.1 EXPERIMENTAL SETTINGS", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Dataset: We conduct experiments to compare the robustness of ODENets with CNN models on", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "three datasets, i.e., the MNIST (LeCun et al., 1998), the SVHN (Netzer et al., 2011), and a subset of", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "the ImageNet datset (Deng et al., 2009). We call the subset ImgNet10 since it is collected from 10", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "synsets of ImageNet: dog, bird, car, fish, monkey, turtle, lizard, bridge, cow, and crab. We selected", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 495, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 448, + 733 + ], + "score": 1.0, + "content": "3,000 training images and 300 test images from each synset and resized all images to", + "type": "text" + }, + { + "bbox": [ + 448, + 721, + 491, + 731 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 720, + 495, + 733 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 47 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 301, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 302, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 302, + 96 + ], + "score": 1.0, + "content": "2 PRELIMINARIES ON NEURAL ODE", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 105, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "score": 1.0, + "content": "It has been shown that a residual block (He et al., 2016) can be interpreted as the discrete approxima-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "score": 1.0, + "content": "tion of an ODE by setting the discretization step to be one. When the discretization step approaches", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "score": 1.0, + "content": "zero, it yields a family of neural networks, which are called neural ODEs (Chen et al., 2018). For-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 506, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 506, + 152 + ], + "score": 1.0, + "content": "mally, in a neural ODE, the relation between input and output is characterized by the following set", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 150, + 161, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 150, + 161, + 162 + ], + "score": 1.0, + "content": "of equations:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 106, + 506, + 162 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 201, + 159, + 409, + 184 + ], + "lines": [ + { + "bbox": [ + 201, + 159, + 409, + 184 + ], + "spans": [ + { + "bbox": [ + 201, + 159, + 409, + 184 + ], + "score": 0.91, + "content": "\\frac { \\mathrm { d } { \\mathbf z } ( t ) } { \\mathrm { d } t } = f _ { \\boldsymbol \\theta } ( { \\mathbf z } ( t ) , t ) , \\quad { \\mathbf z } ( 0 ) = { \\mathbf z } _ { \\mathrm { i n } } , \\quad { \\mathbf z } _ { \\mathrm { o u t } } = { \\mathbf z } ( T ) ,", + "type": "interline_equation", + "image_path": "97910f4443c6f9c371655e158097ca883dd18d4e639302bc0560ca17790de48a.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 201, + 159, + 409, + 171.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 201, + 171.5, + 409, + 184.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 185, + 504, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 134, + 200 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 187, + 239, + 199 + ], + "score": 0.92, + "content": "f _ { \\theta } : \\mathbb { R } ^ { d } \\times [ 0 , \\infty ) \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 185, + 497, + 200 + ], + "score": 1.0, + "content": "denotes the trainable layers that are parameterized by weights", + "type": "text" + }, + { + "bbox": [ + 498, + 188, + 504, + 198 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 123, + 211 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 199, + 195, + 210 + ], + "score": 0.9, + "content": "\\mathbf { z } : [ 0 , \\infty ) \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 197, + 255, + 211 + ], + "score": 1.0, + "content": "represents the", + "type": "text" + }, + { + "bbox": [ + 256, + 199, + 262, + 208 + ], + "score": 0.82, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 197, + 483, + 211 + ], + "score": 1.0, + "content": "-dimensional state of the neural ODE. We assume that", + "type": "text" + }, + { + "bbox": [ + 484, + 199, + 494, + 210 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 197, + 506, + 211 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 209, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 163, + 221 + ], + "score": 1.0, + "content": "continuous in", + "type": "text" + }, + { + "bbox": [ + 163, + 211, + 168, + 219 + ], + "score": 0.79, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 209, + 316, + 221 + ], + "score": 1.0, + "content": "and globally Lipschitz continuous in", + "type": "text" + }, + { + "bbox": [ + 317, + 212, + 323, + 219 + ], + "score": 0.52, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 209, + 414, + 221 + ], + "score": 1.0, + "content": ". In this case, the input", + "type": "text" + }, + { + "bbox": [ + 415, + 211, + 428, + 221 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { \\mathrm { i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 209, + 506, + 221 + ], + "score": 1.0, + "content": "of the neural ODE", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 219, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 213, + 234 + ], + "score": 1.0, + "content": "corresponds to the state at", + "type": "text" + }, + { + "bbox": [ + 213, + 221, + 237, + 231 + ], + "score": 0.9, + "content": "t = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 219, + 300, + 234 + ], + "score": 1.0, + "content": ", and the output", + "type": "text" + }, + { + "bbox": [ + 300, + 222, + 319, + 232 + ], + "score": 0.89, + "content": "\\mathbf { z } _ { \\mathrm { o u t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 219, + 453, + 234 + ], + "score": 1.0, + "content": "is associated to the state at some", + "type": "text" + }, + { + "bbox": [ + 453, + 221, + 501, + 232 + ], + "score": 0.93, + "content": "T \\in ( 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 219, + 506, + 234 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 231, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 143, + 244 + ], + "score": 1.0, + "content": "Because", + "type": "text" + }, + { + "bbox": [ + 144, + 231, + 154, + 243 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 231, + 372, + 244 + ], + "score": 1.0, + "content": "governs how the state changes with respect to time", + "type": "text" + }, + { + "bbox": [ + 372, + 232, + 378, + 241 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 231, + 435, + 244 + ], + "score": 1.0, + "content": ", we also use", + "type": "text" + }, + { + "bbox": [ + 435, + 232, + 446, + 243 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 231, + 505, + 244 + ], + "score": 1.0, + "content": "to denote the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 243, + 225, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 225, + 253 + ], + "score": 1.0, + "content": "dynamics of the neural ODE.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 185, + 506, + 253 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 259, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 155, + 272 + ], + "score": 1.0, + "content": "Given input", + "type": "text" + }, + { + "bbox": [ + 155, + 261, + 168, + 270 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { \\mathrm { i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 259, + 214, + 272 + ], + "score": 1.0, + "content": ", the output", + "type": "text" + }, + { + "bbox": [ + 214, + 261, + 233, + 270 + ], + "score": 0.89, + "content": "\\mathbf { z } _ { \\mathrm { o u t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 259, + 420, + 272 + ], + "score": 1.0, + "content": "can be computed by solving the ODE in (1). If", + "type": "text" + }, + { + "bbox": [ + 420, + 260, + 428, + 269 + ], + "score": 0.83, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 259, + 506, + 272 + ], + "score": 1.0, + "content": "is fixed, the output", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 270, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 124, + 282 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { \\mathrm { o u t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 270, + 229, + 283 + ], + "score": 1.0, + "content": "only depends on the input", + "type": "text" + }, + { + "bbox": [ + 230, + 272, + 243, + 281 + ], + "score": 0.89, + "content": "\\mathbf { z } _ { \\mathrm { i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 270, + 315, + 283 + ], + "score": 1.0, + "content": "and the dynamics", + "type": "text" + }, + { + "bbox": [ + 315, + 271, + 325, + 282 + ], + "score": 0.88, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 270, + 505, + 283 + ], + "score": 1.0, + "content": ", which also corresponds to the weighted lay-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 281, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 409, + 293 + ], + "score": 1.0, + "content": "ers in the neural ODE. Therefore, the neural ODE can be represented as the", + "type": "text" + }, + { + "bbox": [ + 410, + 282, + 416, + 291 + ], + "score": 0.82, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 281, + 505, + 293 + ], + "score": 1.0, + "content": "-dimensional function", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 292, + 307, + 305 + ], + "spans": [ + { + "bbox": [ + 107, + 292, + 137, + 304 + ], + "score": 0.91, + "content": "\\phi _ { T } ( \\cdot , \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 292, + 187, + 305 + ], + "score": 1.0, + "content": "of the input", + "type": "text" + }, + { + "bbox": [ + 187, + 294, + 201, + 303 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { \\mathrm { i n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 292, + 274, + 305 + ], + "score": 1.0, + "content": "and the dynamics", + "type": "text" + }, + { + "bbox": [ + 275, + 293, + 285, + 304 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 292, + 307, + 305 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 259, + 506, + 305 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 191, + 308, + 420, + 337 + ], + "lines": [ + { + "bbox": [ + 191, + 308, + 420, + 337 + ], + "spans": [ + { + "bbox": [ + 191, + 308, + 420, + 337 + ], + "score": 0.92, + "content": "\\mathbf { z } _ { \\mathrm { o u t } } = \\mathbf { z } ( T ) = \\mathbf { z } ( 0 ) + \\int _ { 0 } ^ { T } f _ { \\theta } ( \\mathbf { z } ( t ) , t ) \\mathrm { d } t = \\phi _ { T } ( \\mathbf { z } _ { \\mathrm { i n } } , f _ { \\theta } ) .", + "type": "interline_equation", + "image_path": "ee04e9b71d2db1ef8c0ab434f849a920d8934ab30d56fbcaea2c2fdc1e2122ec.jpg" + } + ] + } + ], + "index": 18.5, + "virtual_lines": [ + { + "bbox": [ + 191, + 308, + 420, + 322.5 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 191, + 322.5, + 420, + 337.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 342, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 185, + 354 + ], + "score": 1.0, + "content": "The terminal time", + "type": "text" + }, + { + "bbox": [ + 185, + 343, + 194, + 353 + ], + "score": 0.82, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 342, + 276, + 354 + ], + "score": 1.0, + "content": "of the output state", + "type": "text" + }, + { + "bbox": [ + 276, + 343, + 297, + 354 + ], + "score": 0.92, + "content": "{ \\mathbf z } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "is set to be 1 in practice. Several methods have", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 354, + 504, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 504, + 366 + ], + "score": 1.0, + "content": "been proposed for training neural ODEs, such as the adjoint sensitivity method (Chen et al., 2018),", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "score": 1.0, + "content": "SNet (Quaglino et al., 2019), and the auto-differentiation technique (Paszke et al., 2017). In", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 375, + 437, + 388 + ], + "score": 1.0, + "content": "this work, we use the most straightforward technique, i.e., updating the weights", + "type": "text" + }, + { + "bbox": [ + 437, + 376, + 443, + 385 + ], + "score": 0.78, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "with the auto-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 386, + 316, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 386, + 316, + 398 + ], + "score": 1.0, + "content": "differentiation technique in the PyTorch framework.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 342, + 505, + 398 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 414, + 424, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 413, + 426, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 426, + 428 + ], + "score": 1.0, + "content": "3 AN EMPIRICAL STUDY ON THE ROBUSTNESS OF ODENETS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 106, + 438, + 505, + 538 + ], + "lines": [ + { + "bbox": [ + 106, + 439, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 504, + 451 + ], + "score": 1.0, + "content": "Robustness of deep models has gained increased attention, as it is imperative that deep models em-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "ployed in critical applications, such as healthcare, are robust. The robustness of a model is measured", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 460, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 474 + ], + "score": 1.0, + "content": "by the sensitivity of the prediction with respect to small perturbations on the inputs. In this study,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 472, + 504, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 472, + 504, + 484 + ], + "score": 1.0, + "content": "we consider three commonly-used perturbation schemes, namely random Gaussian perturbations,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "FGSM (Goodfellow et al., 2014) adversarial examples, and PGD (Madry et al., 2017) adversarial", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 493, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 505, + 506 + ], + "score": 1.0, + "content": "examples. These perturbation schemes reflect noise and adversarial robustness properties of the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 505, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 505, + 517 + ], + "score": 1.0, + "content": "investigated models respectively. We evaluate the robustness via the classification accuracies on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 515, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 104, + 515, + 505, + 529 + ], + "score": 1.0, + "content": "perturbed images, in which the original non-perturbed versions of these images are all correctly", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 527, + 148, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 148, + 538 + ], + "score": 1.0, + "content": "classified.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 30, + "bbox_fs": [ + 104, + 439, + 505, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 543, + 505, + 642 + ], + "lines": [ + { + "bbox": [ + 106, + 544, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 556 + ], + "score": 1.0, + "content": "For a fair comparison with conventional CNN models, we made sure that the number of parameters", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "score": 1.0, + "content": "of an ODENet is close to that of its counterpart CNN model. Specifically, the ODENet shares the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 565, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 578 + ], + "score": 1.0, + "content": "same network architecture with the CNN model for the FE and FCC parts. The only difference", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 577, + 504, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 504, + 588 + ], + "score": 1.0, + "content": "is that, for the RM part, the input of the ODE-based RM is concatenated with one more channel", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 588, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 214, + 599 + ], + "score": 1.0, + "content": "which represents the time", + "type": "text" + }, + { + "bbox": [ + 214, + 588, + 220, + 597 + ], + "score": 0.58, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 588, + 506, + 599 + ], + "score": 1.0, + "content": ", while the RM in a CNN model has a skip connection and serves as", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 597, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 505, + 612 + ], + "score": 1.0, + "content": "a residual block. During the training phase, all the hyperparameters are kept the same, including", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "score": 1.0, + "content": "training epochs, learning rate schedules, and weight decay coefficients. Each model is trained three", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 619, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 505, + 633 + ], + "score": 1.0, + "content": "times with different random seeds, and we report the average performance (classification accuracy)", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 632, + 254, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 254, + 644 + ], + "score": 1.0, + "content": "together with the standard deviation.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 544, + 506, + 644 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 656, + 243, + 667 + ], + "lines": [ + { + "bbox": [ + 106, + 656, + 244, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 244, + 668 + ], + "score": 1.0, + "content": "3.1 EXPERIMENTAL SETTINGS", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Dataset: We conduct experiments to compare the robustness of ODENets with CNN models on", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "three datasets, i.e., the MNIST (LeCun et al., 1998), the SVHN (Netzer et al., 2011), and a subset of", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "the ImageNet datset (Deng et al., 2009). We call the subset ImgNet10 since it is collected from 10", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "synsets of ImageNet: dog, bird, car, fish, monkey, turtle, lizard, bridge, cow, and crab. We selected", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 495, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 448, + 733 + ], + "score": 1.0, + "content": "3,000 training images and 300 test images from each synset and resized all images to", + "type": "text" + }, + { + "bbox": [ + 448, + 721, + 491, + 731 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 720, + 495, + 733 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 47, + "bbox_fs": [ + 105, + 677, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 203 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Architectures: On the MNIST dataset, both the ODENet and the CNN model consists of four", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "convolutional layers and one fully-connected layer. The total number of parameters of the two", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 177, + 117 + ], + "score": 1.0, + "content": "models is around", + "type": "text" + }, + { + "bbox": [ + 178, + 105, + 199, + 115 + ], + "score": 0.46, + "content": "1 4 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 104, + 506, + 117 + ], + "score": 1.0, + "content": ". On the SVHN dataset, the networks are similar to those for the MNIST; we", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "only changed the input channels of the first convolutional layer to three. On the ImgNet10 dataset,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "there are nine convolutional layers and one fully-connected layer for both the ODENet and the CNN", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 328, + 150 + ], + "score": 1.0, + "content": "model. The numbers of parameters is approximately", + "type": "text" + }, + { + "bbox": [ + 328, + 137, + 349, + 148 + ], + "score": 0.36, + "content": "2 8 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 137, + 505, + 150 + ], + "score": 1.0, + "content": ". In practice, the neural ODE can be", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 147, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 147, + 505, + 160 + ], + "score": 1.0, + "content": "solved with different numerical solvers such as the Euler method and the Runge-Kutta methods", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 172 + ], + "score": 1.0, + "content": "(Chen et al., 2018). Here, we use the easily-implemented Euler method in the experiments. To", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 504, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 504, + 182 + ], + "score": 1.0, + "content": "balance the computation and the continuity of the flow, we solve the ODE initial value problem in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 181, + 505, + 193 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 505, + 193 + ], + "score": 1.0, + "content": "equation (1) by the Euler method with step size 0.1. Our implementation builds on the open-source", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 191, + 456, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 456, + 204 + ], + "score": 1.0, + "content": "neural ODE codes. 1 Details on the network architectures are included in the Appendix.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 209, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 105, + 208, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 505, + 222 + ], + "score": 1.0, + "content": "Training: The experiments are conducted using two settings on each dataset—training models only", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "with original non-perturbed images and training models on original images together with their per-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "turbed versions. In both settings, we added a weight decay term into the training objective to reg-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 254 + ], + "score": 1.0, + "content": "ularize the norm of the weights, since this can help control the model’s representation capacity and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 506, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 266 + ], + "score": 1.0, + "content": "improve the robustness of a neural network (Sokolic´ et al., 2017). In the second setting, images", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 277 + ], + "score": 1.0, + "content": "perturbed with random Gaussian noise are used to fine-tune the models, because augmenting the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 273, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 505, + 288 + ], + "score": 1.0, + "content": "dataset with small perturbations can possibly improve the robustness of models and synthesizing", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 286, + 344, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 344, + 298 + ], + "score": 1.0, + "content": "Gaussian noise does not incur excessive computation time.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 109, + 312, + 442, + 323 + ], + "lines": [ + { + "bbox": [ + 106, + 312, + 444, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 444, + 324 + ], + "score": 1.0, + "content": "3.2 ROBUSTNESS OF ODENETS TRAINED ONLY ON NON-PERTURBED IMAGES", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 333, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 345 + ], + "score": 1.0, + "content": "The first question we are interested in is how robust ODENets are against perturbations if the model", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 343, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 505, + 357 + ], + "score": 1.0, + "content": "is only trained on original non-perturbed images. We train CNNs and ODEnets to perform clas-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 354, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 505, + 368 + ], + "score": 1.0, + "content": "sification on three datasets and set the weight decay parameters for all models to be 0.0005. We", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 365, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 379 + ], + "score": 1.0, + "content": "make sure that both the well-trained ODENets and CNN models have satisfactory performances on", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 377, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 377, + 279, + 389 + ], + "score": 1.0, + "content": "original non-perturbed images, i.e., around", + "type": "text" + }, + { + "bbox": [ + 279, + 377, + 307, + 388 + ], + "score": 0.86, + "content": "9 9 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 377, + 357, + 389 + ], + "score": 1.0, + "content": "for MNIST,", + "type": "text" + }, + { + "bbox": [ + 357, + 377, + 384, + 388 + ], + "score": 0.89, + "content": "9 5 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 377, + 462, + 389 + ], + "score": 1.0, + "content": "for the SVHN, and", + "type": "text" + }, + { + "bbox": [ + 463, + 377, + 490, + 388 + ], + "score": 0.88, + "content": "8 0 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 377, + 506, + 389 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 387, + 152, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 152, + 401 + ], + "score": 1.0, + "content": "ImgNet10.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 504, + 470 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 506, + 417 + ], + "score": 1.0, + "content": "Since Gaussian noise is ubiquitous in modeling image degradation, we first evaluated the robustness", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "of the models in the presence of zero-mean random Gaussian perturbations. It has also been shown", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 427, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 505, + 438 + ], + "score": 1.0, + "content": "that a deep model is vulnerable to harmful adversarial examples, such as the FGSM (Goodfellow", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "score": 1.0, + "content": "et al., 2014). We are also interested in how robust ODENets are in the presence of adversarial", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 448, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 245, + 461 + ], + "score": 1.0, + "content": "examples. The standard deviation", + "type": "text" + }, + { + "bbox": [ + 246, + 451, + 253, + 459 + ], + "score": 0.78, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 448, + 362, + 461 + ], + "score": 1.0, + "content": "of Gaussian noise and the", + "type": "text" + }, + { + "bbox": [ + 362, + 449, + 374, + 460 + ], + "score": 0.88, + "content": "l _ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 448, + 400, + 461 + ], + "score": 1.0, + "content": "-norm", + "type": "text" + }, + { + "bbox": [ + 401, + 451, + 406, + 459 + ], + "score": 0.51, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 448, + 506, + 461 + ], + "score": 1.0, + "content": "of the FGSM attack for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 460, + 245, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 245, + 471 + ], + "score": 1.0, + "content": "each dataset are shown in Table 1.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + }, + { + "type": "table", + "bbox": [ + 117, + 546, + 491, + 674 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 480, + 504, + 536 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "Table 1: Robustness comparison of different models. We report their mean classification accuracies", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 107, + 492, + 123, + 502 + ], + "score": 0.8, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 491, + 250, + 504 + ], + "score": 1.0, + "content": "and standard deviations (mean", + "type": "text" + }, + { + "bbox": [ + 250, + 492, + 261, + 502 + ], + "score": 0.52, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 491, + 506, + 504 + ], + "score": 1.0, + "content": "std) on perturbed images from the MNIST, the SVHN, and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 502, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 505, + 514 + ], + "score": 1.0, + "content": "the ImgNet10 datasets. Two types of perturbations are used—zero-mean Gaussian noise and FGSM", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 513, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 506, + 526 + ], + "score": 1.0, + "content": "adversarial attack. The results show that ODENets are much more robust in comparison to CNN", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 523, + 140, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 140, + 536 + ], + "score": 1.0, + "content": "models.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34 + }, + { + "type": "table_body", + "bbox": [ + 117, + 546, + 491, + 674 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 546, + 491, + 674 + ], + "spans": [ + { + "bbox": [ + 117, + 546, + 491, + 674 + ], + "score": 0.986, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=50σ=75σ=100FGSM-0.15FGSM-0.3FGSM-0.5
CNN98.1±0.785.8±4.356.4±5.663.4±2.324.0±8.98.3±3.2
ODENet98.7±0.690.6±5.473.2±8.683.5±0.942.1±2.414.3±2.1
SVHNg=15g=25σ=35FGSM-3/255FGSM-5/255FGSM-8/255
CNN90.0±1.276.3±2.760.9±3.929.2±2.913.7±1.95.4±1.5
ODENet95.7±0.788.1±1.578.2±2.158.2±2.343.0±1.330.9±1.4
ImgNet10σ=10σ=15σ = 25FGSM-5/255FGSM-8/255FGSM-16/255
CNN80.1±1.863.3±2.040.8±2.728.5±0.518.1±0.79.4±1.2
ODENet81.9±2.067.5±2.048.7±2.636.2±1.027.2±1.114.4±1.7
", + "type": "table", + "image_path": "5c76587be27929f6a5c670ff5d42c94d3fd7b15e85b1409c548bdb5f3efde2a1.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 117, + 546, + 491, + 588.6666666666666 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 117, + 588.6666666666666, + 491, + 631.3333333333333 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 117, + 631.3333333333333, + 491, + 673.9999999999999 + ], + "spans": [], + "index": 39 + } + ] + } + ], + "index": 36.0 + }, + { + "type": "text", + "bbox": [ + 107, + 688, + 504, + 711 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 702 + ], + "score": 1.0, + "content": "From the results in Table 1, we observe that the ODENets demonstrate superior robustness com-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "pared to CNNs for all types of perturbations. On the MNIST dataset, in the presence of Gaussian", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 722, + 335, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 720, + 337, + 733 + ], + "spans": [ + { + "bbox": [ + 119, + 720, + 337, + 733 + ], + "score": 1.0, + "content": "1 https://github.com/rtqichen/torchdiffeq.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 203 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Architectures: On the MNIST dataset, both the ODENet and the CNN model consists of four", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "convolutional layers and one fully-connected layer. The total number of parameters of the two", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 177, + 117 + ], + "score": 1.0, + "content": "models is around", + "type": "text" + }, + { + "bbox": [ + 178, + 105, + 199, + 115 + ], + "score": 0.46, + "content": "1 4 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 104, + 506, + 117 + ], + "score": 1.0, + "content": ". On the SVHN dataset, the networks are similar to those for the MNIST; we", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "only changed the input channels of the first convolutional layer to three. On the ImgNet10 dataset,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "there are nine convolutional layers and one fully-connected layer for both the ODENet and the CNN", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 328, + 150 + ], + "score": 1.0, + "content": "model. The numbers of parameters is approximately", + "type": "text" + }, + { + "bbox": [ + 328, + 137, + 349, + 148 + ], + "score": 0.36, + "content": "2 8 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 137, + 505, + 150 + ], + "score": 1.0, + "content": ". In practice, the neural ODE can be", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 147, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 147, + 505, + 160 + ], + "score": 1.0, + "content": "solved with different numerical solvers such as the Euler method and the Runge-Kutta methods", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 172 + ], + "score": 1.0, + "content": "(Chen et al., 2018). Here, we use the easily-implemented Euler method in the experiments. To", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 504, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 504, + 182 + ], + "score": 1.0, + "content": "balance the computation and the continuity of the flow, we solve the ODE initial value problem in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 181, + 505, + 193 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 505, + 193 + ], + "score": 1.0, + "content": "equation (1) by the Euler method with step size 0.1. Our implementation builds on the open-source", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 191, + 456, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 456, + 204 + ], + "score": 1.0, + "content": "neural ODE codes. 1 Details on the network architectures are included in the Appendix.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 82, + 506, + 204 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 209, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 105, + 208, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 505, + 222 + ], + "score": 1.0, + "content": "Training: The experiments are conducted using two settings on each dataset—training models only", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "with original non-perturbed images and training models on original images together with their per-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "turbed versions. In both settings, we added a weight decay term into the training objective to reg-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 254 + ], + "score": 1.0, + "content": "ularize the norm of the weights, since this can help control the model’s representation capacity and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 506, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 266 + ], + "score": 1.0, + "content": "improve the robustness of a neural network (Sokolic´ et al., 2017). In the second setting, images", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 277 + ], + "score": 1.0, + "content": "perturbed with random Gaussian noise are used to fine-tune the models, because augmenting the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 273, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 505, + 288 + ], + "score": 1.0, + "content": "dataset with small perturbations can possibly improve the robustness of models and synthesizing", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 286, + 344, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 344, + 298 + ], + "score": 1.0, + "content": "Gaussian noise does not incur excessive computation time.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 208, + 506, + 298 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 312, + 442, + 323 + ], + "lines": [ + { + "bbox": [ + 106, + 312, + 444, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 444, + 324 + ], + "score": 1.0, + "content": "3.2 ROBUSTNESS OF ODENETS TRAINED ONLY ON NON-PERTURBED IMAGES", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 333, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 345 + ], + "score": 1.0, + "content": "The first question we are interested in is how robust ODENets are against perturbations if the model", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 343, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 505, + 357 + ], + "score": 1.0, + "content": "is only trained on original non-perturbed images. We train CNNs and ODEnets to perform clas-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 354, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 505, + 368 + ], + "score": 1.0, + "content": "sification on three datasets and set the weight decay parameters for all models to be 0.0005. We", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 365, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 379 + ], + "score": 1.0, + "content": "make sure that both the well-trained ODENets and CNN models have satisfactory performances on", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 377, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 377, + 279, + 389 + ], + "score": 1.0, + "content": "original non-perturbed images, i.e., around", + "type": "text" + }, + { + "bbox": [ + 279, + 377, + 307, + 388 + ], + "score": 0.86, + "content": "9 9 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 377, + 357, + 389 + ], + "score": 1.0, + "content": "for MNIST,", + "type": "text" + }, + { + "bbox": [ + 357, + 377, + 384, + 388 + ], + "score": 0.89, + "content": "9 5 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 377, + 462, + 389 + ], + "score": 1.0, + "content": "for the SVHN, and", + "type": "text" + }, + { + "bbox": [ + 463, + 377, + 490, + 388 + ], + "score": 0.88, + "content": "8 0 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 377, + 506, + 389 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 387, + 152, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 152, + 401 + ], + "score": 1.0, + "content": "ImgNet10.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5, + "bbox_fs": [ + 104, + 333, + 506, + 401 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 504, + 470 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 506, + 417 + ], + "score": 1.0, + "content": "Since Gaussian noise is ubiquitous in modeling image degradation, we first evaluated the robustness", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "of the models in the presence of zero-mean random Gaussian perturbations. It has also been shown", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 427, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 505, + 438 + ], + "score": 1.0, + "content": "that a deep model is vulnerable to harmful adversarial examples, such as the FGSM (Goodfellow", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "score": 1.0, + "content": "et al., 2014). We are also interested in how robust ODENets are in the presence of adversarial", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 448, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 245, + 461 + ], + "score": 1.0, + "content": "examples. The standard deviation", + "type": "text" + }, + { + "bbox": [ + 246, + 451, + 253, + 459 + ], + "score": 0.78, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 448, + 362, + 461 + ], + "score": 1.0, + "content": "of Gaussian noise and the", + "type": "text" + }, + { + "bbox": [ + 362, + 449, + 374, + 460 + ], + "score": 0.88, + "content": "l _ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 448, + 400, + 461 + ], + "score": 1.0, + "content": "-norm", + "type": "text" + }, + { + "bbox": [ + 401, + 451, + 406, + 459 + ], + "score": 0.51, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 448, + 506, + 461 + ], + "score": 1.0, + "content": "of the FGSM attack for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 460, + 245, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 245, + 471 + ], + "score": 1.0, + "content": "each dataset are shown in Table 1.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 405, + 506, + 471 + ] + }, + { + "type": "table", + "bbox": [ + 117, + 546, + 491, + 674 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 480, + 504, + 536 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "Table 1: Robustness comparison of different models. We report their mean classification accuracies", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 107, + 492, + 123, + 502 + ], + "score": 0.8, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 491, + 250, + 504 + ], + "score": 1.0, + "content": "and standard deviations (mean", + "type": "text" + }, + { + "bbox": [ + 250, + 492, + 261, + 502 + ], + "score": 0.52, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 491, + 506, + 504 + ], + "score": 1.0, + "content": "std) on perturbed images from the MNIST, the SVHN, and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 502, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 505, + 514 + ], + "score": 1.0, + "content": "the ImgNet10 datasets. Two types of perturbations are used—zero-mean Gaussian noise and FGSM", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 513, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 506, + 526 + ], + "score": 1.0, + "content": "adversarial attack. The results show that ODENets are much more robust in comparison to CNN", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 523, + 140, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 140, + 536 + ], + "score": 1.0, + "content": "models.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34 + }, + { + "type": "table_body", + "bbox": [ + 117, + 546, + 491, + 674 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 546, + 491, + 674 + ], + "spans": [ + { + "bbox": [ + 117, + 546, + 491, + 674 + ], + "score": 0.986, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=50σ=75σ=100FGSM-0.15FGSM-0.3FGSM-0.5
CNN98.1±0.785.8±4.356.4±5.663.4±2.324.0±8.98.3±3.2
ODENet98.7±0.690.6±5.473.2±8.683.5±0.942.1±2.414.3±2.1
SVHNg=15g=25σ=35FGSM-3/255FGSM-5/255FGSM-8/255
CNN90.0±1.276.3±2.760.9±3.929.2±2.913.7±1.95.4±1.5
ODENet95.7±0.788.1±1.578.2±2.158.2±2.343.0±1.330.9±1.4
ImgNet10σ=10σ=15σ = 25FGSM-5/255FGSM-8/255FGSM-16/255
CNN80.1±1.863.3±2.040.8±2.728.5±0.518.1±0.79.4±1.2
ODENet81.9±2.067.5±2.048.7±2.636.2±1.027.2±1.114.4±1.7
", + "type": "table", + "image_path": "5c76587be27929f6a5c670ff5d42c94d3fd7b15e85b1409c548bdb5f3efde2a1.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 117, + 546, + 491, + 588.6666666666666 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 117, + 588.6666666666666, + 491, + 631.3333333333333 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 117, + 631.3333333333333, + 491, + 673.9999999999999 + ], + "spans": [], + "index": 39 + } + ] + } + ], + "index": 36.0 + }, + { + "type": "text", + "bbox": [ + 107, + 688, + 504, + 711 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 702 + ], + "score": 1.0, + "content": "From the results in Table 1, we observe that the ODENets demonstrate superior robustness com-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "pared to CNNs for all types of perturbations. On the MNIST dataset, in the presence of Gaussian", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 212, + 95 + ], + "score": 1.0, + "content": "perturbations with a large", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 212, + 84, + 219, + 92 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 220, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "of 100, the ODENet produces much higher accuracy on perturbed im-", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 254, + 106 + ], + "score": 1.0, + "content": "ages compared to the CNN model (", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 254, + 94, + 281, + 105 + ], + "score": 0.83, + "content": "7 3 . 2 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 282, + 93, + 298, + 106 + ], + "score": 1.0, + "content": "vs.", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 298, + 94, + 326, + 105 + ], + "score": 0.85, + "content": "5 6 . 4 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 326, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "). For the FGSM-0.3 adversarial examples,", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "the accuracy of ONEnet is around twice as high as that of the CNN model. On the SVHN dataset,", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 129 + ], + "score": 1.0, + "content": "ODENets significantly outperform CNN models, e.g., for the FGSM-5/255 examples, the accuracy", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 183, + 139 + ], + "score": 1.0, + "content": "of the ODENet is", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 184, + 127, + 211, + 137 + ], + "score": 0.88, + "content": "4 3 . 0 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 211, + 126, + 433, + 139 + ], + "score": 1.0, + "content": ", which is much higher than that of the CNN model", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 433, + 127, + 465, + 138 + ], + "score": 0.86, + "content": "( 1 3 . 7 \\% )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 466, + 126, + 505, + 139 + ], + "score": 1.0, + "content": ". On the", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 226, + 151 + ], + "score": 1.0, + "content": "ImgNet10, for both cases of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 227, + 137, + 262, + 147 + ], + "score": 0.9, + "content": "\\sigma = 2 5", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 262, + 136, + 505, + 151 + ], + "score": 1.0, + "content": "and FGSM-8/255, ODENet outperforms CNNs by a large", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 149, + 197, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 177, + 160 + ], + "score": 1.0, + "content": "margin of around", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 178, + 149, + 192, + 159 + ], + "score": 0.86, + "content": "9 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 192, + 149, + 197, + 160 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 6 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 687, + 506, + 713 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 159 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 212, + 95 + ], + "score": 1.0, + "content": "perturbations with a large", + "type": "text" + }, + { + "bbox": [ + 212, + 84, + 219, + 92 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "of 100, the ODENet produces much higher accuracy on perturbed im-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 254, + 106 + ], + "score": 1.0, + "content": "ages compared to the CNN model (", + "type": "text" + }, + { + "bbox": [ + 254, + 94, + 281, + 105 + ], + "score": 0.83, + "content": "7 3 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 93, + 298, + 106 + ], + "score": 1.0, + "content": "vs.", + "type": "text" + }, + { + "bbox": [ + 298, + 94, + 326, + 105 + ], + "score": 0.85, + "content": "5 6 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "). For the FGSM-0.3 adversarial examples,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "the accuracy of ONEnet is around twice as high as that of the CNN model. On the SVHN dataset,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 129 + ], + "score": 1.0, + "content": "ODENets significantly outperform CNN models, e.g., for the FGSM-5/255 examples, the accuracy", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 183, + 139 + ], + "score": 1.0, + "content": "of the ODENet is", + "type": "text" + }, + { + "bbox": [ + 184, + 127, + 211, + 137 + ], + "score": 0.88, + "content": "4 3 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 126, + 433, + 139 + ], + "score": 1.0, + "content": ", which is much higher than that of the CNN model", + "type": "text" + }, + { + "bbox": [ + 433, + 127, + 465, + 138 + ], + "score": 0.86, + "content": "( 1 3 . 7 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 126, + 505, + 139 + ], + "score": 1.0, + "content": ". On the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 226, + 151 + ], + "score": 1.0, + "content": "ImgNet10, for both cases of", + "type": "text" + }, + { + "bbox": [ + 227, + 137, + 262, + 147 + ], + "score": 0.9, + "content": "\\sigma = 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 136, + 505, + 151 + ], + "score": 1.0, + "content": "and FGSM-8/255, ODENet outperforms CNNs by a large", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 149, + 197, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 177, + 160 + ], + "score": 1.0, + "content": "margin of around", + "type": "text" + }, + { + "bbox": [ + 178, + 149, + 192, + 159 + ], + "score": 0.86, + "content": "9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 149, + 197, + 160 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 108, + 174, + 459, + 196 + ], + "lines": [ + { + "bbox": [ + 106, + 174, + 461, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 124, + 187 + ], + "score": 1.0, + "content": "3.3", + "type": "text" + }, + { + "bbox": [ + 129, + 175, + 461, + 186 + ], + "score": 1.0, + "content": "ROBUSTNESS OF ODENETS TRAINED ON ORIGINAL IMAGES TOGETHER WITH", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 185, + 251, + 197 + ], + "spans": [ + { + "bbox": [ + 130, + 185, + 251, + 197 + ], + "score": 1.0, + "content": "GAUSSIAN PERTURBATIONS", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 107, + 206, + 505, + 294 + ], + "lines": [ + { + "bbox": [ + 105, + 205, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 506, + 219 + ], + "score": 1.0, + "content": "Training a model on original images together with their perturbed versions can improve the robust-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "ness of the model. As mentioned previously, Gaussian noise is commonly assumed to be present", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 229, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 505, + 241 + ], + "score": 1.0, + "content": "in real-world images. Synthesizing Gaussian noise is also fast and easy. Thus, we add random", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "score": 1.0, + "content": "Gaussian noise into the original images to generate their perturbed versions. ODENets and CNN", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 251, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 505, + 263 + ], + "score": 1.0, + "content": "models are both trained on original images together with their perturbed versions. The standard de-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 355, + 273 + ], + "score": 1.0, + "content": "viation of the added Gaussian noise is randomly chosen from", + "type": "text" + }, + { + "bbox": [ + 355, + 261, + 410, + 273 + ], + "score": 0.89, + "content": "\\bar { \\{ 5 0 , 7 5 , 1 0 0 \\} }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 262, + 505, + 273 + ], + "score": 1.0, + "content": "on the MNIST dataset,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 272, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 107, + 272, + 156, + 285 + ], + "score": 0.83, + "content": "\\{ 1 5 , 2 5 , 3 5 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 272, + 268, + 285 + ], + "score": 1.0, + "content": "on the SVHN dataset, and", + "type": "text" + }, + { + "bbox": [ + 269, + 272, + 318, + 285 + ], + "score": 0.88, + "content": "\\{ 1 0 , 1 5 , 2 5 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 272, + 505, + 285 + ], + "score": 1.0, + "content": "on the ImgNet10. All other hyperparameters", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 284, + 218, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 218, + 295 + ], + "score": 1.0, + "content": "are kept the same as above.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5 + }, + { + "type": "table", + "bbox": [ + 136, + 371, + 473, + 496 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 306, + 504, + 361 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 307, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 318 + ], + "score": 1.0, + "content": "Table 2: Robustness comparison of different models. We report their mean classification accuracies", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 107, + 318, + 123, + 329 + ], + "score": 0.8, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 316, + 250, + 330 + ], + "score": 1.0, + "content": "and standard deviations (mean", + "type": "text" + }, + { + "bbox": [ + 251, + 318, + 261, + 328 + ], + "score": 0.46, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "std) on perturbed images from the MNIST, the SVHN, and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "the ImgNet10 datsets. Three types of perturbations are used—zero-mean Gaussian noise, FGSM", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 339, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 506, + 351 + ], + "score": 1.0, + "content": "adversarial attack, and PGD adversarial attack. The results show that ODENets are more robust", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 351, + 216, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 216, + 363 + ], + "score": 1.0, + "content": "compared to CNN models.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + }, + { + "type": "table_body", + "bbox": [ + 136, + 371, + 473, + 496 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 136, + 371, + 473, + 496 + ], + "spans": [ + { + "bbox": [ + 136, + 371, + 473, + 496 + ], + "score": 0.962, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
ImgNet10σ= 25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
", + "type": "table", + "image_path": "c863f23ea70d5ead2741c14baf42496133f48535a60953861fb5e41251daade3.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 136, + 371, + 473, + 412.6666666666667 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 136, + 412.6666666666667, + 473, + 454.33333333333337 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 136, + 454.33333333333337, + 473, + 496.00000000000006 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 21.0 + }, + { + "type": "text", + "bbox": [ + 106, + 508, + 505, + 640 + ], + "lines": [ + { + "bbox": [ + 106, + 509, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 505, + 521 + ], + "score": 1.0, + "content": "The robustness of the models is evaluated under Gaussian perturbations, FGSM adversarial exam-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "score": 1.0, + "content": "ples, and PGD (Madry et al., 2017) adversarial examples. The latter is a stronger attacker compared", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 531, + 504, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 182, + 543 + ], + "score": 1.0, + "content": "to the FGSM. The", + "type": "text" + }, + { + "bbox": [ + 182, + 531, + 195, + 542 + ], + "score": 0.89, + "content": "l _ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 531, + 220, + 543 + ], + "score": 1.0, + "content": "-norm", + "type": "text" + }, + { + "bbox": [ + 220, + 533, + 226, + 541 + ], + "score": 0.6, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 531, + 504, + 543 + ], + "score": 1.0, + "content": "of the PGD attack for each dataset is shown in Table 2. Based on the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "results, we observe that ODENets consistently outperform CNN models on both two datasets. On", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "the MNIST dataset, the ODENet outperforms the CNN against all types of perturbations. In par-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 408, + 576 + ], + "score": 1.0, + "content": "ticular, for the PGD-0.2 adversarial examples, the accuracy of the ODENet", + "type": "text" + }, + { + "bbox": [ + 409, + 564, + 442, + 575 + ], + "score": 0.88, + "content": "( 6 4 . 7 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "is much higher", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 198, + 587 + ], + "score": 1.0, + "content": "than that of the CNN", + "type": "text" + }, + { + "bbox": [ + 199, + 575, + 231, + 586 + ], + "score": 0.87, + "content": "( 3 2 . 9 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 575, + 505, + 587 + ], + "score": 1.0, + "content": ". Besides, for the PGD-0.3 attack, the CNN is completely misled", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 585, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 598 + ], + "score": 1.0, + "content": "by the adversarial examples, but the ODENet can still classify perturbed images with an accuracy", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 596, + 118, + 608 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 596, + 145, + 607 + ], + "score": 0.86, + "content": "1 3 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 596, + 505, + 608 + ], + "score": 1.0, + "content": ". On the SVHN dataset, ODENets also show superior robustness in comparison to CNN", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 608, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 506, + 620 + ], + "score": 1.0, + "content": "models. For all the adversarial examples, ODENets outperform CNN models by a margin of at least", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 618, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 506, + 632 + ], + "score": 1.0, + "content": "10 percentage points. On the ImgNet10 dataset, the ODENet also performs better than CNN models", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 629, + 275, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 275, + 642 + ], + "score": 1.0, + "content": "against all forms of adversarial examples.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30.5 + }, + { + "type": "title", + "bbox": [ + 108, + 655, + 326, + 667 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 328, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 328, + 668 + ], + "score": 1.0, + "content": "3.4 INSIGHTS ON THE ROBUSTNESS OF ODENETS", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "From the results in Sections 3.2 and 3.3, we find ODENets are more robust compared to CNN", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "models. Here, we attempt to provide an intuitive understanding of the robustness of the neural ODE.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "In an ODENet, given some datum, the FE extracts an informative feature map from the datum. The", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "neural ODE, serving as the RM, takes as input the feature map and performs a nonlinear mapping. In", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "practice, we use the weight decay technique during training which regularizes the norm of weights", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 159 + ], + "lines": [], + "index": 3, + "bbox_fs": [ + 105, + 82, + 505, + 160 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 174, + 459, + 196 + ], + "lines": [ + { + "bbox": [ + 106, + 174, + 461, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 124, + 187 + ], + "score": 1.0, + "content": "3.3", + "type": "text" + }, + { + "bbox": [ + 129, + 175, + 461, + 186 + ], + "score": 1.0, + "content": "ROBUSTNESS OF ODENETS TRAINED ON ORIGINAL IMAGES TOGETHER WITH", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 185, + 251, + 197 + ], + "spans": [ + { + "bbox": [ + 130, + 185, + 251, + 197 + ], + "score": 1.0, + "content": "GAUSSIAN PERTURBATIONS", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 107, + 206, + 505, + 294 + ], + "lines": [ + { + "bbox": [ + 105, + 205, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 506, + 219 + ], + "score": 1.0, + "content": "Training a model on original images together with their perturbed versions can improve the robust-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "ness of the model. As mentioned previously, Gaussian noise is commonly assumed to be present", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 229, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 505, + 241 + ], + "score": 1.0, + "content": "in real-world images. Synthesizing Gaussian noise is also fast and easy. Thus, we add random", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "score": 1.0, + "content": "Gaussian noise into the original images to generate their perturbed versions. ODENets and CNN", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 251, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 505, + 263 + ], + "score": 1.0, + "content": "models are both trained on original images together with their perturbed versions. The standard de-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 355, + 273 + ], + "score": 1.0, + "content": "viation of the added Gaussian noise is randomly chosen from", + "type": "text" + }, + { + "bbox": [ + 355, + 261, + 410, + 273 + ], + "score": 0.89, + "content": "\\bar { \\{ 5 0 , 7 5 , 1 0 0 \\} }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 262, + 505, + 273 + ], + "score": 1.0, + "content": "on the MNIST dataset,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 272, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 107, + 272, + 156, + 285 + ], + "score": 0.83, + "content": "\\{ 1 5 , 2 5 , 3 5 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 272, + 268, + 285 + ], + "score": 1.0, + "content": "on the SVHN dataset, and", + "type": "text" + }, + { + "bbox": [ + 269, + 272, + 318, + 285 + ], + "score": 0.88, + "content": "\\{ 1 0 , 1 5 , 2 5 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 272, + 505, + 285 + ], + "score": 1.0, + "content": "on the ImgNet10. All other hyperparameters", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 284, + 218, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 218, + 295 + ], + "score": 1.0, + "content": "are kept the same as above.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 205, + 506, + 295 + ] + }, + { + "type": "table", + "bbox": [ + 136, + 371, + 473, + 496 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 306, + 504, + 361 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 307, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 318 + ], + "score": 1.0, + "content": "Table 2: Robustness comparison of different models. We report their mean classification accuracies", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 107, + 318, + 123, + 329 + ], + "score": 0.8, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 316, + 250, + 330 + ], + "score": 1.0, + "content": "and standard deviations (mean", + "type": "text" + }, + { + "bbox": [ + 251, + 318, + 261, + 328 + ], + "score": 0.46, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "std) on perturbed images from the MNIST, the SVHN, and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "the ImgNet10 datsets. Three types of perturbations are used—zero-mean Gaussian noise, FGSM", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 339, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 506, + 351 + ], + "score": 1.0, + "content": "adversarial attack, and PGD adversarial attack. The results show that ODENets are more robust", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 351, + 216, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 216, + 363 + ], + "score": 1.0, + "content": "compared to CNN models.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + }, + { + "type": "table_body", + "bbox": [ + 136, + 371, + 473, + 496 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 136, + 371, + 473, + 496 + ], + "spans": [ + { + "bbox": [ + 136, + 371, + 473, + 496 + ], + "score": 0.962, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
ImgNet10σ= 25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
", + "type": "table", + "image_path": "c863f23ea70d5ead2741c14baf42496133f48535a60953861fb5e41251daade3.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 136, + 371, + 473, + 412.6666666666667 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 136, + 412.6666666666667, + 473, + 454.33333333333337 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 136, + 454.33333333333337, + 473, + 496.00000000000006 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 21.0 + }, + { + "type": "text", + "bbox": [ + 106, + 508, + 505, + 640 + ], + "lines": [ + { + "bbox": [ + 106, + 509, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 505, + 521 + ], + "score": 1.0, + "content": "The robustness of the models is evaluated under Gaussian perturbations, FGSM adversarial exam-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "score": 1.0, + "content": "ples, and PGD (Madry et al., 2017) adversarial examples. The latter is a stronger attacker compared", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 531, + 504, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 182, + 543 + ], + "score": 1.0, + "content": "to the FGSM. The", + "type": "text" + }, + { + "bbox": [ + 182, + 531, + 195, + 542 + ], + "score": 0.89, + "content": "l _ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 531, + 220, + 543 + ], + "score": 1.0, + "content": "-norm", + "type": "text" + }, + { + "bbox": [ + 220, + 533, + 226, + 541 + ], + "score": 0.6, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 531, + 504, + 543 + ], + "score": 1.0, + "content": "of the PGD attack for each dataset is shown in Table 2. Based on the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "results, we observe that ODENets consistently outperform CNN models on both two datasets. On", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "the MNIST dataset, the ODENet outperforms the CNN against all types of perturbations. In par-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 408, + 576 + ], + "score": 1.0, + "content": "ticular, for the PGD-0.2 adversarial examples, the accuracy of the ODENet", + "type": "text" + }, + { + "bbox": [ + 409, + 564, + 442, + 575 + ], + "score": 0.88, + "content": "( 6 4 . 7 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "is much higher", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 198, + 587 + ], + "score": 1.0, + "content": "than that of the CNN", + "type": "text" + }, + { + "bbox": [ + 199, + 575, + 231, + 586 + ], + "score": 0.87, + "content": "( 3 2 . 9 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 575, + 505, + 587 + ], + "score": 1.0, + "content": ". Besides, for the PGD-0.3 attack, the CNN is completely misled", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 585, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 598 + ], + "score": 1.0, + "content": "by the adversarial examples, but the ODENet can still classify perturbed images with an accuracy", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 596, + 118, + 608 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 596, + 145, + 607 + ], + "score": 0.86, + "content": "1 3 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 596, + 505, + 608 + ], + "score": 1.0, + "content": ". On the SVHN dataset, ODENets also show superior robustness in comparison to CNN", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 608, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 506, + 620 + ], + "score": 1.0, + "content": "models. For all the adversarial examples, ODENets outperform CNN models by a margin of at least", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 618, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 506, + 632 + ], + "score": 1.0, + "content": "10 percentage points. On the ImgNet10 dataset, the ODENet also performs better than CNN models", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 629, + 275, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 275, + 642 + ], + "score": 1.0, + "content": "against all forms of adversarial examples.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 509, + 506, + 642 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 655, + 326, + 667 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 328, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 328, + 668 + ], + "score": 1.0, + "content": "3.4 INSIGHTS ON THE ROBUSTNESS OF ODENETS", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "From the results in Sections 3.2 and 3.3, we find ODENets are more robust compared to CNN", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "models. Here, we attempt to provide an intuitive understanding of the robustness of the neural ODE.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "In an ODENet, given some datum, the FE extracts an informative feature map from the datum. The", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "neural ODE, serving as the RM, takes as input the feature map and performs a nonlinear mapping. In", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "practice, we use the weight decay technique during training which regularizes the norm of weights", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "in the FE part, so that the change of feature map in terms of a small perturbation on the input can be", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "controlled. We aim to show that, in the neural ODE, a small change on the feature map will not lead", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 414, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 414, + 118 + ], + "score": 1.0, + "content": "to a large deviation from the original output associated with the feature map.", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 677, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "in the FE part, so that the change of feature map in terms of a small perturbation on the input can be", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "controlled. We aim to show that, in the neural ODE, a small change on the feature map will not lead", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 414, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 414, + 118 + ], + "score": 1.0, + "content": "to a large deviation from the original output associated with the feature map.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 121, + 505, + 167 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "score": 1.0, + "content": "Theorem 1 (ODE integral curves do not intersect (Coddington & Levinson, 1955; Younes, 2010;", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 214, + 144 + ], + "score": 1.0, + "content": "Dupont et al., 2019)). Let", + "type": "text" + }, + { + "bbox": [ + 214, + 132, + 236, + 144 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 132, + 255, + 144 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 256, + 132, + 278, + 144 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 2 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 132, + 505, + 144 + ], + "score": 1.0, + "content": "be two solutions of the ODE in (1) with different initial", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 168, + 156 + ], + "score": 1.0, + "content": "conditions, i.e.", + "type": "text" + }, + { + "bbox": [ + 169, + 144, + 228, + 155 + ], + "score": 0.9, + "content": "{ \\bf z } _ { 1 } ( 0 ) \\neq { \\bf z } _ { 2 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 143, + 262, + 156 + ], + "score": 1.0, + "content": ". In (1),", + "type": "text" + }, + { + "bbox": [ + 262, + 144, + 272, + 155 + ], + "score": 0.72, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "is continuous in t and globally Lipschitz continuous in z.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 153, + 313, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 181, + 167 + ], + "score": 1.0, + "content": "Then, it holds that", + "type": "text" + }, + { + "bbox": [ + 182, + 155, + 237, + 167 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( t ) \\neq { \\bf z } _ { 2 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 153, + 266, + 167 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 266, + 155, + 309, + 167 + ], + "score": 0.93, + "content": "t \\in [ 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 153, + 313, + 167 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 177, + 336, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 177, + 336, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 336, + 189 + ], + "score": 1.0, + "content": "To illustrate this theorem, considering a simple 1-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 188, + 337, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 337, + 200 + ], + "score": 1.0, + "content": "dimensional system in which the state is a scalar. As", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 199, + 336, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 291, + 211 + ], + "score": 1.0, + "content": "shown in Figure 2, equation (1) has a solution", + "type": "text" + }, + { + "bbox": [ + 291, + 199, + 312, + 212 + ], + "score": 0.92, + "content": "z _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 199, + 336, + 211 + ], + "score": 1.0, + "content": "start-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 210, + 338, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 145, + 222 + ], + "score": 1.0, + "content": "ing from", + "type": "text" + }, + { + "bbox": [ + 145, + 210, + 212, + 222 + ], + "score": 0.93, + "content": "\\bar { A _ { 1 } } = ( 0 , \\bar { z _ { 1 } } ( 0 ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 210, + 244, + 222 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 245, + 210, + 268, + 222 + ], + "score": 0.92, + "content": "z _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 210, + 338, + 222 + ], + "score": 1.0, + "content": "is the feature of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 221, + 337, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 337, + 233 + ], + "score": 1.0, + "content": "some datum. Equation (1) also has another two solutions", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 231, + 334, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 128, + 244 + ], + "score": 0.91, + "content": "z _ { 2 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 231, + 148, + 245 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 149, + 232, + 170, + 244 + ], + "score": 0.89, + "content": "z _ { 3 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 231, + 266, + 245 + ], + "score": 1.0, + "content": ", whose starting points", + "type": "text" + }, + { + "bbox": [ + 266, + 232, + 334, + 244 + ], + "score": 0.91, + "content": "A _ { 2 } = ( 0 , z _ { 2 } ( 0 ) )", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 241, + 338, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 125, + 256 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 125, + 244, + 196, + 255 + ], + "score": 0.92, + "content": "A _ { 3 } ~ = ~ ( 0 , z _ { 3 } ( 0 ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 241, + 319, + 256 + ], + "score": 1.0, + "content": ", both of which are close to", + "type": "text" + }, + { + "bbox": [ + 320, + 244, + 333, + 254 + ], + "score": 0.87, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 241, + 338, + 256 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 254, + 337, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 143, + 266 + ], + "score": 1.0, + "content": "Suppose", + "type": "text" + }, + { + "bbox": [ + 144, + 255, + 156, + 265 + ], + "score": 0.88, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 254, + 205, + 266 + ], + "score": 1.0, + "content": "is between", + "type": "text" + }, + { + "bbox": [ + 205, + 255, + 218, + 265 + ], + "score": 0.88, + "content": "A _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 254, + 238, + 266 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 238, + 254, + 251, + 265 + ], + "score": 0.87, + "content": "A _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 254, + 337, + 266 + ], + "score": 1.0, + "content": ". By Theorem 1, we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 264, + 336, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 264, + 223, + 277 + ], + "score": 1.0, + "content": "know that the integral curve", + "type": "text" + }, + { + "bbox": [ + 223, + 265, + 245, + 277 + ], + "score": 0.91, + "content": "z _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 264, + 336, + 277 + ], + "score": 1.0, + "content": "is always sandwiched", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 275, + 284, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 218, + 290 + ], + "score": 1.0, + "content": "between the integral curves", + "type": "text" + }, + { + "bbox": [ + 218, + 277, + 239, + 288 + ], + "score": 0.91, + "content": "z _ { 2 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 275, + 258, + 290 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 258, + 276, + 279, + 288 + ], + "score": 0.92, + "content": "z _ { 3 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 275, + 284, + 290 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 293, + 336, + 351 + ], + "lines": [ + { + "bbox": [ + 105, + 292, + 337, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 142, + 306 + ], + "score": 1.0, + "content": "Now, let", + "type": "text" + }, + { + "bbox": [ + 142, + 293, + 311, + 305 + ], + "score": 0.71, + "content": "\\epsilon < \\operatorname* { m i n } \\{ | z _ { 2 } ( 0 ) - z _ { 1 } ( 0 ) | , | z _ { 3 } ( 0 ) - z _ { 1 } ( 0 ) | \\} .", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 292, + 337, + 306 + ], + "score": 1.0, + "content": ". Con-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 303, + 336, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 171, + 316 + ], + "score": 1.0, + "content": "sider a solution", + "type": "text" + }, + { + "bbox": [ + 171, + 304, + 192, + 316 + ], + "score": 0.91, + "content": "\\widetilde { z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 303, + 336, + 316 + ], + "score": 1.0, + "content": "of equation (1). The integral curve", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 315, + 338, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 128, + 329 + ], + "score": 0.91, + "content": "\\widetilde { z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 315, + 203, + 330 + ], + "score": 1.0, + "content": "starts from a point", + "type": "text" + }, + { + "bbox": [ + 204, + 315, + 268, + 329 + ], + "score": 0.92, + "content": "\\widetilde { A } _ { 1 } = ( 0 , \\widetilde { z } _ { 1 } ( 0 ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 315, + 313, + 330 + ], + "score": 1.0, + "content": ". The point", + "type": "text" + }, + { + "bbox": [ + 313, + 315, + 326, + 328 + ], + "score": 0.9, + "content": "\\widetilde { A } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 315, + 338, + 330 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 327, + 336, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 131, + 340 + ], + "score": 1.0, + "content": "!in the", + "type": "text" + }, + { + "bbox": [ + 131, + 330, + 136, + 338 + ], + "score": 0.71, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 327, + 206, + 340 + ], + "score": 1.0, + "content": "-neighborhood of", + "type": "text" + }, + { + "bbox": [ + 207, + 329, + 219, + 339 + ], + "score": 0.88, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 327, + 241, + 340 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 241, + 328, + 319, + 340 + ], + "score": 0.9, + "content": "| \\widetilde z _ { 1 } ( 0 ) - z _ { 1 } ( 0 ) | < \\epsilon .", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 327, + 336, + 340 + ], + "score": 1.0, + "content": ". By", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 338, + 336, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 214, + 352 + ], + "score": 1.0, + "content": "Theorem 1, we know that", + "type": "text" + }, + { + "bbox": [ + 215, + 339, + 336, + 351 + ], + "score": 0.9, + "content": "| \\tilde { z } _ { 1 } ( T ) - z _ { 1 } ( T ) | \\leq | z _ { 3 } ( T ) -", + "type": "inline_equation" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "image", + "bbox": [ + 348, + 174, + 489, + 279 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 348, + 174, + 489, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 348, + 174, + 489, + 279 + ], + "spans": [ + { + "bbox": [ + 348, + 174, + 489, + 279 + ], + "score": 0.971, + "type": "image", + "image_path": "901235eac73c549d6bdb2646bd2f84cf3a82d0581b3c2860c0361d6a9e2870e9.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 348, + 174, + 489, + 226.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 348, + 226.5, + 489, + 279.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 344, + 289, + 504, + 336 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 343, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 343, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "Figure 2: No integral curves intersect.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 343, + 300, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 343, + 302, + 480, + 315 + ], + "score": 1.0, + "content": "The integral curve starting from", + "type": "text" + }, + { + "bbox": [ + 480, + 300, + 493, + 314 + ], + "score": 0.9, + "content": "\\widetilde { A _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 302, + 505, + 315 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 343, + 314, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 343, + 314, + 505, + 325 + ], + "score": 1.0, + "content": "always sandwiched between two inte-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 343, + 324, + 493, + 337 + ], + "spans": [ + { + "bbox": [ + 343, + 324, + 444, + 337 + ], + "score": 1.0, + "content": "gral curves starting from", + "type": "text" + }, + { + "bbox": [ + 445, + 325, + 458, + 336 + ], + "score": 0.88, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 324, + 476, + 337 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 476, + 325, + 489, + 336 + ], + "score": 0.88, + "content": "A _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 324, + 493, + 337 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 351, + 505, + 395 + ], + "lines": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 134, + 362 + ], + "score": 0.91, + "content": "z _ { 2 } ( T ) |", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 349, + 339, + 362 + ], + "score": 1.0, + "content": "!. In other words, if any perturbation smaller than", + "type": "text" + }, + { + "bbox": [ + 339, + 352, + 345, + 360 + ], + "score": 0.71, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 349, + 436, + 362 + ], + "score": 1.0, + "content": "is added to the scalar", + "type": "text" + }, + { + "bbox": [ + 436, + 350, + 459, + 362 + ], + "score": 0.91, + "content": "z _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 349, + 472, + 362 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 473, + 351, + 485, + 361 + ], + "score": 0.87, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 349, + 505, + 362 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 361, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 245, + 374 + ], + "score": 1.0, + "content": "deviation from the original output", + "type": "text" + }, + { + "bbox": [ + 246, + 361, + 271, + 373 + ], + "score": 0.94, + "content": "z _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 361, + 419, + 374 + ], + "score": 1.0, + "content": "is bounded by the distance between", + "type": "text" + }, + { + "bbox": [ + 419, + 361, + 444, + 373 + ], + "score": 0.92, + "content": "z _ { 2 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 361, + 463, + 374 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 463, + 361, + 488, + 373 + ], + "score": 0.92, + "content": "z _ { 3 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 361, + 505, + 374 + ], + "score": 1.0, + "content": ". In", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 372, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 506, + 385 + ], + "score": 1.0, + "content": "contrast, in a CNN model, there is no such bound on the deviation from the original output. Thus,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 383, + 448, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 448, + 396 + ], + "score": 1.0, + "content": "we opine that due to this non-intersecting property, ODENets are intrinsically robust.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "title", + "bbox": [ + 107, + 416, + 420, + 428 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 421, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 421, + 430 + ], + "score": 1.0, + "content": "4 TISODE: BOOSTING THE ROBUSTNESS OF NEURAL ODES", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 487 + ], + "lines": [ + { + "bbox": [ + 105, + 443, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 505, + 455 + ], + "score": 1.0, + "content": "In the previous section, we presented an empirical study on the robustness of ODENets and observed", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "score": 1.0, + "content": "that ODENets are more robust compared to CNN models. In this section, we explore how to boost", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "the robustness of the vanilla neural ODE model further. This motivates the proposal of time-invariant", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 476, + 238, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 238, + 488 + ], + "score": 1.0, + "content": "steady neural ODEs (TisODEs).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "title", + "bbox": [ + 106, + 506, + 309, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 505, + 311, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 311, + 518 + ], + "score": 1.0, + "content": "4.1 TIME-INVARIANT STEADY NEURAL ODES", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 275, + 528, + 504, + 671 + ], + "lines": [ + { + "bbox": [ + 274, + 527, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 274, + 527, + 505, + 540 + ], + "score": 1.0, + "content": "From the discussion in Section 3.4, the key to improv-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 274, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 274, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "ing the robustness of neural ODEs is to control the dif-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 274, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 274, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "ference between neighboring integral curves. By Grow-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 275, + 561, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 275, + 561, + 505, + 573 + ], + "score": 1.0, + "content": "nall’s inequality (Howard, 1998) (see Theorem 2 in the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 274, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 274, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "Appendix), we know that the difference between two ter-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 275, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 275, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "minal states is bounded by the difference between initial", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 274, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 274, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "states multiplied by the exponential of the dynamics’ Lip-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 274, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 274, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "schitz constant. However, it is very difficult to bound the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 274, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 274, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "Lipschitz constant of the dynamics directly. Alternatively,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 274, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 274, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "we propose to achieve the goal of controlling the output", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 274, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 274, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "deviation by following two steps: (i) removing the time", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 274, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 274, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "dependence of the dynamics and (ii) imposing a certain", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 275, + 661, + 368, + 672 + ], + "spans": [ + { + "bbox": [ + 275, + 661, + 368, + 672 + ], + "score": 1.0, + "content": "steady-state constraint.", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 45 + }, + { + "type": "image", + "bbox": [ + 108, + 531, + 255, + 648 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 531, + 255, + 648 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 108, + 531, + 255, + 648 + ], + "spans": [ + { + "bbox": [ + 108, + 531, + 255, + 648 + ], + "score": 0.967, + "type": "image", + "image_path": "b7cae751cc80ff17df6fe622bceb23becc612933d297ca2ae9a8307363b3a95d.jpg" + } + ] + } + ], + "index": 43.0, + "virtual_lines": [ + { + "bbox": [ + 108, + 531, + 255, + 589.5 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 108, + 589.5, + 255, + 648.0 + ], + "spans": [], + "index": 48 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 660, + 266, + 716 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 660, + 268, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 268, + 672 + ], + "score": 1.0, + "content": "Figure 3: An illustration of the time-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 671, + 267, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 267, + 683 + ], + "score": 1.0, + "content": "invariant property of ODEs. We can see", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 681, + 267, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 164, + 694 + ], + "score": 1.0, + "content": "that the curve", + "type": "text" + }, + { + "bbox": [ + 164, + 682, + 187, + 694 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 681, + 267, + 694 + ], + "score": 1.0, + "content": "is exactly the hori-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 693, + 267, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 190, + 705 + ], + "score": 1.0, + "content": "zontal translation of", + "type": "text" + }, + { + "bbox": [ + 191, + 693, + 213, + 705 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 693, + 267, + 705 + ], + "score": 1.0, + "content": "on the inter-", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 702, + 158, + 717 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 121, + 717 + ], + "score": 1.0, + "content": "val", + "type": "text" + }, + { + "bbox": [ + 121, + 704, + 153, + 716 + ], + "score": 0.92, + "content": "[ T ^ { \\prime } , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 702, + 158, + 717 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 51 + } + ], + "index": 47.0 + }, + { + "type": "text", + "bbox": [ + 275, + 677, + 504, + 721 + ], + "lines": [ + { + "bbox": [ + 274, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 274, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "In the neural ODE characterized by equation (1), the dy-", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 274, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 274, + 687, + 306, + 699 + ], + "score": 1.0, + "content": "namics", + "type": "text" + }, + { + "bbox": [ + 306, + 687, + 349, + 700 + ], + "score": 0.93, + "content": "f _ { \\theta } ( \\mathbf { z } ( t ) , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 687, + 455, + 699 + ], + "score": 1.0, + "content": "depends on both the state", + "type": "text" + }, + { + "bbox": [ + 455, + 688, + 473, + 699 + ], + "score": 0.9, + "content": "{ \\bf z } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "at time", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 275, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 275, + 700, + 280, + 709 + ], + "score": 0.71, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 699, + 335, + 710 + ], + "score": 1.0, + "content": "and the time", + "type": "text" + }, + { + "bbox": [ + 336, + 700, + 341, + 709 + ], + "score": 0.72, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "itself. In contrast, if the neural ODE is", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 274, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 274, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "modified to be time-invariant, the time dependence of the", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 59.5 + }, + { + "type": "text", + "bbox": [ + 102, + 721, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "dynamics is removed. Consequently, the dynamics depends only on the state z. So, we can rewrite", + "type": "text" + } + ], + "index": 62 + } + ], + "index": 62 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 116 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 105, + 82, + 505, + 118 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 121, + 505, + 167 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "score": 1.0, + "content": "Theorem 1 (ODE integral curves do not intersect (Coddington & Levinson, 1955; Younes, 2010;", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 214, + 144 + ], + "score": 1.0, + "content": "Dupont et al., 2019)). Let", + "type": "text" + }, + { + "bbox": [ + 214, + 132, + 236, + 144 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 132, + 255, + 144 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 256, + 132, + 278, + 144 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 2 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 132, + 505, + 144 + ], + "score": 1.0, + "content": "be two solutions of the ODE in (1) with different initial", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 168, + 156 + ], + "score": 1.0, + "content": "conditions, i.e.", + "type": "text" + }, + { + "bbox": [ + 169, + 144, + 228, + 155 + ], + "score": 0.9, + "content": "{ \\bf z } _ { 1 } ( 0 ) \\neq { \\bf z } _ { 2 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 143, + 262, + 156 + ], + "score": 1.0, + "content": ". In (1),", + "type": "text" + }, + { + "bbox": [ + 262, + 144, + 272, + 155 + ], + "score": 0.72, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "is continuous in t and globally Lipschitz continuous in z.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 153, + 313, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 181, + 167 + ], + "score": 1.0, + "content": "Then, it holds that", + "type": "text" + }, + { + "bbox": [ + 182, + 155, + 237, + 167 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( t ) \\neq { \\bf z } _ { 2 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 153, + 266, + 167 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 266, + 155, + 309, + 167 + ], + "score": 0.93, + "content": "t \\in [ 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 153, + 313, + 167 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 121, + 505, + 167 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 177, + 336, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 177, + 336, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 336, + 189 + ], + "score": 1.0, + "content": "To illustrate this theorem, considering a simple 1-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 188, + 337, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 337, + 200 + ], + "score": 1.0, + "content": "dimensional system in which the state is a scalar. As", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 199, + 336, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 291, + 211 + ], + "score": 1.0, + "content": "shown in Figure 2, equation (1) has a solution", + "type": "text" + }, + { + "bbox": [ + 291, + 199, + 312, + 212 + ], + "score": 0.92, + "content": "z _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 199, + 336, + 211 + ], + "score": 1.0, + "content": "start-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 210, + 338, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 145, + 222 + ], + "score": 1.0, + "content": "ing from", + "type": "text" + }, + { + "bbox": [ + 145, + 210, + 212, + 222 + ], + "score": 0.93, + "content": "\\bar { A _ { 1 } } = ( 0 , \\bar { z _ { 1 } } ( 0 ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 210, + 244, + 222 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 245, + 210, + 268, + 222 + ], + "score": 0.92, + "content": "z _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 210, + 338, + 222 + ], + "score": 1.0, + "content": "is the feature of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 221, + 337, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 337, + 233 + ], + "score": 1.0, + "content": "some datum. Equation (1) also has another two solutions", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 231, + 334, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 128, + 244 + ], + "score": 0.91, + "content": "z _ { 2 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 231, + 148, + 245 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 149, + 232, + 170, + 244 + ], + "score": 0.89, + "content": "z _ { 3 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 231, + 266, + 245 + ], + "score": 1.0, + "content": ", whose starting points", + "type": "text" + }, + { + "bbox": [ + 266, + 232, + 334, + 244 + ], + "score": 0.91, + "content": "A _ { 2 } = ( 0 , z _ { 2 } ( 0 ) )", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 241, + 338, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 125, + 256 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 125, + 244, + 196, + 255 + ], + "score": 0.92, + "content": "A _ { 3 } ~ = ~ ( 0 , z _ { 3 } ( 0 ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 241, + 319, + 256 + ], + "score": 1.0, + "content": ", both of which are close to", + "type": "text" + }, + { + "bbox": [ + 320, + 244, + 333, + 254 + ], + "score": 0.87, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 241, + 338, + 256 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 254, + 337, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 143, + 266 + ], + "score": 1.0, + "content": "Suppose", + "type": "text" + }, + { + "bbox": [ + 144, + 255, + 156, + 265 + ], + "score": 0.88, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 254, + 205, + 266 + ], + "score": 1.0, + "content": "is between", + "type": "text" + }, + { + "bbox": [ + 205, + 255, + 218, + 265 + ], + "score": 0.88, + "content": "A _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 254, + 238, + 266 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 238, + 254, + 251, + 265 + ], + "score": 0.87, + "content": "A _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 254, + 337, + 266 + ], + "score": 1.0, + "content": ". By Theorem 1, we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 264, + 336, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 264, + 223, + 277 + ], + "score": 1.0, + "content": "know that the integral curve", + "type": "text" + }, + { + "bbox": [ + 223, + 265, + 245, + 277 + ], + "score": 0.91, + "content": "z _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 264, + 336, + 277 + ], + "score": 1.0, + "content": "is always sandwiched", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 275, + 284, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 218, + 290 + ], + "score": 1.0, + "content": "between the integral curves", + "type": "text" + }, + { + "bbox": [ + 218, + 277, + 239, + 288 + ], + "score": 0.91, + "content": "z _ { 2 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 275, + 258, + 290 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 258, + 276, + 279, + 288 + ], + "score": 0.92, + "content": "z _ { 3 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 275, + 284, + 290 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 177, + 338, + 290 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 293, + 336, + 351 + ], + "lines": [ + { + "bbox": [ + 105, + 292, + 337, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 142, + 306 + ], + "score": 1.0, + "content": "Now, let", + "type": "text" + }, + { + "bbox": [ + 142, + 293, + 311, + 305 + ], + "score": 0.71, + "content": "\\epsilon < \\operatorname* { m i n } \\{ | z _ { 2 } ( 0 ) - z _ { 1 } ( 0 ) | , | z _ { 3 } ( 0 ) - z _ { 1 } ( 0 ) | \\} .", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 292, + 337, + 306 + ], + "score": 1.0, + "content": ". Con-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 303, + 336, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 171, + 316 + ], + "score": 1.0, + "content": "sider a solution", + "type": "text" + }, + { + "bbox": [ + 171, + 304, + 192, + 316 + ], + "score": 0.91, + "content": "\\widetilde { z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 303, + 336, + 316 + ], + "score": 1.0, + "content": "of equation (1). The integral curve", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 315, + 338, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 128, + 329 + ], + "score": 0.91, + "content": "\\widetilde { z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 315, + 203, + 330 + ], + "score": 1.0, + "content": "starts from a point", + "type": "text" + }, + { + "bbox": [ + 204, + 315, + 268, + 329 + ], + "score": 0.92, + "content": "\\widetilde { A } _ { 1 } = ( 0 , \\widetilde { z } _ { 1 } ( 0 ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 315, + 313, + 330 + ], + "score": 1.0, + "content": ". The point", + "type": "text" + }, + { + "bbox": [ + 313, + 315, + 326, + 328 + ], + "score": 0.9, + "content": "\\widetilde { A } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 315, + 338, + 330 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 327, + 336, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 131, + 340 + ], + "score": 1.0, + "content": "!in the", + "type": "text" + }, + { + "bbox": [ + 131, + 330, + 136, + 338 + ], + "score": 0.71, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 327, + 206, + 340 + ], + "score": 1.0, + "content": "-neighborhood of", + "type": "text" + }, + { + "bbox": [ + 207, + 329, + 219, + 339 + ], + "score": 0.88, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 327, + 241, + 340 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 241, + 328, + 319, + 340 + ], + "score": 0.9, + "content": "| \\widetilde z _ { 1 } ( 0 ) - z _ { 1 } ( 0 ) | < \\epsilon .", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 327, + 336, + 340 + ], + "score": 1.0, + "content": ". By", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 338, + 336, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 214, + 352 + ], + "score": 1.0, + "content": "Theorem 1, we know that", + "type": "text" + }, + { + "bbox": [ + 215, + 339, + 336, + 351 + ], + "score": 0.9, + "content": "| \\tilde { z } _ { 1 } ( T ) - z _ { 1 } ( T ) | \\leq | z _ { 3 } ( T ) -", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 134, + 362 + ], + "score": 0.91, + "content": "z _ { 2 } ( T ) |", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 349, + 339, + 362 + ], + "score": 1.0, + "content": "!. In other words, if any perturbation smaller than", + "type": "text" + }, + { + "bbox": [ + 339, + 352, + 345, + 360 + ], + "score": 0.71, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 349, + 436, + 362 + ], + "score": 1.0, + "content": "is added to the scalar", + "type": "text" + }, + { + "bbox": [ + 436, + 350, + 459, + 362 + ], + "score": 0.91, + "content": "z _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 349, + 472, + 362 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 473, + 351, + 485, + 361 + ], + "score": 0.87, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 349, + 505, + 362 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 361, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 245, + 374 + ], + "score": 1.0, + "content": "deviation from the original output", + "type": "text" + }, + { + "bbox": [ + 246, + 361, + 271, + 373 + ], + "score": 0.94, + "content": "z _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 361, + 419, + 374 + ], + "score": 1.0, + "content": "is bounded by the distance between", + "type": "text" + }, + { + "bbox": [ + 419, + 361, + 444, + 373 + ], + "score": 0.92, + "content": "z _ { 2 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 361, + 463, + 374 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 463, + 361, + 488, + 373 + ], + "score": 0.92, + "content": "z _ { 3 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 361, + 505, + 374 + ], + "score": 1.0, + "content": ". In", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 372, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 506, + 385 + ], + "score": 1.0, + "content": "contrast, in a CNN model, there is no such bound on the deviation from the original output. Thus,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 383, + 448, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 448, + 396 + ], + "score": 1.0, + "content": "we opine that due to this non-intersecting property, ODENets are intrinsically robust.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 292, + 338, + 352 + ] + }, + { + "type": "image", + "bbox": [ + 348, + 174, + 489, + 279 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 348, + 174, + 489, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 348, + 174, + 489, + 279 + ], + "spans": [ + { + "bbox": [ + 348, + 174, + 489, + 279 + ], + "score": 0.971, + "type": "image", + "image_path": "901235eac73c549d6bdb2646bd2f84cf3a82d0581b3c2860c0361d6a9e2870e9.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 348, + 174, + 489, + 226.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 348, + 226.5, + 489, + 279.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 344, + 289, + 504, + 336 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 343, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 343, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "Figure 2: No integral curves intersect.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 343, + 300, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 343, + 302, + 480, + 315 + ], + "score": 1.0, + "content": "The integral curve starting from", + "type": "text" + }, + { + "bbox": [ + 480, + 300, + 493, + 314 + ], + "score": 0.9, + "content": "\\widetilde { A _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 302, + 505, + 315 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 343, + 314, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 343, + 314, + 505, + 325 + ], + "score": 1.0, + "content": "always sandwiched between two inte-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 343, + 324, + 493, + 337 + ], + "spans": [ + { + "bbox": [ + 343, + 324, + 444, + 337 + ], + "score": 1.0, + "content": "gral curves starting from", + "type": "text" + }, + { + "bbox": [ + 445, + 325, + 458, + 336 + ], + "score": 0.88, + "content": "A _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 324, + 476, + 337 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 476, + 325, + 489, + 336 + ], + "score": 0.88, + "content": "A _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 324, + 493, + 337 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 351, + 505, + 395 + ], + "lines": [], + "index": 29.5, + "bbox_fs": [ + 105, + 349, + 506, + 396 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 416, + 420, + 428 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 421, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 421, + 430 + ], + "score": 1.0, + "content": "4 TISODE: BOOSTING THE ROBUSTNESS OF NEURAL ODES", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 487 + ], + "lines": [ + { + "bbox": [ + 105, + 443, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 505, + 455 + ], + "score": 1.0, + "content": "In the previous section, we presented an empirical study on the robustness of ODENets and observed", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "score": 1.0, + "content": "that ODENets are more robust compared to CNN models. In this section, we explore how to boost", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "the robustness of the vanilla neural ODE model further. This motivates the proposal of time-invariant", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 476, + 238, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 238, + 488 + ], + "score": 1.0, + "content": "steady neural ODEs (TisODEs).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 443, + 506, + 488 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 506, + 309, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 505, + 311, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 311, + 518 + ], + "score": 1.0, + "content": "4.1 TIME-INVARIANT STEADY NEURAL ODES", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 275, + 528, + 504, + 671 + ], + "lines": [ + { + "bbox": [ + 274, + 527, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 274, + 527, + 505, + 540 + ], + "score": 1.0, + "content": "From the discussion in Section 3.4, the key to improv-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 274, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 274, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "ing the robustness of neural ODEs is to control the dif-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 274, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 274, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "ference between neighboring integral curves. By Grow-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 275, + 561, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 275, + 561, + 505, + 573 + ], + "score": 1.0, + "content": "nall’s inequality (Howard, 1998) (see Theorem 2 in the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 274, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 274, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "Appendix), we know that the difference between two ter-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 275, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 275, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "minal states is bounded by the difference between initial", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 274, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 274, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "states multiplied by the exponential of the dynamics’ Lip-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 274, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 274, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "schitz constant. However, it is very difficult to bound the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 274, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 274, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "Lipschitz constant of the dynamics directly. Alternatively,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 274, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 274, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "we propose to achieve the goal of controlling the output", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 274, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 274, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "deviation by following two steps: (i) removing the time", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 274, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 274, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "dependence of the dynamics and (ii) imposing a certain", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 275, + 661, + 368, + 672 + ], + "spans": [ + { + "bbox": [ + 275, + 661, + 368, + 672 + ], + "score": 1.0, + "content": "steady-state constraint.", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 45, + "bbox_fs": [ + 274, + 527, + 505, + 672 + ] + }, + { + "type": "image", + "bbox": [ + 108, + 531, + 255, + 648 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 531, + 255, + 648 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 108, + 531, + 255, + 648 + ], + "spans": [ + { + "bbox": [ + 108, + 531, + 255, + 648 + ], + "score": 0.967, + "type": "image", + "image_path": "b7cae751cc80ff17df6fe622bceb23becc612933d297ca2ae9a8307363b3a95d.jpg" + } + ] + } + ], + "index": 43.0, + "virtual_lines": [ + { + "bbox": [ + 108, + 531, + 255, + 589.5 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 108, + 589.5, + 255, + 648.0 + ], + "spans": [], + "index": 48 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 660, + 266, + 716 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 660, + 268, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 268, + 672 + ], + "score": 1.0, + "content": "Figure 3: An illustration of the time-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 671, + 267, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 267, + 683 + ], + "score": 1.0, + "content": "invariant property of ODEs. We can see", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 681, + 267, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 164, + 694 + ], + "score": 1.0, + "content": "that the curve", + "type": "text" + }, + { + "bbox": [ + 164, + 682, + 187, + 694 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 681, + 267, + 694 + ], + "score": 1.0, + "content": "is exactly the hori-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 693, + 267, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 190, + 705 + ], + "score": 1.0, + "content": "zontal translation of", + "type": "text" + }, + { + "bbox": [ + 191, + 693, + 213, + 705 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 693, + 267, + 705 + ], + "score": 1.0, + "content": "on the inter-", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 702, + 158, + 717 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 121, + 717 + ], + "score": 1.0, + "content": "val", + "type": "text" + }, + { + "bbox": [ + 121, + 704, + 153, + 716 + ], + "score": 0.92, + "content": "[ T ^ { \\prime } , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 702, + 158, + 717 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 51 + } + ], + "index": 47.0 + }, + { + "type": "text", + "bbox": [ + 275, + 677, + 504, + 721 + ], + "lines": [ + { + "bbox": [ + 274, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 274, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "In the neural ODE characterized by equation (1), the dy-", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 274, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 274, + 687, + 306, + 699 + ], + "score": 1.0, + "content": "namics", + "type": "text" + }, + { + "bbox": [ + 306, + 687, + 349, + 700 + ], + "score": 0.93, + "content": "f _ { \\theta } ( \\mathbf { z } ( t ) , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 687, + 455, + 699 + ], + "score": 1.0, + "content": "depends on both the state", + "type": "text" + }, + { + "bbox": [ + 455, + 688, + 473, + 699 + ], + "score": 0.9, + "content": "{ \\bf z } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "at time", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 275, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 275, + 700, + 280, + 709 + ], + "score": 0.71, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 699, + 335, + 710 + ], + "score": 1.0, + "content": "and the time", + "type": "text" + }, + { + "bbox": [ + 336, + 700, + 341, + 709 + ], + "score": 0.72, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "itself. In contrast, if the neural ODE is", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 274, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 274, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "modified to be time-invariant, the time dependence of the", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 106, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "dynamics is removed. Consequently, the dynamics depends only on the state z. So, we can rewrite", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 81, + 394, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 208, + 96 + ], + "score": 1.0, + "content": "the dynamics function as", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 208, + 82, + 231, + 95 + ], + "score": 0.92, + "content": "f _ { \\boldsymbol { \\theta } } ( \\mathbf { z } )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 232, + 81, + 394, + 96 + ], + "score": 1.0, + "content": ", and the neural ODE is characterized as", + "type": "text", + "cross_page": true + } + ], + "index": 0 + } + ], + "index": 59.5, + "bbox_fs": [ + 274, + 676, + 505, + 722 + ] + }, + { + "type": "text", + "bbox": [ + 102, + 721, + 504, + 732 + ], + "lines": [], + "index": 62, + "bbox_fs": [ + 106, + 719, + 505, + 734 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 392, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 394, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 208, + 96 + ], + "score": 1.0, + "content": "the dynamics function as", + "type": "text" + }, + { + "bbox": [ + 208, + 82, + 231, + 95 + ], + "score": 0.92, + "content": "f _ { \\boldsymbol { \\theta } } ( \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 81, + 394, + 96 + ], + "score": 1.0, + "content": ", and the neural ODE is characterized as", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 101, + 343, + 153 + ], + "lines": [ + { + "bbox": [ + 257, + 101, + 343, + 153 + ], + "spans": [ + { + "bbox": [ + 257, + 101, + 343, + 153 + ], + "score": 0.92, + "content": "\\left\\{ \\begin{array} { l l } { \\displaystyle \\frac { \\mathrm { d } \\mathbf { z } ( t ) } { \\mathrm { d } t } = f _ { \\boldsymbol { \\theta } } ( \\mathbf { z } ( t ) ) ; } \\\\ { \\mathbf { z } ( 0 ) = \\mathbf { z } _ { \\mathrm { i n } } ; } \\\\ { \\mathbf { z } _ { \\mathrm { o u t } } = \\mathbf { z } ( T ) . } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "f46c4a1faca9cc61c86100722af41f1f7581c94ec9c4520e6d2130757a216c58.jpg" + } + ] + } + ], + "index": 1.5, + "virtual_lines": [ + { + "bbox": [ + 257, + 101, + 343, + 127.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 257, + 127.0, + 343, + 153.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 505, + 210 + ], + "lines": [ + { + "bbox": [ + 104, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 104, + 164, + 123, + 178 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 165, + 145, + 177 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 164, + 244, + 178 + ], + "score": 1.0, + "content": "be a solution of (2) on", + "type": "text" + }, + { + "bbox": [ + 245, + 166, + 271, + 177 + ], + "score": 0.9, + "content": "[ 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 164, + 291, + 178 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 291, + 167, + 319, + 176 + ], + "score": 0.89, + "content": "\\epsilon > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "be a small positive value. We define the set", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 175, + 504, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 218, + 189 + ], + "score": 0.6, + "content": "\\mathbb { M } _ { 1 } = \\{ ( \\mathbf { z } _ { 1 } ( t ) , t ) | t \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 175, + 222, + 190 + ], + "score": 1.0, + "content": ", \"", + "type": "text" + }, + { + "bbox": [ + 222, + 177, + 308, + 189 + ], + "score": 0.82, + "content": "\\| \\mathbf { z } _ { 1 } ( t ) - \\mathbf { z } _ { 1 } ( 0 ) \\| \\leq \\epsilon \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 175, + 481, + 190 + ], + "score": 1.0, + "content": ". This set contains all points on the curve of", + "type": "text" + }, + { + "bbox": [ + 482, + 176, + 504, + 188 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 186, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 135, + 200 + ], + "score": 1.0, + "content": "during", + "type": "text" + }, + { + "bbox": [ + 135, + 188, + 158, + 199 + ], + "score": 0.89, + "content": "[ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 186, + 249, + 200 + ], + "score": 1.0, + "content": "that are also inside the", + "type": "text" + }, + { + "bbox": [ + 249, + 190, + 254, + 197 + ], + "score": 0.7, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 186, + 324, + 200 + ], + "score": 1.0, + "content": "-neighborhood of", + "type": "text" + }, + { + "bbox": [ + 324, + 187, + 347, + 199 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 186, + 425, + 200 + ], + "score": 1.0, + "content": ". For some element", + "type": "text" + }, + { + "bbox": [ + 425, + 187, + 501, + 199 + ], + "score": 0.94, + "content": "( { \\bf z } _ { 1 } ( T ^ { \\prime } ) , T ^ { \\prime } ) \\in \\mathbb { M } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 186, + 505, + 200 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 198, + 428, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 118, + 211 + ], + "score": 1.0, + "content": "let", + "type": "text" + }, + { + "bbox": [ + 119, + 199, + 141, + 210 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 198, + 300, + 211 + ], + "score": 1.0, + "content": "be the solution of (2) which starts from", + "type": "text" + }, + { + "bbox": [ + 301, + 199, + 364, + 210 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf z } _ { 1 } ( 0 ) = \\mathbf z _ { 1 } ( T ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 198, + 428, + 211 + ], + "score": 1.0, + "content": ". Then we have", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 265, + 216, + 345, + 230 + ], + "lines": [ + { + "bbox": [ + 265, + 216, + 345, + 230 + ], + "spans": [ + { + "bbox": [ + 265, + 216, + 345, + 230 + ], + "score": 0.91, + "content": "\\widetilde { { \\mathbf z } } _ { 1 } ( t ) = { \\mathbf z } _ { 1 } ( t + T ^ { \\prime } )", + "type": "interline_equation", + "image_path": "99ba3d941298bb987e3bce08ab9b157bb13e597a10070baaba0ae0a88f784c91.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 265, + 216, + 345, + 230 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 506, + 261 + ], + "lines": [ + { + "bbox": [ + 106, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 134, + 250 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 134, + 239, + 139, + 248 + ], + "score": 0.76, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 237, + 151, + 250 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 151, + 238, + 178, + 250 + ], + "score": 0.93, + "content": "[ 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 237, + 506, + 250 + ], + "score": 1.0, + "content": ". The property shown in equation (3) is known as the time-invariant property. It", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 247, + 401, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 234, + 261 + ], + "score": 1.0, + "content": "indicates that the integral curve", + "type": "text" + }, + { + "bbox": [ + 234, + 249, + 255, + 261 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 247, + 281, + 261 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 281, + 249, + 300, + 259 + ], + "score": 0.9, + "content": "- T ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 247, + 332, + 261 + ], + "score": 1.0, + "content": "shift of", + "type": "text" + }, + { + "bbox": [ + 333, + 248, + 354, + 261 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 247, + 401, + 261 + ], + "score": 1.0, + "content": "(Figure 3).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 265, + 505, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 168, + 278 + ], + "score": 1.0, + "content": "We can regard", + "type": "text" + }, + { + "bbox": [ + 169, + 265, + 192, + 277 + ], + "score": 0.91, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 264, + 333, + 278 + ], + "score": 1.0, + "content": "as a slightly perturbed version of", + "type": "text" + }, + { + "bbox": [ + 333, + 265, + 356, + 277 + ], + "score": 0.93, + "content": "{ \\bf z } _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 264, + 506, + 278 + ], + "score": 1.0, + "content": ", and we are interested in how large", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 200, + 289 + ], + "score": 1.0, + "content": "the difference between", + "type": "text" + }, + { + "bbox": [ + 200, + 276, + 226, + 288 + ], + "score": 0.91, + "content": "\\widetilde { \\mathbf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 276, + 245, + 289 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 245, + 276, + 271, + 288 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 276, + 505, + 289 + ], + "score": 1.0, + "content": "is. In a robust model, the difference should be small. By", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 195, + 300 + ], + "score": 1.0, + "content": "equation (3), we have", + "type": "text" + }, + { + "bbox": [ + 195, + 288, + 370, + 299 + ], + "score": 0.91, + "content": "\\| \\mathbf { \\tilde { z } } _ { 1 } ( T ) - \\mathbf { z } _ { 1 } ( T ) \\| = \\| \\mathbf { z } _ { 1 } ( T + T ^ { \\prime } ) - \\mathbf { z } _ { 1 } ( T ) \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 287, + 398, + 300 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 399, + 287, + 444, + 299 + ], + "score": 0.93, + "content": "T ^ { \\prime } \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 287, + 506, + 300 + ], + "score": 1.0, + "content": ", the difference", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 297, + 324, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 142, + 311 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 142, + 299, + 168, + 310 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 297, + 186, + 311 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 186, + 299, + 212, + 310 + ], + "score": 0.91, + "content": "\\widetilde { \\mathbf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 297, + 324, + 311 + ], + "score": 1.0, + "content": "can be bounded as follows,", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 316, + 495, + 351 + ], + "lines": [ + { + "bbox": [ + 114, + 316, + 495, + 351 + ], + "spans": [ + { + "bbox": [ + 114, + 316, + 495, + 351 + ], + "score": 0.94, + "content": "\\| \\widetilde { \\mathbf z } _ { 1 } ( T ) - \\mathbf z _ { 1 } ( T ) \\| = \\left\\| \\int _ { T } ^ { T + T ^ { \\prime } } f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) \\mathrm { d } t \\right\\| \\leq \\left\\| \\int _ { T } ^ { T + T ^ { \\prime } } | f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) | \\mathrm { d } t \\right\\| \\leq \\left\\| \\int _ { T } ^ { 2 T } | f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) | \\mathrm { d } t \\right\\| ,", + "type": "interline_equation", + "image_path": "dda5a5162f628120cc1d9f8b14d9625d581af070a04ac8f089aaa2d9e4fe900a.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 114, + 316, + 495, + 327.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 114, + 327.6666666666667, + 495, + 339.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 114, + 339.33333333333337, + 495, + 351.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 358, + 505, + 415 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 192, + 371 + ], + "score": 1.0, + "content": "where all norms are", + "type": "text" + }, + { + "bbox": [ + 192, + 360, + 202, + 370 + ], + "score": 0.88, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 357, + 251, + 371 + ], + "score": 1.0, + "content": "norms and", + "type": "text" + }, + { + "bbox": [ + 252, + 359, + 268, + 371 + ], + "score": 0.9, + "content": "| f _ { \\theta } |", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 357, + 505, + 371 + ], + "score": 1.0, + "content": "denotes the element-wise absolute operation of a vector-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 369, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 171, + 383 + ], + "score": 1.0, + "content": "valued function", + "type": "text" + }, + { + "bbox": [ + 171, + 370, + 181, + 381 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 369, + 336, + 383 + ], + "score": 1.0, + "content": ". That is to say, the difference between", + "type": "text" + }, + { + "bbox": [ + 336, + 370, + 362, + 381 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 369, + 380, + 383 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 380, + 370, + 406, + 381 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 369, + 505, + 383 + ], + "score": 1.0, + "content": "can be bounded by only", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 380, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 243, + 393 + ], + "score": 1.0, + "content": "using the information of the curve", + "type": "text" + }, + { + "bbox": [ + 243, + 381, + 265, + 392 + ], + "score": 0.9, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 380, + 302, + 393 + ], + "score": 1.0, + "content": ". For any", + "type": "text" + }, + { + "bbox": [ + 302, + 381, + 344, + 393 + ], + "score": 0.93, + "content": "t ^ { \\prime } \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 380, + 396, + 393 + ], + "score": 1.0, + "content": "and element", + "type": "text" + }, + { + "bbox": [ + 396, + 381, + 465, + 393 + ], + "score": 0.92, + "content": "( { \\bf z } _ { 1 } ( t ^ { \\prime } ) , t ^ { \\prime } ) \\in \\mathbb { M } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 380, + 506, + 393 + ], + "score": 1.0, + "content": ", consider", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 391, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 243, + 405 + ], + "score": 1.0, + "content": "the integral curve that starts from", + "type": "text" + }, + { + "bbox": [ + 243, + 393, + 268, + 404 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 391, + 506, + 405 + ], + "score": 1.0, + "content": ". The difference between the output state of this curve and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 402, + 226, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 132, + 415 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 402, + 226, + 416 + ], + "score": 1.0, + "content": "satisfies inequality (4).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 419, + 503, + 441 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 308, + 433 + ], + "score": 1.0, + "content": "Therefore, we propose to add an additional term", + "type": "text" + }, + { + "bbox": [ + 308, + 420, + 323, + 431 + ], + "score": 0.9, + "content": "L _ { \\mathrm { s s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 418, + 505, + 433 + ], + "score": 1.0, + "content": "to the loss function when training the time-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 430, + 197, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 197, + 442 + ], + "score": 1.0, + "content": "invariant neural ODE:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "interline_equation", + "bbox": [ + 237, + 441, + 373, + 476 + ], + "lines": [ + { + "bbox": [ + 237, + 441, + 373, + 476 + ], + "spans": [ + { + "bbox": [ + 237, + 441, + 373, + 476 + ], + "score": 0.94, + "content": "L _ { \\mathrm { s s } } = \\sum _ { i = 1 } ^ { N } \\left\\| \\int _ { T } ^ { 2 T } | f _ { \\theta } ( \\mathbf { z } _ { i } ( t ) ) | \\mathrm { d } t \\right\\| ,", + "type": "interline_equation", + "image_path": "dead5c3a609cf3ee30da32ebeb2ca3d5179c956d041557ad30986bda72525deb.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 237, + 441, + 373, + 458.5 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 237, + 458.5, + 373, + 476.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 479, + 505, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 480, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 134, + 492 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 481, + 145, + 490 + ], + "score": 0.81, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 480, + 345, + 492 + ], + "score": 1.0, + "content": "is the number of samples in the training set and", + "type": "text" + }, + { + "bbox": [ + 345, + 480, + 366, + 492 + ], + "score": 0.92, + "content": "{ \\bf z } _ { i } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 480, + 506, + 492 + ], + "score": 1.0, + "content": "is the solution whose initial state", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 220, + 504 + ], + "score": 1.0, + "content": "equals to the feature of the", + "type": "text" + }, + { + "bbox": [ + 220, + 491, + 233, + 501 + ], + "score": 0.86, + "content": "i ^ { \\mathrm { t h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 490, + 370, + 504 + ], + "score": 1.0, + "content": "sample. The regularization term", + "type": "text" + }, + { + "bbox": [ + 370, + 492, + 385, + 502 + ], + "score": 0.89, + "content": "L _ { \\mathrm { s s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "is termed as the steady-state", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "loss. This terminology “steady state” is borrowed from the dynamical systems literature. In a stable", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 514, + 504, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 504, + 525 + ], + "score": 1.0, + "content": "dynamical system, the states stabilize around a fixed point, known as the steady-state, as time tends", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 234, + 537 + ], + "score": 1.0, + "content": "to infinity. If we can ensure that", + "type": "text" + }, + { + "bbox": [ + 234, + 524, + 249, + 535 + ], + "score": 0.9, + "content": "L _ { \\mathrm { s s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 524, + 471, + 537 + ], + "score": 1.0, + "content": "is small, for each sample, the outputs of all the points in", + "type": "text" + }, + { + "bbox": [ + 472, + 524, + 486, + 535 + ], + "score": 0.88, + "content": "\\mathbb { M } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "will", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 535, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 172, + 547 + ], + "score": 1.0, + "content": "stabilize around", + "type": "text" + }, + { + "bbox": [ + 173, + 535, + 197, + 547 + ], + "score": 0.92, + "content": "{ \\bf z } _ { i } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 535, + 506, + 547 + ], + "score": 1.0, + "content": ". Consequently, the model is robust. This modification of the neural ODE is", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 546, + 281, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 256, + 558 + ], + "score": 1.0, + "content": "dubbed Time-invariant steady neural", + "type": "text" + }, + { + "bbox": [ + 256, + 546, + 278, + 556 + ], + "score": 0.36, + "content": "O D E", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 546, + 281, + 558 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29 + }, + { + "type": "title", + "bbox": [ + 108, + 572, + 387, + 584 + ], + "lines": [ + { + "bbox": [ + 106, + 573, + 388, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 388, + 585 + ], + "score": 1.0, + "content": "4.2 EVALUATING ROBUSTNESS OF TISODE-BASED CLASSIFIERS", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 594, + 505, + 638 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "Here, we conduct experiments to evaluate the robustness of our proposed TisODE, and compare", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "TisODE-based models with the vanilla ODENets. We train all models with original non-perturbed", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 617, + 504, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 504, + 628 + ], + "score": 1.0, + "content": "images together with their Gaussian perturbed versions. The regularization parameter for the steady-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 626, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 144, + 639 + ], + "score": 1.0, + "content": "state loss", + "type": "text" + }, + { + "bbox": [ + 145, + 627, + 160, + 638 + ], + "score": 0.9, + "content": "L _ { \\mathrm { s s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 626, + 504, + 639 + ], + "score": 1.0, + "content": "is set to be 0.1. All other hyperparameters are exactly the same as those in Section 3.3.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "From the results in Table 3, we can see that our proposed TisODE-based models are clearly more", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "robust compared to vanilla ODENets. On the MNIST dataset, when combating FGSM-0.3 attacks,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "the TisODE-based models outperform vanilla ODENets by more than 4 percentage points. For the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 104, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "FGSM-0.5 adversarial examples, the accuracy of the TisODE-based model is 6 percentage points", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "better. On the SVHN dataset, the TisODE-based models perform better in terms of all forms of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "adversarial examples. On the ImgNet10 dataset, the TisODE-based models also outperform vanilla", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "ODE-based models on all types of perturbations. In the presence of FGSM and PGD-5/255 exam-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 721, + 379, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 379, + 733 + ], + "score": 1.0, + "content": "ples, the accuracies are enhanced by more than 2 percentage points.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 392, + 95 + ], + "lines": [], + "index": 0, + "bbox_fs": [ + 105, + 81, + 394, + 96 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 101, + 343, + 153 + ], + "lines": [ + { + "bbox": [ + 257, + 101, + 343, + 153 + ], + "spans": [ + { + "bbox": [ + 257, + 101, + 343, + 153 + ], + "score": 0.92, + "content": "\\left\\{ \\begin{array} { l l } { \\displaystyle \\frac { \\mathrm { d } \\mathbf { z } ( t ) } { \\mathrm { d } t } = f _ { \\boldsymbol { \\theta } } ( \\mathbf { z } ( t ) ) ; } \\\\ { \\mathbf { z } ( 0 ) = \\mathbf { z } _ { \\mathrm { i n } } ; } \\\\ { \\mathbf { z } _ { \\mathrm { o u t } } = \\mathbf { z } ( T ) . } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "f46c4a1faca9cc61c86100722af41f1f7581c94ec9c4520e6d2130757a216c58.jpg" + } + ] + } + ], + "index": 1.5, + "virtual_lines": [ + { + "bbox": [ + 257, + 101, + 343, + 127.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 257, + 127.0, + 343, + 153.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 505, + 210 + ], + "lines": [ + { + "bbox": [ + 104, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 104, + 164, + 123, + 178 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 165, + 145, + 177 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 164, + 244, + 178 + ], + "score": 1.0, + "content": "be a solution of (2) on", + "type": "text" + }, + { + "bbox": [ + 245, + 166, + 271, + 177 + ], + "score": 0.9, + "content": "[ 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 164, + 291, + 178 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 291, + 167, + 319, + 176 + ], + "score": 0.89, + "content": "\\epsilon > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "be a small positive value. We define the set", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 175, + 504, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 218, + 189 + ], + "score": 0.6, + "content": "\\mathbb { M } _ { 1 } = \\{ ( \\mathbf { z } _ { 1 } ( t ) , t ) | t \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 175, + 222, + 190 + ], + "score": 1.0, + "content": ", \"", + "type": "text" + }, + { + "bbox": [ + 222, + 177, + 308, + 189 + ], + "score": 0.82, + "content": "\\| \\mathbf { z } _ { 1 } ( t ) - \\mathbf { z } _ { 1 } ( 0 ) \\| \\leq \\epsilon \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 175, + 481, + 190 + ], + "score": 1.0, + "content": ". This set contains all points on the curve of", + "type": "text" + }, + { + "bbox": [ + 482, + 176, + 504, + 188 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 186, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 135, + 200 + ], + "score": 1.0, + "content": "during", + "type": "text" + }, + { + "bbox": [ + 135, + 188, + 158, + 199 + ], + "score": 0.89, + "content": "[ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 186, + 249, + 200 + ], + "score": 1.0, + "content": "that are also inside the", + "type": "text" + }, + { + "bbox": [ + 249, + 190, + 254, + 197 + ], + "score": 0.7, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 186, + 324, + 200 + ], + "score": 1.0, + "content": "-neighborhood of", + "type": "text" + }, + { + "bbox": [ + 324, + 187, + 347, + 199 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 186, + 425, + 200 + ], + "score": 1.0, + "content": ". For some element", + "type": "text" + }, + { + "bbox": [ + 425, + 187, + 501, + 199 + ], + "score": 0.94, + "content": "( { \\bf z } _ { 1 } ( T ^ { \\prime } ) , T ^ { \\prime } ) \\in \\mathbb { M } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 186, + 505, + 200 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 198, + 428, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 118, + 211 + ], + "score": 1.0, + "content": "let", + "type": "text" + }, + { + "bbox": [ + 119, + 199, + 141, + 210 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 198, + 300, + 211 + ], + "score": 1.0, + "content": "be the solution of (2) which starts from", + "type": "text" + }, + { + "bbox": [ + 301, + 199, + 364, + 210 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf z } _ { 1 } ( 0 ) = \\mathbf z _ { 1 } ( T ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 198, + 428, + 211 + ], + "score": 1.0, + "content": ". Then we have", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 104, + 164, + 506, + 211 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 265, + 216, + 345, + 230 + ], + "lines": [ + { + "bbox": [ + 265, + 216, + 345, + 230 + ], + "spans": [ + { + "bbox": [ + 265, + 216, + 345, + 230 + ], + "score": 0.91, + "content": "\\widetilde { { \\mathbf z } } _ { 1 } ( t ) = { \\mathbf z } _ { 1 } ( t + T ^ { \\prime } )", + "type": "interline_equation", + "image_path": "99ba3d941298bb987e3bce08ab9b157bb13e597a10070baaba0ae0a88f784c91.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 265, + 216, + 345, + 230 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 506, + 261 + ], + "lines": [ + { + "bbox": [ + 106, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 134, + 250 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 134, + 239, + 139, + 248 + ], + "score": 0.76, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 237, + 151, + 250 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 151, + 238, + 178, + 250 + ], + "score": 0.93, + "content": "[ 0 , \\infty )", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 237, + 506, + 250 + ], + "score": 1.0, + "content": ". The property shown in equation (3) is known as the time-invariant property. It", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 247, + 401, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 234, + 261 + ], + "score": 1.0, + "content": "indicates that the integral curve", + "type": "text" + }, + { + "bbox": [ + 234, + 249, + 255, + 261 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 247, + 281, + 261 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 281, + 249, + 300, + 259 + ], + "score": 0.9, + "content": "- T ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 247, + 332, + 261 + ], + "score": 1.0, + "content": "shift of", + "type": "text" + }, + { + "bbox": [ + 333, + 248, + 354, + 261 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 247, + 401, + 261 + ], + "score": 1.0, + "content": "(Figure 3).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 237, + 506, + 261 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 265, + 505, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 168, + 278 + ], + "score": 1.0, + "content": "We can regard", + "type": "text" + }, + { + "bbox": [ + 169, + 265, + 192, + 277 + ], + "score": 0.91, + "content": "\\widetilde { \\mathbf { z } } _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 264, + 333, + 278 + ], + "score": 1.0, + "content": "as a slightly perturbed version of", + "type": "text" + }, + { + "bbox": [ + 333, + 265, + 356, + 277 + ], + "score": 0.93, + "content": "{ \\bf z } _ { 1 } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 264, + 506, + 278 + ], + "score": 1.0, + "content": ", and we are interested in how large", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 200, + 289 + ], + "score": 1.0, + "content": "the difference between", + "type": "text" + }, + { + "bbox": [ + 200, + 276, + 226, + 288 + ], + "score": 0.91, + "content": "\\widetilde { \\mathbf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 276, + 245, + 289 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 245, + 276, + 271, + 288 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 276, + 505, + 289 + ], + "score": 1.0, + "content": "is. In a robust model, the difference should be small. By", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 195, + 300 + ], + "score": 1.0, + "content": "equation (3), we have", + "type": "text" + }, + { + "bbox": [ + 195, + 288, + 370, + 299 + ], + "score": 0.91, + "content": "\\| \\mathbf { \\tilde { z } } _ { 1 } ( T ) - \\mathbf { z } _ { 1 } ( T ) \\| = \\| \\mathbf { z } _ { 1 } ( T + T ^ { \\prime } ) - \\mathbf { z } _ { 1 } ( T ) \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 287, + 398, + 300 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 399, + 287, + 444, + 299 + ], + "score": 0.93, + "content": "T ^ { \\prime } \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 287, + 506, + 300 + ], + "score": 1.0, + "content": ", the difference", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 297, + 324, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 142, + 311 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 142, + 299, + 168, + 310 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 297, + 186, + 311 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 186, + 299, + 212, + 310 + ], + "score": 0.91, + "content": "\\widetilde { \\mathbf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 297, + 324, + 311 + ], + "score": 1.0, + "content": "can be bounded as follows,", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 264, + 506, + 311 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 316, + 495, + 351 + ], + "lines": [ + { + "bbox": [ + 114, + 316, + 495, + 351 + ], + "spans": [ + { + "bbox": [ + 114, + 316, + 495, + 351 + ], + "score": 0.94, + "content": "\\| \\widetilde { \\mathbf z } _ { 1 } ( T ) - \\mathbf z _ { 1 } ( T ) \\| = \\left\\| \\int _ { T } ^ { T + T ^ { \\prime } } f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) \\mathrm { d } t \\right\\| \\leq \\left\\| \\int _ { T } ^ { T + T ^ { \\prime } } | f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) | \\mathrm { d } t \\right\\| \\leq \\left\\| \\int _ { T } ^ { 2 T } | f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) | \\mathrm { d } t \\right\\| ,", + "type": "interline_equation", + "image_path": "dda5a5162f628120cc1d9f8b14d9625d581af070a04ac8f089aaa2d9e4fe900a.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 114, + 316, + 495, + 327.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 114, + 327.6666666666667, + 495, + 339.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 114, + 339.33333333333337, + 495, + 351.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 358, + 505, + 415 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 192, + 371 + ], + "score": 1.0, + "content": "where all norms are", + "type": "text" + }, + { + "bbox": [ + 192, + 360, + 202, + 370 + ], + "score": 0.88, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 357, + 251, + 371 + ], + "score": 1.0, + "content": "norms and", + "type": "text" + }, + { + "bbox": [ + 252, + 359, + 268, + 371 + ], + "score": 0.9, + "content": "| f _ { \\theta } |", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 357, + 505, + 371 + ], + "score": 1.0, + "content": "denotes the element-wise absolute operation of a vector-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 369, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 171, + 383 + ], + "score": 1.0, + "content": "valued function", + "type": "text" + }, + { + "bbox": [ + 171, + 370, + 181, + 381 + ], + "score": 0.87, + "content": "f _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 369, + 336, + 383 + ], + "score": 1.0, + "content": ". That is to say, the difference between", + "type": "text" + }, + { + "bbox": [ + 336, + 370, + 362, + 381 + ], + "score": 0.92, + "content": "\\widetilde { \\mathbf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 369, + 380, + 383 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 380, + 370, + 406, + 381 + ], + "score": 0.92, + "content": "{ \\bf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 369, + 505, + 383 + ], + "score": 1.0, + "content": "can be bounded by only", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 380, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 243, + 393 + ], + "score": 1.0, + "content": "using the information of the curve", + "type": "text" + }, + { + "bbox": [ + 243, + 381, + 265, + 392 + ], + "score": 0.9, + "content": "{ \\bf z } _ { 1 } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 380, + 302, + 393 + ], + "score": 1.0, + "content": ". For any", + "type": "text" + }, + { + "bbox": [ + 302, + 381, + 344, + 393 + ], + "score": 0.93, + "content": "t ^ { \\prime } \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 380, + 396, + 393 + ], + "score": 1.0, + "content": "and element", + "type": "text" + }, + { + "bbox": [ + 396, + 381, + 465, + 393 + ], + "score": 0.92, + "content": "( { \\bf z } _ { 1 } ( t ^ { \\prime } ) , t ^ { \\prime } ) \\in \\mathbb { M } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 380, + 506, + 393 + ], + "score": 1.0, + "content": ", consider", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 391, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 243, + 405 + ], + "score": 1.0, + "content": "the integral curve that starts from", + "type": "text" + }, + { + "bbox": [ + 243, + 393, + 268, + 404 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( t ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 391, + 506, + 405 + ], + "score": 1.0, + "content": ". The difference between the output state of this curve and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 402, + 226, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 132, + 415 + ], + "score": 0.91, + "content": "{ \\bf z } _ { 1 } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 402, + 226, + 416 + ], + "score": 1.0, + "content": "satisfies inequality (4).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 357, + 506, + 416 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 419, + 503, + 441 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 308, + 433 + ], + "score": 1.0, + "content": "Therefore, we propose to add an additional term", + "type": "text" + }, + { + "bbox": [ + 308, + 420, + 323, + 431 + ], + "score": 0.9, + "content": "L _ { \\mathrm { s s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 418, + 505, + 433 + ], + "score": 1.0, + "content": "to the loss function when training the time-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 430, + 197, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 197, + 442 + ], + "score": 1.0, + "content": "invariant neural ODE:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 418, + 505, + 442 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 237, + 441, + 373, + 476 + ], + "lines": [ + { + "bbox": [ + 237, + 441, + 373, + 476 + ], + "spans": [ + { + "bbox": [ + 237, + 441, + 373, + 476 + ], + "score": 0.94, + "content": "L _ { \\mathrm { s s } } = \\sum _ { i = 1 } ^ { N } \\left\\| \\int _ { T } ^ { 2 T } | f _ { \\theta } ( \\mathbf { z } _ { i } ( t ) ) | \\mathrm { d } t \\right\\| ,", + "type": "interline_equation", + "image_path": "dead5c3a609cf3ee30da32ebeb2ca3d5179c956d041557ad30986bda72525deb.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 237, + 441, + 373, + 458.5 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 237, + 458.5, + 373, + 476.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 479, + 505, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 480, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 134, + 492 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 481, + 145, + 490 + ], + "score": 0.81, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 480, + 345, + 492 + ], + "score": 1.0, + "content": "is the number of samples in the training set and", + "type": "text" + }, + { + "bbox": [ + 345, + 480, + 366, + 492 + ], + "score": 0.92, + "content": "{ \\bf z } _ { i } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 480, + 506, + 492 + ], + "score": 1.0, + "content": "is the solution whose initial state", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 220, + 504 + ], + "score": 1.0, + "content": "equals to the feature of the", + "type": "text" + }, + { + "bbox": [ + 220, + 491, + 233, + 501 + ], + "score": 0.86, + "content": "i ^ { \\mathrm { t h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 490, + 370, + 504 + ], + "score": 1.0, + "content": "sample. The regularization term", + "type": "text" + }, + { + "bbox": [ + 370, + 492, + 385, + 502 + ], + "score": 0.89, + "content": "L _ { \\mathrm { s s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "is termed as the steady-state", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "loss. This terminology “steady state” is borrowed from the dynamical systems literature. In a stable", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 514, + 504, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 504, + 525 + ], + "score": 1.0, + "content": "dynamical system, the states stabilize around a fixed point, known as the steady-state, as time tends", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 234, + 537 + ], + "score": 1.0, + "content": "to infinity. If we can ensure that", + "type": "text" + }, + { + "bbox": [ + 234, + 524, + 249, + 535 + ], + "score": 0.9, + "content": "L _ { \\mathrm { s s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 524, + 471, + 537 + ], + "score": 1.0, + "content": "is small, for each sample, the outputs of all the points in", + "type": "text" + }, + { + "bbox": [ + 472, + 524, + 486, + 535 + ], + "score": 0.88, + "content": "\\mathbb { M } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "will", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 535, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 172, + 547 + ], + "score": 1.0, + "content": "stabilize around", + "type": "text" + }, + { + "bbox": [ + 173, + 535, + 197, + 547 + ], + "score": 0.92, + "content": "{ \\bf z } _ { i } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 535, + 506, + 547 + ], + "score": 1.0, + "content": ". Consequently, the model is robust. This modification of the neural ODE is", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 546, + 281, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 256, + 558 + ], + "score": 1.0, + "content": "dubbed Time-invariant steady neural", + "type": "text" + }, + { + "bbox": [ + 256, + 546, + 278, + 556 + ], + "score": 0.36, + "content": "O D E", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 546, + 281, + 558 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 480, + 506, + 558 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 572, + 387, + 584 + ], + "lines": [ + { + "bbox": [ + 106, + 573, + 388, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 388, + 585 + ], + "score": 1.0, + "content": "4.2 EVALUATING ROBUSTNESS OF TISODE-BASED CLASSIFIERS", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 594, + 505, + 638 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "Here, we conduct experiments to evaluate the robustness of our proposed TisODE, and compare", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "TisODE-based models with the vanilla ODENets. We train all models with original non-perturbed", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 617, + 504, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 504, + 628 + ], + "score": 1.0, + "content": "images together with their Gaussian perturbed versions. The regularization parameter for the steady-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 626, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 144, + 639 + ], + "score": 1.0, + "content": "state loss", + "type": "text" + }, + { + "bbox": [ + 145, + 627, + 160, + 638 + ], + "score": 0.9, + "content": "L _ { \\mathrm { s s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 626, + 504, + 639 + ], + "score": 1.0, + "content": "is set to be 0.1. All other hyperparameters are exactly the same as those in Section 3.3.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 594, + 505, + 639 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "From the results in Table 3, we can see that our proposed TisODE-based models are clearly more", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "robust compared to vanilla ODENets. On the MNIST dataset, when combating FGSM-0.3 attacks,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "the TisODE-based models outperform vanilla ODENets by more than 4 percentage points. For the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 104, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "FGSM-0.5 adversarial examples, the accuracy of the TisODE-based model is 6 percentage points", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "better. On the SVHN dataset, the TisODE-based models perform better in terms of all forms of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "adversarial examples. On the ImgNet10 dataset, the TisODE-based models also outperform vanilla", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "ODE-based models on all types of perturbations. In the presence of FGSM and PGD-5/255 exam-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 721, + 379, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 379, + 733 + ], + "score": 1.0, + "content": "ples, the accuracies are enhanced by more than 2 percentage points.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5, + "bbox_fs": [ + 104, + 643, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 136, + 133, + 473, + 288 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 505, + 125 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 265, + 93 + ], + "score": 1.0, + "content": "Table 3: Classification accuracy (mean", + "type": "text" + }, + { + "bbox": [ + 265, + 81, + 275, + 91 + ], + "score": 0.69, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 80, + 301, + 93 + ], + "score": 1.0, + "content": "std in", + "type": "text" + }, + { + "bbox": [ + 301, + 81, + 311, + 91 + ], + "score": 0.77, + "content": "\\%", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 80, + 506, + 93 + ], + "score": 1.0, + "content": ") on perturbed images from MNIST, SVHN and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 91, + 505, + 104 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 505, + 104 + ], + "score": 1.0, + "content": "ImgNet10. To evaluate the robustness of classifiers, we use three types of perturbations, namely", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 103, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 103, + 317, + 114 + ], + "score": 1.0, + "content": "zero-mean Gaussian noise with standard deviation", + "type": "text" + }, + { + "bbox": [ + 318, + 104, + 325, + 112 + ], + "score": 0.64, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 103, + 505, + 114 + ], + "score": 1.0, + "content": ", FGSM attack and PGD attack. From the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 113, + 473, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 473, + 126 + ], + "score": 1.0, + "content": "results, the proposed TisODE effectively improve the robustness of the vanilla neural ODE.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 136, + 133, + 473, + 288 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 136, + 133, + 473, + 288 + ], + "spans": [ + { + "bbox": [ + 136, + 133, + 473, + 288 + ], + "score": 0.984, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
TisODE99.6±0.075.7±1.426.5±3.867.4±1.513.2±1.0
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
TisODE94.9±0.151.6±1.238.2±1.952.0±0.928.2±0.3
ImgNet10g=25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
TisODE92.8±0.444.3±0.731.4±1.131.1±1.214.5±1.1
", + "type": "table", + "image_path": "564d88a3d2f0ff8bcf0003170d92128d6d885f175ce78aa9082a0b2fbed99adb.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 136, + 133, + 473, + 184.66666666666666 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 136, + 184.66666666666666, + 473, + 236.33333333333331 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 136, + 236.33333333333331, + 473, + 288.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 105, + 300, + 475, + 321 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 298, + 477, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 477, + 312 + ], + "score": 1.0, + "content": "4.3 TISODE - A GENERALLY APPLICABLE DROP-IN TECHNIQUE FOR IMPROVING THE", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 312, + 276, + 322 + ], + "spans": [ + { + "bbox": [ + 130, + 312, + 276, + 322 + ], + "score": 1.0, + "content": "ROBUSTNESS OF DEEP NETWORKS", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 330, + 505, + 430 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "score": 1.0, + "content": "In view of the excellent robustness of the TisODE, we claim that the proposed TisODE can be used", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "as a general drop-in module for improving the robustness of deep networks. We support this claim", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 353, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 505, + 364 + ], + "score": 1.0, + "content": "by showing the TisODE can work in conjunction with other state-of-the-art techniques and further", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "score": 1.0, + "content": "boost the models’ robustness. These techniques include the feature denoising (FDn) method (Xie", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 374, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 387 + ], + "score": 1.0, + "content": "et al., 2019) and the input randomization (IR) method (Xie et al., 2017). We conduct experiments on", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 385, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 506, + 399 + ], + "score": 1.0, + "content": "the MNIST and SVHN datasets. All models are trained with original non-perturbed images together", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "score": 1.0, + "content": "with their Gaussian perturbed versions. We show that models using the FDn/IRd technique becomes", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 407, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 420 + ], + "score": 1.0, + "content": "much more robust when equipped with the TisODE. In the FDn experiments, the dot-product non-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 417, + 474, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 474, + 431 + ], + "score": 1.0, + "content": "local denoising layer (Xie et al., 2019) is added to the head of the fully-connected classifier.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 13 + }, + { + "type": "table", + "bbox": [ + 130, + 494, + 478, + 643 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 440, + 505, + 485 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 440, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 264, + 453 + ], + "score": 1.0, + "content": "Table 4: Classification accuracy (mean", + "type": "text" + }, + { + "bbox": [ + 265, + 442, + 275, + 451 + ], + "score": 0.66, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 440, + 300, + 453 + ], + "score": 1.0, + "content": "std in", + "type": "text" + }, + { + "bbox": [ + 301, + 442, + 310, + 451 + ], + "score": 0.79, + "content": "\\%", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 440, + 505, + 453 + ], + "score": 1.0, + "content": ") on perturbed images from MNIST and SVHN.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 451, + 506, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 464 + ], + "score": 1.0, + "content": "We evaluate against three types of perturbations, namely zero-mean Gaussian noise with standard", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 146, + 475 + ], + "score": 1.0, + "content": "deviation", + "type": "text" + }, + { + "bbox": [ + 146, + 465, + 154, + 473 + ], + "score": 0.73, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 463, + 505, + 475 + ], + "score": 1.0, + "content": ", FGSM attack and PGD attack. From the results, upon the CNNs modified with FDn", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 473, + 345, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 345, + 486 + ], + "score": 1.0, + "content": "and IRd, using TisODE can further improve the robustness.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "table_body", + "bbox": [ + 130, + 494, + 478, + 643 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 130, + 494, + 478, + 643 + ], + "spans": [ + { + "bbox": [ + 130, + 494, + 478, + 643 + ], + "score": 0.985, + "html": "
Gaussian noiseAdversarial attack
MNISTσ= 100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
CNN-FDn TisODE-FDn99.0±0.174.0±4.132.6±5.358.9±4.08.2±2.6
CNN-IRd99.4±0.0 95.3±0.980.6±2.3 78.1±2.240.4±5.7 36.7±2.172.6±2.4 79.6±1.928.2±3.6 55.5±2.9
TisODE-IRd97.6±0.186.8±2.349.1±0.288.8±0.966.0±0.9
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN
90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
CNN-FDn92.4±0.143.8±1.431.5±3.040.0±2.619.6±3.4
TisODE-FDn95.2±0.157.8±1.748.2±2.053.4±2.932.3±1.0
CNN-IRd84.9±1.265.8±0.454.7±1.274.0±0.564.5±0.8
TisODE-IRd91.7±0.574.4±1.261.9±1.881.6±0.871.0±0.5
", + "type": "table", + "image_path": "b0ade5e612e54414b3aa94f7b78718bad3bcceec3355deacbdb5d82dc1e7092d.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 130, + 494, + 478, + 543.6666666666666 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 130, + 543.6666666666666, + 478, + 593.3333333333333 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 130, + 593.3333333333333, + 478, + 642.9999999999999 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 21.25 + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "From Table 4, we observe that both FDn and IRd can effectively improve the adversarial robustness", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 665, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 679 + ], + "score": 1.0, + "content": "of vanilla CNN models (CNN-FDn, CNN-IRd). Furthermore, combining our proposed TisODE", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 677, + 505, + 688 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 688 + ], + "score": 1.0, + "content": "with FDn or IRd (TisODE-FDn, TisODE-IRd), the adversarial robustness of the resultant model", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 104, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "is significantly enhanced. For example, on the MNIST dataset, the additional use of our TisODE", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "increases the accuracies on the PGD-0.3 examples by at least 10 percentage points for both FDn", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 110, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 110, + 710, + 132, + 720 + ], + "score": 0.85, + "content": "8 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 710, + 144, + 722 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 145, + 710, + 174, + 721 + ], + "score": 0.84, + "content": "2 8 . 2 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 710, + 216, + 722 + ], + "score": 1.0, + "content": "and IRd", + "type": "text" + }, + { + "bbox": [ + 216, + 710, + 244, + 721 + ], + "score": 0.82, + "content": "5 5 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 710, + 257, + 722 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 257, + 710, + 285, + 721 + ], + "score": 0.86, + "content": "6 6 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "). However, on both MNIST and SVHN datasets, the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 104, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "IRd technique improves the robustness against adversarial examples, but its performance is worse", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 136, + 133, + 473, + 288 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 505, + 125 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 265, + 93 + ], + "score": 1.0, + "content": "Table 3: Classification accuracy (mean", + "type": "text" + }, + { + "bbox": [ + 265, + 81, + 275, + 91 + ], + "score": 0.69, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 80, + 301, + 93 + ], + "score": 1.0, + "content": "std in", + "type": "text" + }, + { + "bbox": [ + 301, + 81, + 311, + 91 + ], + "score": 0.77, + "content": "\\%", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 80, + 506, + 93 + ], + "score": 1.0, + "content": ") on perturbed images from MNIST, SVHN and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 91, + 505, + 104 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 505, + 104 + ], + "score": 1.0, + "content": "ImgNet10. To evaluate the robustness of classifiers, we use three types of perturbations, namely", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 103, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 103, + 317, + 114 + ], + "score": 1.0, + "content": "zero-mean Gaussian noise with standard deviation", + "type": "text" + }, + { + "bbox": [ + 318, + 104, + 325, + 112 + ], + "score": 0.64, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 103, + 505, + 114 + ], + "score": 1.0, + "content": ", FGSM attack and PGD attack. From the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 113, + 473, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 473, + 126 + ], + "score": 1.0, + "content": "results, the proposed TisODE effectively improve the robustness of the vanilla neural ODE.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 136, + 133, + 473, + 288 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 136, + 133, + 473, + 288 + ], + "spans": [ + { + "bbox": [ + 136, + 133, + 473, + 288 + ], + "score": 0.984, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
TisODE99.6±0.075.7±1.426.5±3.867.4±1.513.2±1.0
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
TisODE94.9±0.151.6±1.238.2±1.952.0±0.928.2±0.3
ImgNet10g=25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
TisODE92.8±0.444.3±0.731.4±1.131.1±1.214.5±1.1
", + "type": "table", + "image_path": "564d88a3d2f0ff8bcf0003170d92128d6d885f175ce78aa9082a0b2fbed99adb.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 136, + 133, + 473, + 184.66666666666666 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 136, + 184.66666666666666, + 473, + 236.33333333333331 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 136, + 236.33333333333331, + 473, + 288.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 105, + 300, + 475, + 321 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 298, + 477, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 477, + 312 + ], + "score": 1.0, + "content": "4.3 TISODE - A GENERALLY APPLICABLE DROP-IN TECHNIQUE FOR IMPROVING THE", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 312, + 276, + 322 + ], + "spans": [ + { + "bbox": [ + 130, + 312, + 276, + 322 + ], + "score": 1.0, + "content": "ROBUSTNESS OF DEEP NETWORKS", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 330, + 505, + 430 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "score": 1.0, + "content": "In view of the excellent robustness of the TisODE, we claim that the proposed TisODE can be used", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "as a general drop-in module for improving the robustness of deep networks. We support this claim", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 353, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 505, + 364 + ], + "score": 1.0, + "content": "by showing the TisODE can work in conjunction with other state-of-the-art techniques and further", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "score": 1.0, + "content": "boost the models’ robustness. These techniques include the feature denoising (FDn) method (Xie", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 374, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 387 + ], + "score": 1.0, + "content": "et al., 2019) and the input randomization (IR) method (Xie et al., 2017). We conduct experiments on", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 385, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 506, + 399 + ], + "score": 1.0, + "content": "the MNIST and SVHN datasets. All models are trained with original non-perturbed images together", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "score": 1.0, + "content": "with their Gaussian perturbed versions. We show that models using the FDn/IRd technique becomes", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 407, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 420 + ], + "score": 1.0, + "content": "much more robust when equipped with the TisODE. In the FDn experiments, the dot-product non-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 417, + 474, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 474, + 431 + ], + "score": 1.0, + "content": "local denoising layer (Xie et al., 2019) is added to the head of the fully-connected classifier.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 331, + 506, + 431 + ] + }, + { + "type": "table", + "bbox": [ + 130, + 494, + 478, + 643 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 440, + 505, + 485 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 440, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 264, + 453 + ], + "score": 1.0, + "content": "Table 4: Classification accuracy (mean", + "type": "text" + }, + { + "bbox": [ + 265, + 442, + 275, + 451 + ], + "score": 0.66, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 440, + 300, + 453 + ], + "score": 1.0, + "content": "std in", + "type": "text" + }, + { + "bbox": [ + 301, + 442, + 310, + 451 + ], + "score": 0.79, + "content": "\\%", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 440, + 505, + 453 + ], + "score": 1.0, + "content": ") on perturbed images from MNIST and SVHN.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 451, + 506, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 464 + ], + "score": 1.0, + "content": "We evaluate against three types of perturbations, namely zero-mean Gaussian noise with standard", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 146, + 475 + ], + "score": 1.0, + "content": "deviation", + "type": "text" + }, + { + "bbox": [ + 146, + 465, + 154, + 473 + ], + "score": 0.73, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 463, + 505, + 475 + ], + "score": 1.0, + "content": ", FGSM attack and PGD attack. From the results, upon the CNNs modified with FDn", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 473, + 345, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 345, + 486 + ], + "score": 1.0, + "content": "and IRd, using TisODE can further improve the robustness.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "table_body", + "bbox": [ + 130, + 494, + 478, + 643 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 130, + 494, + 478, + 643 + ], + "spans": [ + { + "bbox": [ + 130, + 494, + 478, + 643 + ], + "score": 0.985, + "html": "
Gaussian noiseAdversarial attack
MNISTσ= 100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
CNN-FDn TisODE-FDn99.0±0.174.0±4.132.6±5.358.9±4.08.2±2.6
CNN-IRd99.4±0.0 95.3±0.980.6±2.3 78.1±2.240.4±5.7 36.7±2.172.6±2.4 79.6±1.928.2±3.6 55.5±2.9
TisODE-IRd97.6±0.186.8±2.349.1±0.288.8±0.966.0±0.9
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN
90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
CNN-FDn92.4±0.143.8±1.431.5±3.040.0±2.619.6±3.4
TisODE-FDn95.2±0.157.8±1.748.2±2.053.4±2.932.3±1.0
CNN-IRd84.9±1.265.8±0.454.7±1.274.0±0.564.5±0.8
TisODE-IRd91.7±0.574.4±1.261.9±1.881.6±0.871.0±0.5
", + "type": "table", + "image_path": "b0ade5e612e54414b3aa94f7b78718bad3bcceec3355deacbdb5d82dc1e7092d.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 130, + 494, + 478, + 543.6666666666666 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 130, + 543.6666666666666, + 478, + 593.3333333333333 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 130, + 593.3333333333333, + 478, + 642.9999999999999 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 21.25 + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "From Table 4, we observe that both FDn and IRd can effectively improve the adversarial robustness", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 665, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 679 + ], + "score": 1.0, + "content": "of vanilla CNN models (CNN-FDn, CNN-IRd). Furthermore, combining our proposed TisODE", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 677, + 505, + 688 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 688 + ], + "score": 1.0, + "content": "with FDn or IRd (TisODE-FDn, TisODE-IRd), the adversarial robustness of the resultant model", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 104, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "is significantly enhanced. For example, on the MNIST dataset, the additional use of our TisODE", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "increases the accuracies on the PGD-0.3 examples by at least 10 percentage points for both FDn", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 110, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 110, + 710, + 132, + 720 + ], + "score": 0.85, + "content": "8 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 710, + 144, + 722 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 145, + 710, + 174, + 721 + ], + "score": 0.84, + "content": "2 8 . 2 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 710, + 216, + 722 + ], + "score": 1.0, + "content": "and IRd", + "type": "text" + }, + { + "bbox": [ + 216, + 710, + 244, + 721 + ], + "score": 0.82, + "content": "5 5 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 710, + 257, + 722 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 257, + 710, + 285, + 721 + ], + "score": 0.86, + "content": "6 6 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "). However, on both MNIST and SVHN datasets, the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 104, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "IRd technique improves the robustness against adversarial examples, but its performance is worse", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "on random Gaussian noise. With the help of the TisODE, the degradation in the robustness against", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 326, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 326, + 105 + ], + "score": 1.0, + "content": "random Gaussian noise can be effectively ameliorated.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 28, + "bbox_fs": [ + 104, + 655, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "on random Gaussian noise. With the help of the TisODE, the degradation in the robustness against", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 326, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 326, + 105 + ], + "score": 1.0, + "content": "random Gaussian noise can be effectively ameliorated.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "title", + "bbox": [ + 108, + 140, + 214, + 153 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 217, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 217, + 154 + ], + "score": 1.0, + "content": "5 RELATED WORKS", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 176, + 503, + 199 + ], + "lines": [ + { + "bbox": [ + 104, + 173, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 104, + 173, + 505, + 191 + ], + "score": 1.0, + "content": "In this section, we briefly review related works on the neural ODE and works concerning improving", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 188, + 266, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 266, + 199 + ], + "score": 1.0, + "content": "the robustness of deep neural networks.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 505, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 204, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 505, + 217 + ], + "score": 1.0, + "content": "Neural ODE: The neural ODE (Chen et al., 2018) method models the input and output as two", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "score": 1.0, + "content": "states of a continuous-time dynamical system by approximating the dynamics of this system with", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 225, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 240 + ], + "score": 1.0, + "content": "trainable layers. Before the proposal of neural ODE, the idea of modeling nonlinear mappings using", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "continuous-time dynamical systems was proposed in Weinan (2017). Lu et al. (2017) also showed", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 249, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 505, + 261 + ], + "score": 1.0, + "content": "that several popular network architectures could be interpreted as the discretization of a continuous-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 259, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 505, + 271 + ], + "score": 1.0, + "content": "time ODE. For example, the ResNet (He et al., 2016) and PolyNet (Zhang et al., 2017) are associated", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 270, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 505, + 283 + ], + "score": 1.0, + "content": "with the Euler scheme and the FractalNet (Larsson et al., 2016) is related to the Runge-Kutta scheme.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "In contrast to these discretization models, neural ODEs are endowed with an intrinsic invertibility", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 292, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 505, + 304 + ], + "score": 1.0, + "content": "property, which yields a family of invertible models for solving inverse problems (Ardizzone et al.,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 303, + 320, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 320, + 315 + ], + "score": 1.0, + "content": "2018), such as the FFJORD (Grathwohl et al., 2018).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 321, + 505, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 504, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 504, + 333 + ], + "score": 1.0, + "content": "Recently, many researchers have conducted studies on neural ODEs from the perspectives of opti-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 332, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 343 + ], + "score": 1.0, + "content": "mization techniques, approximation capabilities, and generalization. Concerning the optimization", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 342, + 504, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 504, + 353 + ], + "score": 1.0, + "content": "of neural ODEs, the auto-differentiation techniques can effectively train ODENets, but the train-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 353, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 365 + ], + "score": 1.0, + "content": "ing procedure is computationally and memory inefficient. To address this problem, Chen et al.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "score": 1.0, + "content": "(2018) proposed to compute gradients using the adjoint sensitivity method (Pontryagin, 2018), in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 374, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 387 + ], + "score": 1.0, + "content": "which there is no need to store any intermediate quantities of the forward pass. Also in Quaglino", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 385, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 399 + ], + "score": 1.0, + "content": "et al. (2019), the authors proposed the SNet which accelerates the neural ODEs by expressing their", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 397, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 410 + ], + "score": 1.0, + "content": "dynamics as truncated series of Legendre polynomials. Concerning the approximation capability,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "score": 1.0, + "content": "Dupont et al. (2019) pointed out the limitations in approximation capabilities of neural ODEs be-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "score": 1.0, + "content": "cause of the preserving of input topology. The authors proposed an augmented neural ODE which", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "score": 1.0, + "content": "increases the dimension of states by concatenating zeros so that complex mappings can be learned", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 441, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 505, + 453 + ], + "score": 1.0, + "content": "with simple flow. The most relevant work to ours concerns strategies to improve the generalization", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "of neural ODEs. In Liu et al. (2019), the authors proposed the neural stochastic differential equation", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 463, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 505, + 474 + ], + "score": 1.0, + "content": "(SDE) by injecting random noise to the dynamics function and showed that the generalization and", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 474, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 505, + 486 + ], + "score": 1.0, + "content": "robustness of vanilla neural ODEs could be improved. However, our improvement on the neural", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "ODEs is explored from a different perspective by introducing constraints on the flow. We empiri-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 495, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 506, + 508 + ], + "score": 1.0, + "content": "cally found that our proposal and the neural SDE can work in tandem to further boost the robustness", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 506, + 174, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 174, + 518 + ], + "score": 1.0, + "content": "of neural ODEs.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 536 + ], + "score": 1.0, + "content": "Robust Improvement: A straightforward way of improving the robustness of a model is to smooth", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "the loss surface by controlling the spectral norm of the Jacobian matrix of the loss function (Sokolic´", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 544, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 505, + 558 + ], + "score": 1.0, + "content": "et al., 2017). In terms of adversarial examples (Carlini & Wagner, 2017; Chen et al., 2017), re-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "score": 1.0, + "content": "searchers have proposed adversarial training strategies (Madry et al., 2017; Elsayed et al., 2018;", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "Trame`r et al., 2017) in which the model is fine-tuned with adversarial examples generated in real-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 579, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 590 + ], + "score": 1.0, + "content": "time. However, generating adversarial examples is not computationally efficient, and there exists", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 588, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 505, + 601 + ], + "score": 1.0, + "content": "a trade-off between the adversarial robustness and the performance on original non-perturbed im-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "ages (Yan et al., 2018; Tsipras et al., 2018). In Wang et al. (2018a), the authors model the ResNet as", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 611, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 623 + ], + "score": 1.0, + "content": "a transport equation, in which the adversarial vulnerability can be interpreted as the irregularity of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "score": 1.0, + "content": "the decision boundary. Consequently, a diffusion term is introduced to enhance the robustness of the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "neural nets. Besides, there are also some works that propose novel architectural defense mechanisms", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "against adversarial examples. For example, Xie et al. (2017) utilized random resizing and random", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "padding to destroy the specific structure of adversarial perturbations; Wang et al. (2018b) and Wang", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "et al. (2018c) improved the robustness of neural networks by replacing the output layers with novel", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "interpolating functions; In Xie et al. (2019), the authors designed a feature denoising filter that can", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 689, + 504, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 504, + 700 + ], + "score": 1.0, + "content": "remove the perturbation’s pattern from feature maps. In this work, we explore the intrinsic robust-", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "ness of a specific novel architecture (neural ODE), and show that the proposed TisODE can improve", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "the robustness of deep networks and can also work in tandem with these state-of-the-art methods", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 720, + 334, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 334, + 732 + ], + "score": 1.0, + "content": "Xie et al. (2017; 2019) to achieve further improvements.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 42 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 505, + 105 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 140, + 214, + 153 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 217, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 217, + 154 + ], + "score": 1.0, + "content": "5 RELATED WORKS", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 176, + 503, + 199 + ], + "lines": [ + { + "bbox": [ + 104, + 173, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 104, + 173, + 505, + 191 + ], + "score": 1.0, + "content": "In this section, we briefly review related works on the neural ODE and works concerning improving", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 188, + 266, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 266, + 199 + ], + "score": 1.0, + "content": "the robustness of deep neural networks.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 173, + 505, + 199 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 505, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 204, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 505, + 217 + ], + "score": 1.0, + "content": "Neural ODE: The neural ODE (Chen et al., 2018) method models the input and output as two", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "score": 1.0, + "content": "states of a continuous-time dynamical system by approximating the dynamics of this system with", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 225, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 240 + ], + "score": 1.0, + "content": "trainable layers. Before the proposal of neural ODE, the idea of modeling nonlinear mappings using", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "continuous-time dynamical systems was proposed in Weinan (2017). Lu et al. (2017) also showed", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 249, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 505, + 261 + ], + "score": 1.0, + "content": "that several popular network architectures could be interpreted as the discretization of a continuous-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 259, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 505, + 271 + ], + "score": 1.0, + "content": "time ODE. For example, the ResNet (He et al., 2016) and PolyNet (Zhang et al., 2017) are associated", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 270, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 505, + 283 + ], + "score": 1.0, + "content": "with the Euler scheme and the FractalNet (Larsson et al., 2016) is related to the Runge-Kutta scheme.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "In contrast to these discretization models, neural ODEs are endowed with an intrinsic invertibility", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 292, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 505, + 304 + ], + "score": 1.0, + "content": "property, which yields a family of invertible models for solving inverse problems (Ardizzone et al.,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 303, + 320, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 320, + 315 + ], + "score": 1.0, + "content": "2018), such as the FFJORD (Grathwohl et al., 2018).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 204, + 506, + 315 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 321, + 505, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 504, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 504, + 333 + ], + "score": 1.0, + "content": "Recently, many researchers have conducted studies on neural ODEs from the perspectives of opti-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 332, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 343 + ], + "score": 1.0, + "content": "mization techniques, approximation capabilities, and generalization. Concerning the optimization", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 342, + 504, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 504, + 353 + ], + "score": 1.0, + "content": "of neural ODEs, the auto-differentiation techniques can effectively train ODENets, but the train-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 353, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 365 + ], + "score": 1.0, + "content": "ing procedure is computationally and memory inefficient. To address this problem, Chen et al.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "score": 1.0, + "content": "(2018) proposed to compute gradients using the adjoint sensitivity method (Pontryagin, 2018), in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 374, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 387 + ], + "score": 1.0, + "content": "which there is no need to store any intermediate quantities of the forward pass. Also in Quaglino", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 385, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 399 + ], + "score": 1.0, + "content": "et al. (2019), the authors proposed the SNet which accelerates the neural ODEs by expressing their", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 397, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 410 + ], + "score": 1.0, + "content": "dynamics as truncated series of Legendre polynomials. Concerning the approximation capability,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "score": 1.0, + "content": "Dupont et al. (2019) pointed out the limitations in approximation capabilities of neural ODEs be-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "score": 1.0, + "content": "cause of the preserving of input topology. The authors proposed an augmented neural ODE which", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "score": 1.0, + "content": "increases the dimension of states by concatenating zeros so that complex mappings can be learned", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 441, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 505, + 453 + ], + "score": 1.0, + "content": "with simple flow. The most relevant work to ours concerns strategies to improve the generalization", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "of neural ODEs. In Liu et al. (2019), the authors proposed the neural stochastic differential equation", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 463, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 505, + 474 + ], + "score": 1.0, + "content": "(SDE) by injecting random noise to the dynamics function and showed that the generalization and", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 474, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 505, + 486 + ], + "score": 1.0, + "content": "robustness of vanilla neural ODEs could be improved. However, our improvement on the neural", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "ODEs is explored from a different perspective by introducing constraints on the flow. We empiri-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 495, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 506, + 508 + ], + "score": 1.0, + "content": "cally found that our proposal and the neural SDE can work in tandem to further boost the robustness", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 506, + 174, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 174, + 518 + ], + "score": 1.0, + "content": "of neural ODEs.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 320, + 506, + 518 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 536 + ], + "score": 1.0, + "content": "Robust Improvement: A straightforward way of improving the robustness of a model is to smooth", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "the loss surface by controlling the spectral norm of the Jacobian matrix of the loss function (Sokolic´", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 544, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 505, + 558 + ], + "score": 1.0, + "content": "et al., 2017). In terms of adversarial examples (Carlini & Wagner, 2017; Chen et al., 2017), re-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "score": 1.0, + "content": "searchers have proposed adversarial training strategies (Madry et al., 2017; Elsayed et al., 2018;", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "Trame`r et al., 2017) in which the model is fine-tuned with adversarial examples generated in real-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 579, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 590 + ], + "score": 1.0, + "content": "time. However, generating adversarial examples is not computationally efficient, and there exists", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 588, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 505, + 601 + ], + "score": 1.0, + "content": "a trade-off between the adversarial robustness and the performance on original non-perturbed im-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "ages (Yan et al., 2018; Tsipras et al., 2018). In Wang et al. (2018a), the authors model the ResNet as", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 611, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 623 + ], + "score": 1.0, + "content": "a transport equation, in which the adversarial vulnerability can be interpreted as the irregularity of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "score": 1.0, + "content": "the decision boundary. Consequently, a diffusion term is introduced to enhance the robustness of the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "neural nets. Besides, there are also some works that propose novel architectural defense mechanisms", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "against adversarial examples. For example, Xie et al. (2017) utilized random resizing and random", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "padding to destroy the specific structure of adversarial perturbations; Wang et al. (2018b) and Wang", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "et al. (2018c) improved the robustness of neural networks by replacing the output layers with novel", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "interpolating functions; In Xie et al. (2019), the authors designed a feature denoising filter that can", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 689, + 504, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 504, + 700 + ], + "score": 1.0, + "content": "remove the perturbation’s pattern from feature maps. In this work, we explore the intrinsic robust-", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "ness of a specific novel architecture (neural ODE), and show that the proposed TisODE can improve", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "the robustness of deep networks and can also work in tandem with these state-of-the-art methods", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 720, + 334, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 334, + 732 + ], + "score": 1.0, + "content": "Xie et al. (2017; 2019) to achieve further improvements.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 523, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 195, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 197, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 197, + 97 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "In this paper, we first empirically study the robustness of neural ODEs. Our studies reveal that neural", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 116, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 505, + 130 + ], + "score": 1.0, + "content": "ODE-based models are superior in terms of robustness compared to CNN models. We then explore", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "score": 1.0, + "content": "how to further boost the robustness of vanilla neural ODEs and propose the TisODE. Finally, we", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 505, + 152 + ], + "score": 1.0, + "content": "show that the proposed TisODE outperforms the vanilla neural ODE and also can work in conjunc-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 150, + 505, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 505, + 163 + ], + "score": 1.0, + "content": "tion with other state-of-the-art techniques to further improve the robustness of deep networks. Thus,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 160, + 442, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 160, + 442, + 173 + ], + "score": 1.0, + "content": "the TisODE method is an effective drop-in module for building robust deep models.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 108, + 189, + 219, + 201 + ], + "lines": [ + { + "bbox": [ + 106, + 189, + 220, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 220, + 203 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENT", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 104, + 213, + 503, + 235 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 504, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 504, + 226 + ], + "score": 1.0, + "content": "This work is funded by a Singapore National Research Foundation (NRF) Fellowship (R-263-000-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 222, + 151, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 151, + 237 + ], + "score": 1.0, + "content": "D02-281).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 104, + 241, + 504, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "score": 1.0, + "content": "Jiashi Feng was partially supported by NUS IDS R-263-000-C67-646, ECRA R-263-000-C87-133,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 253, + 373, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 373, + 264 + ], + "score": 1.0, + "content": "MOE Tier-II R-263-000-D17-112 and AI.SG R-263-000-D97-490", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 752, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 195, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 197, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 197, + 97 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "In this paper, we first empirically study the robustness of neural ODEs. Our studies reveal that neural", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 116, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 505, + 130 + ], + "score": 1.0, + "content": "ODE-based models are superior in terms of robustness compared to CNN models. We then explore", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 505, + 141 + ], + "score": 1.0, + "content": "how to further boost the robustness of vanilla neural ODEs and propose the TisODE. Finally, we", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 505, + 152 + ], + "score": 1.0, + "content": "show that the proposed TisODE outperforms the vanilla neural ODE and also can work in conjunc-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 150, + 505, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 505, + 163 + ], + "score": 1.0, + "content": "tion with other state-of-the-art techniques to further improve the robustness of deep networks. Thus,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 160, + 442, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 160, + 442, + 173 + ], + "score": 1.0, + "content": "the TisODE method is an effective drop-in module for building robust deep models.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 106, + 505, + 173 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 189, + 219, + 201 + ], + "lines": [ + { + "bbox": [ + 106, + 189, + 220, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 220, + 203 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENT", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 104, + 213, + 503, + 235 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 504, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 504, + 226 + ], + "score": 1.0, + "content": "This work is funded by a Singapore National Research Foundation (NRF) Fellowship (R-263-000-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 222, + 151, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 151, + 237 + ], + "score": 1.0, + "content": "D02-281).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 213, + 504, + 237 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 241, + 504, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "score": 1.0, + "content": "Jiashi Feng was partially supported by NUS IDS R-263-000-C67-646, ECRA R-263-000-C87-133,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 253, + 373, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 373, + 264 + ], + "score": 1.0, + "content": "MOE Tier-II R-263-000-D17-112 and AI.SG R-263-000-D97-490", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 241, + 506, + 264 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 100, + 504, + 134 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 505, + 113 + ], + "score": 1.0, + "content": "Lynton Ardizzone, Jakob Kruse, Sebastian Wirkert, Daniel Rahner, Eric W Pellegrini, Ralf S", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 111, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 505, + 124 + ], + "score": 1.0, + "content": "Klessen, Lena Maier-Hein, Carsten Rother, and Ullrich Ko¨the. Analyzing inverse problems with", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 122, + 388, + 135 + ], + "spans": [ + { + "bbox": [ + 116, + 122, + 388, + 135 + ], + "score": 1.0, + "content": "invertible neural networks. arXiv preprint arXiv:1808.04730, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 105, + 142, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "Nicholas Carlini and David Wagner. Towards evaluating the robustness of neural networks. In 2017", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 153, + 407, + 165 + ], + "spans": [ + { + "bbox": [ + 116, + 153, + 407, + 165 + ], + "score": 1.0, + "content": "IEEE Symposium on Security and Privacy (SP), pp. 39–57. IEEE, 2017.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 505, + 217 + ], + "lines": [ + { + "bbox": [ + 105, + 172, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 505, + 185 + ], + "score": 1.0, + "content": "Pin-Yu Chen, Huan Zhang, Yash Sharma, Jinfeng Yi, and Cho-Jui Hsieh. Zoo: Zeroth order opti-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 183, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 116, + 183, + 505, + 195 + ], + "score": 1.0, + "content": "mization based black-box attacks to deep neural networks without training substitute models. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 194, + 506, + 207 + ], + "spans": [ + { + "bbox": [ + 115, + 194, + 506, + 207 + ], + "score": 1.0, + "content": "Proceedings of the 10th ACM Workshop on Artificial Intelligence and Security, pp. 15–26. ACM,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 205, + 143, + 217 + ], + "spans": [ + { + "bbox": [ + 115, + 205, + 143, + 217 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 105, + 225, + 502, + 248 + ], + "lines": [ + { + "bbox": [ + 106, + 225, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 505, + 238 + ], + "score": 1.0, + "content": "Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differ-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 236, + 495, + 249 + ], + "spans": [ + { + "bbox": [ + 116, + 236, + 495, + 249 + ], + "score": 1.0, + "content": "ential equations. In Advances in neural information processing systems, pp. 6571–6583, 2018.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 255, + 504, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 255, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 505, + 269 + ], + "score": 1.0, + "content": "Earl A Coddington and Norman Levinson. Theory of ordinary differential equations. Tata McGraw-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 267, + 204, + 278 + ], + "spans": [ + { + "bbox": [ + 116, + 267, + 204, + 278 + ], + "score": 1.0, + "content": "Hill Education, 1955.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 285, + 505, + 320 + ], + "lines": [ + { + "bbox": [ + 106, + 286, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 504, + 299 + ], + "score": 1.0, + "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hi-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 297, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 115, + 297, + 505, + 311 + ], + "score": 1.0, + "content": "erarchical image database. In 2009 IEEE conference on computer vision and pattern recognition,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 309, + 218, + 320 + ], + "spans": [ + { + "bbox": [ + 115, + 309, + 218, + 320 + ], + "score": 1.0, + "content": "pp. 248–255. Ieee, 2009.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 104, + 327, + 505, + 351 + ], + "lines": [ + { + "bbox": [ + 105, + 326, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 505, + 342 + ], + "score": 1.0, + "content": "Emilien Dupont, Arnaud Doucet, and Yee Whye Teh. Augmented neural odes. arXiv preprint", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 339, + 220, + 350 + ], + "spans": [ + { + "bbox": [ + 115, + 339, + 220, + 350 + ], + "score": 1.0, + "content": "arXiv:1904.01681, 2019.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 358, + 504, + 393 + ], + "lines": [ + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "score": 1.0, + "content": "Gamaleldin F Elsayed, Shreya Shankar, Brian Cheung, Nicolas Papernot, Alex Kurakin, Ian Good-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 115, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "fellow, and Jascha Sohl-Dickstein. Adversarial examples that fool both human and computer", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 381, + 324, + 393 + ], + "spans": [ + { + "bbox": [ + 116, + 381, + 324, + 393 + ], + "score": 1.0, + "content": "vision. arXiv preprint arXiv:1802.08195, 10, 2018.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 104, + 400, + 504, + 423 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "score": 1.0, + "content": "Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 412, + 317, + 424 + ], + "spans": [ + { + "bbox": [ + 115, + 412, + 317, + 424 + ], + "score": 1.0, + "content": "examples. arXiv preprint arXiv:1412.6572, 2014.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 430, + 505, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 431, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 505, + 443 + ], + "score": 1.0, + "content": "Will Grathwohl, Ricky TQ Chen, Jesse Betterncourt, Ilya Sutskever, and David Duvenaud. Ffjord:", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 115, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "Free-form continuous dynamics for scalable reversible generative models. arXiv preprint", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 453, + 220, + 465 + ], + "spans": [ + { + "bbox": [ + 115, + 453, + 220, + 465 + ], + "score": 1.0, + "content": "arXiv:1810.01367, 2018.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 472, + 505, + 506 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 488 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 482, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 115, + 482, + 505, + 498 + ], + "score": 1.0, + "content": "nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 493, + 182, + 507 + ], + "spans": [ + { + "bbox": [ + 115, + 493, + 182, + 507 + ], + "score": 1.0, + "content": "770–778, 2016.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 352, + 526 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 352, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 352, + 528 + ], + "score": 1.0, + "content": "Ralph Howard. The gronwall inequality. lecture notes, 1998.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 504, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 533, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 505, + 547 + ], + "score": 1.0, + "content": "Gustav Larsson, Michael Maire, and Gregory Shakhnarovich. Fractalnet: Ultra-deep neural net-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 545, + 380, + 558 + ], + "spans": [ + { + "bbox": [ + 116, + 545, + 380, + 558 + ], + "score": 1.0, + "content": "works without residuals. arXiv preprint arXiv:1605.07648, 2016.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 564, + 504, + 588 + ], + "lines": [ + { + "bbox": [ + 107, + 564, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 107, + 564, + 505, + 577 + ], + "score": 1.0, + "content": "Yann LeCun, Le´on Bottou, Yoshua Bengio, Patrick Haffner, et al. Gradient-based learning applied", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 576, + 427, + 588 + ], + "spans": [ + { + "bbox": [ + 115, + 576, + 427, + 588 + ], + "score": 1.0, + "content": "to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "score": 1.0, + "content": "Xuanqing Liu, Si Si, Qin Cao, Sanjiv Kumar, and Cho-Jui Hsieh. Neural sde: Stabilizing neural ode", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 606, + 406, + 619 + ], + "spans": [ + { + "bbox": [ + 115, + 606, + 406, + 619 + ], + "score": 1.0, + "content": "networks with stochastic noise. arXiv preprint arXiv:1906.02355, 2019.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 107, + 626, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "Yiping Lu, Aoxiao Zhong, Quanzheng Li, and Bin Dong. Beyond finite layer neural net-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 115, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "works: Bridging deep architectures and numerical differential equations. arXiv preprint", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 648, + 219, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 648, + 219, + 660 + ], + "score": 1.0, + "content": "arXiv:1710.10121, 2017.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 504, + 701 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "score": 1.0, + "content": "Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 679, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 116, + 679, + 505, + 691 + ], + "score": 1.0, + "content": "Towards deep learning models resistant to adversarial attacks. arXiv preprint arXiv:1706.06083,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 114, + 689, + 144, + 702 + ], + "spans": [ + { + "bbox": [ + 114, + 689, + 144, + 702 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, and Andrew Y Ng. Reading", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 720, + 382, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 382, + 733 + ], + "score": 1.0, + "content": "digits in natural images with unsupervised feature learning. 2011.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 100, + 504, + 134 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 505, + 113 + ], + "score": 1.0, + "content": "Lynton Ardizzone, Jakob Kruse, Sebastian Wirkert, Daniel Rahner, Eric W Pellegrini, Ralf S", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 111, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 505, + 124 + ], + "score": 1.0, + "content": "Klessen, Lena Maier-Hein, Carsten Rother, and Ullrich Ko¨the. Analyzing inverse problems with", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 122, + 388, + 135 + ], + "spans": [ + { + "bbox": [ + 116, + 122, + 388, + 135 + ], + "score": 1.0, + "content": "invertible neural networks. arXiv preprint arXiv:1808.04730, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 100, + 505, + 135 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 142, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "Nicholas Carlini and David Wagner. Towards evaluating the robustness of neural networks. In 2017", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 153, + 407, + 165 + ], + "spans": [ + { + "bbox": [ + 116, + 153, + 407, + 165 + ], + "score": 1.0, + "content": "IEEE Symposium on Security and Privacy (SP), pp. 39–57. IEEE, 2017.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 142, + 505, + 165 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 505, + 217 + ], + "lines": [ + { + "bbox": [ + 105, + 172, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 505, + 185 + ], + "score": 1.0, + "content": "Pin-Yu Chen, Huan Zhang, Yash Sharma, Jinfeng Yi, and Cho-Jui Hsieh. Zoo: Zeroth order opti-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 183, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 116, + 183, + 505, + 195 + ], + "score": 1.0, + "content": "mization based black-box attacks to deep neural networks without training substitute models. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 194, + 506, + 207 + ], + "spans": [ + { + "bbox": [ + 115, + 194, + 506, + 207 + ], + "score": 1.0, + "content": "Proceedings of the 10th ACM Workshop on Artificial Intelligence and Security, pp. 15–26. ACM,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 205, + 143, + 217 + ], + "spans": [ + { + "bbox": [ + 115, + 205, + 143, + 217 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 172, + 506, + 217 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 225, + 502, + 248 + ], + "lines": [ + { + "bbox": [ + 106, + 225, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 505, + 238 + ], + "score": 1.0, + "content": "Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differ-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 236, + 495, + 249 + ], + "spans": [ + { + "bbox": [ + 116, + 236, + 495, + 249 + ], + "score": 1.0, + "content": "ential equations. In Advances in neural information processing systems, pp. 6571–6583, 2018.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 106, + 225, + 505, + 249 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 255, + 504, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 255, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 505, + 269 + ], + "score": 1.0, + "content": "Earl A Coddington and Norman Levinson. Theory of ordinary differential equations. Tata McGraw-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 267, + 204, + 278 + ], + "spans": [ + { + "bbox": [ + 116, + 267, + 204, + 278 + ], + "score": 1.0, + "content": "Hill Education, 1955.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 255, + 505, + 278 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 285, + 505, + 320 + ], + "lines": [ + { + "bbox": [ + 106, + 286, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 504, + 299 + ], + "score": 1.0, + "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hi-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 297, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 115, + 297, + 505, + 311 + ], + "score": 1.0, + "content": "erarchical image database. In 2009 IEEE conference on computer vision and pattern recognition,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 309, + 218, + 320 + ], + "spans": [ + { + "bbox": [ + 115, + 309, + 218, + 320 + ], + "score": 1.0, + "content": "pp. 248–255. Ieee, 2009.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 106, + 286, + 505, + 320 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 327, + 505, + 351 + ], + "lines": [ + { + "bbox": [ + 105, + 326, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 505, + 342 + ], + "score": 1.0, + "content": "Emilien Dupont, Arnaud Doucet, and Yee Whye Teh. Augmented neural odes. arXiv preprint", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 339, + 220, + 350 + ], + "spans": [ + { + "bbox": [ + 115, + 339, + 220, + 350 + ], + "score": 1.0, + "content": "arXiv:1904.01681, 2019.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 326, + 505, + 350 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 358, + 504, + 393 + ], + "lines": [ + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "score": 1.0, + "content": "Gamaleldin F Elsayed, Shreya Shankar, Brian Cheung, Nicolas Papernot, Alex Kurakin, Ian Good-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 115, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "fellow, and Jascha Sohl-Dickstein. Adversarial examples that fool both human and computer", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 381, + 324, + 393 + ], + "spans": [ + { + "bbox": [ + 116, + 381, + 324, + 393 + ], + "score": 1.0, + "content": "vision. arXiv preprint arXiv:1802.08195, 10, 2018.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 106, + 359, + 506, + 393 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 400, + 504, + 423 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "score": 1.0, + "content": "Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 412, + 317, + 424 + ], + "spans": [ + { + "bbox": [ + 115, + 412, + 317, + 424 + ], + "score": 1.0, + "content": "examples. arXiv preprint arXiv:1412.6572, 2014.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 399, + 505, + 424 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 430, + 505, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 431, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 505, + 443 + ], + "score": 1.0, + "content": "Will Grathwohl, Ricky TQ Chen, Jesse Betterncourt, Ilya Sutskever, and David Duvenaud. Ffjord:", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 115, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "Free-form continuous dynamics for scalable reversible generative models. arXiv preprint", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 453, + 220, + 465 + ], + "spans": [ + { + "bbox": [ + 115, + 453, + 220, + 465 + ], + "score": 1.0, + "content": "arXiv:1810.01367, 2018.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25, + "bbox_fs": [ + 106, + 431, + 505, + 465 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 472, + 505, + 506 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 488 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 482, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 115, + 482, + 505, + 498 + ], + "score": 1.0, + "content": "nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 493, + 182, + 507 + ], + "spans": [ + { + "bbox": [ + 115, + 493, + 182, + 507 + ], + "score": 1.0, + "content": "770–778, 2016.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 471, + 505, + 507 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 352, + 526 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 352, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 352, + 528 + ], + "score": 1.0, + "content": "Ralph Howard. The gronwall inequality. lecture notes, 1998.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 513, + 352, + 528 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 504, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 533, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 505, + 547 + ], + "score": 1.0, + "content": "Gustav Larsson, Michael Maire, and Gregory Shakhnarovich. Fractalnet: Ultra-deep neural net-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 545, + 380, + 558 + ], + "spans": [ + { + "bbox": [ + 116, + 545, + 380, + 558 + ], + "score": 1.0, + "content": "works without residuals. arXiv preprint arXiv:1605.07648, 2016.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 533, + 505, + 558 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 564, + 504, + 588 + ], + "lines": [ + { + "bbox": [ + 107, + 564, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 107, + 564, + 505, + 577 + ], + "score": 1.0, + "content": "Yann LeCun, Le´on Bottou, Yoshua Bengio, Patrick Haffner, et al. Gradient-based learning applied", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 576, + 427, + 588 + ], + "spans": [ + { + "bbox": [ + 115, + 576, + 427, + 588 + ], + "score": 1.0, + "content": "to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 107, + 564, + 505, + 588 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "score": 1.0, + "content": "Xuanqing Liu, Si Si, Qin Cao, Sanjiv Kumar, and Cho-Jui Hsieh. Neural sde: Stabilizing neural ode", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 606, + 406, + 619 + ], + "spans": [ + { + "bbox": [ + 115, + 606, + 406, + 619 + ], + "score": 1.0, + "content": "networks with stochastic noise. arXiv preprint arXiv:1906.02355, 2019.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 594, + 505, + 619 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 626, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "Yiping Lu, Aoxiao Zhong, Quanzheng Li, and Bin Dong. Beyond finite layer neural net-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 115, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "works: Bridging deep architectures and numerical differential equations. arXiv preprint", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 648, + 219, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 648, + 219, + 660 + ], + "score": 1.0, + "content": "arXiv:1710.10121, 2017.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38, + "bbox_fs": [ + 106, + 626, + 506, + 660 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 504, + 701 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "score": 1.0, + "content": "Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 679, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 116, + 679, + 505, + 691 + ], + "score": 1.0, + "content": "Towards deep learning models resistant to adversarial attacks. arXiv preprint arXiv:1706.06083,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 114, + 689, + 144, + 702 + ], + "spans": [ + { + "bbox": [ + 114, + 689, + 144, + 702 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 667, + 505, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, and Andrew Y Ng. Reading", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 720, + 382, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 382, + 733 + ], + "score": 1.0, + "content": "digits in natural images with unsupervised feature learning. 2011.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 106, + 708, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 105, + 178, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 105, + 178, + 116 + ], + "score": 1.0, + "content": "pytorch. 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 104, + 123, + 470, + 136 + ], + "lines": [ + { + "bbox": [ + 105, + 123, + 470, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 470, + 137 + ], + "score": 1.0, + "content": "Lev Semenovich Pontryagin. Mathematical theory of optimal processes. Routledge, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 142, + 503, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "Alessio Quaglino, Marco Gallieri, Jonathan Masci, and Jan Koutn´ık. Accelerating neural odes with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 154, + 354, + 165 + ], + "spans": [ + { + "bbox": [ + 115, + 154, + 354, + 165 + ], + "score": 1.0, + "content": "spectral elements. arXiv preprint arXiv:1906.07038, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 171, + 504, + 195 + ], + "lines": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "score": 1.0, + "content": "Jure Sokolic´, Raja Giryes, Guillermo Sapiro, and Miguel RD Rodrigues. Robust large margin deep", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 458, + 195 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 458, + 195 + ], + "score": 1.0, + "content": "neural networks. IEEE Transactions on Signal Processing, 65(16):4265–4280, 2017.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 201, + 504, + 225 + ], + "lines": [ + { + "bbox": [ + 106, + 201, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 505, + 214 + ], + "score": 1.0, + "content": "Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 213, + 506, + 226 + ], + "spans": [ + { + "bbox": [ + 115, + 213, + 506, + 226 + ], + "score": 1.0, + "content": "and Rob Fergus. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 108, + 231, + 504, + 265 + ], + "lines": [ + { + "bbox": [ + 105, + 231, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 245 + ], + "score": 1.0, + "content": "Florian Trame`r, Alexey Kurakin, Nicolas Papernot, Ian Goodfellow, Dan Boneh, and Patrick Mc-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 243, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 115, + 243, + 506, + 255 + ], + "score": 1.0, + "content": "Daniel. Ensemble adversarial training: Attacks and defenses. arXiv preprint arXiv:1705.07204,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 253, + 145, + 266 + ], + "spans": [ + { + "bbox": [ + 115, + 253, + 145, + 266 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 105, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 106, + 272, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 505, + 286 + ], + "score": 1.0, + "content": "Dimitris Tsipras, Shibani Santurkar, Logan Engstrom, Alexander Turner, and Aleksander Madry.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 283, + 450, + 296 + ], + "spans": [ + { + "bbox": [ + 115, + 283, + 450, + 296 + ], + "score": 1.0, + "content": "Robustness may be at odds with accuracy. arXiv preprint arXiv:1805.12152, 2018.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 302, + 503, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "score": 1.0, + "content": "B. Wang, B. Yuan, Z. Shi, and S. Osher. ResNets Ensemble via the Feynman-Kac Formalism to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 313, + 484, + 326 + ], + "spans": [ + { + "bbox": [ + 116, + 313, + 484, + 326 + ], + "score": 1.0, + "content": "Improve Natural and Robust Accuracies. arXiv e-prints, art. arXiv:1811.10745, Nov 2018a.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 504, + 366 + ], + "lines": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "score": 1.0, + "content": "Bao Wang, Alex T Lin, Zuoqiang Shi, Wei Zhu, Penghang Yin, Andrea L Bertozzi, and Stanley J Os-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 116, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "her. Adversarial defense via data dependent activation function and total variation minimization.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 354, + 284, + 366 + ], + "spans": [ + { + "bbox": [ + 116, + 354, + 284, + 366 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1809.08516, 2018b.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 372, + 504, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 372, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 505, + 386 + ], + "score": 1.0, + "content": "Bao Wang, Xiyang Luo, Zhen Li, Wei Zhu, Zuoqiang Shi, and Stanley Osher. Deep neural nets", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 383, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 115, + 383, + 505, + 398 + ], + "score": 1.0, + "content": "with interpolating function as output activation. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 396, + 237, + 408 + ], + "spans": [ + { + "bbox": [ + 116, + 396, + 237, + 408 + ], + "score": 1.0, + "content": "Systems, pp. 743–753, 2018c.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 108, + 414, + 504, + 437 + ], + "lines": [ + { + "bbox": [ + 105, + 413, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 505, + 427 + ], + "score": 1.0, + "content": "E Weinan. A proposal on machine learning via dynamical systems. Communications in Mathematics", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 425, + 244, + 436 + ], + "spans": [ + { + "bbox": [ + 116, + 425, + 244, + 436 + ], + "score": 1.0, + "content": "and Statistics, 5(1):1–11, 2017.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 467 + ], + "lines": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "Cihang Xie, Jianyu Wang, Zhishuai Zhang, Zhou Ren, and Alan Yuille. Mitigating adversarial", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 455, + 405, + 467 + ], + "spans": [ + { + "bbox": [ + 116, + 455, + 405, + 467 + ], + "score": 1.0, + "content": "effects through randomization. arXiv preprint arXiv:1711.01991, 2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 504, + 508 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 506, + 488 + ], + "score": 1.0, + "content": "Cihang Xie, Yuxin Wu, Laurens van der Maaten, Alan L Yuille, and Kaiming He. Feature denoising", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 485, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 116, + 485, + 505, + 498 + ], + "score": 1.0, + "content": "for improving adversarial robustness. In Proceedings of the IEEE Conference on Computer Vision", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 496, + 298, + 508 + ], + "spans": [ + { + "bbox": [ + 115, + 496, + 298, + 508 + ], + "score": 1.0, + "content": "and Pattern Recognition, pp. 501–509, 2019.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 504, + 538 + ], + "lines": [ + { + "bbox": [ + 106, + 514, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 505, + 527 + ], + "score": 1.0, + "content": "Ziang Yan, Yiwen Guo, and Changshui Zhang. Deep defense: Training dnns with improved adver-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 525, + 493, + 538 + ], + "spans": [ + { + "bbox": [ + 115, + 525, + 493, + 538 + ], + "score": 1.0, + "content": "sarial robustness. In Advances in Neural Information Processing Systems, pp. 419–428, 2018.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 544, + 412, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 412, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 412, + 558 + ], + "score": 1.0, + "content": "Laurent Younes. Shapes and diffeomorphisms, volume 171. Springer, 2010.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 564, + 506, + 597 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 506, + 578 + ], + "score": 1.0, + "content": "Xingcheng Zhang, Zhizhong Li, Chen Change Loy, and Dahua Lin. Polynet: A pursuit of structural", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 574, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 115, + 574, + 506, + 588 + ], + "score": 1.0, + "content": "diversity in very deep networks. In Proceedings of the IEEE Conference on Computer Vision and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 585, + 281, + 599 + ], + "spans": [ + { + "bbox": [ + 116, + 585, + 281, + 599 + ], + "score": 1.0, + "content": "Pattern Recognition, pp. 718–726, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 105, + 178, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 105, + 178, + 116 + ], + "score": 1.0, + "content": "pytorch. 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 505, + 116 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 123, + 470, + 136 + ], + "lines": [ + { + "bbox": [ + 105, + 123, + 470, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 470, + 137 + ], + "score": 1.0, + "content": "Lev Semenovich Pontryagin. Mathematical theory of optimal processes. Routledge, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 123, + 470, + 137 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 142, + 503, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "Alessio Quaglino, Marco Gallieri, Jonathan Masci, and Jan Koutn´ık. Accelerating neural odes with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 154, + 354, + 165 + ], + "spans": [ + { + "bbox": [ + 115, + 154, + 354, + 165 + ], + "score": 1.0, + "content": "spectral elements. arXiv preprint arXiv:1906.07038, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 142, + 505, + 165 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 171, + 504, + 195 + ], + "lines": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "score": 1.0, + "content": "Jure Sokolic´, Raja Giryes, Guillermo Sapiro, and Miguel RD Rodrigues. Robust large margin deep", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 458, + 195 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 458, + 195 + ], + "score": 1.0, + "content": "neural networks. IEEE Transactions on Signal Processing, 65(16):4265–4280, 2017.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 171, + 505, + 195 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 201, + 504, + 225 + ], + "lines": [ + { + "bbox": [ + 106, + 201, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 505, + 214 + ], + "score": 1.0, + "content": "Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 213, + 506, + 226 + ], + "spans": [ + { + "bbox": [ + 115, + 213, + 506, + 226 + ], + "score": 1.0, + "content": "and Rob Fergus. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 106, + 201, + 506, + 226 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 231, + 504, + 265 + ], + "lines": [ + { + "bbox": [ + 105, + 231, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 245 + ], + "score": 1.0, + "content": "Florian Trame`r, Alexey Kurakin, Nicolas Papernot, Ian Goodfellow, Dan Boneh, and Patrick Mc-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 243, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 115, + 243, + 506, + 255 + ], + "score": 1.0, + "content": "Daniel. Ensemble adversarial training: Attacks and defenses. arXiv preprint arXiv:1705.07204,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 253, + 145, + 266 + ], + "spans": [ + { + "bbox": [ + 115, + 253, + 145, + 266 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 231, + 506, + 266 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 106, + 272, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 505, + 286 + ], + "score": 1.0, + "content": "Dimitris Tsipras, Shibani Santurkar, Logan Engstrom, Alexander Turner, and Aleksander Madry.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 283, + 450, + 296 + ], + "spans": [ + { + "bbox": [ + 115, + 283, + 450, + 296 + ], + "score": 1.0, + "content": "Robustness may be at odds with accuracy. arXiv preprint arXiv:1805.12152, 2018.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 272, + 505, + 296 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 302, + 503, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "score": 1.0, + "content": "B. Wang, B. Yuan, Z. Shi, and S. Osher. ResNets Ensemble via the Feynman-Kac Formalism to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 313, + 484, + 326 + ], + "spans": [ + { + "bbox": [ + 116, + 313, + 484, + 326 + ], + "score": 1.0, + "content": "Improve Natural and Robust Accuracies. arXiv e-prints, art. arXiv:1811.10745, Nov 2018a.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 302, + 505, + 326 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 504, + 366 + ], + "lines": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "score": 1.0, + "content": "Bao Wang, Alex T Lin, Zuoqiang Shi, Wei Zhu, Penghang Yin, Andrea L Bertozzi, and Stanley J Os-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 116, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "her. Adversarial defense via data dependent activation function and total variation minimization.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 354, + 284, + 366 + ], + "spans": [ + { + "bbox": [ + 116, + 354, + 284, + 366 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1809.08516, 2018b.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 332, + 505, + 366 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 372, + 504, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 372, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 505, + 386 + ], + "score": 1.0, + "content": "Bao Wang, Xiyang Luo, Zhen Li, Wei Zhu, Zuoqiang Shi, and Stanley Osher. Deep neural nets", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 383, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 115, + 383, + 505, + 398 + ], + "score": 1.0, + "content": "with interpolating function as output activation. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 396, + 237, + 408 + ], + "spans": [ + { + "bbox": [ + 116, + 396, + 237, + 408 + ], + "score": 1.0, + "content": "Systems, pp. 743–753, 2018c.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 372, + 505, + 408 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 414, + 504, + 437 + ], + "lines": [ + { + "bbox": [ + 105, + 413, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 505, + 427 + ], + "score": 1.0, + "content": "E Weinan. A proposal on machine learning via dynamical systems. Communications in Mathematics", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 425, + 244, + 436 + ], + "spans": [ + { + "bbox": [ + 116, + 425, + 244, + 436 + ], + "score": 1.0, + "content": "and Statistics, 5(1):1–11, 2017.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 413, + 505, + 436 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 467 + ], + "lines": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "Cihang Xie, Jianyu Wang, Zhishuai Zhang, Zhou Ren, and Alan Yuille. Mitigating adversarial", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 455, + 405, + 467 + ], + "spans": [ + { + "bbox": [ + 116, + 455, + 405, + 467 + ], + "score": 1.0, + "content": "effects through randomization. arXiv preprint arXiv:1711.01991, 2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 444, + 505, + 467 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 504, + 508 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 506, + 488 + ], + "score": 1.0, + "content": "Cihang Xie, Yuxin Wu, Laurens van der Maaten, Alan L Yuille, and Kaiming He. Feature denoising", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 485, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 116, + 485, + 505, + 498 + ], + "score": 1.0, + "content": "for improving adversarial robustness. In Proceedings of the IEEE Conference on Computer Vision", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 496, + 298, + 508 + ], + "spans": [ + { + "bbox": [ + 115, + 496, + 298, + 508 + ], + "score": 1.0, + "content": "and Pattern Recognition, pp. 501–509, 2019.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 471, + 506, + 508 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 504, + 538 + ], + "lines": [ + { + "bbox": [ + 106, + 514, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 505, + 527 + ], + "score": 1.0, + "content": "Ziang Yan, Yiwen Guo, and Changshui Zhang. Deep defense: Training dnns with improved adver-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 525, + 493, + 538 + ], + "spans": [ + { + "bbox": [ + 115, + 525, + 493, + 538 + ], + "score": 1.0, + "content": "sarial robustness. In Advances in Neural Information Processing Systems, pp. 419–428, 2018.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 106, + 514, + 505, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 544, + 412, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 412, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 412, + 558 + ], + "score": 1.0, + "content": "Laurent Younes. Shapes and diffeomorphisms, volume 171. Springer, 2010.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 543, + 412, + 558 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 564, + 506, + 597 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 506, + 578 + ], + "score": 1.0, + "content": "Xingcheng Zhang, Zhizhong Li, Chen Change Loy, and Dahua Lin. Polynet: A pursuit of structural", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 574, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 115, + 574, + 506, + 588 + ], + "score": 1.0, + "content": "diversity in very deep networks. In Proceedings of the IEEE Conference on Computer Vision and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 585, + 281, + 599 + ], + "spans": [ + { + "bbox": [ + 116, + 585, + 281, + 599 + ], + "score": 1.0, + "content": "Pattern Recognition, pp. 718–726, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 563, + 506, + 599 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 180, + 93 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 182, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 182, + 96 + ], + "score": 1.0, + "content": "7 APPENDIX", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 103, + 108, + 469, + 120 + ], + "lines": [ + { + "bbox": [ + 105, + 108, + 470, + 121 + ], + "spans": [ + { + "bbox": [ + 105, + 108, + 470, + 121 + ], + "score": 1.0, + "content": "7.1 NETWORKS USED ON THE MNIST, THE SVHN, AND THE IMGNET10 DATASETS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "table", + "bbox": [ + 176, + 159, + 434, + 224 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 176, + 137, + 434, + 150 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 136, + 435, + 151 + ], + "spans": [ + { + "bbox": [ + 176, + 136, + 435, + 151 + ], + "score": 1.0, + "content": "Table 5: The architectures of the ODENets on different datasets.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 176, + 159, + 434, + 224 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 159, + 434, + 224 + ], + "spans": [ + { + "bbox": [ + 176, + 159, + 434, + 224 + ], + "score": 0.112, + "html": "
MNISTRepetitionLayer
FE×1×1Conv(1,64,3,1) +GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "type": "table", + "image_path": "d883b2f0037322e3b140127861d9b4cbcecf0aad825dcfc8ae8472d4037da268.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 176, + 159, + 434, + 180.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 176, + 180.66666666666666, + 434, + 202.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 176, + 202.33333333333331, + 434, + 223.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 3.0 + }, + { + "type": "table", + "bbox": [ + 175, + 234, + 433, + 298 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 175, + 234, + 433, + 298 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 175, + 234, + 433, + 298 + ], + "spans": [ + { + "bbox": [ + 175, + 234, + 433, + 298 + ], + "score": 0.239, + "html": "
SVHNRepetitionLayer
FE×1×1Conv(3,64,3,1) + GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "type": "table", + "image_path": "faf3961048327cb024192b9d8e2992e6f976673ff550b4ad755809e130397b17.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 175, + 234, + 433, + 255.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 175, + 255.33333333333334, + 433, + 276.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 175, + 276.6666666666667, + 433, + 298.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 7 + }, + { + "type": "table", + "bbox": [ + 175, + 308, + 434, + 393 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 175, + 308, + 434, + 393 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 175, + 308, + 434, + 393 + ], + "spans": [ + { + "bbox": [ + 175, + 308, + 434, + 393 + ], + "score": 0.257, + "html": "
ImgNet10RepetitionLayer
FE×1Conv(3,32,5,2)+GroupNorm
×1MaxPooling(2)
×1BaiscBlock(32, 64,2)
×1MaxPooling(2)
RM×3BaiscBlock(64,64,1)
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "type": "table", + "image_path": "716551303bb6992cbdca9e029eee5f1d7cf0b6fc48418513eee4e672c4c78e95.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 175, + 308, + 434, + 336.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 175, + 336.3333333333333, + 434, + 364.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 175, + 364.66666666666663, + 434, + 392.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 408, + 505, + 486 + ], + "lines": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "score": 1.0, + "content": "In Table 5, the four arguments of the Conv layer represent the input channel, output channel, kernel", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 419, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 505, + 432 + ], + "score": 1.0, + "content": "size, and the stride. The two arguments of the Linear layer represents the input dimension and the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "score": 1.0, + "content": "output dimension of this fully-connected layer. In the network on the ImgNet10, the BasicBlock", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 441, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 505, + 453 + ], + "score": 1.0, + "content": "refers to the standard architecture in (He et al., 2016), the three arguments of the BasicBlock repre-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 452, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 505, + 464 + ], + "score": 1.0, + "content": "sent the input channel, output channel and the stride of the Conv layers inside the block. Note that", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 505, + 475 + ], + "score": 1.0, + "content": "we replace the BatchNorm layers in BasicBlocks as the GroupNorm to guarantee that the dynamics", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 474, + 374, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 374, + 485 + ], + "score": 1.0, + "content": "of each datum is independent of other data in the same mini-batch.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 108, + 502, + 325, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 502, + 325, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 325, + 515 + ], + "score": 1.0, + "content": "7.2 THE CONSTRUCTION OF IMGNET10 DATASET", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "table", + "bbox": [ + 204, + 551, + 404, + 667 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 141, + 531, + 469, + 543 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 140, + 528, + 471, + 546 + ], + "spans": [ + { + "bbox": [ + 140, + 528, + 471, + 546 + ], + "score": 1.0, + "content": "Table 6: The corresponding indexes to each class in the original ImageNet dataset", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "table_body", + "bbox": [ + 204, + 551, + 404, + 667 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 204, + 551, + 404, + 667 + ], + "spans": [ + { + "bbox": [ + 204, + 551, + 404, + 667 + ], + "score": 0.98, + "html": "
ClassIndexing
dogn02090721, n02091032, n02088094
birdn01532829, n01558993, n01534433
carn02814533, n03930630, n03100240
fishn01484850,n01491361, n01494475
monkeyn02483708,n02484975, n02486261
turtlen01664065,n01665541, n01667114
lizardn01677366,n01682714,n01685808
bridgen03933933,n04366367,n04311004
cown02403003,n02408429,n02410509
crabn01980166,n01978455,n01981276
", + "type": "table", + "image_path": "1a44f870fa6c81d744953453aa8fe7c73defdee5af67ea0f098fac0f2d07d1c9.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 204, + 551, + 404, + 565.5 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 204, + 565.5, + 404, + 580.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 204, + 580.0, + 404, + 594.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 204, + 594.5, + 404, + 609.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 204, + 609.0, + 404, + 623.5 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 204, + 623.5, + 404, + 638.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 204, + 638.0, + 404, + 652.5 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 204, + 652.5, + 404, + 667.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "index": 22.25 + }, + { + "type": "title", + "bbox": [ + 107, + 696, + 246, + 708 + ], + "lines": [ + { + "bbox": [ + 105, + 695, + 247, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 695, + 247, + 709 + ], + "score": 1.0, + "content": "7.3 GRONWALL’S INEQUALITY", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 103, + 718, + 471, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 718, + 473, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 718, + 473, + 732 + ], + "score": 1.0, + "content": "We formally state the Gronwall’s Inequality here, following the version in (Howard, 1998).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 294, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 180, + 93 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 182, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 182, + 96 + ], + "score": 1.0, + "content": "7 APPENDIX", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 103, + 108, + 469, + 120 + ], + "lines": [ + { + "bbox": [ + 105, + 108, + 470, + 121 + ], + "spans": [ + { + "bbox": [ + 105, + 108, + 470, + 121 + ], + "score": 1.0, + "content": "7.1 NETWORKS USED ON THE MNIST, THE SVHN, AND THE IMGNET10 DATASETS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 108, + 470, + 121 + ] + }, + { + "type": "table", + "bbox": [ + 176, + 159, + 434, + 224 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 176, + 137, + 434, + 150 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 136, + 435, + 151 + ], + "spans": [ + { + "bbox": [ + 176, + 136, + 435, + 151 + ], + "score": 1.0, + "content": "Table 5: The architectures of the ODENets on different datasets.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 176, + 159, + 434, + 224 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 159, + 434, + 224 + ], + "spans": [ + { + "bbox": [ + 176, + 159, + 434, + 224 + ], + "score": 0.112, + "html": "
MNISTRepetitionLayer
FE×1×1Conv(1,64,3,1) +GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "type": "table", + "image_path": "d883b2f0037322e3b140127861d9b4cbcecf0aad825dcfc8ae8472d4037da268.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 176, + 159, + 434, + 180.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 176, + 180.66666666666666, + 434, + 202.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 176, + 202.33333333333331, + 434, + 223.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 3.0 + }, + { + "type": "table", + "bbox": [ + 175, + 234, + 433, + 298 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 175, + 234, + 433, + 298 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 175, + 234, + 433, + 298 + ], + "spans": [ + { + "bbox": [ + 175, + 234, + 433, + 298 + ], + "score": 0.239, + "html": "
SVHNRepetitionLayer
FE×1×1Conv(3,64,3,1) + GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "type": "table", + "image_path": "faf3961048327cb024192b9d8e2992e6f976673ff550b4ad755809e130397b17.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 175, + 234, + 433, + 255.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 175, + 255.33333333333334, + 433, + 276.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 175, + 276.6666666666667, + 433, + 298.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 7 + }, + { + "type": "table", + "bbox": [ + 175, + 308, + 434, + 393 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 175, + 308, + 434, + 393 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 175, + 308, + 434, + 393 + ], + "spans": [ + { + "bbox": [ + 175, + 308, + 434, + 393 + ], + "score": 0.257, + "html": "
ImgNet10RepetitionLayer
FE×1Conv(3,32,5,2)+GroupNorm
×1MaxPooling(2)
×1BaiscBlock(32, 64,2)
×1MaxPooling(2)
RM×3BaiscBlock(64,64,1)
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "type": "table", + "image_path": "716551303bb6992cbdca9e029eee5f1d7cf0b6fc48418513eee4e672c4c78e95.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 175, + 308, + 434, + 336.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 175, + 336.3333333333333, + 434, + 364.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 175, + 364.66666666666663, + 434, + 392.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 408, + 505, + 486 + ], + "lines": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "score": 1.0, + "content": "In Table 5, the four arguments of the Conv layer represent the input channel, output channel, kernel", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 419, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 505, + 432 + ], + "score": 1.0, + "content": "size, and the stride. The two arguments of the Linear layer represents the input dimension and the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "score": 1.0, + "content": "output dimension of this fully-connected layer. In the network on the ImgNet10, the BasicBlock", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 441, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 505, + 453 + ], + "score": 1.0, + "content": "refers to the standard architecture in (He et al., 2016), the three arguments of the BasicBlock repre-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 452, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 505, + 464 + ], + "score": 1.0, + "content": "sent the input channel, output channel and the stride of the Conv layers inside the block. Note that", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 505, + 475 + ], + "score": 1.0, + "content": "we replace the BatchNorm layers in BasicBlocks as the GroupNorm to guarantee that the dynamics", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 474, + 374, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 374, + 485 + ], + "score": 1.0, + "content": "of each datum is independent of other data in the same mini-batch.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 407, + 505, + 485 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 502, + 325, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 502, + 325, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 325, + 515 + ], + "score": 1.0, + "content": "7.2 THE CONSTRUCTION OF IMGNET10 DATASET", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "table", + "bbox": [ + 204, + 551, + 404, + 667 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 141, + 531, + 469, + 543 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 140, + 528, + 471, + 546 + ], + "spans": [ + { + "bbox": [ + 140, + 528, + 471, + 546 + ], + "score": 1.0, + "content": "Table 6: The corresponding indexes to each class in the original ImageNet dataset", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "table_body", + "bbox": [ + 204, + 551, + 404, + 667 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 204, + 551, + 404, + 667 + ], + "spans": [ + { + "bbox": [ + 204, + 551, + 404, + 667 + ], + "score": 0.98, + "html": "
ClassIndexing
dogn02090721, n02091032, n02088094
birdn01532829, n01558993, n01534433
carn02814533, n03930630, n03100240
fishn01484850,n01491361, n01494475
monkeyn02483708,n02484975, n02486261
turtlen01664065,n01665541, n01667114
lizardn01677366,n01682714,n01685808
bridgen03933933,n04366367,n04311004
cown02403003,n02408429,n02410509
crabn01980166,n01978455,n01981276
", + "type": "table", + "image_path": "1a44f870fa6c81d744953453aa8fe7c73defdee5af67ea0f098fac0f2d07d1c9.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 204, + 551, + 404, + 565.5 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 204, + 565.5, + 404, + 580.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 204, + 580.0, + 404, + 594.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 204, + 594.5, + 404, + 609.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 204, + 609.0, + 404, + 623.5 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 204, + 623.5, + 404, + 638.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 204, + 638.0, + 404, + 652.5 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 204, + 652.5, + 404, + 667.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "index": 22.25 + }, + { + "type": "title", + "bbox": [ + 107, + 696, + 246, + 708 + ], + "lines": [ + { + "bbox": [ + 105, + 695, + 247, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 695, + 247, + 709 + ], + "score": 1.0, + "content": "7.3 GRONWALL’S INEQUALITY", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 103, + 718, + 471, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 718, + 473, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 718, + 473, + 732 + ], + "score": 1.0, + "content": "We formally state the Gronwall’s Inequality here, following the version in (Howard, 1998).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 718, + 473, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 81, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 175, + 95 + ], + "score": 1.0, + "content": "Theorem 2. Let", + "type": "text" + }, + { + "bbox": [ + 175, + 82, + 209, + 93 + ], + "score": 0.91, + "content": "U \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 81, + 289, + 95 + ], + "score": 1.0, + "content": "be an open set. Let", + "type": "text" + }, + { + "bbox": [ + 289, + 82, + 374, + 95 + ], + "score": 0.94, + "content": "f : U \\times [ 0 , T ] \\to \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "be a continuous function and let", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 93, + 321, + 107 + ], + "spans": [ + { + "bbox": [ + 107, + 96, + 117, + 105 + ], + "score": 0.56, + "content": "\\mathbf { z } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 93, + 120, + 107 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 121, + 95, + 132, + 105 + ], + "score": 0.58, + "content": "\\mathbf { z } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 93, + 137, + 107 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 137, + 93, + 184, + 106 + ], + "score": 0.82, + "content": "[ 0 , T ] \\to U", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 93, + 321, + 107 + ], + "score": 1.0, + "content": "satisfy the initial value problems:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 113, + 379, + 164 + ], + "lines": [ + { + "bbox": [ + 232, + 113, + 379, + 164 + ], + "spans": [ + { + "bbox": [ + 232, + 113, + 379, + 164 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\frac { \\mathrm { d } { \\mathbf z } _ { 1 } ( t ) } { \\mathrm { d } t } = f ( { \\mathbf z } _ { 1 } ( t ) , t ) , \\quad { \\mathbf z } _ { 1 } ( t ) = { \\mathbf x } _ { 1 } } \\\\ & { \\frac { \\mathrm { d } { \\mathbf z } _ { 2 } ( t ) } { \\mathrm { d } t } = f ( { \\mathbf z } _ { 2 } ( t ) , t ) , \\quad { \\mathbf z } _ { 2 } ( t ) = { \\mathbf x } _ { 2 } } \\end{array}", + "type": "interline_equation", + "image_path": "7f6eeb32253c8cad90d543707eb2749cd6d45ec22143c726afbc82c5047bbd0d.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 232, + 113, + 379, + 138.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 232, + 138.5, + 379, + 164.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 169, + 354, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 169, + 354, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 215, + 183 + ], + "score": 1.0, + "content": "Assume there is a constant", + "type": "text" + }, + { + "bbox": [ + 215, + 171, + 242, + 181 + ], + "score": 0.9, + "content": "C \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 169, + 311, + 183 + ], + "score": 1.0, + "content": "such that, for all", + "type": "text" + }, + { + "bbox": [ + 312, + 170, + 351, + 182 + ], + "score": 0.92, + "content": "t \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 169, + 354, + 183 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 189, + 404, + 204 + ], + "lines": [ + { + "bbox": [ + 206, + 189, + 404, + 204 + ], + "spans": [ + { + "bbox": [ + 206, + 189, + 404, + 204 + ], + "score": 0.86, + "content": "\\| f ( \\mathbf { z } _ { 2 } ( t ) , t ) - f ( \\mathbf { z } _ { 1 } ( t ) , t ) ) \\| \\leq C \\| \\mathbf { z } _ { 2 } ( t ) - \\mathbf { z } _ { 1 } ( t ) \\|", + "type": "interline_equation", + "image_path": "d44d70833a33bcbc6c66109b2ee246c42ecdacd3538edc2233bfcfea280df71a.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 206, + 189, + 404, + 204 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 210, + 204, + 223 + ], + "lines": [ + { + "bbox": [ + 106, + 209, + 205, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 209, + 162, + 225 + ], + "score": 1.0, + "content": "Then, for any", + "type": "text" + }, + { + "bbox": [ + 163, + 211, + 201, + 223 + ], + "score": 0.92, + "content": "t \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 209, + 205, + 225 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 225, + 379, + 240 + ], + "lines": [ + { + "bbox": [ + 232, + 225, + 379, + 240 + ], + "spans": [ + { + "bbox": [ + 232, + 225, + 379, + 240 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\| \\mathbf { z } _ { 1 } ( t ) - \\mathbf { z } _ { 2 } ( t ) \\| \\leq \\| \\mathbf { x } _ { 2 } - \\mathbf { x } _ { 1 } \\| \\cdot e ^ { C t } . } \\end{array}", + "type": "interline_equation", + "image_path": "9e1ae5bce98edee3c029a1c7316e4e9f396d1d4a5d6b60e76e8363662abe1da1.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 232, + 225, + 379, + 240 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 255, + 267, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 268, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 268, + 267 + ], + "score": 1.0, + "content": "7.4 MORE EXPERIMENTAL RESULTS", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 107, + 276, + 393, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 275, + 393, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 393, + 289 + ], + "score": 1.0, + "content": "7.4.1 COMPARISON IN THE SETTING OF ADVERSARIAL TRAINING", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 296, + 506, + 352 + ], + "lines": [ + { + "bbox": [ + 106, + 296, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 296, + 505, + 308 + ], + "score": 1.0, + "content": "We implement the adversarial training of the models on the MNIST dataset, and the adversarial", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 430, + 320 + ], + "score": 1.0, + "content": "examples for training are generated in real-time via the FGSM method (epsilon", + "type": "text" + }, + { + "bbox": [ + 430, + 308, + 450, + 318 + ], + "score": 0.69, + "content": "_ { 1 = 0 . 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 307, + 505, + 320 + ], + "score": 1.0, + "content": ") during each", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "epoch (Madry et al., 2017). The results of the adversarially trained models are shown in Table 7. We", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 329, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 342 + ], + "score": 1.0, + "content": "can observe that the neural ODE-based models are consistently more robust than CNN models. The", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 340, + 347, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 347, + 353 + ], + "score": 1.0, + "content": "proposed TisODE also outperforms the vanilla neural ODE.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12 + }, + { + "type": "table", + "bbox": [ + 178, + 408, + 430, + 469 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 364, + 504, + 398 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 365, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 237, + 377 + ], + "score": 1.0, + "content": "Table 7: Classification accuracy", + "type": "text" + }, + { + "bbox": [ + 238, + 365, + 253, + 376 + ], + "score": 0.79, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 365, + 506, + 377 + ], + "score": 1.0, + "content": "on perturbed images from MNIST. To evaluate the robustness", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 375, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 506, + 389 + ], + "score": 1.0, + "content": "of classifiers, we use three types of perturbations, namely zero-mean Gaussian noise with standard", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 387, + 280, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 145, + 398 + ], + "score": 1.0, + "content": "deviation", + "type": "text" + }, + { + "bbox": [ + 146, + 388, + 153, + 397 + ], + "score": 0.73, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 387, + 280, + 398 + ], + "score": 1.0, + "content": ", FGSM attack and PGD attack.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "table_body", + "bbox": [ + 178, + 408, + 430, + 469 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 178, + 408, + 430, + 469 + ], + "spans": [ + { + "bbox": [ + 178, + 408, + 430, + 469 + ], + "score": 0.981, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.3
CNN58.098.421.15.3
ODENet84.299.136.012.3
TisODE87.999.166.578.9
", + "type": "table", + "image_path": "55430c76602428adbc3f3d6582221d15213646ccc38819102c5ce51011e89650.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 178, + 408, + 430, + 428.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 178, + 428.3333333333333, + 430, + 448.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 178, + 448.66666666666663, + 430, + 468.99999999999994 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 107, + 492, + 321, + 504 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 322, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 322, + 505 + ], + "score": 1.0, + "content": "7.4.2 EXPERIMENTS ON THE CIFAR10 DATASET", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 512, + 506, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 506, + 524 + ], + "score": 1.0, + "content": "We conduct experiments on CIFAR10 to compare the robustness of CNN and neural ODE-based", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 523, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 536 + ], + "score": 1.0, + "content": "models. We train all the models only with original non-perturbed images and evaluate the robustness", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 534, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 506, + 547 + ], + "score": 1.0, + "content": "of models against random Gaussian noise and FGSM adversarial attacks. The results are shown in", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 546, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 505, + 557 + ], + "score": 1.0, + "content": "Table 8. We can observe that the ONENet is more robust than the CNN model in terms of both the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "score": 1.0, + "content": "random noise and the FGSM attack. Besides, our proposal, TisODE, can improve the robustness of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 568, + 203, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 203, + 578 + ], + "score": 1.0, + "content": "the vanilla neural ODE.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24.5 + }, + { + "type": "table", + "bbox": [ + 182, + 632, + 426, + 694 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 589, + 505, + 623 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 235, + 601 + ], + "score": 1.0, + "content": "Table 8: Classification accuracy", + "type": "text" + }, + { + "bbox": [ + 235, + 590, + 250, + 600 + ], + "score": 0.64, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "on perturbed images from CIFAR10. To evaluate the robustness", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "of classifiers, we use two types of perturbations, namely zero-mean Gaussian noise with standard", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 612, + 229, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 145, + 622 + ], + "score": 1.0, + "content": "deviation", + "type": "text" + }, + { + "bbox": [ + 146, + 613, + 153, + 621 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 612, + 229, + 622 + ], + "score": 1.0, + "content": "and FGSM attack.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "table_body", + "bbox": [ + 182, + 632, + 426, + 694 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 182, + 632, + 426, + 694 + ], + "spans": [ + { + "bbox": [ + 182, + 632, + 426, + 694 + ], + "score": 0.976, + "html": "
Gaussian noiseAdversarial attack
CIFAR10σ=15σ=20FGSM-8/255FGSM-10/255
CNN70.257.624.318.4
ODENet72.660.631.226.0
TisODE74.362.033.626.8
", + "type": "table", + "image_path": "2bad224570b817a9b629ba9432e5070888e5d4466432b80c75d325e9d0218670.jpg" + } + ] + } + ], + "index": 32.5, + "virtual_lines": [ + { + "bbox": [ + 182, + 632, + 426, + 647.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 182, + 647.5, + 426, + 663.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 182, + 663.0, + 426, + 678.5 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 182, + 678.5, + 426, + 694.0 + ], + "spans": [], + "index": 34 + } + ] + } + ], + "index": 30.75 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 710, + 504, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 504, + 721 + ], + "score": 1.0, + "content": "Here, we control the number of parameters to be the same for all kinds of models. We use a small", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 720, + 399, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 399, + 733 + ], + "score": 1.0, + "content": "network, which consists of five convolutional layers and one linear layer.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 81, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 175, + 95 + ], + "score": 1.0, + "content": "Theorem 2. Let", + "type": "text" + }, + { + "bbox": [ + 175, + 82, + 209, + 93 + ], + "score": 0.91, + "content": "U \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 81, + 289, + 95 + ], + "score": 1.0, + "content": "be an open set. Let", + "type": "text" + }, + { + "bbox": [ + 289, + 82, + 374, + 95 + ], + "score": 0.94, + "content": "f : U \\times [ 0 , T ] \\to \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "be a continuous function and let", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 93, + 321, + 107 + ], + "spans": [ + { + "bbox": [ + 107, + 96, + 117, + 105 + ], + "score": 0.56, + "content": "\\mathbf { z } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 93, + 120, + 107 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 121, + 95, + 132, + 105 + ], + "score": 0.58, + "content": "\\mathbf { z } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 93, + 137, + 107 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 137, + 93, + 184, + 106 + ], + "score": 0.82, + "content": "[ 0 , T ] \\to U", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 93, + 321, + 107 + ], + "score": 1.0, + "content": "satisfy the initial value problems:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 81, + 505, + 107 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 113, + 379, + 164 + ], + "lines": [ + { + "bbox": [ + 232, + 113, + 379, + 164 + ], + "spans": [ + { + "bbox": [ + 232, + 113, + 379, + 164 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\frac { \\mathrm { d } { \\mathbf z } _ { 1 } ( t ) } { \\mathrm { d } t } = f ( { \\mathbf z } _ { 1 } ( t ) , t ) , \\quad { \\mathbf z } _ { 1 } ( t ) = { \\mathbf x } _ { 1 } } \\\\ & { \\frac { \\mathrm { d } { \\mathbf z } _ { 2 } ( t ) } { \\mathrm { d } t } = f ( { \\mathbf z } _ { 2 } ( t ) , t ) , \\quad { \\mathbf z } _ { 2 } ( t ) = { \\mathbf x } _ { 2 } } \\end{array}", + "type": "interline_equation", + "image_path": "7f6eeb32253c8cad90d543707eb2749cd6d45ec22143c726afbc82c5047bbd0d.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 232, + 113, + 379, + 138.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 232, + 138.5, + 379, + 164.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 169, + 354, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 169, + 354, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 215, + 183 + ], + "score": 1.0, + "content": "Assume there is a constant", + "type": "text" + }, + { + "bbox": [ + 215, + 171, + 242, + 181 + ], + "score": 0.9, + "content": "C \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 169, + 311, + 183 + ], + "score": 1.0, + "content": "such that, for all", + "type": "text" + }, + { + "bbox": [ + 312, + 170, + 351, + 182 + ], + "score": 0.92, + "content": "t \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 169, + 354, + 183 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 169, + 354, + 183 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 189, + 404, + 204 + ], + "lines": [ + { + "bbox": [ + 206, + 189, + 404, + 204 + ], + "spans": [ + { + "bbox": [ + 206, + 189, + 404, + 204 + ], + "score": 0.86, + "content": "\\| f ( \\mathbf { z } _ { 2 } ( t ) , t ) - f ( \\mathbf { z } _ { 1 } ( t ) , t ) ) \\| \\leq C \\| \\mathbf { z } _ { 2 } ( t ) - \\mathbf { z } _ { 1 } ( t ) \\|", + "type": "interline_equation", + "image_path": "d44d70833a33bcbc6c66109b2ee246c42ecdacd3538edc2233bfcfea280df71a.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 206, + 189, + 404, + 204 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 210, + 204, + 223 + ], + "lines": [ + { + "bbox": [ + 106, + 209, + 205, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 209, + 162, + 225 + ], + "score": 1.0, + "content": "Then, for any", + "type": "text" + }, + { + "bbox": [ + 163, + 211, + 201, + 223 + ], + "score": 0.92, + "content": "t \\in [ 0 , T ]", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 209, + 205, + 225 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 106, + 209, + 205, + 225 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 225, + 379, + 240 + ], + "lines": [ + { + "bbox": [ + 232, + 225, + 379, + 240 + ], + "spans": [ + { + "bbox": [ + 232, + 225, + 379, + 240 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\| \\mathbf { z } _ { 1 } ( t ) - \\mathbf { z } _ { 2 } ( t ) \\| \\leq \\| \\mathbf { x } _ { 2 } - \\mathbf { x } _ { 1 } \\| \\cdot e ^ { C t } . } \\end{array}", + "type": "interline_equation", + "image_path": "9e1ae5bce98edee3c029a1c7316e4e9f396d1d4a5d6b60e76e8363662abe1da1.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 232, + 225, + 379, + 240 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 255, + 267, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 268, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 268, + 267 + ], + "score": 1.0, + "content": "7.4 MORE EXPERIMENTAL RESULTS", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 107, + 276, + 393, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 275, + 393, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 393, + 289 + ], + "score": 1.0, + "content": "7.4.1 COMPARISON IN THE SETTING OF ADVERSARIAL TRAINING", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 296, + 506, + 352 + ], + "lines": [ + { + "bbox": [ + 106, + 296, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 296, + 505, + 308 + ], + "score": 1.0, + "content": "We implement the adversarial training of the models on the MNIST dataset, and the adversarial", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 430, + 320 + ], + "score": 1.0, + "content": "examples for training are generated in real-time via the FGSM method (epsilon", + "type": "text" + }, + { + "bbox": [ + 430, + 308, + 450, + 318 + ], + "score": 0.69, + "content": "_ { 1 = 0 . 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 307, + 505, + 320 + ], + "score": 1.0, + "content": ") during each", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "epoch (Madry et al., 2017). The results of the adversarially trained models are shown in Table 7. We", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 329, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 342 + ], + "score": 1.0, + "content": "can observe that the neural ODE-based models are consistently more robust than CNN models. The", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 340, + 347, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 347, + 353 + ], + "score": 1.0, + "content": "proposed TisODE also outperforms the vanilla neural ODE.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 296, + 506, + 353 + ] + }, + { + "type": "table", + "bbox": [ + 178, + 408, + 430, + 469 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 364, + 504, + 398 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 365, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 237, + 377 + ], + "score": 1.0, + "content": "Table 7: Classification accuracy", + "type": "text" + }, + { + "bbox": [ + 238, + 365, + 253, + 376 + ], + "score": 0.79, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 365, + 506, + 377 + ], + "score": 1.0, + "content": "on perturbed images from MNIST. To evaluate the robustness", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 375, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 506, + 389 + ], + "score": 1.0, + "content": "of classifiers, we use three types of perturbations, namely zero-mean Gaussian noise with standard", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 387, + 280, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 145, + 398 + ], + "score": 1.0, + "content": "deviation", + "type": "text" + }, + { + "bbox": [ + 146, + 388, + 153, + 397 + ], + "score": 0.73, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 387, + 280, + 398 + ], + "score": 1.0, + "content": ", FGSM attack and PGD attack.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "table_body", + "bbox": [ + 178, + 408, + 430, + 469 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 178, + 408, + 430, + 469 + ], + "spans": [ + { + "bbox": [ + 178, + 408, + 430, + 469 + ], + "score": 0.981, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.3
CNN58.098.421.15.3
ODENet84.299.136.012.3
TisODE87.999.166.578.9
", + "type": "table", + "image_path": "55430c76602428adbc3f3d6582221d15213646ccc38819102c5ce51011e89650.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 178, + 408, + 430, + 428.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 178, + 428.3333333333333, + 430, + 448.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 178, + 448.66666666666663, + 430, + 468.99999999999994 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 107, + 492, + 321, + 504 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 322, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 322, + 505 + ], + "score": 1.0, + "content": "7.4.2 EXPERIMENTS ON THE CIFAR10 DATASET", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 512, + 506, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 506, + 524 + ], + "score": 1.0, + "content": "We conduct experiments on CIFAR10 to compare the robustness of CNN and neural ODE-based", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 523, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 536 + ], + "score": 1.0, + "content": "models. We train all the models only with original non-perturbed images and evaluate the robustness", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 534, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 506, + 547 + ], + "score": 1.0, + "content": "of models against random Gaussian noise and FGSM adversarial attacks. The results are shown in", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 546, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 505, + 557 + ], + "score": 1.0, + "content": "Table 8. We can observe that the ONENet is more robust than the CNN model in terms of both the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "score": 1.0, + "content": "random noise and the FGSM attack. Besides, our proposal, TisODE, can improve the robustness of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 568, + 203, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 203, + 578 + ], + "score": 1.0, + "content": "the vanilla neural ODE.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 512, + 506, + 578 + ] + }, + { + "type": "table", + "bbox": [ + 182, + 632, + 426, + 694 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 589, + 505, + 623 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 235, + 601 + ], + "score": 1.0, + "content": "Table 8: Classification accuracy", + "type": "text" + }, + { + "bbox": [ + 235, + 590, + 250, + 600 + ], + "score": 0.64, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "on perturbed images from CIFAR10. To evaluate the robustness", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "of classifiers, we use two types of perturbations, namely zero-mean Gaussian noise with standard", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 612, + 229, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 145, + 622 + ], + "score": 1.0, + "content": "deviation", + "type": "text" + }, + { + "bbox": [ + 146, + 613, + 153, + 621 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 612, + 229, + 622 + ], + "score": 1.0, + "content": "and FGSM attack.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "table_body", + "bbox": [ + 182, + 632, + 426, + 694 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 182, + 632, + 426, + 694 + ], + "spans": [ + { + "bbox": [ + 182, + 632, + 426, + 694 + ], + "score": 0.976, + "html": "
Gaussian noiseAdversarial attack
CIFAR10σ=15σ=20FGSM-8/255FGSM-10/255
CNN70.257.624.318.4
ODENet72.660.631.226.0
TisODE74.362.033.626.8
", + "type": "table", + "image_path": "2bad224570b817a9b629ba9432e5070888e5d4466432b80c75d325e9d0218670.jpg" + } + ] + } + ], + "index": 32.5, + "virtual_lines": [ + { + "bbox": [ + 182, + 632, + 426, + 647.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 182, + 647.5, + 426, + 663.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 182, + 663.0, + 426, + 678.5 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 182, + 678.5, + 426, + 694.0 + ], + "spans": [], + "index": 34 + } + ] + } + ], + "index": 30.75 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 710, + 504, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 504, + 721 + ], + "score": 1.0, + "content": "Here, we control the number of parameters to be the same for all kinds of models. We use a small", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 720, + 399, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 399, + 733 + ], + "score": 1.0, + "content": "network, which consists of five convolutional layers and one linear layer.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 106, + 710, + 504, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 185, + 101, + 424, + 177 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 195, + 80, + 415, + 92 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 194, + 79, + 416, + 93 + ], + "spans": [ + { + "bbox": [ + 194, + 79, + 416, + 93 + ], + "score": 1.0, + "content": "Table 9: The architecture of the ODENet on CIFAR10.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 185, + 101, + 424, + 177 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 185, + 101, + 424, + 177 + ], + "spans": [ + { + "bbox": [ + 185, + 101, + 424, + 177 + ], + "score": 0.979, + "html": "
RepetitionLayer
FE×1Conv(3,16,3,1)+GroupNorm+ReLU
×1Conv(16,32,3,2) + GroupNorm+ReLU
×1Conv(32,64,3,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "type": "table", + "image_path": "43938662417564eb0ebd0d23ae05bc447cbb134bb8df69de1457887746166857.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 185, + 101, + 424, + 116.2 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 185, + 116.2, + 424, + 131.4 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 185, + 131.4, + 424, + 146.6 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 185, + 146.6, + 424, + 161.79999999999998 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 185, + 161.79999999999998, + 424, + 176.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 1.5 + }, + { + "type": "title", + "bbox": [ + 106, + 197, + 442, + 209 + ], + "lines": [ + { + "bbox": [ + 106, + 197, + 443, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 197, + 443, + 210 + ], + "score": 1.0, + "content": "7.4.3 AN EXTENSION ON THE COMPARISON BETWEEN CNNS AND ODENETS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 108, + 216, + 505, + 250 + ], + "lines": [ + { + "bbox": [ + 105, + 216, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 505, + 229 + ], + "score": 1.0, + "content": "Here, we compare CNN and neural ODE-based models by controlling both the number of parameters", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 227, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 505, + 240 + ], + "score": 1.0, + "content": "and the number of function evaluations. We conduct experiments on the MNIST dataset, and all the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 237, + 349, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 349, + 252 + ], + "score": 1.0, + "content": "models are trained only with original non-perturbed images.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 255, + 505, + 321 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 506, + 267 + ], + "score": 1.0, + "content": "For the neural ODE-based models, the time range is set from 0 to 1. We use the Euler method, and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 266, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 363, + 278 + ], + "score": 1.0, + "content": "the step size is set to be 0.05. Thus the number of evaluations is", + "type": "text" + }, + { + "bbox": [ + 363, + 266, + 415, + 278 + ], + "score": 0.9, + "content": "1 / 0 . 0 5 = 2 0 ", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 267, + 505, + 278 + ], + "score": 1.0, + "content": ". For the CNN models", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "(specifically ResNet), we repeatedly concatenate the residual block for 20 times, and these 20 blocks", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 289, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 300 + ], + "score": 1.0, + "content": "share the same weights. Our experiments show that, in this condition, the neural ODE-based models", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 299, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 302, + 311 + ], + "score": 1.0, + "content": "still outperform the CNN models (FGSM-0.15:", + "type": "text" + }, + { + "bbox": [ + 302, + 299, + 330, + 310 + ], + "score": 0.84, + "content": "8 7 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 299, + 347, + 311 + ], + "score": 1.0, + "content": "vs.", + "type": "text" + }, + { + "bbox": [ + 347, + 299, + 374, + 310 + ], + "score": 0.85, + "content": "8 1 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 299, + 428, + 311 + ], + "score": 1.0, + "content": ", FGSM-0.3:", + "type": "text" + }, + { + "bbox": [ + 429, + 299, + 456, + 310 + ], + "score": 0.84, + "content": "5 3 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 299, + 474, + 311 + ], + "score": 1.0, + "content": "vs.", + "type": "text" + }, + { + "bbox": [ + 474, + 299, + 501, + 310 + ], + "score": 0.85, + "content": "4 9 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 299, + 505, + 311 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 310, + 219, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 148, + 321 + ], + "score": 1.0, + "content": "PGD-0.2:", + "type": "text" + }, + { + "bbox": [ + 148, + 310, + 175, + 321 + ], + "score": 0.61, + "content": "1 1 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 310, + 190, + 321 + ], + "score": 1.0, + "content": "vs.", + "type": "text" + }, + { + "bbox": [ + 191, + 311, + 213, + 321 + ], + "score": 0.83, + "content": "4 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 310, + 219, + 321 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 185, + 101, + 424, + 177 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 195, + 80, + 415, + 92 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 194, + 79, + 416, + 93 + ], + "spans": [ + { + "bbox": [ + 194, + 79, + 416, + 93 + ], + "score": 1.0, + "content": "Table 9: The architecture of the ODENet on CIFAR10.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 185, + 101, + 424, + 177 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 185, + 101, + 424, + 177 + ], + "spans": [ + { + "bbox": [ + 185, + 101, + 424, + 177 + ], + "score": 0.979, + "html": "
RepetitionLayer
FE×1Conv(3,16,3,1)+GroupNorm+ReLU
×1Conv(16,32,3,2) + GroupNorm+ReLU
×1Conv(32,64,3,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
", + "type": "table", + "image_path": "43938662417564eb0ebd0d23ae05bc447cbb134bb8df69de1457887746166857.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 185, + 101, + 424, + 116.2 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 185, + 116.2, + 424, + 131.4 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 185, + 131.4, + 424, + 146.6 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 185, + 146.6, + 424, + 161.79999999999998 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 185, + 161.79999999999998, + 424, + 176.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 1.5 + }, + { + "type": "title", + "bbox": [ + 106, + 197, + 442, + 209 + ], + "lines": [ + { + "bbox": [ + 106, + 197, + 443, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 197, + 443, + 210 + ], + "score": 1.0, + "content": "7.4.3 AN EXTENSION ON THE COMPARISON BETWEEN CNNS AND ODENETS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 108, + 216, + 505, + 250 + ], + "lines": [ + { + "bbox": [ + 105, + 216, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 505, + 229 + ], + "score": 1.0, + "content": "Here, we compare CNN and neural ODE-based models by controlling both the number of parameters", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 227, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 505, + 240 + ], + "score": 1.0, + "content": "and the number of function evaluations. We conduct experiments on the MNIST dataset, and all the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 237, + 349, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 349, + 252 + ], + "score": 1.0, + "content": "models are trained only with original non-perturbed images.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 216, + 505, + 252 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 255, + 505, + 321 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 506, + 267 + ], + "score": 1.0, + "content": "For the neural ODE-based models, the time range is set from 0 to 1. We use the Euler method, and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 266, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 363, + 278 + ], + "score": 1.0, + "content": "the step size is set to be 0.05. Thus the number of evaluations is", + "type": "text" + }, + { + "bbox": [ + 363, + 266, + 415, + 278 + ], + "score": 0.9, + "content": "1 / 0 . 0 5 = 2 0 ", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 267, + 505, + 278 + ], + "score": 1.0, + "content": ". For the CNN models", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "(specifically ResNet), we repeatedly concatenate the residual block for 20 times, and these 20 blocks", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 289, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 300 + ], + "score": 1.0, + "content": "share the same weights. Our experiments show that, in this condition, the neural ODE-based models", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 299, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 302, + 311 + ], + "score": 1.0, + "content": "still outperform the CNN models (FGSM-0.15:", + "type": "text" + }, + { + "bbox": [ + 302, + 299, + 330, + 310 + ], + "score": 0.84, + "content": "8 7 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 299, + 347, + 311 + ], + "score": 1.0, + "content": "vs.", + "type": "text" + }, + { + "bbox": [ + 347, + 299, + 374, + 310 + ], + "score": 0.85, + "content": "8 1 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 299, + 428, + 311 + ], + "score": 1.0, + "content": ", FGSM-0.3:", + "type": "text" + }, + { + "bbox": [ + 429, + 299, + 456, + 310 + ], + "score": 0.84, + "content": "5 3 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 299, + 474, + 311 + ], + "score": 1.0, + "content": "vs.", + "type": "text" + }, + { + "bbox": [ + 474, + 299, + 501, + 310 + ], + "score": 0.85, + "content": "4 9 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 299, + 505, + 311 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 310, + 219, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 148, + 321 + ], + "score": 1.0, + "content": "PGD-0.2:", + "type": "text" + }, + { + "bbox": [ + 148, + 310, + 175, + 321 + ], + "score": 0.61, + "content": "1 1 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 310, + 190, + 321 + ], + "score": 1.0, + "content": "vs.", + "type": "text" + }, + { + "bbox": [ + 191, + 311, + 213, + 321 + ], + "score": 0.83, + "content": "4 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 310, + 219, + 321 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 254, + 506, + 321 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_model.json b/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_model.json new file mode 100644 index 0000000000000000000000000000000000000000..8384783ee84d3529ee1f2f6eec31b826072a413d --- /dev/null +++ b/parse/train/B1e9Y2NYvS/B1e9Y2NYvS_model.json @@ -0,0 +1,18998 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 658, + 1302, + 658, + 1302, + 1269, + 398, + 1269 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1406, + 1402, + 1406, + 1402, + 1741, + 298, + 1741 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1759, + 1403, + 1759, + 1403, + 2034, + 298, + 2034 + ], + "score": 0.981 + }, + { + "category_id": 0, + "poly": [ + 300, + 219, + 1397, + 219, + 1397, + 323, + 300, + 323 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 314, + 417, + 1302, + 417, + 1302, + 506, + 314, + 506 + ], + "score": 0.901 + }, + { + "category_id": 0, + "poly": [ + 302, + 1336, + 573, + 1336, + 573, + 1370, + 302, + 1370 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 104, + 299, + 104 + ], + "score": 0.883 + }, + { + "category_id": 0, + "poly": [ + 773, + 587, + 926, + 587, + 926, + 620, + 773, + 620 + ], + "score": 0.883 + }, + { + "category_id": 1, + "poly": [ + 313, + 381, + 1055, + 381, + 1055, + 414, + 313, + 414 + ], + "score": 0.839 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 857, + 2088, + 857, + 2112, + 842, + 2112 + ], + "score": 0.66 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 217.0, + 1402.0, + 217.0, + 1402.0, + 272.0, + 295.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 272.0, + 654.0, + 272.0, + 654.0, + 332.0, + 294.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1332.0, + 579.0, + 1332.0, + 579.0, + 1378.0, + 294.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 586.0, + 932.0, + 586.0, + 932.0, + 623.0, + 769.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 2088.0, + 858.0, + 2088.0, + 858.0, + 2116.0, + 841.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 659.0, + 1304.0, + 659.0, + 1304.0, + 694.0, + 394.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 687.0, + 1305.0, + 687.0, + 1305.0, + 725.0, + 393.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 722.0, + 1305.0, + 722.0, + 1305.0, + 754.0, + 393.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 748.0, + 1306.0, + 748.0, + 1306.0, + 788.0, + 392.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 782.0, + 1306.0, + 782.0, + 1306.0, + 814.0, + 393.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 813.0, + 1306.0, + 813.0, + 1306.0, + 844.0, + 392.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 842.0, + 1306.0, + 842.0, + 1306.0, + 878.0, + 394.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 874.0, + 1305.0, + 874.0, + 1305.0, + 906.0, + 394.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 903.0, + 1305.0, + 903.0, + 1305.0, + 936.0, + 393.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 935.0, + 1305.0, + 935.0, + 1305.0, + 966.0, + 393.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 964.0, + 1306.0, + 964.0, + 1306.0, + 998.0, + 393.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 994.0, + 1304.0, + 994.0, + 1304.0, + 1030.0, + 394.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1024.0, + 1305.0, + 1024.0, + 1305.0, + 1059.0, + 393.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1052.0, + 1306.0, + 1052.0, + 1306.0, + 1090.0, + 393.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1086.0, + 1306.0, + 1086.0, + 1306.0, + 1118.0, + 393.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1116.0, + 1306.0, + 1116.0, + 1306.0, + 1152.0, + 394.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1148.0, + 1305.0, + 1148.0, + 1305.0, + 1180.0, + 394.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1175.0, + 1305.0, + 1175.0, + 1305.0, + 1211.0, + 394.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1208.0, + 1305.0, + 1208.0, + 1305.0, + 1241.0, + 395.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1236.0, + 1047.0, + 1236.0, + 1047.0, + 1271.0, + 394.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1407.0, + 1404.0, + 1407.0, + 1404.0, + 1441.0, + 294.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1439.0, + 1406.0, + 1439.0, + 1406.0, + 1472.0, + 292.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1467.0, + 1403.0, + 1467.0, + 1403.0, + 1503.0, + 293.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1498.0, + 1404.0, + 1498.0, + 1404.0, + 1533.0, + 296.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1530.0, + 1402.0, + 1530.0, + 1402.0, + 1561.0, + 296.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1560.0, + 1406.0, + 1560.0, + 1406.0, + 1595.0, + 294.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1592.0, + 1403.0, + 1592.0, + 1403.0, + 1623.0, + 296.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1620.0, + 1404.0, + 1620.0, + 1404.0, + 1655.0, + 294.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1653.0, + 1404.0, + 1653.0, + 1404.0, + 1683.0, + 296.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1680.0, + 1404.0, + 1680.0, + 1404.0, + 1716.0, + 294.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1713.0, + 726.0, + 1713.0, + 726.0, + 1744.0, + 296.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1758.0, + 1407.0, + 1758.0, + 1407.0, + 1795.0, + 296.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1786.0, + 1405.0, + 1786.0, + 1405.0, + 1828.0, + 293.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1854.0, + 296.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1851.0, + 1404.0, + 1851.0, + 1404.0, + 1883.0, + 296.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1878.0, + 1405.0, + 1878.0, + 1405.0, + 1916.0, + 293.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1405.0, + 1911.0, + 1405.0, + 1947.0, + 294.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1938.0, + 1404.0, + 1938.0, + 1404.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1969.0, + 1405.0, + 1969.0, + 1405.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 424.0, + 2003.0, + 424.0, + 2033.0, + 294.0, + 2033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 409.0, + 901.0, + 409.0, + 901.0, + 451.0, + 309.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 443.0, + 689.0, + 443.0, + 689.0, + 479.0, + 311.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 466.0, + 1309.0, + 466.0, + 1309.0, + 517.0, + 308.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 381.0, + 1057.0, + 381.0, + 1057.0, + 418.0, + 311.0, + 418.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 947, + 1404, + 947, + 1404, + 1374, + 298, + 1374 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1390, + 1403, + 1390, + 1403, + 1634, + 298, + 1634 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1649, + 1404, + 1649, + 1404, + 1895, + 297, + 1895 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 232, + 1045, + 232, + 1045, + 778, + 298, + 778 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 780, + 1403, + 780, + 1403, + 930, + 298, + 930 + ], + "score": 0.96 + }, + { + "category_id": 4, + "poly": [ + 1065, + 604, + 1402, + 604, + 1402, + 758, + 1065, + 758 + ], + "score": 0.959 + }, + { + "category_id": 3, + "poly": [ + 1122, + 205, + 1347, + 205, + 1347, + 584, + 1122, + 584 + ], + "score": 0.95 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 816, + 76, + 816, + 104, + 299, + 104 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.707 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.151 + }, + { + "category_id": 13, + "poly": [ + 787, + 1099, + 971, + 1099, + 971, + 1133, + 787, + 1133 + ], + "score": 0.94, + "latex": "\\tilde { { \\bf z } } _ { 1 } ( 0 ) \\bar { { \\bf \\phi } } = { \\bf \\bar { z } } _ { 1 } ( T ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 411, + 1221, + 682, + 1221, + 682, + 1255, + 411, + 1255 + ], + "score": 0.93, + "latex": "\\| { \\bf z } _ { 1 } ( T + T ^ { \\prime } ) - { \\bf z } _ { 1 } ( T ) \\|" + }, + { + "category_id": 13, + "poly": [ + 1193, + 1190, + 1393, + 1190, + 1393, + 1224, + 1193, + 1224 + ], + "score": 0.93, + "latex": "\\| \\widetilde { \\mathbf z } _ { 1 } ( \\bar { T } ) - \\mathbf z _ { 1 } ( T ) \\|" + }, + { + "category_id": 13, + "poly": [ + 576, + 1129, + 636, + 1129, + 636, + 1163, + 576, + 1163 + ], + "score": 0.92, + "latex": "{ \\bf z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 486, + 1099, + 548, + 1099, + 548, + 1133, + 486, + 1133 + ], + "score": 0.92, + "latex": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1085, + 1100, + 1174, + 1100, + 1174, + 1129, + 1085, + 1129 + ], + "score": 0.92, + "latex": "T ^ { \\prime } > 0" + }, + { + "category_id": 13, + "poly": [ + 1332, + 1069, + 1393, + 1069, + 1393, + 1103, + 1332, + 1103 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 383, + 1191, + 462, + 1191, + 462, + 1219, + 383, + 1219 + ], + "score": 0.89, + "latex": "T > 0" + }, + { + "category_id": 13, + "poly": [ + 338, + 1129, + 390, + 1129, + 390, + 1158, + 338, + 1158 + ], + "score": 0.87, + "latex": "- T ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1265, + 1253, + 1288, + 1253, + 1288, + 1279, + 1265, + 1279 + ], + "score": 0.71, + "latex": "T" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 602.0, + 1405.0, + 602.0, + 1405.0, + 636.0, + 1065.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 634.0, + 1405.0, + 634.0, + 1405.0, + 666.0, + 1065.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 665.0, + 1405.0, + 665.0, + 1405.0, + 697.0, + 1064.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 696.0, + 1403.0, + 696.0, + 1403.0, + 731.0, + 1065.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 726.0, + 1219.0, + 726.0, + 1219.0, + 764.0, + 1063.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 211.0, + 1264.0, + 211.0, + 1264.0, + 242.0, + 1206.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 270.0, + 1273.0, + 270.0, + 1273.0, + 300.0, + 1197.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 293.0, + 1279.0, + 293.0, + 1279.0, + 322.0, + 1192.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 357.0, + 1304.0, + 357.0, + 1304.0, + 387.0, + 1167.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 377.0, + 1280.0, + 377.0, + 1280.0, + 412.0, + 1191.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 403.0, + 1298.0, + 403.0, + 1298.0, + 429.0, + 1173.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 465.0, + 1308.0, + 465.0, + 1308.0, + 494.0, + 1159.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1190.0, + 486.0, + 1280.0, + 486.0, + 1280.0, + 519.0, + 1190.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 529.0, + 1243.0, + 529.0, + 1243.0, + 545.0, + 1226.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 546.0, + 1284.0, + 546.0, + 1284.0, + 578.0, + 1187.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 946.0, + 1405.0, + 946.0, + 1405.0, + 982.0, + 294.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 978.0, + 1406.0, + 978.0, + 1406.0, + 1014.0, + 294.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1006.0, + 1405.0, + 1006.0, + 1405.0, + 1044.0, + 293.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1036.0, + 1405.0, + 1036.0, + 1405.0, + 1074.0, + 293.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1068.0, + 1331.0, + 1068.0, + 1331.0, + 1107.0, + 293.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1068.0, + 1405.0, + 1068.0, + 1405.0, + 1107.0, + 1394.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1100.0, + 485.0, + 1100.0, + 485.0, + 1135.0, + 294.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1100.0, + 786.0, + 1100.0, + 786.0, + 1135.0, + 549.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 1100.0, + 1084.0, + 1100.0, + 1084.0, + 1135.0, + 972.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 1100.0, + 1406.0, + 1100.0, + 1406.0, + 1135.0, + 1175.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1125.0, + 337.0, + 1125.0, + 337.0, + 1167.0, + 292.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1125.0, + 575.0, + 1125.0, + 575.0, + 1167.0, + 391.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1125.0, + 1408.0, + 1125.0, + 1408.0, + 1167.0, + 637.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1162.0, + 1405.0, + 1162.0, + 1405.0, + 1193.0, + 296.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1191.0, + 382.0, + 1191.0, + 382.0, + 1226.0, + 293.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1191.0, + 1192.0, + 1191.0, + 1192.0, + 1226.0, + 463.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1191.0, + 1404.0, + 1191.0, + 1404.0, + 1226.0, + 1394.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1220.0, + 410.0, + 1220.0, + 410.0, + 1258.0, + 293.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1220.0, + 1407.0, + 1220.0, + 1407.0, + 1258.0, + 683.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1250.0, + 1264.0, + 1250.0, + 1264.0, + 1286.0, + 294.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 1250.0, + 1406.0, + 1250.0, + 1406.0, + 1286.0, + 1289.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1282.0, + 1405.0, + 1282.0, + 1405.0, + 1316.0, + 293.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1313.0, + 1405.0, + 1313.0, + 1405.0, + 1348.0, + 294.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1344.0, + 1175.0, + 1344.0, + 1175.0, + 1376.0, + 296.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1388.0, + 1405.0, + 1388.0, + 1405.0, + 1425.0, + 293.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1422.0, + 1404.0, + 1422.0, + 1404.0, + 1452.0, + 296.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1451.0, + 1403.0, + 1451.0, + 1403.0, + 1485.0, + 294.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1480.0, + 1404.0, + 1480.0, + 1404.0, + 1514.0, + 293.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1512.0, + 1405.0, + 1512.0, + 1405.0, + 1545.0, + 294.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1543.0, + 1405.0, + 1543.0, + 1405.0, + 1576.0, + 294.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1569.0, + 1407.0, + 1569.0, + 1407.0, + 1610.0, + 293.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1601.0, + 480.0, + 1601.0, + 480.0, + 1639.0, + 295.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1651.0, + 1404.0, + 1651.0, + 1404.0, + 1684.0, + 295.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1682.0, + 1405.0, + 1682.0, + 1405.0, + 1712.0, + 296.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1712.0, + 1405.0, + 1712.0, + 1405.0, + 1746.0, + 294.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1741.0, + 1402.0, + 1741.0, + 1402.0, + 1777.0, + 294.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1773.0, + 1404.0, + 1773.0, + 1404.0, + 1807.0, + 295.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1803.0, + 1405.0, + 1803.0, + 1405.0, + 1837.0, + 294.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1832.0, + 1407.0, + 1832.0, + 1407.0, + 1867.0, + 294.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1862.0, + 1072.0, + 1862.0, + 1072.0, + 1901.0, + 294.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 1045.0, + 229.0, + 1045.0, + 263.0, + 296.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 260.0, + 1046.0, + 260.0, + 1046.0, + 296.0, + 295.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 288.0, + 1046.0, + 288.0, + 1046.0, + 328.0, + 294.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 321.0, + 1046.0, + 321.0, + 1046.0, + 356.0, + 295.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 352.0, + 1046.0, + 352.0, + 1046.0, + 386.0, + 296.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 382.0, + 1047.0, + 382.0, + 1047.0, + 416.0, + 296.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 415.0, + 1045.0, + 415.0, + 1045.0, + 446.0, + 296.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 443.0, + 1046.0, + 443.0, + 1046.0, + 478.0, + 295.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 475.0, + 1047.0, + 475.0, + 1047.0, + 506.0, + 295.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 504.0, + 1047.0, + 504.0, + 1047.0, + 538.0, + 295.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 534.0, + 1046.0, + 534.0, + 1046.0, + 569.0, + 293.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 565.0, + 1047.0, + 565.0, + 1047.0, + 600.0, + 295.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 596.0, + 1047.0, + 596.0, + 1047.0, + 629.0, + 295.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 627.0, + 1046.0, + 627.0, + 1046.0, + 658.0, + 294.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 656.0, + 1046.0, + 656.0, + 1046.0, + 689.0, + 295.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 687.0, + 1045.0, + 687.0, + 1045.0, + 720.0, + 295.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 719.0, + 1045.0, + 719.0, + 1045.0, + 750.0, + 296.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 748.0, + 1046.0, + 748.0, + 1046.0, + 783.0, + 294.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 778.0, + 1403.0, + 778.0, + 1403.0, + 811.0, + 296.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 807.0, + 1404.0, + 807.0, + 1404.0, + 843.0, + 294.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 838.0, + 1405.0, + 838.0, + 1405.0, + 876.0, + 292.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 865.0, + 1407.0, + 865.0, + 1407.0, + 907.0, + 293.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 901.0, + 1053.0, + 901.0, + 1053.0, + 933.0, + 294.0, + 933.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1510, + 1405, + 1510, + 1405, + 1786, + 298, + 1786 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 520, + 1403, + 520, + 1403, + 706, + 298, + 706 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 951, + 1404, + 951, + 1404, + 1107, + 298, + 1107 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 294, + 1403, + 294, + 1403, + 447, + 297, + 447 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 720, + 1403, + 720, + 1403, + 846, + 298, + 846 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1881, + 1403, + 1881, + 1403, + 2034, + 299, + 2034 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1218, + 1404, + 1218, + 1404, + 1496, + 297, + 1496 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 530, + 861, + 1165, + 861, + 1165, + 938, + 530, + 938 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 563, + 450, + 1136, + 450, + 1136, + 509, + 563, + 509 + ], + "score": 0.937 + }, + { + "category_id": 0, + "poly": [ + 299, + 1151, + 1180, + 1151, + 1180, + 1186, + 299, + 1186 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 299, + 225, + 838, + 225, + 838, + 262, + 299, + 262 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 300, + 1824, + 677, + 1824, + 677, + 1855, + 300, + 1855 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 105, + 299, + 105 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1366, + 467, + 1400, + 467, + 1400, + 498, + 1366, + 498 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.6 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.508 + }, + { + "category_id": 13, + "poly": [ + 1260, + 614, + 1393, + 614, + 1393, + 647, + 1260, + 647 + ], + "score": 0.93, + "latex": "T \\in ( 0 , \\infty )" + }, + { + "category_id": 14, + "poly": [ + 531, + 858, + 1167, + 858, + 1167, + 938, + 531, + 938 + ], + "score": 0.92, + "latex": "\\mathbf { z } _ { \\mathrm { o u t } } = \\mathbf { z } ( T ) = \\mathbf { z } ( 0 ) + \\int _ { 0 } ^ { T } f _ { \\theta } ( \\mathbf { z } ( t ) , t ) \\mathrm { d } t = \\phi _ { T } ( \\mathbf { z } _ { \\mathrm { i n } } , f _ { \\theta } ) ." + }, + { + "category_id": 13, + "poly": [ + 768, + 953, + 827, + 953, + 827, + 986, + 768, + 986 + ], + "score": 0.92, + "latex": "{ \\mathbf z } ( T )" + }, + { + "category_id": 13, + "poly": [ + 375, + 520, + 665, + 520, + 665, + 554, + 375, + 554 + ], + "score": 0.92, + "latex": "f _ { \\theta } : \\mathbb { R } ^ { d } \\times [ 0 , \\infty ) \\mathbb { R } ^ { d }" + }, + { + "category_id": 14, + "poly": [ + 561, + 444, + 1137, + 444, + 1137, + 513, + 561, + 513 + ], + "score": 0.91, + "latex": "\\frac { \\mathrm { d } { \\mathbf z } ( t ) } { \\mathrm { d } t } = f _ { \\boldsymbol \\theta } ( { \\mathbf z } ( t ) , t ) , \\quad { \\mathbf z } ( 0 ) = { \\mathbf z } _ { \\mathrm { i n } } , \\quad { \\mathbf z } _ { \\mathrm { o u t } } = { \\mathbf z } ( T ) ," + }, + { + "category_id": 13, + "poly": [ + 298, + 813, + 383, + 813, + 383, + 846, + 298, + 846 + ], + "score": 0.91, + "latex": "\\phi _ { T } ( \\cdot , \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 345, + 554, + 544, + 554, + 544, + 585, + 345, + 585 + ], + "score": 0.9, + "latex": "\\mathbf { z } : [ 0 , \\infty ) \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 594, + 616, + 659, + 616, + 659, + 642, + 594, + 642 + ], + "score": 0.9, + "latex": "t = 0" + }, + { + "category_id": 13, + "poly": [ + 836, + 617, + 888, + 617, + 888, + 645, + 836, + 645 + ], + "score": 0.89, + "latex": "\\mathbf { z } _ { \\mathrm { o u t } }" + }, + { + "category_id": 13, + "poly": [ + 1246, + 2003, + 1366, + 2003, + 1366, + 2032, + 1246, + 2032 + ], + "score": 0.89, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 639, + 756, + 676, + 756, + 676, + 783, + 639, + 783 + ], + "score": 0.89, + "latex": "\\mathbf { z } _ { \\mathrm { i n } }" + }, + { + "category_id": 13, + "poly": [ + 597, + 727, + 648, + 727, + 648, + 752, + 597, + 752 + ], + "score": 0.89, + "latex": "\\mathbf { z } _ { \\mathrm { o u t } }" + }, + { + "category_id": 13, + "poly": [ + 877, + 753, + 905, + 753, + 905, + 784, + 877, + 784 + ], + "score": 0.88, + "latex": "f _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1345, + 554, + 1373, + 554, + 1373, + 584, + 1345, + 584 + ], + "score": 0.87, + "latex": "f _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1211, + 645, + 1240, + 645, + 1240, + 676, + 1211, + 676 + ], + "score": 0.87, + "latex": "f _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1153, + 587, + 1190, + 587, + 1190, + 614, + 1153, + 614 + ], + "score": 0.87, + "latex": "\\mathbf { z } _ { \\mathrm { i n } }" + }, + { + "category_id": 13, + "poly": [ + 764, + 814, + 792, + 814, + 792, + 845, + 764, + 845 + ], + "score": 0.87, + "latex": "f _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 433, + 726, + 469, + 726, + 469, + 752, + 433, + 752 + ], + "score": 0.87, + "latex": "\\mathbf { z } _ { \\mathrm { i n } }" + }, + { + "category_id": 13, + "poly": [ + 522, + 817, + 559, + 817, + 559, + 843, + 522, + 843 + ], + "score": 0.87, + "latex": "\\mathbf { z } _ { \\mathrm { i n } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 755, + 347, + 755, + 347, + 784, + 297, + 784 + ], + "score": 0.87, + "latex": "\\mathbf { z } _ { \\mathrm { o u t } }" + }, + { + "category_id": 13, + "poly": [ + 400, + 644, + 429, + 644, + 429, + 676, + 400, + 676 + ], + "score": 0.87, + "latex": "f _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1168, + 723, + 1191, + 723, + 1191, + 749, + 1168, + 749 + ], + "score": 0.83, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 1139, + 784, + 1157, + 784, + 1157, + 810, + 1139, + 810 + ], + "score": 0.82, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 515, + 955, + 539, + 955, + 539, + 981, + 515, + 981 + ], + "score": 0.82, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 712, + 554, + 730, + 554, + 730, + 580, + 712, + 580 + ], + "score": 0.82, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 454, + 588, + 467, + 588, + 467, + 611, + 454, + 611 + ], + "score": 0.79, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1215, + 1046, + 1233, + 1046, + 1233, + 1072, + 1215, + 1072 + ], + "score": 0.78, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 1384, + 523, + 1401, + 523, + 1401, + 550, + 1384, + 550 + ], + "score": 0.76, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 1036, + 647, + 1050, + 647, + 1050, + 671, + 1036, + 671 + ], + "score": 0.74, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 597, + 1635, + 612, + 1635, + 612, + 1661, + 597, + 1661 + ], + "score": 0.58, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 881, + 589, + 899, + 589, + 899, + 611, + 881, + 611 + ], + "score": 0.52, + "latex": "\\mathbf { z }" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1148.0, + 1184.0, + 1148.0, + 1184.0, + 1191.0, + 292.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 224.0, + 839.0, + 224.0, + 839.0, + 267.0, + 292.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1824.0, + 679.0, + 1824.0, + 679.0, + 1857.0, + 296.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 860.0, + 2085.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1512.0, + 1405.0, + 1512.0, + 1405.0, + 1545.0, + 296.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1542.0, + 1403.0, + 1542.0, + 1403.0, + 1575.0, + 296.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1571.0, + 1405.0, + 1571.0, + 1405.0, + 1608.0, + 293.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1605.0, + 1402.0, + 1605.0, + 1402.0, + 1634.0, + 296.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1634.0, + 596.0, + 1634.0, + 596.0, + 1666.0, + 296.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 1634.0, + 1406.0, + 1634.0, + 1406.0, + 1666.0, + 613.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1661.0, + 1405.0, + 1661.0, + 1405.0, + 1700.0, + 292.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1694.0, + 1403.0, + 1694.0, + 1403.0, + 1726.0, + 296.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1722.0, + 1405.0, + 1722.0, + 1405.0, + 1760.0, + 293.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1756.0, + 707.0, + 1756.0, + 707.0, + 1789.0, + 297.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 516.0, + 374.0, + 516.0, + 374.0, + 558.0, + 293.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 516.0, + 1383.0, + 516.0, + 1383.0, + 558.0, + 666.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 516.0, + 1408.0, + 516.0, + 1408.0, + 558.0, + 1402.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 549.0, + 344.0, + 549.0, + 344.0, + 587.0, + 294.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 549.0, + 711.0, + 549.0, + 711.0, + 587.0, + 545.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 549.0, + 1344.0, + 549.0, + 1344.0, + 587.0, + 731.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 549.0, + 1407.0, + 549.0, + 1407.0, + 587.0, + 1374.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 582.0, + 453.0, + 582.0, + 453.0, + 616.0, + 293.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 582.0, + 880.0, + 582.0, + 880.0, + 616.0, + 468.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 582.0, + 1152.0, + 582.0, + 1152.0, + 616.0, + 900.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 582.0, + 1407.0, + 582.0, + 1407.0, + 616.0, + 1191.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 610.0, + 593.0, + 610.0, + 593.0, + 650.0, + 293.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 610.0, + 835.0, + 610.0, + 835.0, + 650.0, + 660.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 610.0, + 1259.0, + 610.0, + 1259.0, + 650.0, + 889.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 610.0, + 1407.0, + 610.0, + 1407.0, + 650.0, + 1394.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 642.0, + 399.0, + 642.0, + 399.0, + 678.0, + 294.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 642.0, + 1035.0, + 642.0, + 1035.0, + 678.0, + 430.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 642.0, + 1210.0, + 642.0, + 1210.0, + 678.0, + 1051.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 642.0, + 1405.0, + 642.0, + 1405.0, + 678.0, + 1241.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 676.0, + 626.0, + 676.0, + 626.0, + 705.0, + 296.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 952.0, + 514.0, + 952.0, + 514.0, + 985.0, + 297.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 952.0, + 767.0, + 952.0, + 767.0, + 985.0, + 540.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 952.0, + 1404.0, + 952.0, + 1404.0, + 985.0, + 828.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 984.0, + 1402.0, + 984.0, + 1402.0, + 1017.0, + 296.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1013.0, + 1405.0, + 1013.0, + 1405.0, + 1046.0, + 296.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1042.0, + 1214.0, + 1042.0, + 1214.0, + 1078.0, + 295.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 1042.0, + 1403.0, + 1042.0, + 1403.0, + 1078.0, + 1234.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1074.0, + 879.0, + 1074.0, + 879.0, + 1108.0, + 296.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 297.0, + 1404.0, + 297.0, + 1404.0, + 330.0, + 295.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 327.0, + 1405.0, + 327.0, + 1405.0, + 360.0, + 296.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 356.0, + 1405.0, + 356.0, + 1405.0, + 392.0, + 293.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 386.0, + 1407.0, + 386.0, + 1407.0, + 423.0, + 293.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 418.0, + 448.0, + 418.0, + 448.0, + 452.0, + 295.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 721.0, + 432.0, + 721.0, + 432.0, + 757.0, + 294.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 721.0, + 596.0, + 721.0, + 596.0, + 757.0, + 470.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 721.0, + 1167.0, + 721.0, + 1167.0, + 757.0, + 649.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 721.0, + 1407.0, + 721.0, + 1407.0, + 757.0, + 1192.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 751.0, + 296.0, + 751.0, + 296.0, + 787.0, + 292.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 751.0, + 638.0, + 751.0, + 638.0, + 787.0, + 348.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 751.0, + 876.0, + 751.0, + 876.0, + 787.0, + 677.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 751.0, + 1404.0, + 751.0, + 1404.0, + 787.0, + 906.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 783.0, + 1138.0, + 783.0, + 1138.0, + 816.0, + 293.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 783.0, + 1403.0, + 783.0, + 1403.0, + 816.0, + 1158.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 812.0, + 297.0, + 812.0, + 297.0, + 849.0, + 294.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 812.0, + 521.0, + 812.0, + 521.0, + 849.0, + 384.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 812.0, + 763.0, + 812.0, + 763.0, + 849.0, + 560.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 812.0, + 854.0, + 812.0, + 854.0, + 849.0, + 793.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1881.0, + 1405.0, + 1881.0, + 1405.0, + 1914.0, + 295.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1911.0, + 1405.0, + 1911.0, + 1405.0, + 1944.0, + 297.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1940.0, + 1407.0, + 1940.0, + 1407.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1969.0, + 1408.0, + 1969.0, + 1408.0, + 2008.0, + 292.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 1245.0, + 2001.0, + 1245.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1367.0, + 2001.0, + 1376.0, + 2001.0, + 1376.0, + 2038.0, + 1367.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1221.0, + 1402.0, + 1221.0, + 1402.0, + 1253.0, + 296.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1251.0, + 1405.0, + 1251.0, + 1405.0, + 1285.0, + 293.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1278.0, + 1404.0, + 1278.0, + 1404.0, + 1318.0, + 294.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1312.0, + 1401.0, + 1312.0, + 1401.0, + 1345.0, + 296.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1343.0, + 1404.0, + 1343.0, + 1404.0, + 1375.0, + 295.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1371.0, + 1405.0, + 1371.0, + 1405.0, + 1407.0, + 294.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1404.0, + 1405.0, + 1404.0, + 1405.0, + 1437.0, + 295.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1432.0, + 1405.0, + 1432.0, + 1405.0, + 1470.0, + 291.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1464.0, + 412.0, + 1464.0, + 412.0, + 1495.0, + 294.0, + 1495.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 327, + 1517, + 1366, + 1517, + 1366, + 1873, + 327, + 1873 + ], + "score": 0.986, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=50σ=75σ=100FGSM-0.15FGSM-0.3FGSM-0.5
CNN98.1±0.785.8±4.356.4±5.663.4±2.324.0±8.98.3±3.2
ODENet98.7±0.690.6±5.473.2±8.683.5±0.942.1±2.414.3±2.1
SVHNg=15g=25σ=35FGSM-3/255FGSM-5/255FGSM-8/255
CNN90.0±1.276.3±2.760.9±3.929.2±2.913.7±1.95.4±1.5
ODENet95.7±0.788.1±1.578.2±2.158.2±2.343.0±1.330.9±1.4
ImgNet10σ=10σ=15σ = 25FGSM-5/255FGSM-8/255FGSM-16/255
CNN80.1±1.863.3±2.040.8±2.728.5±0.518.1±0.79.4±1.2
ODENet81.9±2.067.5±2.048.7±2.636.2±1.027.2±1.114.4±1.7
" + }, + { + "category_id": 1, + "poly": [ + 298, + 230, + 1404, + 230, + 1404, + 565, + 298, + 565 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 582, + 1403, + 582, + 1403, + 826, + 298, + 826 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1124, + 1402, + 1124, + 1402, + 1308, + 298, + 1308 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 926, + 1403, + 926, + 1403, + 1109, + 298, + 1109 + ], + "score": 0.978 + }, + { + "category_id": 2, + "poly": [ + 332, + 2006, + 933, + 2006, + 933, + 2034, + 332, + 2034 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 299, + 1913, + 1401, + 1913, + 1401, + 1976, + 299, + 1976 + ], + "score": 0.91 + }, + { + "category_id": 6, + "poly": [ + 298, + 1334, + 1402, + 1334, + 1402, + 1489, + 298, + 1489 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 816, + 76, + 816, + 104, + 299, + 104 + ], + "score": 0.888 + }, + { + "category_id": 0, + "poly": [ + 304, + 868, + 1229, + 868, + 1229, + 899, + 304, + 899 + ], + "score": 0.824 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2111, + 840, + 2111 + ], + "score": 0.779 + }, + { + "category_id": 1, + "poly": [ + 298, + 1334, + 1402, + 1334, + 1402, + 1489, + 298, + 1489 + ], + "score": 0.091 + }, + { + "category_id": 13, + "poly": [ + 993, + 1049, + 1068, + 1049, + 1068, + 1078, + 993, + 1078 + ], + "score": 0.89, + "latex": "9 5 . 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1287, + 1048, + 1362, + 1048, + 1362, + 1078, + 1287, + 1078 + ], + "score": 0.88, + "latex": "8 0 . 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1248, + 1041, + 1248, + 1041, + 1278, + 1007, + 1278 + ], + "score": 0.88, + "latex": "l _ { \\infty }" + }, + { + "category_id": 13, + "poly": [ + 777, + 1049, + 853, + 1049, + 853, + 1078, + 777, + 1078 + ], + "score": 0.86, + "latex": "9 9 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 298, + 1367, + 342, + 1367, + 342, + 1397, + 298, + 1397 + ], + "score": 0.8, + "latex": "( \\% )" + }, + { + "category_id": 13, + "poly": [ + 684, + 1253, + 705, + 1253, + 705, + 1275, + 684, + 1275 + ], + "score": 0.78, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 697, + 1368, + 727, + 1368, + 727, + 1396, + 697, + 1396 + ], + "score": 0.52, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 1114, + 1253, + 1130, + 1253, + 1130, + 1275, + 1114, + 1275 + ], + "score": 0.51, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 495, + 292, + 553, + 292, + 553, + 321, + 495, + 321 + ], + "score": 0.46, + "latex": "1 4 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 913, + 383, + 972, + 383, + 972, + 412, + 913, + 412 + ], + "score": 0.36, + "latex": "2 8 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 746, + 1779, + 825, + 1779, + 825, + 1804, + 746, + 1804 + ], + "score": 0.28, + "latex": "\\sigma = 2 5" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 2001.0, + 937.0, + 2001.0, + 937.0, + 2038.0, + 333.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1336.0, + 1404.0, + 1336.0, + 1404.0, + 1369.0, + 296.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1365.0, + 297.0, + 1365.0, + 297.0, + 1402.0, + 294.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1365.0, + 696.0, + 1365.0, + 696.0, + 1402.0, + 343.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1365.0, + 1406.0, + 1365.0, + 1406.0, + 1402.0, + 728.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1397.0, + 1404.0, + 1397.0, + 1404.0, + 1430.0, + 296.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1426.0, + 1406.0, + 1426.0, + 1406.0, + 1463.0, + 293.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1455.0, + 391.0, + 1455.0, + 391.0, + 1490.0, + 292.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 867.0, + 1236.0, + 867.0, + 1236.0, + 902.0, + 296.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 863.0, + 2086.0, + 863.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 262.0, + 1406.0, + 262.0, + 1406.0, + 295.0, + 293.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 289.0, + 494.0, + 289.0, + 494.0, + 326.0, + 293.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 289.0, + 1406.0, + 289.0, + 1406.0, + 326.0, + 554.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 321.0, + 1404.0, + 321.0, + 1404.0, + 355.0, + 294.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 351.0, + 1404.0, + 351.0, + 1404.0, + 385.0, + 294.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 383.0, + 912.0, + 383.0, + 912.0, + 417.0, + 293.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 383.0, + 1405.0, + 383.0, + 1405.0, + 417.0, + 973.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 411.0, + 1405.0, + 411.0, + 1405.0, + 447.0, + 293.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 443.0, + 1405.0, + 443.0, + 1405.0, + 478.0, + 296.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 475.0, + 1402.0, + 475.0, + 1402.0, + 506.0, + 296.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 504.0, + 1405.0, + 504.0, + 1405.0, + 538.0, + 296.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 532.0, + 1267.0, + 532.0, + 1267.0, + 569.0, + 293.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 580.0, + 1404.0, + 580.0, + 1404.0, + 618.0, + 293.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 612.0, + 1405.0, + 612.0, + 1405.0, + 647.0, + 294.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 639.0, + 1404.0, + 639.0, + 1404.0, + 680.0, + 293.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 672.0, + 1405.0, + 672.0, + 1405.0, + 708.0, + 293.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 702.0, + 1407.0, + 702.0, + 1407.0, + 740.0, + 292.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 734.0, + 1407.0, + 734.0, + 1407.0, + 770.0, + 292.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 761.0, + 1405.0, + 761.0, + 1405.0, + 802.0, + 292.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 796.0, + 956.0, + 796.0, + 956.0, + 829.0, + 296.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1125.0, + 1406.0, + 1125.0, + 1406.0, + 1160.0, + 296.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1158.0, + 1404.0, + 1158.0, + 1404.0, + 1189.0, + 296.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1187.0, + 1403.0, + 1187.0, + 1403.0, + 1219.0, + 296.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1216.0, + 1404.0, + 1216.0, + 1404.0, + 1252.0, + 293.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1246.0, + 683.0, + 1246.0, + 683.0, + 1282.0, + 294.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1246.0, + 1006.0, + 1246.0, + 1006.0, + 1282.0, + 706.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 1246.0, + 1113.0, + 1246.0, + 1113.0, + 1282.0, + 1042.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1246.0, + 1407.0, + 1246.0, + 1407.0, + 1282.0, + 1131.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1278.0, + 683.0, + 1278.0, + 683.0, + 1309.0, + 294.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 925.0, + 1405.0, + 925.0, + 1405.0, + 961.0, + 294.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 953.0, + 1405.0, + 953.0, + 1405.0, + 994.0, + 291.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 985.0, + 1405.0, + 985.0, + 1405.0, + 1023.0, + 293.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1015.0, + 1407.0, + 1015.0, + 1407.0, + 1053.0, + 293.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1048.0, + 776.0, + 1048.0, + 776.0, + 1083.0, + 293.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1048.0, + 992.0, + 1048.0, + 992.0, + 1083.0, + 854.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 1048.0, + 1286.0, + 1048.0, + 1286.0, + 1083.0, + 1069.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 1048.0, + 1407.0, + 1048.0, + 1407.0, + 1083.0, + 1363.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1077.0, + 424.0, + 1077.0, + 424.0, + 1114.0, + 293.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1403.0, + 1911.0, + 1403.0, + 1950.0, + 294.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1981.0, + 294.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1336.0, + 1404.0, + 1336.0, + 1404.0, + 1369.0, + 296.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1365.0, + 297.0, + 1365.0, + 297.0, + 1402.0, + 294.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1365.0, + 696.0, + 1365.0, + 696.0, + 1402.0, + 343.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1365.0, + 1406.0, + 1365.0, + 1406.0, + 1402.0, + 728.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1397.0, + 1404.0, + 1397.0, + 1404.0, + 1430.0, + 296.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1426.0, + 1406.0, + 1426.0, + 1406.0, + 1463.0, + 293.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1455.0, + 391.0, + 1455.0, + 391.0, + 1490.0, + 292.0, + 1490.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1413, + 1404, + 1413, + 1404, + 1780, + 297, + 1780 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 574, + 1403, + 574, + 1403, + 819, + 298, + 819 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1403, + 229, + 1403, + 444, + 298, + 444 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1880, + 1402, + 1880, + 1402, + 2034, + 299, + 2034 + ], + "score": 0.976 + }, + { + "category_id": 6, + "poly": [ + 298, + 851, + 1402, + 851, + 1402, + 1005, + 298, + 1005 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 300, + 1821, + 908, + 1821, + 908, + 1853, + 300, + 1853 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 816, + 76, + 816, + 104, + 300, + 104 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 301, + 486, + 1276, + 486, + 1276, + 546, + 301, + 546 + ], + "score": 0.748 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.713 + }, + { + "category_id": 1, + "poly": [ + 298, + 851, + 1402, + 851, + 1402, + 1005, + 298, + 1005 + ], + "score": 0.107 + }, + { + "category_id": 13, + "poly": [ + 631, + 383, + 729, + 383, + 729, + 411, + 631, + 411 + ], + "score": 0.9, + "latex": "\\sigma = 2 5" + }, + { + "category_id": 13, + "poly": [ + 507, + 1477, + 542, + 1477, + 542, + 1506, + 507, + 1506 + ], + "score": 0.89, + "latex": "l _ { \\infty }" + }, + { + "category_id": 13, + "poly": [ + 988, + 727, + 1139, + 727, + 1139, + 761, + 988, + 761 + ], + "score": 0.89, + "latex": "\\bar { \\{ 5 0 , 7 5 , 1 0 0 \\} }" + }, + { + "category_id": 13, + "poly": [ + 1137, + 1567, + 1228, + 1567, + 1228, + 1598, + 1137, + 1598 + ], + "score": 0.88, + "latex": "( 6 4 . 7 \\% )" + }, + { + "category_id": 13, + "poly": [ + 748, + 758, + 885, + 758, + 885, + 792, + 748, + 792 + ], + "score": 0.88, + "latex": "\\{ 1 0 , 1 5 , 2 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 512, + 353, + 587, + 353, + 587, + 382, + 512, + 382 + ], + "score": 0.88, + "latex": "4 3 . 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 553, + 1598, + 644, + 1598, + 644, + 1629, + 553, + 1629 + ], + "score": 0.87, + "latex": "( 3 2 . 9 \\% )" + }, + { + "category_id": 13, + "poly": [ + 329, + 1658, + 403, + 1658, + 403, + 1687, + 329, + 1687 + ], + "score": 0.86, + "latex": "1 3 . 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1204, + 353, + 1294, + 353, + 1294, + 384, + 1204, + 384 + ], + "score": 0.86, + "latex": "( 1 3 . 7 \\% )" + }, + { + "category_id": 13, + "poly": [ + 495, + 414, + 535, + 414, + 535, + 442, + 495, + 442 + ], + "score": 0.86, + "latex": "9 \\%" + }, + { + "category_id": 13, + "poly": [ + 830, + 262, + 907, + 262, + 907, + 292, + 830, + 292 + ], + "score": 0.85, + "latex": "5 6 . 4 \\%" + }, + { + "category_id": 13, + "poly": [ + 707, + 262, + 783, + 262, + 783, + 292, + 707, + 292 + ], + "score": 0.83, + "latex": "7 3 . 2 \\%" + }, + { + "category_id": 13, + "poly": [ + 299, + 758, + 436, + 758, + 436, + 792, + 299, + 792 + ], + "score": 0.83, + "latex": "\\{ 1 5 , 2 5 , 3 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 298, + 885, + 342, + 885, + 342, + 915, + 298, + 915 + ], + "score": 0.8, + "latex": "( \\% )" + }, + { + "category_id": 13, + "poly": [ + 590, + 236, + 611, + 236, + 611, + 258, + 590, + 258 + ], + "score": 0.76, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 613, + 1481, + 630, + 1481, + 630, + 1504, + 613, + 1504 + ], + "score": 0.6, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 566, + 1286, + 643, + 1286, + 643, + 1312, + 566, + 1312 + ], + "score": 0.57, + "latex": "\\sigma = 2 5" + }, + { + "category_id": 13, + "poly": [ + 698, + 885, + 727, + 885, + 727, + 913, + 698, + 913 + ], + "score": 0.46, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 566, + 1185, + 643, + 1185, + 643, + 1209, + 566, + 1209 + ], + "score": 0.33, + "latex": "\\sigma = 3 5" + }, + { + "category_id": 5, + "poly": [ + 380, + 1031, + 1315, + 1031, + 1315, + 1380, + 380, + 1380 + ], + "score": 0.962, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
ImgNet10σ= 25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 853.0, + 1404.0, + 853.0, + 1404.0, + 886.0, + 296.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 880.0, + 297.0, + 880.0, + 297.0, + 919.0, + 293.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 880.0, + 697.0, + 880.0, + 697.0, + 919.0, + 343.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 880.0, + 1406.0, + 880.0, + 1406.0, + 919.0, + 728.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 913.0, + 1404.0, + 913.0, + 1404.0, + 947.0, + 294.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 944.0, + 1406.0, + 944.0, + 1406.0, + 977.0, + 296.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 976.0, + 600.0, + 976.0, + 600.0, + 1009.0, + 294.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1822.0, + 912.0, + 1822.0, + 912.0, + 1857.0, + 294.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 485.0, + 347.0, + 485.0, + 347.0, + 520.0, + 296.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 487.0, + 1281.0, + 487.0, + 1281.0, + 518.0, + 361.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 516.0, + 698.0, + 516.0, + 698.0, + 549.0, + 363.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1414.0, + 1404.0, + 1414.0, + 1404.0, + 1448.0, + 295.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1443.0, + 1406.0, + 1443.0, + 1406.0, + 1480.0, + 292.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1475.0, + 506.0, + 1475.0, + 506.0, + 1509.0, + 295.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1475.0, + 612.0, + 1475.0, + 612.0, + 1509.0, + 543.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 1475.0, + 1402.0, + 1475.0, + 1402.0, + 1509.0, + 631.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1505.0, + 1404.0, + 1505.0, + 1404.0, + 1539.0, + 294.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1534.0, + 1404.0, + 1534.0, + 1404.0, + 1571.0, + 294.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1566.0, + 1136.0, + 1566.0, + 1136.0, + 1600.0, + 294.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1566.0, + 1405.0, + 1566.0, + 1405.0, + 1600.0, + 1229.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1598.0, + 552.0, + 1598.0, + 552.0, + 1632.0, + 294.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1598.0, + 1405.0, + 1598.0, + 1405.0, + 1632.0, + 645.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1626.0, + 1404.0, + 1626.0, + 1404.0, + 1663.0, + 294.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1657.0, + 328.0, + 1657.0, + 328.0, + 1691.0, + 295.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1657.0, + 1404.0, + 1657.0, + 1404.0, + 1691.0, + 404.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1689.0, + 1406.0, + 1689.0, + 1406.0, + 1723.0, + 294.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1718.0, + 1406.0, + 1718.0, + 1406.0, + 1756.0, + 295.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1749.0, + 764.0, + 1749.0, + 764.0, + 1786.0, + 295.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 572.0, + 1407.0, + 572.0, + 1407.0, + 611.0, + 293.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 604.0, + 1405.0, + 604.0, + 1405.0, + 640.0, + 293.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 638.0, + 1405.0, + 638.0, + 1405.0, + 671.0, + 294.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 665.0, + 1405.0, + 665.0, + 1405.0, + 702.0, + 294.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 698.0, + 1403.0, + 698.0, + 1403.0, + 731.0, + 294.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 728.0, + 987.0, + 728.0, + 987.0, + 761.0, + 294.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 728.0, + 1403.0, + 728.0, + 1403.0, + 761.0, + 1140.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 758.0, + 747.0, + 758.0, + 747.0, + 794.0, + 437.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 758.0, + 1405.0, + 758.0, + 1405.0, + 794.0, + 886.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 789.0, + 607.0, + 789.0, + 607.0, + 822.0, + 294.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 229.0, + 589.0, + 229.0, + 589.0, + 266.0, + 292.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 229.0, + 1404.0, + 229.0, + 1404.0, + 266.0, + 612.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 260.0, + 706.0, + 260.0, + 706.0, + 297.0, + 292.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 260.0, + 829.0, + 260.0, + 829.0, + 297.0, + 784.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 260.0, + 1405.0, + 260.0, + 1405.0, + 297.0, + 908.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 291.0, + 1405.0, + 291.0, + 1405.0, + 325.0, + 294.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 317.0, + 1405.0, + 317.0, + 1405.0, + 360.0, + 292.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 350.0, + 511.0, + 350.0, + 511.0, + 387.0, + 293.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 350.0, + 1203.0, + 350.0, + 1203.0, + 387.0, + 588.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 350.0, + 1405.0, + 350.0, + 1405.0, + 387.0, + 1295.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 380.0, + 630.0, + 380.0, + 630.0, + 420.0, + 292.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 380.0, + 1405.0, + 380.0, + 1405.0, + 420.0, + 730.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 414.0, + 494.0, + 414.0, + 494.0, + 447.0, + 293.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 414.0, + 548.0, + 414.0, + 548.0, + 447.0, + 536.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1882.0, + 1404.0, + 1882.0, + 1404.0, + 1915.0, + 297.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1911.0, + 1403.0, + 1911.0, + 1403.0, + 1944.0, + 295.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1977.0, + 293.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2009.0, + 294.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 1404.0, + 2001.0, + 1404.0, + 2040.0, + 294.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 853.0, + 1404.0, + 853.0, + 1404.0, + 886.0, + 296.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 880.0, + 297.0, + 880.0, + 297.0, + 919.0, + 293.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 880.0, + 697.0, + 880.0, + 697.0, + 919.0, + 343.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 880.0, + 1406.0, + 880.0, + 1406.0, + 919.0, + 728.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 913.0, + 1404.0, + 913.0, + 1404.0, + 947.0, + 294.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 944.0, + 1406.0, + 944.0, + 1406.0, + 977.0, + 296.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 976.0, + 600.0, + 976.0, + 600.0, + 1009.0, + 294.0, + 1009.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 766, + 1469, + 1401, + 1469, + 1401, + 1864, + 766, + 1864 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 494, + 935, + 494, + 935, + 800, + 298, + 800 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1231, + 1404, + 1231, + 1404, + 1354, + 298, + 1354 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 969, + 486, + 1360, + 486, + 1360, + 775, + 969, + 775 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 338, + 1403, + 338, + 1403, + 464, + 298, + 464 + ], + "score": 0.97 + }, + { + "category_id": 3, + "poly": [ + 301, + 1477, + 711, + 1477, + 711, + 1800, + 301, + 1800 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 297, + 1835, + 739, + 1835, + 739, + 1991, + 297, + 1991 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 299, + 976, + 1405, + 976, + 1405, + 1098, + 299, + 1098 + ], + "score": 0.965 + }, + { + "category_id": 4, + "poly": [ + 956, + 805, + 1401, + 805, + 1401, + 936, + 956, + 936 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 1401, + 229, + 1401, + 323, + 297, + 323 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 298, + 814, + 936, + 814, + 936, + 975, + 298, + 975 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 765, + 1882, + 1400, + 1882, + 1400, + 2003, + 765, + 2003 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 285, + 2004, + 1401, + 2004, + 1401, + 2035, + 285, + 2035 + ], + "score": 0.91 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 816, + 76, + 816, + 104, + 300, + 104 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 298, + 1156, + 1168, + 1156, + 1168, + 1190, + 298, + 1190 + ], + "score": 0.889 + }, + { + "category_id": 0, + "poly": [ + 297, + 1406, + 861, + 1406, + 861, + 1437, + 297, + 1437 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.786 + }, + { + "category_id": 13, + "poly": [ + 684, + 1005, + 754, + 1005, + 754, + 1037, + 684, + 1037 + ], + "score": 0.94, + "latex": "z _ { 1 } ( T )" + }, + { + "category_id": 13, + "poly": [ + 852, + 1911, + 970, + 1911, + 970, + 1945, + 852, + 1945 + ], + "score": 0.93, + "latex": "f _ { \\theta } ( \\mathbf { z } ( t ) , t )" + }, + { + "category_id": 13, + "poly": [ + 405, + 585, + 591, + 585, + 591, + 619, + 405, + 619 + ], + "score": 0.93, + "latex": "\\bar { A _ { 1 } } = ( 0 , \\bar { z _ { 1 } } ( 0 ) )" + }, + { + "category_id": 13, + "poly": [ + 741, + 431, + 859, + 431, + 859, + 464, + 741, + 464 + ], + "score": 0.93, + "latex": "t \\in [ 0 , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 1165, + 1005, + 1235, + 1005, + 1235, + 1037, + 1165, + 1037 + ], + "score": 0.92, + "latex": "z _ { 2 } ( T )" + }, + { + "category_id": 13, + "poly": [ + 338, + 1957, + 426, + 1957, + 426, + 1991, + 338, + 1991 + ], + "score": 0.92, + "latex": "[ T ^ { \\prime } , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 458, + 1896, + 521, + 1896, + 521, + 1929, + 458, + 1929 + ], + "score": 0.92, + "latex": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 506, + 431, + 661, + 431, + 661, + 464, + 506, + 464 + ], + "score": 0.92, + "latex": "{ \\bf z } _ { 1 } ( t ) \\neq { \\bf z } _ { 2 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 681, + 585, + 746, + 585, + 746, + 619, + 681, + 619 + ], + "score": 0.92, + "latex": "z _ { 1 } ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 349, + 678, + 546, + 678, + 546, + 710, + 349, + 710 + ], + "score": 0.92, + "latex": "A _ { 3 } ~ = ~ ( 0 , z _ { 3 } ( 0 ) )" + }, + { + "category_id": 13, + "poly": [ + 718, + 768, + 776, + 768, + 776, + 802, + 718, + 802 + ], + "score": 0.92, + "latex": "z _ { 3 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1288, + 1005, + 1358, + 1005, + 1358, + 1037, + 1288, + 1037 + ], + "score": 0.92, + "latex": "z _ { 3 } ( T )" + }, + { + "category_id": 13, + "poly": [ + 531, + 1927, + 592, + 1927, + 592, + 1960, + 531, + 1960 + ], + "score": 0.92, + "latex": "{ \\bf z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 810, + 555, + 869, + 555, + 869, + 589, + 810, + 589 + ], + "score": 0.92, + "latex": "z _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 567, + 876, + 745, + 876, + 745, + 914, + 567, + 914 + ], + "score": 0.92, + "latex": "\\widetilde { A } _ { 1 } = ( 0 , \\widetilde { z } _ { 1 } ( 0 ) )" + }, + { + "category_id": 13, + "poly": [ + 297, + 973, + 374, + 973, + 374, + 1007, + 297, + 1007 + ], + "score": 0.91, + "latex": "z _ { 2 } ( T ) |" + }, + { + "category_id": 13, + "poly": [ + 621, + 738, + 681, + 738, + 681, + 770, + 621, + 770 + ], + "score": 0.91, + "latex": "z _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 596, + 369, + 657, + 369, + 657, + 401, + 596, + 401 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 607, + 770, + 666, + 770, + 666, + 802, + 607, + 802 + ], + "score": 0.91, + "latex": "z _ { 2 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 740, + 646, + 930, + 646, + 930, + 679, + 740, + 679 + ], + "score": 0.91, + "latex": "A _ { 2 } = ( 0 , z _ { 2 } ( 0 ) )" + }, + { + "category_id": 13, + "poly": [ + 297, + 646, + 357, + 646, + 357, + 679, + 297, + 679 + ], + "score": 0.91, + "latex": "z _ { 2 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 712, + 369, + 773, + 369, + 773, + 401, + 712, + 401 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 2 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 297, + 880, + 357, + 880, + 357, + 915, + 297, + 915 + ], + "score": 0.91, + "latex": "\\widetilde { z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 476, + 847, + 535, + 847, + 535, + 878, + 476, + 878 + ], + "score": 0.91, + "latex": "\\widetilde { z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1213, + 973, + 1277, + 973, + 1277, + 1006, + 1213, + 1006 + ], + "score": 0.91, + "latex": "z _ { 1 } ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 1266, + 1912, + 1315, + 1912, + 1315, + 1944, + 1266, + 1944 + ], + "score": 0.9, + "latex": "{ \\bf z } ( t )" + }, + { + "category_id": 13, + "poly": [ + 470, + 400, + 635, + 400, + 635, + 431, + 470, + 431 + ], + "score": 0.9, + "latex": "{ \\bf z } _ { 1 } ( 0 ) \\neq { \\bf z } _ { 2 } ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 671, + 913, + 887, + 913, + 887, + 945, + 671, + 945 + ], + "score": 0.9, + "latex": "| \\widetilde z _ { 1 } ( 0 ) - z _ { 1 } ( 0 ) | < \\epsilon ." + }, + { + "category_id": 13, + "poly": [ + 598, + 944, + 934, + 944, + 934, + 977, + 598, + 977 + ], + "score": 0.9, + "latex": "| \\tilde { z } _ { 1 } ( T ) - z _ { 1 } ( T ) | \\leq | z _ { 3 } ( T ) -" + }, + { + "category_id": 13, + "poly": [ + 1335, + 835, + 1372, + 835, + 1372, + 873, + 1335, + 873 + ], + "score": 0.9, + "latex": "\\widetilde { A _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 872, + 876, + 908, + 876, + 908, + 913, + 872, + 913 + ], + "score": 0.9, + "latex": "\\widetilde { A } _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 414, + 647, + 473, + 647, + 473, + 678, + 414, + 678 + ], + "score": 0.89, + "latex": "z _ { 3 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 400, + 710, + 436, + 710, + 436, + 738, + 400, + 738 + ], + "score": 0.88, + "latex": "A _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1324, + 904, + 1360, + 904, + 1360, + 935, + 1324, + 935 + ], + "score": 0.88, + "latex": "A _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1237, + 904, + 1273, + 904, + 1273, + 934, + 1237, + 934 + ], + "score": 0.88, + "latex": "A _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 575, + 915, + 610, + 915, + 610, + 943, + 575, + 943 + ], + "score": 0.88, + "latex": "A _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 571, + 709, + 608, + 709, + 608, + 738, + 571, + 738 + ], + "score": 0.88, + "latex": "A _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 889, + 679, + 925, + 679, + 925, + 708, + 889, + 708 + ], + "score": 0.87, + "latex": "A _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1314, + 975, + 1349, + 975, + 1349, + 1004, + 1314, + 1004 + ], + "score": 0.87, + "latex": "A _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 663, + 708, + 698, + 708, + 698, + 738, + 663, + 738 + ], + "score": 0.87, + "latex": "A _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 681, + 814, + 861, + 814, + 861, + 849, + 681, + 849 + ], + "score": 0.85, + "latex": "| z _ { 3 } ( 0 ) - z _ { 1 } ( 0 ) | \\}" + }, + { + "category_id": 13, + "poly": [ + 729, + 402, + 757, + 402, + 757, + 431, + 729, + 431 + ], + "score": 0.72, + "latex": "f _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 934, + 1947, + 948, + 1947, + 948, + 1970, + 934, + 1970 + ], + "score": 0.72, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 396, + 814, + 864, + 814, + 864, + 848, + 396, + 848 + ], + "score": 0.71, + "latex": "\\epsilon < \\operatorname* { m i n } \\{ | z _ { 2 } ( 0 ) - z _ { 1 } ( 0 ) | , | z _ { 3 } ( 0 ) - z _ { 1 } ( 0 ) | \\} ." + }, + { + "category_id": 13, + "poly": [ + 765, + 1945, + 780, + 1945, + 780, + 1970, + 765, + 1970 + ], + "score": 0.71, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 365, + 918, + 379, + 918, + 379, + 940, + 365, + 940 + ], + "score": 0.71, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 943, + 980, + 959, + 980, + 959, + 1001, + 943, + 1001 + ], + "score": 0.71, + "latex": "\\epsilon" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 491.0, + 1015.0, + 491.0, + 1015.0, + 512.0, + 982.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 494.0, + 1034.0, + 494.0, + 1034.0, + 508.0, + 1020.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 494.0, + 1047.0, + 494.0, + 1047.0, + 521.0, + 1029.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 520.0, + 1014.0, + 520.0, + 1014.0, + 542.0, + 981.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 548.0, + 1015.0, + 548.0, + 1015.0, + 571.0, + 981.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 573.0, + 1121.0, + 573.0, + 1121.0, + 581.0, + 1113.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 578.0, + 1015.0, + 578.0, + 1015.0, + 600.0, + 980.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 602.0, + 1017.0, + 602.0, + 1017.0, + 636.0, + 965.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 604.0, + 1048.0, + 604.0, + 1048.0, + 630.0, + 1018.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 636.0, + 1015.0, + 636.0, + 1015.0, + 659.0, + 981.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 629.0, + 1236.0, + 629.0, + 1236.0, + 658.0, + 1180.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 665.0, + 1015.0, + 665.0, + 1015.0, + 687.0, + 982.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 662.0, + 1048.0, + 662.0, + 1048.0, + 690.0, + 1017.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 654.0, + 1237.0, + 654.0, + 1237.0, + 683.0, + 1181.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 693.0, + 1015.0, + 693.0, + 1015.0, + 717.0, + 982.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 681.0, + 1237.0, + 681.0, + 1237.0, + 709.0, + 1180.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 704.0, + 1237.0, + 704.0, + 1237.0, + 736.0, + 1183.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 723.0, + 1014.0, + 723.0, + 1014.0, + 745.0, + 982.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 745.0, + 1044.0, + 745.0, + 1044.0, + 764.0, + 1018.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 746.0, + 1105.0, + 746.0, + 1105.0, + 765.0, + 1080.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 746.0, + 1169.0, + 746.0, + 1169.0, + 765.0, + 1143.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 745.0, + 1230.0, + 745.0, + 1230.0, + 765.0, + 1204.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 746.0, + 1292.0, + 746.0, + 1292.0, + 765.0, + 1267.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1331.0, + 748.0, + 1353.0, + 748.0, + 1353.0, + 763.0, + 1331.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 762.0, + 1192.0, + 762.0, + 1192.0, + 777.0, + 1180.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 722.0, + 1043.0, + 722.0, + 1043.0, + 745.0, + 1021.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 1490.0, + 718.0, + 1490.0, + 718.0, + 1520.0, + 441.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1528.0, + 356.0, + 1528.0, + 356.0, + 1560.0, + 300.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1555.0, + 354.0, + 1555.0, + 354.0, + 1584.0, + 298.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1704.0, + 356.0, + 1704.0, + 356.0, + 1736.0, + 300.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1710.0, + 574.0, + 1710.0, + 574.0, + 1740.0, + 521.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1745.0, + 358.0, + 1745.0, + 358.0, + 1781.0, + 299.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1733.0, + 576.0, + 1733.0, + 576.0, + 1767.0, + 524.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1777.0, + 366.0, + 1777.0, + 366.0, + 1801.0, + 346.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1782.0, + 589.0, + 1782.0, + 589.0, + 1795.0, + 576.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 1777.0, + 676.0, + 1777.0, + 676.0, + 1801.0, + 623.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.25, + 1780.0, + 435.25, + 1780.0, + 435.25, + 1792.0, + 411.25, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1834.0, + 745.0, + 1834.0, + 745.0, + 1868.0, + 295.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1865.0, + 744.0, + 1865.0, + 744.0, + 1899.0, + 296.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1894.0, + 457.0, + 1894.0, + 457.0, + 1930.0, + 294.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1894.0, + 743.0, + 1894.0, + 743.0, + 1930.0, + 522.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1925.0, + 530.0, + 1925.0, + 530.0, + 1961.0, + 293.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 1925.0, + 744.0, + 1925.0, + 744.0, + 1961.0, + 593.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1952.0, + 337.0, + 1952.0, + 337.0, + 1994.0, + 293.0, + 1994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 1952.0, + 439.0, + 1952.0, + 439.0, + 1994.0, + 427.0, + 1994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 803.0, + 1403.0, + 803.0, + 1403.0, + 840.0, + 955.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 839.0, + 1334.0, + 839.0, + 1334.0, + 876.0, + 953.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 839.0, + 1405.0, + 839.0, + 1405.0, + 876.0, + 1373.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 873.0, + 1403.0, + 873.0, + 1403.0, + 905.0, + 955.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 902.0, + 1236.0, + 902.0, + 1236.0, + 938.0, + 954.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1274.0, + 902.0, + 1323.0, + 902.0, + 1323.0, + 938.0, + 1274.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 902.0, + 1372.0, + 902.0, + 1372.0, + 938.0, + 1361.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1151.0, + 1171.0, + 1151.0, + 1171.0, + 1195.0, + 292.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1404.0, + 864.0, + 1404.0, + 864.0, + 1441.0, + 294.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1466.0, + 1403.0, + 1466.0, + 1403.0, + 1502.0, + 763.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1499.0, + 1404.0, + 1499.0, + 1404.0, + 1532.0, + 762.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1527.0, + 1404.0, + 1527.0, + 1404.0, + 1564.0, + 762.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1560.0, + 1403.0, + 1560.0, + 1403.0, + 1592.0, + 764.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1590.0, + 1405.0, + 1590.0, + 1405.0, + 1623.0, + 763.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1621.0, + 1404.0, + 1621.0, + 1404.0, + 1653.0, + 764.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1651.0, + 1404.0, + 1651.0, + 1404.0, + 1686.0, + 763.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1681.0, + 1405.0, + 1681.0, + 1405.0, + 1716.0, + 763.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1712.0, + 1404.0, + 1712.0, + 1404.0, + 1747.0, + 763.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1743.0, + 1404.0, + 1743.0, + 1404.0, + 1778.0, + 762.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1773.0, + 1403.0, + 1773.0, + 1403.0, + 1808.0, + 763.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1803.0, + 1404.0, + 1803.0, + 1404.0, + 1840.0, + 762.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1837.0, + 1024.0, + 1837.0, + 1024.0, + 1867.0, + 765.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 493.0, + 935.0, + 493.0, + 935.0, + 527.0, + 295.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 523.0, + 937.0, + 523.0, + 937.0, + 556.0, + 296.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 554.0, + 809.0, + 554.0, + 809.0, + 587.0, + 295.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 554.0, + 936.0, + 554.0, + 936.0, + 587.0, + 870.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 585.0, + 404.0, + 585.0, + 404.0, + 619.0, + 296.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 585.0, + 680.0, + 585.0, + 680.0, + 619.0, + 592.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 585.0, + 939.0, + 585.0, + 939.0, + 619.0, + 747.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 616.0, + 938.0, + 616.0, + 938.0, + 648.0, + 294.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 643.0, + 413.0, + 643.0, + 413.0, + 682.0, + 358.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 643.0, + 739.0, + 643.0, + 739.0, + 682.0, + 474.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 643.0, + 937.0, + 643.0, + 937.0, + 682.0, + 931.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 672.0, + 348.0, + 672.0, + 348.0, + 713.0, + 292.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 672.0, + 888.0, + 672.0, + 888.0, + 713.0, + 547.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 672.0, + 939.0, + 672.0, + 939.0, + 713.0, + 926.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 707.0, + 399.0, + 707.0, + 399.0, + 741.0, + 296.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 707.0, + 570.0, + 707.0, + 570.0, + 741.0, + 437.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 707.0, + 662.0, + 707.0, + 662.0, + 741.0, + 609.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 707.0, + 937.0, + 707.0, + 937.0, + 741.0, + 699.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 736.0, + 620.0, + 736.0, + 620.0, + 770.0, + 295.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 736.0, + 936.0, + 736.0, + 936.0, + 770.0, + 682.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 766.0, + 606.0, + 766.0, + 606.0, + 806.0, + 294.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 766.0, + 717.0, + 766.0, + 717.0, + 806.0, + 667.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 766.0, + 789.0, + 766.0, + 789.0, + 806.0, + 777.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1233.0, + 1405.0, + 1233.0, + 1405.0, + 1266.0, + 294.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1261.0, + 1406.0, + 1261.0, + 1406.0, + 1296.0, + 293.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1291.0, + 1405.0, + 1291.0, + 1405.0, + 1330.0, + 293.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1324.0, + 662.0, + 1324.0, + 662.0, + 1356.0, + 293.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 337.0, + 1404.0, + 337.0, + 1404.0, + 370.0, + 296.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 369.0, + 595.0, + 369.0, + 595.0, + 402.0, + 296.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 369.0, + 711.0, + 369.0, + 711.0, + 402.0, + 658.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 369.0, + 1405.0, + 369.0, + 1405.0, + 402.0, + 774.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 399.0, + 469.0, + 399.0, + 469.0, + 434.0, + 293.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 399.0, + 728.0, + 399.0, + 728.0, + 434.0, + 636.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 399.0, + 1405.0, + 399.0, + 1405.0, + 434.0, + 758.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 427.0, + 505.0, + 427.0, + 505.0, + 466.0, + 293.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 427.0, + 740.0, + 427.0, + 740.0, + 466.0, + 662.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 427.0, + 871.0, + 427.0, + 871.0, + 466.0, + 860.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 972.0, + 942.0, + 972.0, + 942.0, + 1008.0, + 375.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 972.0, + 1212.0, + 972.0, + 1212.0, + 1008.0, + 960.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 972.0, + 1313.0, + 972.0, + 1313.0, + 1008.0, + 1278.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 972.0, + 1405.0, + 972.0, + 1405.0, + 1008.0, + 1350.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1003.0, + 683.0, + 1003.0, + 683.0, + 1039.0, + 294.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1003.0, + 1164.0, + 1003.0, + 1164.0, + 1039.0, + 755.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1003.0, + 1287.0, + 1003.0, + 1287.0, + 1039.0, + 1236.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 1003.0, + 1405.0, + 1003.0, + 1405.0, + 1039.0, + 1359.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1035.0, + 1406.0, + 1035.0, + 1406.0, + 1071.0, + 294.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1066.0, + 1245.0, + 1066.0, + 1245.0, + 1101.0, + 294.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 230.0, + 1403.0, + 230.0, + 1403.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 261.0, + 1405.0, + 261.0, + 1405.0, + 295.0, + 295.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 289.0, + 1151.0, + 289.0, + 1151.0, + 329.0, + 292.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 812.0, + 395.0, + 812.0, + 395.0, + 851.0, + 294.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 812.0, + 937.0, + 812.0, + 937.0, + 851.0, + 865.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 844.0, + 475.0, + 844.0, + 475.0, + 878.0, + 295.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 844.0, + 936.0, + 844.0, + 936.0, + 878.0, + 536.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 877.0, + 296.0, + 877.0, + 296.0, + 917.0, + 293.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 877.0, + 566.0, + 877.0, + 566.0, + 917.0, + 358.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 877.0, + 871.0, + 877.0, + 871.0, + 917.0, + 746.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 877.0, + 939.0, + 877.0, + 939.0, + 917.0, + 909.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 909.0, + 364.0, + 909.0, + 364.0, + 947.0, + 294.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 909.0, + 574.0, + 909.0, + 574.0, + 947.0, + 380.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 909.0, + 670.0, + 909.0, + 670.0, + 947.0, + 611.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 909.0, + 936.0, + 909.0, + 936.0, + 947.0, + 888.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 939.0, + 597.0, + 939.0, + 597.0, + 979.0, + 293.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1878.0, + 1403.0, + 1878.0, + 1403.0, + 1916.0, + 762.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1911.0, + 851.0, + 1911.0, + 851.0, + 1944.0, + 762.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1911.0, + 1265.0, + 1911.0, + 1265.0, + 1944.0, + 971.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 1911.0, + 1404.0, + 1911.0, + 1404.0, + 1944.0, + 1316.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1942.0, + 933.0, + 1942.0, + 933.0, + 1973.0, + 781.0, + 1973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1973.0, + 949.0, + 1973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1970.0, + 1404.0, + 1970.0, + 1404.0, + 2007.0, + 762.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1998.0, + 1405.0, + 1998.0, + 1405.0, + 2040.0, + 295.0, + 2040.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1333, + 1405, + 1333, + 1405, + 1549, + 297, + 1549 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 459, + 1404, + 459, + 1404, + 585, + 296, + 585 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 737, + 1403, + 737, + 1403, + 862, + 299, + 862 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1787, + 1404, + 1787, + 1404, + 2035, + 297, + 2035 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 996, + 1404, + 996, + 1404, + 1154, + 298, + 1154 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 1650, + 1404, + 1650, + 1404, + 1774, + 298, + 1774 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 721, + 283, + 954, + 283, + 954, + 425, + 721, + 425 + ], + "score": 0.959 + }, + { + "category_id": 8, + "poly": [ + 660, + 1229, + 1038, + 1229, + 1038, + 1323, + 660, + 1323 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 298, + 660, + 1406, + 660, + 1406, + 725, + 298, + 725 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 737, + 603, + 960, + 603, + 960, + 642, + 737, + 642 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 298, + 1165, + 1398, + 1165, + 1398, + 1227, + 298, + 1227 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 297, + 227, + 1091, + 227, + 1091, + 264, + 297, + 264 + ], + "score": 0.924 + }, + { + "category_id": 8, + "poly": [ + 313, + 878, + 1388, + 878, + 1388, + 976, + 313, + 976 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 817, + 74, + 817, + 106, + 298, + 106 + ], + "score": 0.92 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1259, + 1400, + 1259, + 1400, + 1289, + 1366, + 1289 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1365, + 338, + 1401, + 338, + 1401, + 369, + 1365, + 369 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1366, + 607, + 1400, + 607, + 1400, + 638, + 1366, + 638 + ], + "score": 0.878 + }, + { + "category_id": 0, + "poly": [ + 300, + 1591, + 1076, + 1591, + 1076, + 1624, + 300, + 1624 + ], + "score": 0.863 + }, + { + "category_id": 9, + "poly": [ + 1366, + 972, + 1400, + 972, + 1400, + 998, + 1366, + 998 + ], + "score": 0.83 + }, + { + "category_id": 2, + "poly": [ + 842, + 2087, + 858, + 2087, + 858, + 2111, + 842, + 2111 + ], + "score": 0.655 + }, + { + "category_id": 2, + "poly": [ + 842, + 2087, + 859, + 2087, + 859, + 2111, + 842, + 2111 + ], + "score": 0.295 + }, + { + "category_id": 14, + "poly": [ + 660, + 1225, + 1038, + 1225, + 1038, + 1323, + 660, + 1323 + ], + "score": 0.94, + "latex": "L _ { \\mathrm { s s } } = \\sum _ { i = 1 } ^ { N } \\left\\| \\int _ { T } ^ { 2 T } | f _ { \\theta } ( \\mathbf { z } _ { i } ( t ) ) | \\mathrm { d } t \\right\\| ," + }, + { + "category_id": 14, + "poly": [ + 319, + 879, + 1375, + 879, + 1375, + 975, + 319, + 975 + ], + "score": 0.94, + "latex": "\\| \\widetilde { \\mathbf z } _ { 1 } ( T ) - \\mathbf z _ { 1 } ( T ) \\| = \\left\\| \\int _ { T } ^ { T + T ^ { \\prime } } f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) \\mathrm { d } t \\right\\| \\leq \\left\\| \\int _ { T } ^ { T + T ^ { \\prime } } | f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) | \\mathrm { d } t \\right\\| \\leq \\left\\| \\int _ { T } ^ { 2 T } | f _ { \\theta } ( \\mathbf z _ { 1 } ( t ) ) | \\mathrm { d } t \\right\\| ," + }, + { + "category_id": 13, + "poly": [ + 1183, + 521, + 1394, + 521, + 1394, + 555, + 1183, + 555 + ], + "score": 0.94, + "latex": "( { \\bf z } _ { 1 } ( T ^ { \\prime } ) , T ^ { \\prime } ) \\in \\mathbb { M } _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1109, + 799, + 1235, + 799, + 1235, + 833, + 1109, + 833 + ], + "score": 0.93, + "latex": "T ^ { \\prime } \\in [ 0 , T ]" + }, + { + "category_id": 13, + "poly": [ + 422, + 662, + 496, + 662, + 496, + 695, + 422, + 695 + ], + "score": 0.93, + "latex": "[ 0 , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 926, + 738, + 990, + 738, + 990, + 771, + 926, + 771 + ], + "score": 0.93, + "latex": "{ \\bf z } _ { 1 } ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 841, + 1060, + 956, + 1060, + 956, + 1093, + 841, + 1093 + ], + "score": 0.93, + "latex": "t ^ { \\prime } \\in [ 0 , T ]" + }, + { + "category_id": 13, + "poly": [ + 1102, + 1059, + 1293, + 1059, + 1293, + 1093, + 1102, + 1093 + ], + "score": 0.92, + "latex": "( { \\bf z } _ { 1 } ( t ^ { \\prime } ) , t ^ { \\prime } ) \\in \\mathbb { M } _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 397, + 831, + 467, + 831, + 467, + 863, + 397, + 863 + ], + "score": 0.92, + "latex": "{ \\bf z } _ { 1 } ( T )" + }, + { + "category_id": 14, + "poly": [ + 716, + 281, + 954, + 281, + 954, + 427, + 716, + 427 + ], + "score": 0.92, + "latex": "\\left\\{ \\begin{array} { l l } { \\displaystyle \\frac { \\mathrm { d } \\mathbf { z } ( t ) } { \\mathrm { d } t } = f _ { \\boldsymbol { \\theta } } ( \\mathbf { z } ( t ) ) ; } \\\\ { \\mathbf { z } ( 0 ) = \\mathbf { z } _ { \\mathrm { i n } } ; } \\\\ { \\mathbf { z } _ { \\mathrm { o u t } } = \\mathbf { z } ( T ) . } \\end{array} \\right." + }, + { + "category_id": 13, + "poly": [ + 960, + 1335, + 1018, + 1335, + 1018, + 1368, + 960, + 1368 + ], + "score": 0.92, + "latex": "{ \\bf z } _ { i } ( t )" + }, + { + "category_id": 13, + "poly": [ + 651, + 692, + 711, + 692, + 711, + 725, + 651, + 725 + ], + "score": 0.92, + "latex": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 481, + 1487, + 549, + 1487, + 549, + 1521, + 481, + 1521 + ], + "score": 0.92, + "latex": "{ \\bf z } _ { i } ( T )" + }, + { + "category_id": 13, + "poly": [ + 580, + 230, + 644, + 230, + 644, + 264, + 580, + 264 + ], + "score": 0.92, + "latex": "f _ { \\boldsymbol { \\theta } } ( \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 936, + 1029, + 1007, + 1029, + 1007, + 1061, + 936, + 1061 + ], + "score": 0.92, + "latex": "\\widetilde { \\mathbf z } _ { 1 } ( T )" + }, + { + "category_id": 13, + "poly": [ + 331, + 553, + 392, + 553, + 392, + 585, + 331, + 585 + ], + "score": 0.92, + "latex": "\\widetilde { \\mathbf { z } } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 837, + 553, + 1013, + 553, + 1013, + 586, + 837, + 586 + ], + "score": 0.92, + "latex": "\\widetilde { \\mathbf z } _ { 1 } ( 0 ) = \\mathbf z _ { 1 } ( T ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 925, + 691, + 985, + 691, + 985, + 725, + 925, + 725 + ], + "score": 0.92, + "latex": "{ \\bf z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1058, + 1029, + 1129, + 1029, + 1129, + 1061, + 1058, + 1061 + ], + "score": 0.92, + "latex": "{ \\bf z } _ { 1 } ( T )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1120, + 368, + 1120, + 368, + 1154, + 297, + 1154 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 1 } ( T )" + }, + { + "category_id": 14, + "poly": [ + 737, + 602, + 961, + 602, + 961, + 641, + 737, + 641 + ], + "score": 0.91, + "latex": "\\widetilde { { \\mathbf z } } _ { 1 } ( t ) = { \\mathbf z } _ { 1 } ( t + T ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 677, + 1092, + 745, + 1092, + 745, + 1123, + 677, + 1123 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 1 } ( t ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 902, + 522, + 966, + 522, + 966, + 553, + 902, + 553 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 1 } ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 558, + 769, + 629, + 769, + 629, + 801, + 558, + 801 + ], + "score": 0.91, + "latex": "\\widetilde { \\mathbf z } _ { 1 } ( T )" + }, + { + "category_id": 13, + "poly": [ + 518, + 831, + 589, + 831, + 589, + 863, + 518, + 863 + ], + "score": 0.91, + "latex": "\\widetilde { \\mathbf z } _ { 1 } ( T )" + }, + { + "category_id": 13, + "poly": [ + 470, + 738, + 534, + 738, + 534, + 771, + 470, + 771 + ], + "score": 0.91, + "latex": "\\widetilde { \\mathbf { z } } _ { 1 } ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 543, + 800, + 1028, + 800, + 1028, + 833, + 543, + 833 + ], + "score": 0.91, + "latex": "\\| \\mathbf { \\tilde { z } } _ { 1 } ( T ) - \\mathbf { z } _ { 1 } ( T ) \\| = \\| \\mathbf { z } _ { 1 } ( T + T ^ { \\prime } ) - \\mathbf { z } _ { 1 } ( T ) \\|" + }, + { + "category_id": 13, + "poly": [ + 343, + 461, + 404, + 461, + 404, + 493, + 343, + 493 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1339, + 491, + 1400, + 491, + 1400, + 523, + 1339, + 523 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 683, + 769, + 754, + 769, + 754, + 801, + 683, + 801 + ], + "score": 0.91, + "latex": "{ \\bf z } _ { 1 } ( T )" + }, + { + "category_id": 13, + "poly": [ + 700, + 999, + 745, + 999, + 745, + 1031, + 700, + 1031 + ], + "score": 0.9, + "latex": "| f _ { \\theta } |" + }, + { + "category_id": 13, + "poly": [ + 652, + 1458, + 693, + 1458, + 693, + 1488, + 652, + 1488 + ], + "score": 0.9, + "latex": "L _ { \\mathrm { s s } }" + }, + { + "category_id": 13, + "poly": [ + 782, + 692, + 834, + 692, + 834, + 720, + 782, + 720 + ], + "score": 0.9, + "latex": "- T ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 677, + 1060, + 738, + 1060, + 738, + 1091, + 677, + 1091 + ], + "score": 0.9, + "latex": "{ \\bf z } _ { 1 } ( t )" + }, + { + "category_id": 13, + "poly": [ + 403, + 1743, + 445, + 1743, + 445, + 1773, + 403, + 1773 + ], + "score": 0.9, + "latex": "L _ { \\mathrm { s s } }" + }, + { + "category_id": 13, + "poly": [ + 857, + 1169, + 898, + 1169, + 898, + 1198, + 857, + 1198 + ], + "score": 0.9, + "latex": "L _ { \\mathrm { s s } }" + }, + { + "category_id": 13, + "poly": [ + 681, + 462, + 755, + 462, + 755, + 493, + 681, + 493 + ], + "score": 0.9, + "latex": "[ 0 , \\infty )" + }, + { + "category_id": 13, + "poly": [ + 377, + 524, + 439, + 524, + 439, + 554, + 377, + 554 + ], + "score": 0.89, + "latex": "[ 0 , T ]" + }, + { + "category_id": 13, + "poly": [ + 811, + 464, + 888, + 464, + 888, + 490, + 811, + 490 + ], + "score": 0.89, + "latex": "\\epsilon > 0" + }, + { + "category_id": 13, + "poly": [ + 1030, + 1367, + 1072, + 1367, + 1072, + 1396, + 1030, + 1396 + ], + "score": 0.89, + "latex": "L _ { \\mathrm { s s } }" + }, + { + "category_id": 13, + "poly": [ + 1312, + 1457, + 1351, + 1457, + 1351, + 1487, + 1312, + 1487 + ], + "score": 0.88, + "latex": "\\mathbb { M } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 536, + 1000, + 563, + 1000, + 563, + 1029, + 536, + 1029 + ], + "score": 0.88, + "latex": "\\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 477, + 1030, + 505, + 1030, + 505, + 1061, + 477, + 1061 + ], + "score": 0.87, + "latex": "f _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 613, + 1364, + 648, + 1364, + 648, + 1393, + 613, + 1393 + ], + "score": 0.86, + "latex": "i ^ { \\mathrm { t h } }" + }, + { + "category_id": 13, + "poly": [ + 618, + 492, + 857, + 492, + 857, + 525, + 618, + 525 + ], + "score": 0.82, + "latex": "\\| \\mathbf { z } _ { 1 } ( t ) - \\mathbf { z } _ { 1 } ( 0 ) \\| \\leq \\epsilon \\}" + }, + { + "category_id": 13, + "poly": [ + 374, + 1337, + 403, + 1337, + 403, + 1363, + 374, + 1363 + ], + "score": 0.81, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 374, + 664, + 387, + 664, + 387, + 689, + 374, + 689 + ], + "score": 0.76, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 693, + 529, + 707, + 529, + 707, + 549, + 693, + 549 + ], + "score": 0.7, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 297, + 491, + 606, + 491, + 606, + 525, + 297, + 525 + ], + "score": 0.6, + "latex": "\\mathbb { M } _ { 1 } = \\{ ( \\mathbf { z } _ { 1 } ( t ) , t ) | t \\in [ 0 , T ]" + }, + { + "category_id": 13, + "poly": [ + 713, + 1519, + 773, + 1519, + 773, + 1546, + 713, + 1546 + ], + "score": 0.36, + "latex": "O D E" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 109.0, + 295.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1592.0, + 1079.0, + 1592.0, + 1079.0, + 1627.0, + 295.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2086.0, + 861.0, + 2086.0, + 861.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1335.0, + 373.0, + 1335.0, + 373.0, + 1369.0, + 296.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1335.0, + 959.0, + 1335.0, + 959.0, + 1369.0, + 404.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 1335.0, + 1406.0, + 1335.0, + 1406.0, + 1369.0, + 1019.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1363.0, + 612.0, + 1363.0, + 612.0, + 1401.0, + 294.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 1363.0, + 1029.0, + 1363.0, + 1029.0, + 1401.0, + 649.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1363.0, + 1407.0, + 1363.0, + 1407.0, + 1401.0, + 1073.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1397.0, + 1406.0, + 1397.0, + 1406.0, + 1431.0, + 294.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1429.0, + 1402.0, + 1429.0, + 1402.0, + 1460.0, + 296.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1457.0, + 651.0, + 1457.0, + 651.0, + 1492.0, + 294.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1457.0, + 1311.0, + 1457.0, + 1311.0, + 1492.0, + 694.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1457.0, + 1405.0, + 1457.0, + 1405.0, + 1492.0, + 1352.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1488.0, + 480.0, + 1488.0, + 480.0, + 1522.0, + 294.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 1488.0, + 1406.0, + 1488.0, + 1406.0, + 1522.0, + 550.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1519.0, + 712.0, + 1519.0, + 712.0, + 1550.0, + 296.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 1519.0, + 782.0, + 1519.0, + 782.0, + 1550.0, + 774.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 457.0, + 342.0, + 457.0, + 342.0, + 496.0, + 291.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 457.0, + 680.0, + 457.0, + 680.0, + 496.0, + 405.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 457.0, + 810.0, + 457.0, + 810.0, + 496.0, + 756.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 457.0, + 1407.0, + 457.0, + 1407.0, + 496.0, + 889.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 487.0, + 296.0, + 487.0, + 296.0, + 528.0, + 291.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 487.0, + 617.0, + 487.0, + 617.0, + 528.0, + 607.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 487.0, + 1338.0, + 487.0, + 1338.0, + 528.0, + 858.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 487.0, + 1404.0, + 487.0, + 1404.0, + 528.0, + 1401.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 518.0, + 376.0, + 518.0, + 376.0, + 557.0, + 293.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 518.0, + 692.0, + 518.0, + 692.0, + 557.0, + 440.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 518.0, + 901.0, + 518.0, + 901.0, + 557.0, + 708.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 518.0, + 1182.0, + 518.0, + 1182.0, + 557.0, + 967.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 518.0, + 1405.0, + 518.0, + 1405.0, + 557.0, + 1395.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 550.0, + 330.0, + 550.0, + 330.0, + 588.0, + 293.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 550.0, + 836.0, + 550.0, + 836.0, + 588.0, + 393.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 550.0, + 1190.0, + 550.0, + 1190.0, + 588.0, + 1014.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 735.0, + 469.0, + 735.0, + 469.0, + 774.0, + 294.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 735.0, + 925.0, + 735.0, + 925.0, + 774.0, + 535.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 735.0, + 1407.0, + 735.0, + 1407.0, + 774.0, + 991.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 767.0, + 557.0, + 767.0, + 557.0, + 803.0, + 294.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 767.0, + 682.0, + 767.0, + 682.0, + 803.0, + 630.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 767.0, + 1404.0, + 767.0, + 1404.0, + 803.0, + 755.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 798.0, + 542.0, + 798.0, + 542.0, + 835.0, + 293.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 798.0, + 1108.0, + 798.0, + 1108.0, + 835.0, + 1029.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 798.0, + 1407.0, + 798.0, + 1407.0, + 835.0, + 1236.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 826.0, + 396.0, + 826.0, + 396.0, + 865.0, + 293.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 826.0, + 517.0, + 826.0, + 517.0, + 865.0, + 468.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 826.0, + 902.0, + 826.0, + 902.0, + 865.0, + 590.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1788.0, + 1404.0, + 1788.0, + 1404.0, + 1824.0, + 294.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1820.0, + 1404.0, + 1820.0, + 1404.0, + 1854.0, + 294.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1849.0, + 1405.0, + 1849.0, + 1405.0, + 1886.0, + 292.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1878.0, + 1408.0, + 1878.0, + 1408.0, + 1918.0, + 291.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1406.0, + 1911.0, + 1406.0, + 1945.0, + 294.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 1406.0, + 1973.0, + 1406.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2004.0, + 1053.0, + 2004.0, + 1053.0, + 2038.0, + 295.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 994.0, + 535.0, + 994.0, + 535.0, + 1033.0, + 295.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 994.0, + 699.0, + 994.0, + 699.0, + 1033.0, + 564.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 994.0, + 1405.0, + 994.0, + 1405.0, + 1033.0, + 746.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1025.0, + 476.0, + 1025.0, + 476.0, + 1064.0, + 295.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1025.0, + 935.0, + 1025.0, + 935.0, + 1064.0, + 506.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 1025.0, + 1057.0, + 1025.0, + 1057.0, + 1064.0, + 1008.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1025.0, + 1404.0, + 1025.0, + 1404.0, + 1064.0, + 1130.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1058.0, + 676.0, + 1058.0, + 676.0, + 1094.0, + 293.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1058.0, + 840.0, + 1058.0, + 840.0, + 1094.0, + 739.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1058.0, + 1101.0, + 1058.0, + 1101.0, + 1094.0, + 957.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 1058.0, + 1406.0, + 1058.0, + 1406.0, + 1094.0, + 1294.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1087.0, + 676.0, + 1087.0, + 676.0, + 1125.0, + 294.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1087.0, + 1406.0, + 1087.0, + 1406.0, + 1125.0, + 746.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1118.0, + 629.0, + 1118.0, + 629.0, + 1156.0, + 369.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1650.0, + 1404.0, + 1650.0, + 1404.0, + 1685.0, + 293.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1682.0, + 1405.0, + 1682.0, + 1405.0, + 1714.0, + 296.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1714.0, + 1402.0, + 1714.0, + 1402.0, + 1747.0, + 296.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1741.0, + 402.0, + 1741.0, + 402.0, + 1776.0, + 293.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1741.0, + 1402.0, + 1741.0, + 1402.0, + 1776.0, + 446.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 660.0, + 373.0, + 660.0, + 373.0, + 696.0, + 295.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 660.0, + 421.0, + 660.0, + 421.0, + 696.0, + 388.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 660.0, + 1406.0, + 660.0, + 1406.0, + 696.0, + 497.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 687.0, + 650.0, + 687.0, + 650.0, + 727.0, + 293.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 687.0, + 781.0, + 687.0, + 781.0, + 727.0, + 712.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 687.0, + 924.0, + 687.0, + 924.0, + 727.0, + 835.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 687.0, + 1114.0, + 687.0, + 1114.0, + 727.0, + 986.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1163.0, + 856.0, + 1163.0, + 856.0, + 1204.0, + 293.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1163.0, + 1403.0, + 1163.0, + 1403.0, + 1204.0, + 899.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1195.0, + 549.0, + 1195.0, + 549.0, + 1230.0, + 294.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 227.0, + 579.0, + 227.0, + 579.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 227.0, + 1095.0, + 227.0, + 1095.0, + 267.0, + 645.0, + 267.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 363, + 1374, + 1330, + 1374, + 1330, + 1787, + 363, + 1787 + ], + "score": 0.985, + "html": "
Gaussian noiseAdversarial attack
MNISTσ= 100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
CNN-FDn TisODE-FDn99.0±0.174.0±4.132.6±5.358.9±4.08.2±2.6
CNN-IRd99.4±0.0 95.3±0.980.6±2.3 78.1±2.240.4±5.7 36.7±2.172.6±2.4 79.6±1.928.2±3.6 55.5±2.9
TisODE-IRd97.6±0.186.8±2.349.1±0.288.8±0.966.0±0.9
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN
90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
CNN-FDn92.4±0.143.8±1.431.5±3.040.0±2.619.6±3.4
TisODE-FDn95.2±0.157.8±1.748.2±2.053.4±2.932.3±1.0
CNN-IRd84.9±1.265.8±0.454.7±1.274.0±0.564.5±0.8
TisODE-IRd91.7±0.574.4±1.261.9±1.881.6±0.871.0±0.5
" + }, + { + "category_id": 5, + "poly": [ + 379, + 372, + 1316, + 372, + 1316, + 802, + 379, + 802 + ], + "score": 0.984, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.2PGD-0.3
CNN98.7±0.154.2±1.115.8±1.332.9±3.70.0±0.0
ODENet99.4±0.171.5±1.119.9±1.264.7±1.813.0±0.2
TisODE99.6±0.075.7±1.426.5±3.867.4±1.513.2±1.0
SVHNσ=35FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN90.6±0.225.3±0.612.3±0.732.4±0.414.0±0.5
ODENet95.1±0.149.4±1.034.7±0.550.9±1.327.2±1.4
TisODE94.9±0.151.6±1.238.2±1.952.0±0.928.2±0.3
ImgNet10g=25FGSM-5/255FGSM-8/255PGD-3/255PGD-5/255
CNN92.6±0.640.9±1.826.7±1.728.6±1.511.2±1.2
ODENet92.6±0.542.0±0.429.0±1.029.8±0.412.3±0.6
TisODE92.8±0.444.3±0.731.4±1.131.1±1.214.5±1.1
" + }, + { + "category_id": 1, + "poly": [ + 296, + 918, + 1405, + 918, + 1405, + 1195, + 296, + 1195 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 296, + 1819, + 1404, + 1819, + 1404, + 2035, + 296, + 2035 + ], + "score": 0.959 + }, + { + "category_id": 6, + "poly": [ + 297, + 223, + 1403, + 223, + 1403, + 348, + 297, + 348 + ], + "score": 0.918 + }, + { + "category_id": 6, + "poly": [ + 298, + 1224, + 1404, + 1224, + 1404, + 1348, + 298, + 1348 + ], + "score": 0.91 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.801 + }, + { + "category_id": 7, + "poly": [ + 294, + 834, + 1322, + 834, + 1322, + 894, + 294, + 894 + ], + "score": 0.334 + }, + { + "category_id": 1, + "poly": [ + 298, + 1224, + 1404, + 1224, + 1404, + 1348, + 298, + 1348 + ], + "score": 0.142 + }, + { + "category_id": 13, + "poly": [ + 715, + 1973, + 792, + 1973, + 792, + 2003, + 715, + 2003 + ], + "score": 0.86, + "latex": "6 6 . 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 306, + 1973, + 368, + 1973, + 368, + 2002, + 306, + 2002 + ], + "score": 0.85, + "latex": "8 . 2 \\%" + }, + { + "category_id": 13, + "poly": [ + 403, + 1973, + 484, + 1973, + 484, + 2003, + 403, + 2003 + ], + "score": 0.84, + "latex": "2 8 . 2 \\% )" + }, + { + "category_id": 13, + "poly": [ + 602, + 1973, + 679, + 1973, + 679, + 2003, + 602, + 2003 + ], + "score": 0.82, + "latex": "5 5 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 837, + 1228, + 863, + 1228, + 863, + 1254, + 837, + 1254 + ], + "score": 0.79, + "latex": "\\%" + }, + { + "category_id": 13, + "poly": [ + 838, + 226, + 864, + 226, + 864, + 253, + 838, + 253 + ], + "score": 0.77, + "latex": "\\%" + }, + { + "category_id": 13, + "poly": [ + 408, + 1292, + 428, + 1292, + 428, + 1315, + 408, + 1315 + ], + "score": 0.73, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 583, + 1616, + 659, + 1616, + 659, + 1635, + 583, + 1635 + ], + "score": 0.7, + "latex": "\\sigma = 3 5" + }, + { + "category_id": 13, + "poly": [ + 738, + 226, + 766, + 226, + 766, + 254, + 738, + 254 + ], + "score": 0.69, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 737, + 1228, + 766, + 1228, + 766, + 1255, + 737, + 1255 + ], + "score": 0.66, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 884, + 291, + 904, + 291, + 904, + 313, + 884, + 313 + ], + "score": 0.64, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 566, + 553, + 642, + 553, + 642, + 578, + 566, + 578 + ], + "score": 0.44, + "latex": "\\sigma = 3 5" + }, + { + "category_id": 13, + "poly": [ + 566, + 689, + 642, + 689, + 642, + 707, + 566, + 707 + ], + "score": 0.44, + "latex": "\\sigma = 2 5" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 223.0, + 737.0, + 223.0, + 737.0, + 260.0, + 293.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 223.0, + 837.0, + 223.0, + 837.0, + 260.0, + 767.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 223.0, + 1407.0, + 223.0, + 1407.0, + 260.0, + 865.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 254.0, + 1403.0, + 254.0, + 1403.0, + 289.0, + 294.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 288.0, + 883.0, + 288.0, + 883.0, + 317.0, + 295.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 288.0, + 1404.0, + 288.0, + 1404.0, + 317.0, + 905.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 316.0, + 1316.0, + 316.0, + 1316.0, + 350.0, + 292.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1224.0, + 736.0, + 1224.0, + 736.0, + 1260.0, + 293.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1224.0, + 836.0, + 1224.0, + 836.0, + 1260.0, + 767.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 1224.0, + 1404.0, + 1224.0, + 1404.0, + 1260.0, + 864.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1254.0, + 1406.0, + 1254.0, + 1406.0, + 1291.0, + 294.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1287.0, + 407.0, + 1287.0, + 407.0, + 1320.0, + 294.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1287.0, + 1404.0, + 1287.0, + 1404.0, + 1320.0, + 429.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1316.0, + 960.0, + 1316.0, + 960.0, + 1351.0, + 293.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 829.0, + 1327.0, + 829.0, + 1327.0, + 868.0, + 293.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 868.0, + 769.0, + 868.0, + 769.0, + 895.0, + 363.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 920.0, + 1405.0, + 920.0, + 1405.0, + 953.0, + 295.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 950.0, + 1405.0, + 950.0, + 1405.0, + 986.0, + 293.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 981.0, + 1405.0, + 981.0, + 1405.0, + 1013.0, + 295.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1012.0, + 1403.0, + 1012.0, + 1403.0, + 1045.0, + 295.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1039.0, + 1405.0, + 1039.0, + 1405.0, + 1077.0, + 294.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1070.0, + 1406.0, + 1070.0, + 1406.0, + 1109.0, + 294.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1101.0, + 1405.0, + 1101.0, + 1405.0, + 1139.0, + 294.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1131.0, + 1405.0, + 1131.0, + 1405.0, + 1169.0, + 293.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1161.0, + 1318.0, + 1161.0, + 1318.0, + 1199.0, + 293.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1855.0, + 294.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1849.0, + 1405.0, + 1849.0, + 1405.0, + 1887.0, + 293.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1883.0, + 1404.0, + 1883.0, + 1404.0, + 1913.0, + 296.0, + 1913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1911.0, + 1405.0, + 1911.0, + 1405.0, + 1946.0, + 291.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1944.0, + 1404.0, + 1944.0, + 1404.0, + 1975.0, + 295.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 305.0, + 1973.0, + 305.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1973.0, + 402.0, + 1973.0, + 402.0, + 2007.0, + 369.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 1973.0, + 601.0, + 1973.0, + 601.0, + 2007.0, + 485.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1973.0, + 714.0, + 1973.0, + 714.0, + 2007.0, + 680.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2007.0, + 793.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 2000.0, + 1405.0, + 2000.0, + 1405.0, + 2041.0, + 291.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1224.0, + 736.0, + 1224.0, + 736.0, + 1260.0, + 293.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1224.0, + 836.0, + 1224.0, + 836.0, + 1260.0, + 767.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 1224.0, + 1404.0, + 1224.0, + 1404.0, + 1260.0, + 864.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1254.0, + 1406.0, + 1254.0, + 1406.0, + 1291.0, + 294.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1287.0, + 407.0, + 1287.0, + 407.0, + 1320.0, + 294.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1287.0, + 1404.0, + 1287.0, + 1404.0, + 1320.0, + 429.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1316.0, + 960.0, + 1316.0, + 960.0, + 1351.0, + 293.0, + 1351.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 892, + 1404, + 892, + 1404, + 1437, + 298, + 1437 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1452, + 1404, + 1452, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 568, + 1403, + 568, + 1403, + 874, + 298, + 874 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 490, + 1399, + 490, + 1399, + 553, + 298, + 553 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 294, + 229, + 1402, + 229, + 1402, + 293, + 294, + 293 + ], + "score": 0.948 + }, + { + "category_id": 0, + "poly": [ + 300, + 390, + 597, + 390, + 597, + 425, + 300, + 425 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 105, + 299, + 105 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.793 + }, + { + "category_id": 15, + "poly": [ + 293.0, + 390.0, + 603.0, + 390.0, + 603.0, + 430.0, + 293.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 890.0, + 1402.0, + 890.0, + 1402.0, + 925.0, + 294.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 923.0, + 1404.0, + 923.0, + 1404.0, + 954.0, + 296.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 952.0, + 1401.0, + 952.0, + 1401.0, + 983.0, + 296.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 982.0, + 1404.0, + 982.0, + 1404.0, + 1015.0, + 293.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1012.0, + 1405.0, + 1012.0, + 1405.0, + 1047.0, + 296.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1041.0, + 1404.0, + 1041.0, + 1404.0, + 1076.0, + 294.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1071.0, + 1405.0, + 1071.0, + 1405.0, + 1109.0, + 292.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1104.0, + 1404.0, + 1104.0, + 1404.0, + 1139.0, + 294.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1132.0, + 1405.0, + 1132.0, + 1405.0, + 1171.0, + 293.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1163.0, + 1405.0, + 1163.0, + 1405.0, + 1199.0, + 293.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1195.0, + 1405.0, + 1195.0, + 1405.0, + 1230.0, + 294.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1225.0, + 1404.0, + 1225.0, + 1404.0, + 1259.0, + 296.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1255.0, + 1405.0, + 1255.0, + 1405.0, + 1290.0, + 294.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1288.0, + 1405.0, + 1288.0, + 1405.0, + 1319.0, + 296.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1318.0, + 1405.0, + 1318.0, + 1405.0, + 1350.0, + 296.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1346.0, + 1404.0, + 1346.0, + 1404.0, + 1382.0, + 293.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1377.0, + 1406.0, + 1377.0, + 1406.0, + 1413.0, + 294.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1408.0, + 485.0, + 1408.0, + 485.0, + 1439.0, + 294.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1453.0, + 1404.0, + 1453.0, + 1404.0, + 1489.0, + 294.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1486.0, + 1404.0, + 1486.0, + 1404.0, + 1517.0, + 296.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1512.0, + 1404.0, + 1512.0, + 1404.0, + 1550.0, + 292.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1546.0, + 1404.0, + 1546.0, + 1404.0, + 1579.0, + 292.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1574.0, + 1404.0, + 1574.0, + 1404.0, + 1610.0, + 295.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1609.0, + 1404.0, + 1609.0, + 1404.0, + 1640.0, + 296.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1635.0, + 1404.0, + 1635.0, + 1404.0, + 1672.0, + 292.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1667.0, + 1407.0, + 1667.0, + 1407.0, + 1703.0, + 292.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1698.0, + 1406.0, + 1698.0, + 1406.0, + 1732.0, + 292.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1729.0, + 1405.0, + 1729.0, + 1405.0, + 1761.0, + 296.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1758.0, + 1404.0, + 1758.0, + 1404.0, + 1793.0, + 294.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1788.0, + 1405.0, + 1788.0, + 1405.0, + 1824.0, + 292.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1816.0, + 1408.0, + 1816.0, + 1408.0, + 1858.0, + 292.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1848.0, + 1405.0, + 1848.0, + 1405.0, + 1885.0, + 293.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1881.0, + 1404.0, + 1881.0, + 1404.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1914.0, + 1402.0, + 1914.0, + 1402.0, + 1945.0, + 296.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1944.0, + 1404.0, + 1944.0, + 1404.0, + 1975.0, + 296.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1974.0, + 1404.0, + 1974.0, + 1404.0, + 2005.0, + 296.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2001.0, + 928.0, + 2001.0, + 928.0, + 2035.0, + 296.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 567.0, + 1404.0, + 567.0, + 1404.0, + 603.0, + 294.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 598.0, + 1405.0, + 598.0, + 1405.0, + 637.0, + 293.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 625.0, + 1407.0, + 625.0, + 1407.0, + 669.0, + 292.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 660.0, + 1405.0, + 660.0, + 1405.0, + 694.0, + 293.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 693.0, + 1404.0, + 693.0, + 1404.0, + 725.0, + 296.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 722.0, + 1404.0, + 722.0, + 1404.0, + 754.0, + 296.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 750.0, + 1405.0, + 750.0, + 1405.0, + 787.0, + 294.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 781.0, + 1403.0, + 781.0, + 1403.0, + 817.0, + 293.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 812.0, + 1405.0, + 812.0, + 1405.0, + 847.0, + 293.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 844.0, + 889.0, + 844.0, + 889.0, + 876.0, + 296.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 483.0, + 1405.0, + 483.0, + 1405.0, + 532.0, + 291.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 523.0, + 740.0, + 523.0, + 740.0, + 555.0, + 297.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 226.0, + 1405.0, + 226.0, + 1405.0, + 267.0, + 292.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 262.0, + 906.0, + 262.0, + 906.0, + 294.0, + 296.0, + 294.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 295, + 1403, + 295, + 1403, + 480, + 299, + 480 + ], + "score": 0.976 + }, + { + "category_id": 0, + "poly": [ + 299, + 226, + 543, + 226, + 543, + 261, + 299, + 261 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 301, + 527, + 609, + 527, + 609, + 559, + 301, + 559 + ], + "score": 0.882 + }, + { + "category_id": 1, + "poly": [ + 291, + 670, + 1402, + 670, + 1402, + 733, + 291, + 733 + ], + "score": 0.872 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 816, + 75, + 816, + 105, + 300, + 105 + ], + "score": 0.871 + }, + { + "category_id": 1, + "poly": [ + 291, + 593, + 1399, + 593, + 1399, + 655, + 291, + 655 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 837, + 2089, + 865, + 2089, + 865, + 2112, + 837, + 2112 + ], + "score": 0.833 + }, + { + "category_id": 15, + "poly": [ + 292.0, + 222.0, + 548.0, + 222.0, + 548.0, + 270.0, + 292.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 527.0, + 613.0, + 527.0, + 613.0, + 566.0, + 297.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 296.0, + 1405.0, + 296.0, + 1405.0, + 332.0, + 294.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 324.0, + 1405.0, + 324.0, + 1405.0, + 362.0, + 293.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 356.0, + 1405.0, + 356.0, + 1405.0, + 393.0, + 293.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 385.0, + 1405.0, + 385.0, + 1405.0, + 423.0, + 294.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 417.0, + 1404.0, + 417.0, + 1404.0, + 454.0, + 293.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 446.0, + 1229.0, + 446.0, + 1229.0, + 483.0, + 294.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 670.0, + 1407.0, + 670.0, + 1407.0, + 709.0, + 293.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 703.0, + 1038.0, + 703.0, + 1038.0, + 735.0, + 295.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 592.0, + 1402.0, + 592.0, + 1402.0, + 628.0, + 295.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 619.0, + 421.0, + 619.0, + 421.0, + 660.0, + 293.0, + 660.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 105, + 299, + 105 + ], + "score": 0.891 + }, + { + "category_id": 1, + "poly": [ + 290, + 1112, + 1402, + 1112, + 1402, + 1177, + 290, + 1177 + ], + "score": 0.857 + }, + { + "category_id": 1, + "poly": [ + 296, + 1429, + 979, + 1429, + 979, + 1463, + 296, + 1463 + ], + "score": 0.839 + }, + { + "category_id": 1, + "poly": [ + 298, + 1483, + 1400, + 1483, + 1400, + 1548, + 298, + 1548 + ], + "score": 0.832 + }, + { + "category_id": 1, + "poly": [ + 297, + 996, + 1402, + 996, + 1402, + 1093, + 297, + 1093 + ], + "score": 0.829 + }, + { + "category_id": 1, + "poly": [ + 295, + 710, + 1401, + 710, + 1401, + 776, + 295, + 776 + ], + "score": 0.829 + }, + { + "category_id": 1, + "poly": [ + 291, + 911, + 1403, + 911, + 1403, + 976, + 291, + 976 + ], + "score": 0.829 + }, + { + "category_id": 1, + "poly": [ + 301, + 279, + 1400, + 279, + 1400, + 374, + 301, + 374 + ], + "score": 0.826 + }, + { + "category_id": 1, + "poly": [ + 298, + 1740, + 1404, + 1740, + 1404, + 1835, + 298, + 1835 + ], + "score": 0.825 + }, + { + "category_id": 0, + "poly": [ + 300, + 227, + 489, + 227, + 489, + 261, + 300, + 261 + ], + "score": 0.822 + }, + { + "category_id": 1, + "poly": [ + 298, + 1855, + 1401, + 1855, + 1401, + 1949, + 298, + 1949 + ], + "score": 0.811 + }, + { + "category_id": 1, + "poly": [ + 300, + 1971, + 1398, + 1971, + 1398, + 2034, + 300, + 2034 + ], + "score": 0.808 + }, + { + "category_id": 1, + "poly": [ + 292, + 395, + 1402, + 395, + 1402, + 460, + 292, + 460 + ], + "score": 0.808 + }, + { + "category_id": 1, + "poly": [ + 294, + 625, + 1397, + 625, + 1397, + 691, + 294, + 691 + ], + "score": 0.802 + }, + { + "category_id": 1, + "poly": [ + 297, + 794, + 1405, + 794, + 1405, + 891, + 297, + 891 + ], + "score": 0.801 + }, + { + "category_id": 1, + "poly": [ + 299, + 1197, + 1405, + 1197, + 1405, + 1292, + 299, + 1292 + ], + "score": 0.797 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 862, + 2088, + 862, + 2113, + 836, + 2113 + ], + "score": 0.796 + }, + { + "category_id": 1, + "poly": [ + 297, + 479, + 1405, + 479, + 1405, + 603, + 297, + 603 + ], + "score": 0.786 + }, + { + "category_id": 1, + "poly": [ + 299, + 1568, + 1401, + 1568, + 1401, + 1635, + 299, + 1635 + ], + "score": 0.785 + }, + { + "category_id": 1, + "poly": [ + 297, + 1653, + 1403, + 1653, + 1403, + 1720, + 297, + 1720 + ], + "score": 0.776 + }, + { + "category_id": 1, + "poly": [ + 298, + 1313, + 1404, + 1313, + 1404, + 1408, + 298, + 1408 + ], + "score": 0.751 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 229.0, + 490.0, + 229.0, + 490.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1111.0, + 1404.0, + 1111.0, + 1404.0, + 1150.0, + 294.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1145.0, + 883.0, + 1145.0, + 883.0, + 1178.0, + 322.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1427.0, + 979.0, + 1427.0, + 979.0, + 1467.0, + 295.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1482.0, + 1404.0, + 1482.0, + 1404.0, + 1521.0, + 294.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1514.0, + 1057.0, + 1514.0, + 1057.0, + 1550.0, + 324.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 998.0, + 1404.0, + 998.0, + 1404.0, + 1032.0, + 297.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1027.0, + 1406.0, + 1027.0, + 1406.0, + 1064.0, + 322.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1059.0, + 902.0, + 1059.0, + 902.0, + 1092.0, + 324.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 709.0, + 1403.0, + 709.0, + 1403.0, + 748.0, + 294.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 742.0, + 569.0, + 742.0, + 569.0, + 774.0, + 323.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 907.0, + 1405.0, + 907.0, + 1405.0, + 950.0, + 293.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 942.0, + 612.0, + 942.0, + 612.0, + 973.0, + 322.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 278.0, + 1404.0, + 278.0, + 1404.0, + 316.0, + 293.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 311.0, + 1404.0, + 311.0, + 1404.0, + 345.0, + 323.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 341.0, + 1079.0, + 341.0, + 1079.0, + 375.0, + 323.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1739.0, + 1404.0, + 1739.0, + 1404.0, + 1777.0, + 297.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1769.0, + 1406.0, + 1769.0, + 1406.0, + 1808.0, + 321.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1801.0, + 611.0, + 1801.0, + 611.0, + 1834.0, + 322.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1853.0, + 1405.0, + 1853.0, + 1405.0, + 1891.0, + 294.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1887.0, + 1405.0, + 1887.0, + 1405.0, + 1921.0, + 323.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1914.0, + 400.0, + 1914.0, + 400.0, + 1950.0, + 318.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1967.0, + 1404.0, + 1967.0, + 1404.0, + 2010.0, + 295.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 2001.0, + 1062.0, + 2001.0, + 1062.0, + 2037.0, + 324.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 395.0, + 1404.0, + 395.0, + 1404.0, + 431.0, + 295.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 425.0, + 1131.0, + 425.0, + 1131.0, + 460.0, + 323.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 626.0, + 1403.0, + 626.0, + 1403.0, + 662.0, + 295.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 657.0, + 1377.0, + 657.0, + 1377.0, + 693.0, + 324.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 797.0, + 1402.0, + 797.0, + 1402.0, + 831.0, + 296.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 826.0, + 1405.0, + 826.0, + 1405.0, + 865.0, + 321.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 859.0, + 607.0, + 859.0, + 607.0, + 890.0, + 321.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1198.0, + 1403.0, + 1198.0, + 1403.0, + 1232.0, + 297.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1229.0, + 1405.0, + 1229.0, + 1405.0, + 1264.0, + 322.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1259.0, + 612.0, + 1259.0, + 612.0, + 1292.0, + 322.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 479.0, + 1405.0, + 479.0, + 1405.0, + 515.0, + 292.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 511.0, + 1403.0, + 511.0, + 1403.0, + 543.0, + 324.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 540.0, + 1406.0, + 540.0, + 1406.0, + 577.0, + 320.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 570.0, + 399.0, + 570.0, + 399.0, + 605.0, + 320.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1569.0, + 1405.0, + 1569.0, + 1405.0, + 1605.0, + 298.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1600.0, + 1188.0, + 1600.0, + 1188.0, + 1635.0, + 322.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1651.0, + 1405.0, + 1651.0, + 1405.0, + 1693.0, + 294.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1685.0, + 1130.0, + 1685.0, + 1130.0, + 1720.0, + 322.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1309.0, + 1404.0, + 1309.0, + 1404.0, + 1356.0, + 293.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1341.0, + 1405.0, + 1341.0, + 1405.0, + 1385.0, + 320.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1371.0, + 506.0, + 1371.0, + 506.0, + 1409.0, + 321.0, + 1409.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 105, + 299, + 105 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.834 + }, + { + "category_id": 1, + "poly": [ + 299, + 228, + 1403, + 228, + 1403, + 323, + 299, + 323 + ], + "score": 0.613 + }, + { + "category_id": 1, + "poly": [ + 296, + 1429, + 1400, + 1429, + 1400, + 1495, + 296, + 1495 + ], + "score": 0.576 + }, + { + "category_id": 1, + "poly": [ + 297, + 1512, + 1147, + 1512, + 1147, + 1548, + 297, + 1548 + ], + "score": 0.572 + }, + { + "category_id": 1, + "poly": [ + 297, + 1316, + 1401, + 1316, + 1401, + 1412, + 297, + 1412 + ], + "score": 0.532 + }, + { + "category_id": 1, + "poly": [ + 300, + 643, + 1402, + 643, + 1402, + 738, + 300, + 738 + ], + "score": 0.505 + }, + { + "category_id": 1, + "poly": [ + 298, + 1233, + 1403, + 1233, + 1403, + 1298, + 298, + 1298 + ], + "score": 0.505 + }, + { + "category_id": 1, + "poly": [ + 300, + 1150, + 1402, + 1150, + 1402, + 1215, + 300, + 1215 + ], + "score": 0.496 + }, + { + "category_id": 1, + "poly": [ + 297, + 840, + 1398, + 840, + 1398, + 905, + 297, + 905 + ], + "score": 0.473 + }, + { + "category_id": 1, + "poly": [ + 297, + 1036, + 1400, + 1036, + 1400, + 1132, + 297, + 1132 + ], + "score": 0.462 + }, + { + "category_id": 1, + "poly": [ + 298, + 923, + 1400, + 923, + 1400, + 1018, + 298, + 1018 + ], + "score": 0.46 + }, + { + "category_id": 1, + "poly": [ + 292, + 757, + 1402, + 757, + 1402, + 822, + 292, + 822 + ], + "score": 0.445 + }, + { + "category_id": 1, + "poly": [ + 297, + 561, + 1402, + 561, + 1402, + 626, + 297, + 626 + ], + "score": 0.444 + }, + { + "category_id": 1, + "poly": [ + 291, + 343, + 1308, + 343, + 1308, + 378, + 291, + 378 + ], + "score": 0.431 + }, + { + "category_id": 1, + "poly": [ + 297, + 477, + 1401, + 477, + 1401, + 543, + 297, + 543 + ], + "score": 0.431 + }, + { + "category_id": 1, + "poly": [ + 295, + 395, + 1399, + 395, + 1399, + 460, + 295, + 460 + ], + "score": 0.42 + }, + { + "category_id": 1, + "poly": [ + 299, + 1567, + 1406, + 1567, + 1406, + 1661, + 299, + 1661 + ], + "score": 0.316 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 226.0, + 1404.0, + 226.0, + 1404.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 261.0, + 1405.0, + 261.0, + 1405.0, + 295.0, + 323.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 293.0, + 497.0, + 293.0, + 497.0, + 323.0, + 320.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1430.0, + 1404.0, + 1430.0, + 1404.0, + 1466.0, + 295.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1461.0, + 1370.0, + 1461.0, + 1370.0, + 1496.0, + 321.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1510.0, + 1145.0, + 1510.0, + 1145.0, + 1551.0, + 294.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1311.0, + 1406.0, + 1311.0, + 1406.0, + 1358.0, + 294.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1349.0, + 1405.0, + 1349.0, + 1405.0, + 1384.0, + 324.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1380.0, + 830.0, + 1380.0, + 830.0, + 1413.0, + 322.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 642.0, + 1403.0, + 642.0, + 1403.0, + 681.0, + 293.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 675.0, + 1406.0, + 675.0, + 1406.0, + 711.0, + 321.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 703.0, + 403.0, + 703.0, + 403.0, + 739.0, + 320.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1235.0, + 1403.0, + 1235.0, + 1403.0, + 1271.0, + 296.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1264.0, + 1127.0, + 1264.0, + 1127.0, + 1299.0, + 323.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1149.0, + 1404.0, + 1149.0, + 1404.0, + 1188.0, + 293.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1183.0, + 678.0, + 1183.0, + 678.0, + 1213.0, + 323.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 839.0, + 1404.0, + 839.0, + 1404.0, + 878.0, + 293.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 870.0, + 1347.0, + 870.0, + 1347.0, + 906.0, + 323.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1035.0, + 1405.0, + 1035.0, + 1405.0, + 1074.0, + 295.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1066.0, + 1405.0, + 1066.0, + 1405.0, + 1107.0, + 321.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1100.0, + 661.0, + 1100.0, + 661.0, + 1134.0, + 323.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 923.0, + 1404.0, + 923.0, + 1404.0, + 961.0, + 294.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 955.0, + 1404.0, + 955.0, + 1404.0, + 989.0, + 323.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 986.0, + 789.0, + 986.0, + 789.0, + 1019.0, + 323.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 757.0, + 1404.0, + 757.0, + 1404.0, + 795.0, + 295.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 788.0, + 1250.0, + 788.0, + 1250.0, + 823.0, + 321.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 561.0, + 1404.0, + 561.0, + 1404.0, + 597.0, + 297.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 592.0, + 1406.0, + 592.0, + 1406.0, + 628.0, + 322.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 342.0, + 1307.0, + 342.0, + 1307.0, + 381.0, + 293.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 476.0, + 1405.0, + 476.0, + 1405.0, + 517.0, + 293.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 510.0, + 1274.0, + 510.0, + 1274.0, + 544.0, + 322.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 395.0, + 1403.0, + 395.0, + 1403.0, + 431.0, + 296.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 428.0, + 985.0, + 428.0, + 985.0, + 461.0, + 321.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1564.0, + 1407.0, + 1564.0, + 1407.0, + 1606.0, + 294.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1595.0, + 1407.0, + 1595.0, + 1407.0, + 1635.0, + 322.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1627.0, + 782.0, + 1627.0, + 782.0, + 1665.0, + 323.0, + 1665.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 567, + 1533, + 1123, + 1533, + 1123, + 1855, + 567, + 1855 + ], + "score": 0.98, + "html": "
ClassIndexing
dogn02090721, n02091032, n02088094
birdn01532829, n01558993, n01534433
carn02814533, n03930630, n03100240
fishn01484850,n01491361, n01494475
monkeyn02483708,n02484975, n02486261
turtlen01664065,n01665541, n01667114
lizardn01677366,n01682714,n01685808
bridgen03933933,n04366367,n04311004
cown02403003,n02408429,n02410509
crabn01980166,n01978455,n01981276
" + }, + { + "category_id": 1, + "poly": [ + 296, + 1134, + 1405, + 1134, + 1405, + 1350, + 296, + 1350 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 288, + 1996, + 1310, + 1996, + 1310, + 2031, + 288, + 2031 + ], + "score": 0.914 + }, + { + "category_id": 6, + "poly": [ + 393, + 1475, + 1304, + 1475, + 1304, + 1510, + 393, + 1510 + ], + "score": 0.911 + }, + { + "category_id": 6, + "poly": [ + 489, + 383, + 1208, + 383, + 1208, + 418, + 489, + 418 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 299, + 1935, + 684, + 1935, + 684, + 1968, + 299, + 1968 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 299, + 226, + 501, + 226, + 501, + 261, + 299, + 261 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 300, + 1396, + 903, + 1396, + 903, + 1429, + 300, + 1429 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 817, + 75, + 817, + 105, + 299, + 105 + ], + "score": 0.901 + }, + { + "category_id": 1, + "poly": [ + 288, + 300, + 1305, + 300, + 1305, + 335, + 288, + 335 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2113, + 836, + 2113 + ], + "score": 0.864 + }, + { + "category_id": 5, + "poly": [ + 487, + 856, + 1207, + 856, + 1207, + 1092, + 487, + 1092 + ], + "score": 0.257, + "html": "
ImgNet10RepetitionLayer
FE×1Conv(3,32,5,2)+GroupNorm
×1MaxPooling(2)
×1BaiscBlock(32, 64,2)
×1MaxPooling(2)
RM×3BaiscBlock(64,64,1)
FCC×1AdaptiveAvgPool2d + Linear(64,10)
" + }, + { + "category_id": 5, + "poly": [ + 488, + 651, + 1204, + 651, + 1204, + 829, + 488, + 829 + ], + "score": 0.239, + "html": "
SVHNRepetitionLayer
FE×1×1Conv(3,64,3,1) + GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
" + }, + { + "category_id": 5, + "poly": [ + 489, + 442, + 1206, + 442, + 1206, + 624, + 489, + 624 + ], + "score": 0.112, + "html": "
MNISTRepetitionLayer
FE×1×1Conv(1,64,3,1) +GroupNorm+ReLUConv(64,64,4,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
" + }, + { + "category_id": 13, + "poly": [ + 671, + 800, + 704, + 800, + 704, + 820, + 671, + 820 + ], + "score": 0.85, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 670, + 594, + 704, + 594, + 704, + 615, + 670, + 615 + ], + "score": 0.77, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 671, + 692, + 703, + 692, + 703, + 712, + 671, + 712 + ], + "score": 0.76, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 670, + 1055, + 705, + 1055, + 705, + 1081, + 670, + 1081 + ], + "score": 0.75, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 671, + 1015, + 705, + 1015, + 705, + 1041, + 671, + 1041 + ], + "score": 0.73, + "latex": "\\times 3" + }, + { + "category_id": 13, + "poly": [ + 766, + 758, + 1191, + 758, + 1191, + 783, + 766, + 783 + ], + "score": 0.72, + "latex": "\\mathrm { C o n v } ( 6 4 , 6 4 , 3 , 1 ) + \\mathrm { G r o u p N o r m + R e L U }" + }, + { + "category_id": 13, + "poly": [ + 670, + 482, + 704, + 482, + 704, + 507, + 670, + 507 + ], + "score": 0.7, + "latex": "\\overline { { \\times 1 } }" + }, + { + "category_id": 13, + "poly": [ + 671, + 719, + 705, + 719, + 705, + 741, + 671, + 741 + ], + "score": 0.7, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 766, + 548, + 1193, + 548, + 1193, + 577, + 766, + 577 + ], + "score": 0.7, + "latex": "\\mathrm { C o n v } ( 6 4 , 6 4 , 3 , 1 ) + \\mathrm { G r o u p N o r m + R e L U }" + }, + { + "category_id": 13, + "poly": [ + 671, + 761, + 704, + 761, + 704, + 781, + 671, + 781 + ], + "score": 0.68, + "latex": "\\times 2" + }, + { + "category_id": 13, + "poly": [ + 670, + 510, + 706, + 510, + 706, + 535, + 670, + 535 + ], + "score": 0.67, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 670, + 949, + 705, + 949, + 705, + 973, + 670, + 973 + ], + "score": 0.67, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 772, + 481, + 1188, + 481, + 1188, + 508, + 772, + 508 + ], + "score": 0.67, + "latex": "\\mathrm { C o n v } ( 1 , 6 4 , 3 , 1 ) + \\mathrm { G r o u p N o r m + R e L U }" + }, + { + "category_id": 13, + "poly": [ + 671, + 551, + 705, + 551, + 705, + 575, + 671, + 575 + ], + "score": 0.67, + "latex": "\\times 2" + }, + { + "category_id": 13, + "poly": [ + 670, + 923, + 704, + 923, + 704, + 946, + 670, + 946 + ], + "score": 0.66, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 772, + 691, + 1187, + 691, + 1187, + 714, + 772, + 714 + ], + "score": 0.62, + "latex": "\\mathrm { C o n v } ( 3 , 6 4 , 3 , 1 ) + \\mathrm { G r o u p N o r m } + \\mathrm { R e L U }" + }, + { + "category_id": 13, + "poly": [ + 670, + 976, + 705, + 976, + 705, + 1001, + 670, + 1001 + ], + "score": 0.58, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 671, + 898, + 703, + 898, + 703, + 919, + 671, + 919 + ], + "score": 0.57, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 769, + 510, + 1190, + 510, + 1190, + 537, + 769, + 537 + ], + "score": 0.47, + "latex": "\\mathrm { C o n v } ( 6 4 , 6 4 , 4 , 2 ) + \\mathrm { G r o u p N o r m + R e L U }" + }, + { + "category_id": 13, + "poly": [ + 769, + 719, + 1188, + 719, + 1188, + 743, + 769, + 743 + ], + "score": 0.45, + "latex": "\\mathrm { C o n v } ( 6 4 , 6 4 , 4 , 2 ) + \\mathrm { G r o u p N o r m + R e L U }" + }, + { + "category_id": 13, + "poly": [ + 1096, + 484, + 1117, + 484, + 1117, + 505, + 1096, + 505 + ], + "score": 0.37, + "latex": "^ +" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 1469.0, + 1310.0, + 1469.0, + 1310.0, + 1517.0, + 389.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 380.0, + 1210.0, + 380.0, + 1210.0, + 420.0, + 489.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1933.0, + 688.0, + 1933.0, + 688.0, + 1972.0, + 293.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 221.0, + 507.0, + 221.0, + 507.0, + 269.0, + 291.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1395.0, + 905.0, + 1395.0, + 905.0, + 1432.0, + 294.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1133.0, + 1405.0, + 1133.0, + 1405.0, + 1170.0, + 293.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1165.0, + 1405.0, + 1165.0, + 1405.0, + 1200.0, + 294.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1195.0, + 1405.0, + 1195.0, + 1405.0, + 1230.0, + 294.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1225.0, + 1403.0, + 1225.0, + 1403.0, + 1261.0, + 293.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1256.0, + 1405.0, + 1256.0, + 1405.0, + 1290.0, + 294.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1288.0, + 1405.0, + 1288.0, + 1405.0, + 1322.0, + 295.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1319.0, + 1039.0, + 1319.0, + 1039.0, + 1349.0, + 296.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1995.0, + 1314.0, + 1995.0, + 1314.0, + 2034.0, + 295.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 300.0, + 1308.0, + 300.0, + 1308.0, + 338.0, + 293.0, + 338.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 496, + 1134, + 1196, + 1134, + 1196, + 1304, + 496, + 1304 + ], + "score": 0.981, + "html": "
Gaussian noiseAdversarial attack
MNISTσ=100FGSM-0.3FGSM-0.5PGD-0.3
CNN58.098.421.15.3
ODENet84.299.136.012.3
TisODE87.999.166.578.9
" + }, + { + "category_id": 1, + "poly": [ + 297, + 824, + 1406, + 824, + 1406, + 979, + 297, + 979 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 296, + 1423, + 1406, + 1423, + 1406, + 1608, + 296, + 1608 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 506, + 1758, + 1184, + 1758, + 1184, + 1928, + 506, + 1928 + ], + "score": 0.976, + "html": "
Gaussian noiseAdversarial attack
CIFAR10σ=15σ=20FGSM-8/255FGSM-10/255
CNN70.257.624.318.4
ODENet72.660.631.226.0
TisODE74.362.033.626.8
" + }, + { + "category_id": 1, + "poly": [ + 290, + 226, + 1403, + 226, + 1403, + 294, + 290, + 294 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 646, + 316, + 1054, + 316, + 1054, + 455, + 646, + 455 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 297, + 1972, + 1398, + 1972, + 1398, + 2035, + 297, + 2035 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 645, + 627, + 1051, + 627, + 1051, + 670, + 645, + 670 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 572, + 527, + 1124, + 527, + 1124, + 569, + 572, + 569 + ], + "score": 0.931 + }, + { + "category_id": 6, + "poly": [ + 299, + 1637, + 1403, + 1637, + 1403, + 1731, + 299, + 1731 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 295, + 472, + 985, + 472, + 985, + 508, + 295, + 508 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 299, + 586, + 568, + 586, + 568, + 620, + 299, + 620 + ], + "score": 0.926 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 817, + 74, + 817, + 106, + 298, + 106 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 299, + 1367, + 894, + 1367, + 894, + 1400, + 299, + 1400 + ], + "score": 0.919 + }, + { + "category_id": 6, + "poly": [ + 299, + 1013, + 1402, + 1013, + 1402, + 1108, + 299, + 1108 + ], + "score": 0.905 + }, + { + "category_id": 0, + "poly": [ + 300, + 709, + 743, + 709, + 743, + 740, + 300, + 740 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 299, + 768, + 1092, + 768, + 1092, + 802, + 299, + 802 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2113, + 836, + 2113 + ], + "score": 0.871 + }, + { + "category_id": 13, + "poly": [ + 804, + 228, + 1039, + 228, + 1039, + 264, + 804, + 264 + ], + "score": 0.94, + "latex": "f : U \\times [ 0 , T ] \\to \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 453, + 587, + 560, + 587, + 560, + 621, + 453, + 621 + ], + "score": 0.92, + "latex": "t \\in [ 0 , T ]" + }, + { + "category_id": 14, + "poly": [ + 646, + 315, + 1055, + 315, + 1055, + 458, + 646, + 458 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { \\frac { \\mathrm { d } { \\mathbf z } _ { 1 } ( t ) } { \\mathrm { d } t } = f ( { \\mathbf z } _ { 1 } ( t ) , t ) , \\quad { \\mathbf z } _ { 1 } ( t ) = { \\mathbf x } _ { 1 } } \\\\ & { \\frac { \\mathrm { d } { \\mathbf z } _ { 2 } ( t ) } { \\mathrm { d } t } = f ( { \\mathbf z } _ { 2 } ( t ) , t ) , \\quad { \\mathbf z } _ { 2 } ( t ) = { \\mathbf x } _ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 867, + 474, + 975, + 474, + 975, + 508, + 867, + 508 + ], + "score": 0.92, + "latex": "t \\in [ 0 , T ]" + }, + { + "category_id": 13, + "poly": [ + 488, + 228, + 583, + 228, + 583, + 259, + 488, + 259 + ], + "score": 0.91, + "latex": "U \\subset \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 599, + 475, + 674, + 475, + 674, + 505, + 599, + 505 + ], + "score": 0.9, + "latex": "C \\geq 0" + }, + { + "category_id": 14, + "poly": [ + 645, + 625, + 1054, + 625, + 1054, + 668, + 645, + 668 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\| \\mathbf { z } _ { 1 } ( t ) - \\mathbf { z } _ { 2 } ( t ) \\| \\leq \\| \\mathbf { x } _ { 2 } - \\mathbf { x } _ { 1 } \\| \\cdot e ^ { C t } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 573, + 527, + 1124, + 527, + 1124, + 567, + 573, + 567 + ], + "score": 0.86, + "latex": "\\| f ( \\mathbf { z } _ { 2 } ( t ) , t ) - f ( \\mathbf { z } _ { 1 } ( t ) , t ) ) \\| \\leq C \\| \\mathbf { z } _ { 2 } ( t ) - \\mathbf { z } _ { 1 } ( t ) \\|" + }, + { + "category_id": 13, + "poly": [ + 383, + 261, + 512, + 261, + 512, + 295, + 383, + 295 + ], + "score": 0.82, + "latex": "[ 0 , T ] \\to U" + }, + { + "category_id": 13, + "poly": [ + 662, + 1016, + 705, + 1016, + 705, + 1045, + 662, + 1045 + ], + "score": 0.79, + "latex": "( \\% )" + }, + { + "category_id": 13, + "poly": [ + 406, + 1705, + 427, + 1705, + 427, + 1727, + 406, + 1727 + ], + "score": 0.76, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 406, + 1080, + 426, + 1080, + 426, + 1103, + 406, + 1103 + ], + "score": 0.73, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 1197, + 858, + 1251, + 858, + 1251, + 886, + 1197, + 886 + ], + "score": 0.69, + "latex": "_ { 1 = 0 . 3 }" + }, + { + "category_id": 13, + "poly": [ + 655, + 1640, + 696, + 1640, + 696, + 1669, + 655, + 1669 + ], + "score": 0.64, + "latex": "( \\% )" + }, + { + "category_id": 13, + "poly": [ + 337, + 265, + 367, + 265, + 367, + 292, + 337, + 292 + ], + "score": 0.58, + "latex": "\\mathbf { z } _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 267, + 326, + 267, + 326, + 292, + 298, + 292 + ], + "score": 0.56, + "latex": "\\mathbf { z } _ { 1 }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1638.0, + 654.0, + 1638.0, + 654.0, + 1672.0, + 294.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1638.0, + 1405.0, + 1638.0, + 1405.0, + 1672.0, + 697.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1668.0, + 1404.0, + 1668.0, + 1404.0, + 1702.0, + 294.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1700.0, + 405.0, + 1700.0, + 405.0, + 1730.0, + 297.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 1700.0, + 638.0, + 1700.0, + 638.0, + 1730.0, + 428.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 109.0, + 295.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1366.0, + 897.0, + 1366.0, + 897.0, + 1403.0, + 294.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1014.0, + 661.0, + 1014.0, + 661.0, + 1048.0, + 294.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1014.0, + 1406.0, + 1014.0, + 1406.0, + 1048.0, + 706.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1043.0, + 1406.0, + 1043.0, + 1406.0, + 1081.0, + 294.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1076.0, + 405.0, + 1076.0, + 405.0, + 1107.0, + 297.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 1076.0, + 780.0, + 1076.0, + 780.0, + 1107.0, + 427.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 708.0, + 746.0, + 708.0, + 746.0, + 743.0, + 295.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 766.0, + 1094.0, + 766.0, + 1094.0, + 804.0, + 295.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 824.0, + 1403.0, + 824.0, + 1403.0, + 858.0, + 297.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 855.0, + 1196.0, + 855.0, + 1196.0, + 891.0, + 295.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 855.0, + 1404.0, + 855.0, + 1404.0, + 891.0, + 1252.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 884.0, + 1407.0, + 884.0, + 1407.0, + 921.0, + 292.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 915.0, + 1406.0, + 915.0, + 1406.0, + 951.0, + 292.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 947.0, + 966.0, + 947.0, + 966.0, + 981.0, + 294.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1424.0, + 1406.0, + 1424.0, + 1406.0, + 1458.0, + 294.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1453.0, + 1404.0, + 1453.0, + 1404.0, + 1489.0, + 294.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1485.0, + 1406.0, + 1485.0, + 1406.0, + 1522.0, + 293.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1517.0, + 1404.0, + 1517.0, + 1404.0, + 1549.0, + 295.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1546.0, + 1407.0, + 1546.0, + 1407.0, + 1582.0, + 294.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1578.0, + 565.0, + 1578.0, + 565.0, + 1608.0, + 294.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 225.0, + 487.0, + 225.0, + 487.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 225.0, + 803.0, + 225.0, + 803.0, + 265.0, + 584.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 225.0, + 1405.0, + 225.0, + 1405.0, + 265.0, + 1040.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 259.0, + 297.0, + 259.0, + 297.0, + 298.0, + 292.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 259.0, + 336.0, + 259.0, + 336.0, + 298.0, + 327.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 259.0, + 382.0, + 259.0, + 382.0, + 298.0, + 368.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 259.0, + 892.0, + 259.0, + 892.0, + 298.0, + 513.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1973.0, + 1402.0, + 1973.0, + 1402.0, + 2005.0, + 296.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 1110.0, + 2002.0, + 1110.0, + 2038.0, + 295.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 471.0, + 598.0, + 471.0, + 598.0, + 511.0, + 294.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 471.0, + 866.0, + 471.0, + 866.0, + 511.0, + 675.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 471.0, + 984.0, + 471.0, + 984.0, + 511.0, + 976.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 583.0, + 452.0, + 583.0, + 452.0, + 626.0, + 295.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 583.0, + 570.0, + 583.0, + 570.0, + 626.0, + 561.0, + 626.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 514, + 282, + 1179, + 282, + 1179, + 493, + 514, + 493 + ], + "score": 0.979, + "html": "
RepetitionLayer
FE×1Conv(3,16,3,1)+GroupNorm+ReLU
×1Conv(16,32,3,2) + GroupNorm+ReLU
×1Conv(32,64,3,2) + GroupNorm + ReLU
RM×2Conv(64,64,3,1) + GroupNorm + ReLU
FCC×1AdaptiveAvgPool2d + Linear(64,10)
" + }, + { + "category_id": 1, + "poly": [ + 297, + 709, + 1405, + 709, + 1405, + 894, + 297, + 894 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 300, + 601, + 1404, + 601, + 1404, + 695, + 300, + 695 + ], + "score": 0.974 + }, + { + "category_id": 6, + "poly": [ + 543, + 223, + 1155, + 223, + 1155, + 257, + 543, + 257 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 816, + 75, + 816, + 105, + 298, + 105 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.838 + }, + { + "category_id": 0, + "poly": [ + 296, + 549, + 1229, + 549, + 1229, + 581, + 296, + 581 + ], + "score": 0.818 + }, + { + "category_id": 13, + "poly": [ + 1010, + 741, + 1154, + 741, + 1154, + 774, + 1010, + 774 + ], + "score": 0.9, + "latex": "1 / 0 . 0 5 = 2 0 " + }, + { + "category_id": 13, + "poly": [ + 966, + 833, + 1041, + 833, + 1041, + 862, + 966, + 862 + ], + "score": 0.85, + "latex": "8 1 . 9 \\%" + }, + { + "category_id": 13, + "poly": [ + 1318, + 833, + 1393, + 833, + 1393, + 863, + 1318, + 863 + ], + "score": 0.85, + "latex": "4 9 . 7 \\%" + }, + { + "category_id": 13, + "poly": [ + 1192, + 833, + 1269, + 833, + 1269, + 863, + 1192, + 863 + ], + "score": 0.84, + "latex": "5 3 . 4 \\%" + }, + { + "category_id": 13, + "poly": [ + 841, + 833, + 917, + 833, + 917, + 862, + 841, + 862 + ], + "score": 0.84, + "latex": "8 7 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 531, + 864, + 592, + 864, + 592, + 893, + 531, + 893 + ], + "score": 0.83, + "latex": "4 . 8 \\%" + }, + { + "category_id": 13, + "poly": [ + 644, + 419, + 678, + 419, + 678, + 441, + 644, + 441 + ], + "score": 0.65, + "latex": "\\times 2" + }, + { + "category_id": 13, + "poly": [ + 413, + 863, + 488, + 863, + 488, + 894, + 413, + 894 + ], + "score": 0.61, + "latex": "1 1 . 8 \\%" + }, + { + "category_id": 13, + "poly": [ + 644, + 459, + 678, + 459, + 678, + 482, + 644, + 482 + ], + "score": 0.57, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 740, + 415, + 1165, + 415, + 1165, + 444, + 740, + 444 + ], + "score": 0.49, + "latex": "\\mathrm { C o n v } ( 6 4 , 6 4 , 3 , 1 ) + \\mathrm { G r o u p N o r m + R e L U }" + }, + { + "category_id": 13, + "poly": [ + 644, + 321, + 677, + 321, + 677, + 346, + 644, + 346 + ], + "score": 0.49, + "latex": "\\overline { { \\times 1 } }" + }, + { + "category_id": 13, + "poly": [ + 741, + 350, + 1160, + 350, + 1160, + 375, + 741, + 375 + ], + "score": 0.46, + "latex": "\\mathrm { C o n v } ( 1 6 , 3 2 , 3 , 2 ) + \\mathrm { G r o u p N o r m + R e L U }" + }, + { + "category_id": 13, + "poly": [ + 1069, + 324, + 1090, + 324, + 1090, + 344, + 1069, + 344 + ], + "score": 0.38, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 644, + 376, + 679, + 376, + 679, + 402, + 644, + 402 + ], + "score": 0.36, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 643, + 349, + 679, + 349, + 679, + 373, + 643, + 373 + ], + "score": 0.36, + "latex": "\\times 1" + }, + { + "category_id": 13, + "poly": [ + 744, + 321, + 1158, + 321, + 1158, + 347, + 744, + 347 + ], + "score": 0.31, + "latex": "\\mathrm { C o n v } ( 3 , 1 6 , 3 , 1 ) + \\mathrm { G r o u p N o r m } + \\mathrm { R e L U }" + }, + { + "category_id": 13, + "poly": [ + 742, + 376, + 1166, + 376, + 1166, + 402, + 742, + 402 + ], + "score": 0.25, + "latex": "\\mathrm { C o n v } ( 3 2 , 6 4 , 3 , 2 ) + \\mathrm { G r o u p N o r m + R e L U }" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 221.0, + 1156.0, + 221.0, + 1156.0, + 259.0, + 540.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 549.0, + 1232.0, + 549.0, + 1232.0, + 584.0, + 295.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 707.0, + 1406.0, + 707.0, + 1406.0, + 744.0, + 294.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 742.0, + 1009.0, + 742.0, + 1009.0, + 774.0, + 296.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 742.0, + 1403.0, + 742.0, + 1403.0, + 774.0, + 1155.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 771.0, + 1405.0, + 771.0, + 1405.0, + 805.0, + 294.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 803.0, + 1403.0, + 803.0, + 1403.0, + 835.0, + 294.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 833.0, + 840.0, + 833.0, + 840.0, + 865.0, + 296.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 833.0, + 965.0, + 833.0, + 965.0, + 865.0, + 918.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 833.0, + 1191.0, + 833.0, + 1191.0, + 865.0, + 1042.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 833.0, + 1317.0, + 833.0, + 1317.0, + 865.0, + 1270.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 833.0, + 1405.0, + 833.0, + 1405.0, + 865.0, + 1394.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 863.0, + 412.0, + 863.0, + 412.0, + 894.0, + 295.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 863.0, + 530.0, + 863.0, + 530.0, + 894.0, + 489.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 863.0, + 609.0, + 863.0, + 609.0, + 894.0, + 593.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 601.0, + 1405.0, + 601.0, + 1405.0, + 638.0, + 294.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 633.0, + 1405.0, + 633.0, + 1405.0, + 667.0, + 296.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 661.0, + 971.0, + 661.0, + 971.0, + 701.0, + 294.0, + 701.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/B1lKS2AqtX/B1lKS2AqtX.md b/parse/train/B1lKS2AqtX/B1lKS2AqtX.md new file mode 100644 index 0000000000000000000000000000000000000000..f2d77bfd5770c4bdc014f3555ed3c3464094ac50 --- /dev/null +++ b/parse/train/B1lKS2AqtX/B1lKS2AqtX.md @@ -0,0 +1,299 @@ +# EIDETIC 3D LSTM: A MODEL FOR VIDEO PREDICTION AND BEYOND + +Yunbo Wang1∗, Lu Jiang2, Ming-Hsuan Yang2,3, Li-Jia $\mathbf { L i } ^ { 4 }$ , Mingsheng Long1, Li Fei-Fei4 1Tsinghua University, 2Google AI, 3University of California, Merced, 4Stanford University + +# ABSTRACT + +Spatiotemporal predictive learning, though long considered to be a promising selfsupervised feature learning method, seldom shows its effectiveness beyond future video prediction. The reason is that it is difficult to learn good representations for both short-term frame dependency and long-term high-level relations. We present a new model, Eidetic 3D LSTM (E3D-LSTM), that integrates 3D convolutions into RNNs. The encapsulated 3D-Conv makes local perceptrons of RNNs motion-aware and enables the memory cell to store better short-term features. For long-term relations, we make the present memory state interact with its historical records via a gate-controlled self-attention module. We describe this memory transition mechanism eidetic as it is able to effectively recall the stored memories across multiple time stamps even after long periods of disturbance. We first evaluate the E3D-LSTM network on widely-used future video prediction datasets and achieve the state-of-the-art performance. Then we show that the E3D-LSTM network also performs well on the early activity recognition to infer what is happening or what will happen after observing only limited frames of video. This task aligns well with video prediction in modeling action intentions and tendency. + +# 1 INTRODUCTION + +A fundamental problem in spatiotemporal predictive learning is how to effectively learn good representations for video inference or reasoning. Currently, recurrent neural networks (RNNs) remain to be the most promising models in this field, and have achieved state-of-the-art results on a number of future video prediction benchmarks (Wang et al., 2018b; Oliu et al., 2018). However, beyond frames prediction, RNN based models are less effective in learning high-level video representations or capturing long-term relations. On the other hand, recent studies demonstrate that 3D Convolutional Neural networks (3D-CNNs) surpass RNNs in learning better representations for action classification (Carreira & Zisserman, 2017; Tran et al., 2015). For instance, variants of 3D-CNNs, such as Inflated 3D-CNNs, have significantly increased action classification accuracy over the UCF 101 and Kinetics datasets. These 3D-CNN architectures have no recurrent structures but instead employ 3D convolution (3D-Conv) and 3D pooling operations to preserve temporal information of the input sequences which would be otherwise discarded in classical 2D convolution operations. + +Motivated by the recent success of 3D-CNNs, in this paper we propose a new model for spatiotemporal predictive learning based on both recurrent modeling (for temporal dependency) and feedforward 3D-Conv modeling (for local dynamics). A plausible approach, of course, is to simply stack 3D-Convs and each RNN unit in a feed-forward way using 3D-Convs for either perceiving fine-grained features from raw videos or combining high-level representations. However, as shown in our experiments, these straightforward extensions may not outperform the baseline RNN model. We attribute these findings to that RNNs and 3D-CNNs represent two very different mechanisms for the same purpose of spatiotemporal modeling, and connecting them directly fails to exploit their complementary advantages. Therefore, it remains challenging and requires principled approaches to design an effective spatiotemporal network. + +To this end, we propose a new model called Eidetic 3D LSTM (E3D-LSTM) for spatiotemporal predictive learning. We introduce an eidetic 3D memory to: a) memorize local appearance and motion in a short spatiotemporal volume, and b) recall the long-range historical context by learning to attend to previous memory states. Regarding the short-term dependency, in many cases, spatiotemporal predictive modeling mainly depends on temporally nearby appearances and on-going short-term motions. All the information is encapsulated into the eidetic 3D memory cell with a short time convolution window, and used in recurrent transitions. Our experimental results show that integrating 3D-Conv deep into RNNs is effective for modeling local representations in a consecutive manner. On the other hand, for long-term interactions, which is important for predicting non-stationary or periodical videos as well as learning high-level video representations, we exploit a self-attention mechanism controlled by revised recurrent gates to recall temporally distant memory. The current memory state of E3D-LSTM is learned to attend to all previous relevant moments. Our experimental results verify that this attention mechanism is beneficial for long-term memorization. We describe this memory transition mechanism eidetic as it is able to effectively recall the stored memories across multiple time stamps even after long periods of disturbance. + +To the best of our knowledge, the proposed E3D-LSTM model is among the first approaches that leverage 3D-Conv in RNNs. We empirically validate it on standard spatiotemporal predictive tasks and an early activity recognition task over four benchmarks: a) on future video prediction, it achieves the best-published accuracy on three classical benchmarks; b) on early activity recognition, it outperforms the state-of-the-art action recognition methods. In addition, we show that self-supervised learning can further improve the performance of early activity recognition. We present ablation studies to verify the effectiveness of all modules in the proposed E3D-LSTM model. + +# 2 RELATED WORK AND PROBLEM CONTEXT + +Spatiotemporal Predictive Learning Models. In recent years, RNNs have been extensively used in sequence prediction and future frame prediction. Srivastava et al. (2015) extended the LSTMbased sequence to sequence model (Sutskever et al., 2014) for language modeling to learning video representations. Shi et al. (2015) proposed the convolutional LSTM by integrating convolutions into recurrent state transitions for high-dimensional sequence prediction. The convolutional LSTM model is extended by Finn et al. (2016) to predict future states of robotic environments. Villegas et al. (2017) leveraged optical flow to help capture short-term video dynamics for video prediction. Xu et al. (2018) proposed a two-stream RNN that deals with structural video content in separate streams. Kalchbrenner et al. (2017) introduced a sophisticated model that extends recurrent structures to estimate local dependencies between adjacent pixels. While this video pixel network (VPN) model is able to describe image sequences, the computational load is prohibitively high. + +The above-mentioned recurrent models predict future frame mainly based on sequentially updated memory states. When the memory cell is refreshed, older memories will be discarded immediately. In contrast, the proposed E3D-LSTM model maintains a list of historical memory records and revokes them when necessary, thereby facilitating long-range video reasoning. While in spirit this idea is similar to the self-attention module in feed-forward networks (Vaswani et al., 2017; Wang et al., 2018a), we exploit it to correlate long-term and short-term video representations in this work. + +Another significant difference between the above-mentioned prior work and the proposed model is that we use 3D-Convs as basic operations inside the E3D-LSTM instead of fully-connected or 2D convolution operations. We show using 3D-Convs to model recurrent state-to-state transitions can significantly improve prediction performance. This idea is motivated by recent advances in video classification (a high-level representation learning task) (Ji et al., 2013; Tran et al., 2015; Carreira & Zisserman, 2017). We note that Vondrick et al. (2016) and Tulyakov et al. (2018) also introduced 3D-CNNs for spatiotemporal predictive learning. However, these networks are feed-forward and do not capture temporal consistency effectively. + +Future prediction errors of an imperfect model can be categorized by two factors: a) the “systematic errors” caused by a lack of modeling ability to the deterministic variations; b) the stochastic, inherent uncertainty of the future. We aim to minimize the first factor in this work. For the second factor, numerous methods have applied adversarial training or variational auto-encoders to video prediction, for example (Mathieu et al., 2016; Vondrick et al., 2016; Denton & Fergus, 2018; Bhattacharjee & Das, 2017; Tulyakov et al., 2018; Lu et al., 2017; Wichers et al., 2018). + +Convolutional Recurrent Networks. Our model is closely related to convolutional recurrent networks. In the ConvLSTM network (Shi et al., 2015), all state transitions are implemented with 2D convolutions. As such, the transition function is no longer permutation invariant and able to better perceive relations in a spatiotemporal neighborhood. The spatiotemporal LSTM (ST-LSTM) is characterized by delivering two memory states separately (Wang et al., 2017): memory $\mathcal { M }$ in a zigzag direction and memory $\mathcal { C }$ being passed horizontally (see Appendix A for details). In this model, $\mathcal { M }$ provides greater capability to model short-term motions, and $\mathcal { C }$ is adopted from fully-connected LSTMs (Hochreiter & Schmidhuber, 1997) to ease the vanishing gradient problem. Although the ST-LSTM performs well on video prediction benchmarks, it does not capture long-term video relations effectively. The forget gates of memory $\mathcal { C }$ tend to respond strongly to short-term features, thereby easily falling into a saturated zone (with values between 0 and 0.1) and interrupting longrange information flows. We adopt the zigzag updating route of memory $\mathcal { M }$ from the ST-LSTM, while improving the forgetting mechanism in updating the temporal memory $\mathcal { C }$ . We also increase the dimensions of memory states and take 3D-Convs as the basic operators for state transitions. + +![](images/299a5e3b87a1b0eb569a94f56cb3cd52fca6f4a79109947cd0b9f83db0b77962.jpg) +Figure 1: Three approaches to integrate 3D-Convs into recurrent networks. Blue arrows indicate data transition paths with 3D-Convs (for feed-forward features or recurrent hidden states). The diagrams are simplified for illustration, with fewer layers and RNN states than what are actually used in our experiments. The classifiers are removed when being trained for future video prediction. + +# 3 EIDETIC 3D LSTM + +This section first presents the Eidetic 3D LSTM for perceiving and memorizing both short-term and long-term representations in videos. We then discuss a scheduled multi-task learning strategy that uses predictive learning as an auxiliary self-supervised task for activity recognition. + +# 3.1 3D CONVOLUTIONS IN RECURRENT NETWORK + +An ideal predictive model relies on effective learning of video representations. RNNs and 3D-CNNs are network architectures of different mechanisms for modeling spatiotemporal data. In this work, we aim to leverage the strength of each one in a unified architecture and start the discussion with two plausible extensions of stacking 3D-Convs and RNN units. Figure 1(a) and 1(b) illustrate two hybrid baseline networks which add 3D-CNNs before or after stacked spatiotemporal LSTMs. + +However, we find that integrating the 3D-Convs outside the LSTM unit performs noticeably worse than the baseline RNN model. To this end, we propose a “deeper” integration of 3D-Convs inside the LSTM unit in order to incorporate the convolutional features into the recurrent state transition over time. Figure 1(c) shows the overall encoder-decoder architecture. In this model, a consecutive of $T$ input frames are first encoded by a few layers of 3D-Convs to obtain high-dimension feature maps. The 3D-Conv feature maps are directly fed into a novel E3D-LSTM to model the longterm spatiotemporal interaction. Finally, the E3D-LSTM hidden states are decoded by a number of stacked 3D-Conv layers to get the predicted video frames. For classification tasks, the hidden states can be directly used as the learned video representation. + +# 3.2 EIDETIC MEMORY TRANSITION + +The architecture of the proposed Eidetic 3D LSTM is illustrated in Figure 2, where the red arrows indicate short-term information flow and the blue arrows denote long-term information flow. There are 4 inputs: $\mathcal { X } _ { t }$ , the 3D-Conv feature maps from encoders or hidden states from the previous E3D + +![](images/924cf68bdd48840e3b363ec3c0c9bc6abeab5aaa00eac072e9b283bd4a42e9c3.jpg) +Figure 2: Comparison of (a) the standard memory transition approach in the Spatiotemporal LSTM and (b) the attentive memory transition approach in the Eidetic 3D LSTM. Red arrows indicate the short-term information flow. Blue arrows are the attentive memory flow, which potentially enables our model to capture the long-term relations. Cubes denote higher-dimensional hidden states and memory states. Cylinders denote higher-dimensional gates. $\odot$ is the Hadamard product. $\otimes$ is the matrix product after reshaping matrices into appropriate 2-dimensional forms. + +LSTM layer; $\mathcal { H } _ { t - 1 } ^ { k }$ , the hidden states from previous time stamp; $\mathcal { C } _ { t - 1 } ^ { k }$ , the memory states from previous time stamp; and $\mathcal { M } _ { t } ^ { k - 1 }$ , the previous spatiotemporal memory states described earlier. + +We use recurrent 3D-Convs as motion-aware perceptrons to extract short-term appearance and local motions in continuous space-time fields and store them in a small spatiotemporal volume. As such, video appearance and short-term motions can be encoded in $\mathbb { R } ^ { T \times \mathbf { \dot { H } } \times W \times C }$ tensors, in which each dimension indicates temporal depth, spatial size, and the number of feature map channels, respectively. By inflating the memory state along the time dimension, we found that the proposed E3D-LSTM becomes more capable of characterizing and memorizing local or short-term motions. + +To capture the long-term frame interactions, we improve the recurrent transition function of the memory states by proposing a new memory RECALL mechanism: + +$$ +\begin{array} { r l } & { \mathcal { R } _ { t } = \sigma \big ( W _ { x r } * \mathcal { X } _ { t } + W _ { h r } * \mathcal { H } _ { t - 1 } ^ { k } + b _ { r } \big ) } \\ & { \mathcal { Z } _ { t } = \sigma \big ( W _ { x i } * \mathcal { X } _ { t } + W _ { h i } * \mathcal { H } _ { t - 1 } ^ { k } + b _ { i } \big ) } \\ & { \mathcal { G } _ { t } = \mathrm { t a n h } \big ( W _ { x g } * \mathcal { X } _ { t } + W _ { h g } * \mathcal { H } _ { t - 1 } ^ { k } + b _ { g } \big ) } \\ & { \mathrm { R E C A L L } \big ( \mathcal { R } _ { t } , \mathcal { C } _ { t - \tau : t - 1 } ^ { k } \big ) = \mathrm { s o f t m a x } \big ( \mathcal { R } _ { t } \cdot \big ( \mathcal { C } _ { t - \tau : t - 1 } ^ { k } \big ) ^ { \sf T } \big ) \cdot \mathcal { C } _ { t - \tau : t - 1 } ^ { k } } \\ & { \mathcal { C } _ { t } ^ { k } = \mathcal { T } _ { t } \odot \mathcal { G } _ { t } + \mathrm { L a y e r N o r m } ( \mathcal { C } _ { t - 1 } ^ { k } + \mathrm { R E C A L L } \big ( \mathcal { R } _ { t } , \mathcal { C } _ { t - \tau : t - 1 } ^ { k } \big ) \big ) , } \end{array} +$$ + +where $\sigma$ is the sigmoid function, $^ *$ is the 3D-Conv operation, $\odot$ is the Hadamard product, · is the matrix product after reshaping the recall gate $\mathcal { R } _ { t }$ and memory states $ { \mathcal { C } } _ { t - \tau : t - 1 } ^ { k }$ into $\mathbb { R } ^ { T H W \times C }$ and $\mathbb { R } ^ { \tau T H W \times C }$ matrices, respectively, and $\tau$ is the number of memory states that are concatenated along the temporal dimension. Three terms are involved in computing $\mathcal { C } _ { t } ^ { k }$ . The first one $\mathcal { T } _ { t } \odot \mathcal { G } _ { t }$ encodes local video appearance and motions, where $\mathcal { T } _ { t }$ is the input gate and $\mathcal { G } _ { t }$ is the input modulation gate like standard LSTMs. The second one $\mathcal { C } _ { t - 1 } ^ { k }$ can be viewed as a short-cut connection from the previous memory state, which captures short-term changes between adjacent time stamps. In this process, the accessible memory field is fixed and limited. Therefore, we introduce the third term of memory transition function, modeling long-term video relations according to local motion and appearance (as encoded in $\mathcal { X } _ { t }$ and $\mathcal { H } _ { t - 1 } ^ { k }$ ). The RECALL function is implemented as an attentive module to compute the relationship between the encoded local patterns and the whole memory space. A set of parameterized gates $\mathcal { R } _ { t }$ , acting as memory access instructions, control where and what to attend in historical memory records. These two terms are respectively designed for shortterm and long-term video modeling. We integrate them in a unified network by applying layer normalization (Ba et al., 2016) to their element-wise sum, in order to mitigate the covariant shift and stabilize the training process, as it has been commonly used in RNNs. The hyper-parameter $\tau$ in $\mathcal { C } _ { t - \tau : t - 1 } ^ { k }$ decides how many historical memory states are attended by the recall gate $\mathcal { R } _ { t }$ . To involve more long-term relations, in most experiments, we take as the inputs of the RECALL function and do not fix $\tau$ . Whereas in particular, we enable online recognition by setting $\tau$ to 5. + +Unlike the conventional memory transition function, the RECALL function learns the size of temporal interactions. For longer sequences, this allows attending to distant states containing salient information. Our work is partially motivated by self-attention mechanisms (Lin et al., 2017; Vaswani et al., 2017). However, in our model, the attention mechanism is not applied over the output states but during the memory transitions. It is used to evoke past memories from distant time stamps for memorizing and distilling useful information from what has been perceived. We show that learning attention over previous memory states is beneficial in recalling the long-range historical context. The memory tensor $\mathcal { C } _ { t } ^ { k }$ is named eidetic $3 D$ memory and the entire unit is called E3D-LSTM. We also exploit the same RECALL method to correlate $\dot { \mathcal { M } } _ { t } ^ { 1 : k }$ along the vertical memory transition flow, but it turns out to be less helpful. With the updated memory state $\mathcal { C } _ { t } ^ { k }$ , the output hidden states are: + +$$ +\begin{array} { r l } & { \mathcal { Z } _ { t } ^ { \prime } = \sigma ( W _ { x t } ^ { \prime } \ast \mathcal { X } _ { t } + W _ { m i } \ast \mathcal { M } _ { t } ^ { k - 1 } + b _ { i } ^ { \prime } ) } \\ & { \mathcal { G } _ { t } ^ { \prime } = \operatorname { t a n h } ( W _ { x g } ^ { \prime } \ast \mathcal { X } _ { t } + W _ { m g } \ast \mathcal { M } _ { t } ^ { k - 1 } + b _ { g } ^ { \prime } ) } \\ & { \mathcal { F } _ { t } ^ { \prime } = \sigma ( W _ { x f } ^ { \prime } \ast \mathcal { X } _ { t } + W _ { m f } \ast \mathcal { M } _ { t } ^ { k - 1 } + b _ { f } ^ { \prime } ) } \\ & { \mathcal { M } _ { t } ^ { k } = \mathcal { Z } _ { t } ^ { \prime } \odot \mathcal { G } _ { t } ^ { \prime } + \mathcal { F } _ { t } ^ { \prime } \odot \mathcal { M } _ { t } ^ { k - 1 } } \\ & { \mathcal { O } _ { t } = \sigma ( W _ { x o } \ast \mathcal { X } _ { t } + W _ { h o } \ast \mathcal { H } _ { t - 1 } ^ { k } + W _ { c o } \ast \mathcal { C } _ { t } ^ { k } + W _ { m o } \ast \mathcal { M } _ { t } ^ { k } + b _ { o } ) } \\ & { \mathcal { H } _ { t } ^ { k } = \mathcal { O } _ { t } \odot \operatorname { t a n h } ( W _ { 1 \times 1 \times 1 } \ast [ \mathcal { C } _ { t } ^ { k } , \mathcal { M } _ { t } ^ { k } ] ) , } \end{array} +$$ + +where $W _ { 1 \times 1 \times 1 }$ is the $1 \times 1 \times 1$ convolutions for the transformation of the channel number. $\mathcal { T } _ { t } ^ { \prime } , \mathcal { G } _ { t } ^ { \prime }$ , and $\mathcal { F } _ { t } ^ { \prime }$ are gate structures of the spatiotemporal memory. ${ \mathcal { O } } _ { t }$ is the output gate. + +# 3.3 SELF-SUPERVISED AUXILIARY LEARNING + +For many supervised tasks such as video action recognition, there are often not enough supervisions or annotations over time for training a satisfactory RNN. As an auxiliary measure to this problem, future video prediction is considered as a promising representation learning approach that is more densely supervised over time and might extract useful features to assist video understanding. + +We consider two tasks: the pixel-level future frames prediction and another video-level classification task (early activity recognition in our case). For frames prediction, the objective function is: + +$$ +\mathcal { L } _ { \mathrm { p r e d i c t i o n } } = \left. \mathcal { X } - \widehat { \mathcal { X } } \right. _ { F } ^ { 2 } + \left. \mathcal { X } - \widehat { \mathcal { X } } \right. _ { 1 } , +$$ + +where $\widehat { \mathcal X }$ and $\mathcal { X }$ are respectively predicted and ground truth future frames. $\| \cdot \| _ { F }$ is the Frobenius norm. For early activity recognition, we make the models for these two tasks share the same network backbone in the end-to-end training using a multi-task learning objective: + +$$ +\mathcal { L } _ { \mathrm { r e c o g n i t i o n } } = \lambda \| \mathcal { X } - \widehat { \mathcal { X } } \| _ { F } ^ { 2 } + \mathcal { L } _ { \mathrm { c e } } ( \mathcal { Y } , \widehat { \mathcal { Y } } ) , +$$ + +where $\widehat { \mathcal { V } }$ and $\mathcal { V }$ are high-level predictions and corresponding ground truth classes. $\mathcal { L } _ { \mathrm { c e } }$ is the crossentropy loss for classification, and $\lambda$ is the weight factor. + +Although improving both tasks requires proper long short-term contextual representations, there is no guarantee that features learned with pixel-level supervisions will fully align with any highlevel objectives. We thus introduce a scheduled learning strategy where the objective function is gradually inclined from one task to the other in a curriculum learning manner (Bengio et al., 2009). Specifically, we apply a linear decay to $\lambda$ over the number of iterations $i$ : + +$$ +\lambda ( i ) = \operatorname* { m a x } ( \eta , \lambda ( 0 ) - \epsilon \cdot i ) , +$$ + +where $\lambda ( 0 )$ and $\eta$ are respectively maximum and minimum values of $\lambda ( i )$ , $\epsilon$ controls the decreasing speed of the role of the auxiliary task. We call this approach the Self-supervised Auxiliary Learning. + +# 4 EXPERIMENTS + +We evaluate the proposed E3D-LSTM model on two tasks: future video prediction and early activity recognition. These two tasks are of great importance with numerous applications that require effective spatiotemporal predictive models. We demonstrate that the E3D-LSTM model performs favorably against the state-of-the-art models on four challenging datasets. The source code and trained models will be made available to the public. + +![](images/75fe0e2cd497e4e80eecc0ec9f0adce2a5a55b83ce417f48d788ca7b2d8cfb24.jpg) +Figure 3: Video prediction examples on the Moving MNIST dataset. + +Table 1: Results on the Moving MNIST dataset. All models, except DFN and VPN, are trained with a comparable number of parameters. Higher SSIM or lower MSE scores indicate better results. + +
MODEL10→10COPY
SSIMMSESSIMMSE
CONVLSTM (SHI ET AL., 2015)0.71396.50.539143.2
DFN (DE BRABANDERE ET AL., 2016)0.72689.00.598153.9
CDNA (FINN ET AL., 2016)0.72884.20.671127.1
FRNN (OLIU ET AL., 2018)0.81968.40.694110.5
VPN BASELINE (KALCHBRENNER ET AL., 2017)0.87064.10.73678.0
PREDRNN(WANG ET AL., 2017)0.86956.50.74580.3
PREDRNN++(WANG ET AL.,2018B)0.88546.30.80769.9
E3D-LSTM0.91041.30.85256.8
+ +# 4.1 FUTURE VIDEO PREDICTION: MOVING MNIST + +We first evaluate the E3D-LSTM model against the state-of-the-art video prediction models on a commonly used synthetic benchmark dataset with moving digits. All experiments are conducted using TensorFlow (Abadi et al., 2016) and trained with the ADAM optimizer (Kingma & Ba, 2015) to minimize the $l _ { 1 } + l _ { 2 }$ loss over every pixel in the frame. For fair comparisons, we ensure all models to have comparable numbers of parameters, and apply the same scheduled sampling strategy (Bengio et al., 2015) in order to reduce the difficulty of training recurrent models. + +Dataset and Setup. The moving MNIST dataset is constructed by randomly sampling two digits from the original MNIST dataset and making them float and bounce at boundaries with a constant velocity and angle inside a black canvas of $6 4 \times 6 4$ pixels. The whole dataset has a fixed number of entries, 10, 000 sequences for training, $3 , 0 0 0$ for validation and 5, 000 for test. + +We stack 4 E3D-LSTMs in the architecture illustrated in Figure 1(c), leaving out 3D-CNN encoders for this task. To retain the shape of hidden states over time, the integrated 3D-Conv operators are composed of a $2 \times 5 \times 5$ (time $\times$ height $\times$ width) convolutions and a corresponding transposed convolution with the same filter size. The number of hidden state channels of each E3D-LSTM is 64. The temporal stride is set to 1 and there is one overlapping frame over consecutive time stamps. A single 3D-Conv layer is used as the decoder to map motion-aware hidden states to output frames. + +The E3D-LSTM model is evaluated against the state-of-the-art methods including the ConvLSTM network (Shi et al., 2015), DFN (De Brabandere et al., 2016), CDNA (Finn et al., 2016), VPN baseline model with CNN decoders (Kalchbrenner et al., 2017), PredRNN (Wang et al., 2017), PredRNN $^ { + + }$ (Wang et al., 2018b) and FRNN (Oliu et al., 2018). + +Main Results. Table 1 shows the performance of the evaluated models using a common setting in the literature: generating 10 future frames given the previous 10 observations (denoted as $1 0 1 0$ ). We use the per-frame structural similarity index measure (SSIM) (Wang et al., 2004) and per-frame mean squared error (MSE) for evaluation. The SSIM ranges between $- 1$ and 1, representing the similarity between the generated image and the ground truth. As shown in the second column $1 0 1 0$ ) of Table 1, our model performs well against the state-of-the-art methods in both metrics. The results show that the E3D-LSTM network is effective in modeling spatiotemporal data for video prediction. Figure 3(a) shows the qualitative comparisons in which our model predicts future frames from entangled digits better than other methods. + +Table 2: Ablation study on the Moving MNIST dataset $( 1 0 1 0 )$ ). + +
MODELSSIMMSE
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))0.85950.6
BASELINE 2: 3D-CNN ON TOP(FIGURE 1(B))0.86253.4
BASELINE 3: ：OURS (W/O 3D CONVOLUTIONS)0.89444.2
BASELINE 4: OURS (W/O MEMORY ATTENTION)0.88045.7
E3D-LSTM0.91041.3
+ +Copy Test. We evaluate the proposed model using the Copy Test setting where the task is to memorize useful information in a longer input sequence when the recurrent disturbance is present. The input clip consists of three sub-sequences, as illustrated in Figure 3(b). Seq 1 and Seq 2 are completely irrelevant, and ahead of them, another sub-sequence called prior context is given as the input, which is exactly the same as Seq 2. Frames marked by black arrows are inputs and those marked by red arrows are expected outputs. There are two training objective: a) to predict 10 future frames of Seq 1; and b) to predict 10 future frames of Seq 2. At the test time, we only evaluate the prediction result of Seq 2. The copy test evaluates the modeling capability of long-range video frame relations. A well-designed model should make precise predictions regarding Seq 2, as it has seen all frames of this sequence before. However, this task is difficult for previous LSTM networks. Because Seq 1 is completely irrelevant, the attempt of making predictions of Seq 1 can erase its memory of Seq 2. + +The results are presented in the third column (Copy) of Table 1. All baseline models suffer from the influence brought by irrelevant frames in Seq 2 and tend to gradually forget the salient information in the prior context. However, thanks to the eidetic 3D memory, our E3D-LSTM model captures the long-term video frame interactions and performs well in both metrics. A careful inspection of the attention weight shows that the E3D-LSTM model can better attend to useful historical representations across multiple time stamps. The copy test suggests that the E3D-LSTM network is capable of modeling long-range periodical motions effectively. + +Ablation Study. We conduct a series of ablation studies and summarize the results in Table 2. First, on the first two rows, we show two alternative 3D-LSTM models with 3D-Convs outside the recurrent unit, including 3D-CNN at Bottom (Figure 1(a)) and 3D-CNN on Top (Figure 1(b)). The performance drop validates the integration of 3D-Convs and RNN units via the eidetic 3D memory. Second, the third baseline method is a special case where all 3D convolutional filters in our model are reduced to 2D. The results demonstrate the effect of capturing local spatiotemporal patterns by the 3D memory within an individual recurrent state. Furthermore, the contribution of the memory attention mechanism can be isolated in the fourth baseline method. Note that all evaluated models are trained with a similar number of parameters for fair comparisons, and the performance gain comes from design options rather than increased model parameters. + +# 4.2 FUTURE VIDEO PREDICTION: KTH ACTION + +We evaluate the proposed E3D-LSTM model on video prediction of real-world datasets. + +Dataset and Setup. The KTH action dataset (Schuldt et al., 2004) contains 25 individuals performing 6 types of actions, including walking, jogging, running, boxing, hand waving and hand clapping. On average, each video clip lasts 4 seconds. We follow the experimental setup in (Villegas et al., 2017) by using person 1-16 for training and 17-25 for testing. Each frame is resized to $1 2 8 \times 1 2 8$ pixels. We employ the same E3D-LSTM network architecture detailed in Section 4.1. Models are trained to predict next 10 frames from the previous 10 observations. The prediction horizon at the test time is extended to 20 or 40 time stamps. + +Results. Table 3 shows quantitative results of the proposed model and state-of-the-art methods. Same as prior work, we use SSIM and PSNR as metrics. Consistent with the observations on the moving MNIST dataset, the E3D-LSTM model performs favorably against the state-of-the-art methods across three settings of predicting future 10 frames, 20 frames, and copy test. These empirical results demonstrate the effectiveness of the E3D-LSTM model for modeling spatiotemporal data. + +Figure 4 compares representative generated frames. We select video sequences with relatively complicated spatiotemporal variations (in both moving trajectories and human figure sizes). In the top half (predicting the next 40 frames based on 10 previous frames), E3D-LSTM predicts more accurate motion trajectories into the future, whereas $\mathrm { P r e d R N N + + }$ and ConvLSTM incorrectly predict the person moving out of the scenes. The lower half shows the copy test providing the expected outputs as prior inputs. We directly apply models, which are trained under the first setting, to this test. Without the prior context, it would be difficult to predict human motions for some cases. With prior inputs, E3D-LSTM benefits the most from its memories and responds well to rapid appearance change. In contrast, PredR $\mathrm { N N } { + } { + }$ and ConvLSTM are not able to capture useful spatiotemporal patterns from distant observations due to the lack of modeling long-term data relations. + +![](images/98c7186a972f6792a079875d57cc48e443d0a0c3c46eb79688381a3a07a6d3dc.jpg) +Figure 4: Comparisons of the generated frames on KTH. (Top) predictions of next 40 frames based on 10 previous observations. (Bottom) the copy test that requires to reproducing prior inputs. + +Table 3: Quantitative evaluation of different methods on the KTH human action test set. The metrics are averaged over the predicted frames. Higher scores indicate better prediction results. + +
MODEL10→2010→40COPY(→40)
PSNRSSIMPSNRSSIMPSNRSSIM
CONVLSTM(SHI ET AL.,2015)23.580.71222.850.63923.490.670
DFN (DE BRABANDERE ET AL., 2016)27.260.79423.010.65223.370.664
MCNET(VILLEGAS ET AL.,2017)25.950.804---1
FRNN(OLIU ET AL., 2018)26.120.77123.770.67824.000.685
PREDRNN(WANG ET AL., 2017)27.550.83924.160.70324.450.711
PREDRNN++(WANG ET AL., 2018B)28.470.86525.210.74125.900.759
E3D-LSTM29.310.87927.240.81030.590.874
+ +# 4.3 A REAL VIDEO PREDICTION APPLICATION: TRAFFIC FLOW PREDICTION + +We further evaluate our method on the TaxiBJ dataset, which contains real-world traffic flow data in consecutive heat maps. Predicting urban traffic conditions is a complex setting, as the heat maps are very noisy and we do not have any underlying or additional information that can facilitate this task. + +Table 4: Experimental results on the TaxiBJ dataset. We report MSE at every time stamp. + +
MODELFRAME1FRAME 2FRAME 3FRAME 4
ST-RESNET (ZHANG ET AL., 2017)0.6880.9391.1301.288
VPN(KALCHBRENNER ET AL., 2017)0.7441.0311.2511.444
FRNN(OLIU ET AL., 2018)0.6820.8230.9891.183
PREDRNN(WANG ET AL.,2017)0.6340.9341.0471.263
PREDRNN++(WANG ET AL., 2018B)0.6410.8550.9791.158
E3D-LSTM0.6200.7730.8880.984
+ +Dataset and Setup. The TaxiBJ dataset is collected from the chaotic real-world environment using GPS monitors of taxicabs Beijing. Each frame is a $3 2 \times 3 2 \times 2$ heat map. The last dimension denotes the entering and leaving traffic flow intensities at the same area. We split the whole dataset into a training set and a test set as described in (Zhang et al., 2017). We train the networks to predict 4 frames (the next 2 hours) from 4 observations. We use the same network architecture and training setups as the one on Moving MNIST and KTH datasets. + +![](images/1d1f26e3ee7b1adcf8951a828d333cd1d19a35b5578e98204d3503dab5694b12.jpg) +Figure 5: Prediction results on the TaxiBJ traffic flow dataset. For ease of comparison, we visualize the differences between the generated heat maps and their corresponding ground truth heat maps. + +Results. We report MSE at every time stamp in Table 4 where lower scores indicate better prediction results. We also show a prediction example in Figure 5. Furthermore, we visualize the differences between the generated heat maps and the ground truth heat maps. Overall, the E3D-LSTM model outperforms the other methods, with the lowest differences intensities in most areas. + +# 4.4 EARLY ACTIVITY RECOGNITION: SOMETHING-SOMETHING + +To validate that the E3D-LSTM model can learn high-level video representations effectively, we carry out experiments on early activity recognition. The task is to predict an activity category in a video after only observing a fraction of frames. We choose not to evaluate on the full-length video for the activity recognition task, because when a model sees the full-length video, it may make decisions solely based on the scene information, e.g. seeing only the last frame is enough to recognize many actions. As a result, the full-length video task may not align well with our previous video prediction tasks, in which the sequential tendency and causality are important. + +Dataset and Setup. The something-something dataset (Goyal et al., 2017) is a recent benchmark for activity/action recognition (https://20bn.com/datasets/something-something). We use the standard and official subset which contains 56, 769 short videos for the training set and 7, 503 videos for the validation set on 41 action categories. The video length ranges between 2 and 6 seconds with 24 fps. We adopt the early activity recognition setting (Ma et al., 2016; Zeng et al., 2017; Zhou et al., 2018), where a model predicts an action type after observing the first $2 5 \%$ or $50 \%$ frames of each video. As these actions appear in diverse scenes and involve interaction with different objects, it is challenging to predict actions even for humans (See Figure 6). There are only subtle differences between some actions in this dataset, such as “Poking a stack of [Something] so that the stack collapses” versus “Poking a stack of [Something] without the stack collapsing”, or “Pouring [Something] into [Something]” versus “Trying to pour [Something] into [Something], but missing so it spills next to it”. To make a correct prediction, a model needs to exploit spatiotemporal cues to understand the subtle differences between actions. Namely, one can evaluate the model effectiveness for high-level video tasks. Recognizing early action accurately requires predictions into future frames, which can only be achieved using an effective model based on historical observations. + +Hyper-parameters and Baselines. We use the architecture illustrated in Figure 1(c) as our model, which consists of 2 layers of 3D-CNN encoders, 4 layers of E3D-LSTMs, and 2 layers of 3D-CNN decoders. The 3D-CNN encoders take 4 consecutive $2 2 4 \times 2 2 4$ raw frames, encode them into $2 \times 5 6 \times 5 6 \times 6 4$ feature maps at each time stamp, and then feed them into E3D-LSTM. Each encoder layer has 64 filters (the filter dimensions are $2 \times 5 \times 5 )$ . We use the same hyper-parameters for E3D-LSTMs as for video prediction. The decoder layers map the output of E3D-LSTMs back to RGB space, which is an $1 \times 3$ matrix, predicting the next frame following the inputs. We train the network to predict the next 10 frames using the front $2 5 \%$ or $5 0 \%$ frames of the video. Note that we do not extend any predictive states into the future at the test time. For both training and testing, we concatenate hidden representations of the top recurrent units with respect to the last 16 input time stamps (considering the first $2 5 \%$ video snippets usually have about 20 to 30 frames), and feed them into the classifier for activity recognition. The classifier contains 2 layers of 3D-Convs with 128 filters (filter dimensions: $2 \times 3 \times 3$ , filter strides: $2 \times 2 \times 2$ ) followed by a $2 \times 2 \times 2$ pooling layer. They transform the concatenated recurrent features from $1 6 \times 5 6 \times 5 6 \times 6 4$ to $1 \times 7 \times 7 \times 1 2 8$ , then pass them to a 512-channel fully-connected layer followed by a 41-way classification. We also exploit the self-supervised auxiliary learning approach and train the model with an objective function in Equation 4. We set $\lambda ( i )$ in Equation 5 to 10 in the beginning $( i = 0$ ), and decrease it with a speed of $2 \times 1 0 ^ { - 5 }$ per iteration, lower bounded by $\eta = 0 . 1$ . + +![](images/35208ce44d8e892975f7b7f7278d7cae6d8d5d530e0543611b9a676083f43948.jpg) +Figure 6: Early activity recognition results given the first $2 5 \%$ and $5 0 \%$ frames of videos on the Something-Something validation set. The blue bars indicate making correct classifications and the red bars are incorrect results. The length of the bar denotes the confidence of the result. + +Table 5: Early activity recognition accuracy on the 41-category subset of Something-Something. + +
MODELFRONT 25%FRONT 50%
3D-CNN9.1110.30
SEPARABLE-CNN:SEPARABLE-CONV AT BOTTOM8.949.62
(2+1)D-CNN:SEPARABLE-CONV ON TOP9.0810.17
E(2+1)D-LSTM: SEPARABLE INSIDE UNITS12.4519.86
E3D-LSTM14.5922.73
+ +We evaluate the E3D-LSTM model against the state-of-the-art feed-forward 3D-Conv architectures including C3D/I3D (Diba et al., 2016; Carreira & Zisserman, 2017), Separable 3D-CNN (Xie et al., 2018; Qiu et al., 2017) and (2+1)D-CNN (Tran et al., 2018). These networks achieve the stateof-the-art results on the UCF-101 and Kinetics benchmark datasets for action recognition. For fair comparisons, we train these baseline models using similar backbones to the E3D-LSTM network. + +Results. Table 5 shows the classification accuracy of the E3D-LSTM network against the stateof-the-art feed-forward 3D-CNNs. The E3D-LSTM model performs favorably against the other methods in two settings of using the first $2 5 \%$ and $5 0 \%$ frames, showing its effectiveness in learning high-level spatiotemporal representations. Figure 6 shows two pairs of video activities that are easy to confuse, especially with such limited observations. For instance, our model correctly forecasts the collapse of books, while only a tendency of it has been shown explicitly within the first $2 5 \%$ frames. This reasoning ability comes from the integrated design of our model to capture both shortterm motions and long-term dependencies. On the other hand, as the feed-forward 3D-CNN models long-term relations by sampling and assembling, it does not perform well in finding the temporal dependencies between cause and effect. We note that Zhou et al. (2018) introduced a feed-forward CNN model and also reported early recognition results on the same dataset. It is not meaningful to compare these two methods in terms of accuracy as our model is trained only using $2 5 \% { - } 5 0 \%$ frames of a video instead of the entire video in (Zhou et al., 2018). Moreover, the two methods are trained using different backbone networks and different splits of datasets. + +Table 6: Ablation study of early activity recognition on the Something-Something dataset. + +
MODELFRONT 25%FRONT 50%
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))10.2816.05
BASELINE 2: 3D-CNN ON TOP (FIGURE 1(B))9.6314.82
BASELINE 3:OURS W/O 3D CONVOLUTIONS9.5813.92
BASELINE 4: OURS W/O MEMORY ATTENTION11.3918.84
E3D-LSTM14.5922.73
+ +Table 7: Accuracy comparisons of different training strategies on the Something-Something dataset. + +
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.7820.91
PRE-TRAINED ON THE AUXILIARY TASK14.0022.15
TRAINED ON BOTH TASKS WITH A FIXED LOSS RATIO13.5720.46
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.5922.73
+ +A number of recent studies show that separating temporal and spatial convolution operations in a 3D-CNN model leads to better results (Xie et al., 2018; Qiu et al., 2017; Tran et al., 2018). This observation is validated by our results shown in Table 5. However, it seems counter-intuitive since such separation leads to a pseudo-3D convolution, in which spatial and temporal filters are independent. Interestingly, such separation in our model leads to performance loss, suggesting the 3D convolution in the E3D-LSTM jointly captures the temporal and spatial information. + +Ablation Study. We conduct similar ablation studies as in Section 4.1 and summarize the results in Table 6. The results from the first two rows show that our deeper integration of 3D-Convs inside RNNs is helpful not only for pixel-level video prediction, but also for high-level activity recognition. The results on rows 3 and 4 show the contribution of the two important components in the proposed Eidetic 3D LSTM: a) 3D convolution features, and b) memory attention mechanism. Both components are useful and important for modeling spatiotemporal data effectively. Table 7 shows applying self-supervised training in different settings. The proposed self-supervised auxiliary learning approach performs better than other alternatives, including using video prediction models as network initialization, or training the model under these two tasks with a fixed objective function ratio. + +We enable online early activity recognition by making the classifier only depend on a concatenation of the last 5 recurrent output states. Using Equation 1, we fix the length of the attended memory states by setting $\tau$ to 5. Such settings are applied to both training and testing. Table 8 shows the experimental results. Despite the slight decrease of accuracy, it enables an online prediction. + +Table 8: Online early recognition accuracy: the classifier is built on the last 5 recurrent output states. + +
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.4918.94
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.3020.85
+ +# 5 CONCLUSION + +Spatiotemporal predictive learning has shown significant improvements in a variety of applications, such as weather forecasting, traffic flow prediction, and physical interaction simulation. Although considered to be a promising self-supervised feature learning paradigm, it seldom shows its effectiveness beyond video prediction. In this paper, we presented the E3D-LSTM model based on 3D convolutional recurrent units for this task. In this model, we integrated 3D-Convs into state transitions to perceive short-term motions and designed a memory attentive module controlled by recurrent gates to capture the long-term video frame interaction. Experimental results demonstrate that the E3D-LSTM model performs favorably against the state-of-the-art methods on video prediction and early activity recognition tasks. + +# ACKNOWLEDGMENTS + +We would like to thank anonymous reviewers for useful comments. Mingsheng Long was supported by National Natural Science Foundation of China (61772299, 71690231). + +# REFERENCES + +Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, et al. Tensorflow: Large-scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467, 2016. + +Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016. + +Samy Bengio, Oriol Vinyals, Navdeep Jaitly, and Noam Shazeer. Scheduled sampling for sequence prediction with recurrent neural networks. In NIPS, 2015. + +Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. Curriculum learning. In ICML, 2009. + +Prateep Bhattacharjee and Sukhendu Das. Temporal coherency based criteria for predicting video frames using deep multi-stage generative adversarial networks. In NIPS, 2017. + +Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, 2017. + +Bert De Brabandere, Xu Jia, Tinne Tuytelaars, and Luc Van Gool. Dynamic filter networks. In NIPS, 2016. + +Emily Denton and Rob Fergus. Stochastic video generation with a learned prior. In ICML, 2018. + +Ali Diba, Ali Mohammad Pazandeh, and Luc Van Gool. Efficient two-stream motion and appearance 3d cnns for video classification. In ECCV Workshop, 2016. + +Chelsea Finn, Ian Goodfellow, and Sergey Levine. Unsupervised learning for physical interaction through video prediction. In NIPS, 2016. + +Raghav Goyal, Samira Ebrahimi Kahou, Vincent Michalski, Joanna Materzynska, Susanne Westphal, Heuna Kim, Valentin Haenel, Ingo Fruend, Peter Yianilos, Moritz Mueller-Freitag, et al. The "something something" video database for learning and evaluating visual common sense. In ICCV, 2017. + +Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computaiton, 9(8): 1735–1780, 1997. + +Shuiwang Ji, Wei Xu, Ming Yang, and Kai Yu. 3d convolutional neural networks for human action recognition. TPAMI, 35(1):221–231, 2013. + +Nal Kalchbrenner, Aaron van den Oord, Karen Simonyan, Ivo Danihelka, Oriol Vinyals, Alex Graves, and Koray Kavukcuoglu. Video pixel networks. In ICML, 2017. + +Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015. + +Zhouhan Lin, Minwei Feng, Cicero Nogueira dos Santos, Mo Yu, Bing Xiang, Bowen Zhou, and Yoshua Bengio. A structured self-attentive sentence embedding. ICLR, 2017. + +Chaochao Lu, Michael Hirsch, and Bernhard Schölkopf. Flexible spatio-temporal networks for video prediction. In CVPR, 2017. + +Shugao Ma, Leonid Sigal, and Stan Sclaroff. Learning activity progression in lstms for activity detection and early detection. In CVPR, 2016. + +Michael Mathieu, Camille Couprie, and Yann LeCun. Deep multi-scale video prediction beyond mean square error. In ICLR, 2016. + +Marc Oliu, Javier Selva, and Sergio Escalera. Folded recurrent neural networks for future video prediction. In ECCV, 2018. + +Zhaofan Qiu, Ting Yao, and Tao Mei. Learning spatio-temporal representation with pseudo-3d residual networks. In ICCV, 2017. + +Christian Schuldt, Ivan Laptev, and Barbara Caputo. Recognizing human actions: a local svm approach. In ICPR, 2004. + +Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo. Convolutional lstm network: A machine learning approach for precipitation nowcasting. In NIPS, 2015. + +Nitish Srivastava, Elman Mansimov, and Ruslan Salakhutdinov. Unsupervised learning of video representations using lstms. In ICML, 2015. + +Ilya Sutskever, Oriol Vinyals, and Quoc V Le. Sequence to sequence learning with neural networks. In NIPS, 2014. + +Du Tran, Lubomir Bourdev, Rob Fergus, Lorenzo Torresani, and Manohar Paluri. Learning spatiotemporal features with 3d convolutional networks. In ICCV, 2015. + +Du Tran, Heng Wang, Lorenzo Torresani, Jamie Ray, Yann LeCun, and Manohar Paluri. A closer look at spatiotemporal convolutions for action recognition. In CVPR, 2018. + +Sergey Tulyakov, Ming-Yu Liu, Xiaodong Yang, and Jan Kautz. Mocogan: Decomposing motion and content for video generation. In CVPR, 2018. + +Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In NIPS, 2017. + +Ruben Villegas, Jimei Yang, Seunghoon Hong, Xunyu Lin, and Honglak Lee. Decomposing motion and content for natural video sequence prediction. In ICLR, 2017. + +Carl Vondrick, Hamed Pirsiavash, and Antonio Torralba. Generating videos with scene dynamics. In NIPS, 2016. + +Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In CVPR, 2018a. + +Yunbo Wang, Mingsheng Long, Jianmin Wang, Zhifeng Gao, and S Yu Philip. Predrnn: Recurrent neural networks for predictive learning using spatiotemporal lstms. In NIPS, 2017. + +Yunbo Wang, Zhifeng Gao, Mingsheng Long, Jianmin Wang, and Philip S Yu. Predrnn $^ { + + }$ : Towards a resolution of the deep-in-time dilemma in spatiotemporal predictive learning. In ICML, 2018b. + +Zhou Wang, A. C Bovik, H. R Sheikh, and E. P Simoncelli. Image quality assessment: from error visibility to structural similarity. TIP, 13(4):600, 2004. + +Nevan Wichers, Ruben Villegas, Dumitru Erhan, and Honglak Lee. Hierarchical long-term video prediction without supervision. In ICML, 2018. + +Saining Xie, Chen Sun, Jonathan Huang, Zhuowen Tu, and Kevin Murphy. Rethinking spatiotemporal feature learning for video understanding. In ECCV, 2018. + +Jingwei Xu, Bingbing Ni, Zefan Li, Shuo Cheng, and Xiaokang Yang. Structure preserving video prediction. In CVPR, 2018. + +Kuo-Hao Zeng, William B Shen, De-An Huang, Min Sun, and Juan Carlos Niebles. Visual forecasting by imitating dynamics in natural sequences. In ICCV, 2017. + +Junbo Zhang, Yu Zheng, and Dekang Qi. Deep spatio-temporal residual networks for citywide crowd flows prediction. In AAAI, 2017. + +Bolei Zhou, Alex Andonian, and Antonio Torralba. Temporal relational reasoning in videos. In ECCV, 2018. + +# A KEY EQUATIONS OF SPATIOTEMPORAL LSTM + +Operations inside a Spatiotemporal LSTM unit at time stamp $t$ and layer $k$ are shown as follows: + +$$ +\begin{array} { r l } & { i _ { t } = \sigma ( W _ { r s ^ { t } } + { \bar { X } } _ { t } + { \bar { W } } _ { i h } + { \bar { \mathcal { H } } } _ { i - 1 } ^ { k } + b _ { i } ) } \\ & { g _ { t } = \operatorname { t a n h } ( W _ { r s ^ { t } } + { \bar { X } } _ { t } + W _ { h , p } + { \bar { \mathcal { H } } } _ { i - 1 } ^ { k } + b _ { p } ) } \\ & { f _ { t } = \sigma ( W _ { s ^ { t } } + { \bar { X } } _ { t } + W _ { h , p } + { \bar { \mathcal { H } } } _ { i - 1 } ^ { k } + b _ { f } ) } \\ & { i _ { t } ^ { * } = \sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \bar { X } } _ { t } + W _ { m + 1 } + { \bar { X } } _ { t } ^ { k - 1 } + b _ { i } ) } \\ & { i _ { t } ^ { * } = \operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \bar { X } } _ { t } + W _ { m + 1 } + { \bar { \mathcal { H } } } _ { i } ^ { k - 1 } + b _ { i } ^ { * } ) } \\ & { f _ { t } ^ { * } = \operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \bar { X } } _ { t } + W _ { m + 1 } + { \bar { \mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\ & { c _ { t } ^ { * } = \sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \bar { X } } _ { t } + W _ { m + 1 } + { \bar { \mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\ & { M _ { t } ^ { * } = i _ { t } \odot g _ { t } + f _ { t } \odot { \bar { C } } _ { t - 1 } ^ { k } } \\ & { \boldsymbol { M } _ { t } ^ { * } = i _ { t } ^ { * } \odot g _ { t } ^ { * } + f _ { t } ^ { * } \odot { \bar { M } } _ { t } ^ { k - 1 } } \\ & o _ { s } = \sigma ( W _ { r s ^ { t } } + { \bar { X } } _ { t } + W _ { i s ^ { t } } + { \bar { M } } _ { t - 1 } ^ { k } + W _ { i s ^ { t } } + \bar \end{array} +$$ + +where $\sigma$ is the sigmoid function, $^ *$ is the convolution operator, and $\odot$ denotes the Hadamard product. There are four inputs: $\mathcal { X } _ { t }$ , the raw frame or hidden states from the previous layer; $\mathbf { \mathcal { M } } _ { t } ^ { k - 1 }$ , the previous spatiotemporal memory; $\mathcal { H } _ { t - 1 } ^ { k }$ and $\mathcal { C } _ { t - 1 } ^ { k }$ , the previous hidden states and memory states. Two sets of gate structures, including input gate $i _ { t }$ and $i _ { t } ^ { \prime }$ , forget gate $f _ { t }$ and $f _ { t } ^ { \prime }$ , as well as the output gate $o _ { t }$ , control the information flow in space-time domain. All of them can be presented by $\mathbb { R } ^ { H \times W \times C }$ dimensional tensors, where the first two dimensions are the width and height of feature maps, and the last one is the number of feature map channels. \ No newline at end of file diff --git a/parse/train/B1lKS2AqtX/B1lKS2AqtX_content_list.json b/parse/train/B1lKS2AqtX/B1lKS2AqtX_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..d8f9f1f532b6ec5d30d5d43001dc2b1ceef828e0 --- /dev/null +++ b/parse/train/B1lKS2AqtX/B1lKS2AqtX_content_list.json @@ -0,0 +1,1635 @@ +[ + { + "type": "text", + "text": "EIDETIC 3D LSTM: A MODEL FOR VIDEO PREDICTION AND BEYOND ", + "text_level": 1, + "bbox": [ + 176, + 98, + 769, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Yunbo Wang1∗, Lu Jiang2, Ming-Hsuan Yang2,3, Li-Jia $\\mathbf { L i } ^ { 4 }$ , Mingsheng Long1, Li Fei-Fei4 1Tsinghua University, 2Google AI, 3University of California, Merced, 4Stanford University ", + "bbox": [ + 184, + 167, + 815, + 199 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 236, + 544, + 251 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Spatiotemporal predictive learning, though long considered to be a promising selfsupervised feature learning method, seldom shows its effectiveness beyond future video prediction. The reason is that it is difficult to learn good representations for both short-term frame dependency and long-term high-level relations. We present a new model, Eidetic 3D LSTM (E3D-LSTM), that integrates 3D convolutions into RNNs. The encapsulated 3D-Conv makes local perceptrons of RNNs motion-aware and enables the memory cell to store better short-term features. For long-term relations, we make the present memory state interact with its historical records via a gate-controlled self-attention module. We describe this memory transition mechanism eidetic as it is able to effectively recall the stored memories across multiple time stamps even after long periods of disturbance. We first evaluate the E3D-LSTM network on widely-used future video prediction datasets and achieve the state-of-the-art performance. Then we show that the E3D-LSTM network also performs well on the early activity recognition to infer what is happening or what will happen after observing only limited frames of video. This task aligns well with video prediction in modeling action intentions and tendency. ", + "bbox": [ + 233, + 265, + 764, + 487 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 513, + 336, + 530 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "A fundamental problem in spatiotemporal predictive learning is how to effectively learn good representations for video inference or reasoning. Currently, recurrent neural networks (RNNs) remain to be the most promising models in this field, and have achieved state-of-the-art results on a number of future video prediction benchmarks (Wang et al., 2018b; Oliu et al., 2018). However, beyond frames prediction, RNN based models are less effective in learning high-level video representations or capturing long-term relations. On the other hand, recent studies demonstrate that 3D Convolutional Neural networks (3D-CNNs) surpass RNNs in learning better representations for action classification (Carreira & Zisserman, 2017; Tran et al., 2015). For instance, variants of 3D-CNNs, such as Inflated 3D-CNNs, have significantly increased action classification accuracy over the UCF 101 and Kinetics datasets. These 3D-CNN architectures have no recurrent structures but instead employ 3D convolution (3D-Conv) and 3D pooling operations to preserve temporal information of the input sequences which would be otherwise discarded in classical 2D convolution operations. ", + "bbox": [ + 174, + 540, + 825, + 707 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Motivated by the recent success of 3D-CNNs, in this paper we propose a new model for spatiotemporal predictive learning based on both recurrent modeling (for temporal dependency) and feedforward 3D-Conv modeling (for local dynamics). A plausible approach, of course, is to simply stack 3D-Convs and each RNN unit in a feed-forward way using 3D-Convs for either perceiving fine-grained features from raw videos or combining high-level representations. However, as shown in our experiments, these straightforward extensions may not outperform the baseline RNN model. We attribute these findings to that RNNs and 3D-CNNs represent two very different mechanisms for the same purpose of spatiotemporal modeling, and connecting them directly fails to exploit their complementary advantages. Therefore, it remains challenging and requires principled approaches to design an effective spatiotemporal network. ", + "bbox": [ + 174, + 714, + 825, + 853 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "To this end, we propose a new model called Eidetic 3D LSTM (E3D-LSTM) for spatiotemporal predictive learning. We introduce an eidetic 3D memory to: a) memorize local appearance and motion in a short spatiotemporal volume, and b) recall the long-range historical context by learning to attend to previous memory states. Regarding the short-term dependency, in many cases, spatiotemporal predictive modeling mainly depends on temporally nearby appearances and on-going short-term motions. All the information is encapsulated into the eidetic 3D memory cell with a short time convolution window, and used in recurrent transitions. Our experimental results show that integrating 3D-Conv deep into RNNs is effective for modeling local representations in a consecutive manner. On the other hand, for long-term interactions, which is important for predicting non-stationary or periodical videos as well as learning high-level video representations, we exploit a self-attention mechanism controlled by revised recurrent gates to recall temporally distant memory. The current memory state of E3D-LSTM is learned to attend to all previous relevant moments. Our experimental results verify that this attention mechanism is beneficial for long-term memorization. We describe this memory transition mechanism eidetic as it is able to effectively recall the stored memories across multiple time stamps even after long periods of disturbance. ", + "bbox": [ + 176, + 861, + 823, + 902 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 270 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "To the best of our knowledge, the proposed E3D-LSTM model is among the first approaches that leverage 3D-Conv in RNNs. We empirically validate it on standard spatiotemporal predictive tasks and an early activity recognition task over four benchmarks: a) on future video prediction, it achieves the best-published accuracy on three classical benchmarks; b) on early activity recognition, it outperforms the state-of-the-art action recognition methods. In addition, we show that self-supervised learning can further improve the performance of early activity recognition. We present ablation studies to verify the effectiveness of all modules in the proposed E3D-LSTM model. ", + "bbox": [ + 174, + 277, + 825, + 375 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 RELATED WORK AND PROBLEM CONTEXT ", + "text_level": 1, + "bbox": [ + 174, + 392, + 562, + 409 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Spatiotemporal Predictive Learning Models. In recent years, RNNs have been extensively used in sequence prediction and future frame prediction. Srivastava et al. (2015) extended the LSTMbased sequence to sequence model (Sutskever et al., 2014) for language modeling to learning video representations. Shi et al. (2015) proposed the convolutional LSTM by integrating convolutions into recurrent state transitions for high-dimensional sequence prediction. The convolutional LSTM model is extended by Finn et al. (2016) to predict future states of robotic environments. Villegas et al. (2017) leveraged optical flow to help capture short-term video dynamics for video prediction. Xu et al. (2018) proposed a two-stream RNN that deals with structural video content in separate streams. Kalchbrenner et al. (2017) introduced a sophisticated model that extends recurrent structures to estimate local dependencies between adjacent pixels. While this video pixel network (VPN) model is able to describe image sequences, the computational load is prohibitively high. ", + "bbox": [ + 174, + 420, + 825, + 573 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The above-mentioned recurrent models predict future frame mainly based on sequentially updated memory states. When the memory cell is refreshed, older memories will be discarded immediately. In contrast, the proposed E3D-LSTM model maintains a list of historical memory records and revokes them when necessary, thereby facilitating long-range video reasoning. While in spirit this idea is similar to the self-attention module in feed-forward networks (Vaswani et al., 2017; Wang et al., 2018a), we exploit it to correlate long-term and short-term video representations in this work. ", + "bbox": [ + 174, + 580, + 823, + 664 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Another significant difference between the above-mentioned prior work and the proposed model is that we use 3D-Convs as basic operations inside the E3D-LSTM instead of fully-connected or 2D convolution operations. We show using 3D-Convs to model recurrent state-to-state transitions can significantly improve prediction performance. This idea is motivated by recent advances in video classification (a high-level representation learning task) (Ji et al., 2013; Tran et al., 2015; Carreira & Zisserman, 2017). We note that Vondrick et al. (2016) and Tulyakov et al. (2018) also introduced 3D-CNNs for spatiotemporal predictive learning. However, these networks are feed-forward and do not capture temporal consistency effectively. ", + "bbox": [ + 174, + 670, + 825, + 782 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Future prediction errors of an imperfect model can be categorized by two factors: a) the “systematic errors” caused by a lack of modeling ability to the deterministic variations; b) the stochastic, inherent uncertainty of the future. We aim to minimize the first factor in this work. For the second factor, numerous methods have applied adversarial training or variational auto-encoders to video prediction, for example (Mathieu et al., 2016; Vondrick et al., 2016; Denton & Fergus, 2018; Bhattacharjee & Das, 2017; Tulyakov et al., 2018; Lu et al., 2017; Wichers et al., 2018). ", + "bbox": [ + 174, + 789, + 825, + 872 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Convolutional Recurrent Networks. Our model is closely related to convolutional recurrent networks. In the ConvLSTM network (Shi et al., 2015), all state transitions are implemented with 2D convolutions. As such, the transition function is no longer permutation invariant and able to better perceive relations in a spatiotemporal neighborhood. The spatiotemporal LSTM (ST-LSTM) is characterized by delivering two memory states separately (Wang et al., 2017): memory $\\mathcal { M }$ in a zigzag direction and memory $\\mathcal { C }$ being passed horizontally (see Appendix A for details). In this model, $\\mathcal { M }$ provides greater capability to model short-term motions, and $\\mathcal { C }$ is adopted from fully-connected LSTMs (Hochreiter & Schmidhuber, 1997) to ease the vanishing gradient problem. Although the ST-LSTM performs well on video prediction benchmarks, it does not capture long-term video relations effectively. The forget gates of memory $\\mathcal { C }$ tend to respond strongly to short-term features, thereby easily falling into a saturated zone (with values between 0 and 0.1) and interrupting longrange information flows. We adopt the zigzag updating route of memory $\\mathcal { M }$ from the ST-LSTM, while improving the forgetting mechanism in updating the temporal memory $\\mathcal { C }$ . We also increase the dimensions of memory states and take 3D-Convs as the basic operators for state transitions. ", + "bbox": [ + 176, + 882, + 823, + 924 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/299a5e3b87a1b0eb569a94f56cb3cd52fca6f4a79109947cd0b9f83db0b77962.jpg", + "image_caption": [ + "Figure 1: Three approaches to integrate 3D-Convs into recurrent networks. Blue arrows indicate data transition paths with 3D-Convs (for feed-forward features or recurrent hidden states). The diagrams are simplified for illustration, with fewer layers and RNN states than what are actually used in our experiments. The classifiers are removed when being trained for future video prediction. " + ], + "image_footnote": [], + "bbox": [ + 202, + 88, + 795, + 267 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 348, + 825, + 501 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 EIDETIC 3D LSTM ", + "text_level": 1, + "bbox": [ + 176, + 522, + 367, + 539 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "This section first presents the Eidetic 3D LSTM for perceiving and memorizing both short-term and long-term representations in videos. We then discuss a scheduled multi-task learning strategy that uses predictive learning as an auxiliary self-supervised task for activity recognition. ", + "bbox": [ + 176, + 553, + 823, + 595 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1 3D CONVOLUTIONS IN RECURRENT NETWORK ", + "text_level": 1, + "bbox": [ + 174, + 613, + 540, + 627 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "An ideal predictive model relies on effective learning of video representations. RNNs and 3D-CNNs are network architectures of different mechanisms for modeling spatiotemporal data. In this work, we aim to leverage the strength of each one in a unified architecture and start the discussion with two plausible extensions of stacking 3D-Convs and RNN units. Figure 1(a) and 1(b) illustrate two hybrid baseline networks which add 3D-CNNs before or after stacked spatiotemporal LSTMs. ", + "bbox": [ + 174, + 637, + 823, + 707 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "However, we find that integrating the 3D-Convs outside the LSTM unit performs noticeably worse than the baseline RNN model. To this end, we propose a “deeper” integration of 3D-Convs inside the LSTM unit in order to incorporate the convolutional features into the recurrent state transition over time. Figure 1(c) shows the overall encoder-decoder architecture. In this model, a consecutive of $T$ input frames are first encoded by a few layers of 3D-Convs to obtain high-dimension feature maps. The 3D-Conv feature maps are directly fed into a novel E3D-LSTM to model the longterm spatiotemporal interaction. Finally, the E3D-LSTM hidden states are decoded by a number of stacked 3D-Conv layers to get the predicted video frames. For classification tasks, the hidden states can be directly used as the learned video representation. ", + "bbox": [ + 174, + 713, + 825, + 839 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.2 EIDETIC MEMORY TRANSITION ", + "text_level": 1, + "bbox": [ + 176, + 857, + 434, + 871 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The architecture of the proposed Eidetic 3D LSTM is illustrated in Figure 2, where the red arrows indicate short-term information flow and the blue arrows denote long-term information flow. There are 4 inputs: $\\mathcal { X } _ { t }$ , the 3D-Conv feature maps from encoders or hidden states from the previous E3D", + "bbox": [ + 176, + 882, + 823, + 924 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/924cf68bdd48840e3b363ec3c0c9bc6abeab5aaa00eac072e9b283bd4a42e9c3.jpg", + "image_caption": [ + "Figure 2: Comparison of (a) the standard memory transition approach in the Spatiotemporal LSTM and (b) the attentive memory transition approach in the Eidetic 3D LSTM. Red arrows indicate the short-term information flow. Blue arrows are the attentive memory flow, which potentially enables our model to capture the long-term relations. Cubes denote higher-dimensional hidden states and memory states. Cylinders denote higher-dimensional gates. $\\odot$ is the Hadamard product. $\\otimes$ is the matrix product after reshaping matrices into appropriate 2-dimensional forms. " + ], + "image_footnote": [], + "bbox": [ + 179, + 87, + 820, + 256 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "LSTM layer; $\\mathcal { H } _ { t - 1 } ^ { k }$ , the hidden states from previous time stamp; $\\mathcal { C } _ { t - 1 } ^ { k }$ , the memory states from previous time stamp; and $\\mathcal { M } _ { t } ^ { k - 1 }$ , the previous spatiotemporal memory states described earlier. ", + "bbox": [ + 173, + 363, + 823, + 396 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We use recurrent 3D-Convs as motion-aware perceptrons to extract short-term appearance and local motions in continuous space-time fields and store them in a small spatiotemporal volume. As such, video appearance and short-term motions can be encoded in $\\mathbb { R } ^ { T \\times \\mathbf { \\dot { H } } \\times W \\times C }$ tensors, in which each dimension indicates temporal depth, spatial size, and the number of feature map channels, respectively. By inflating the memory state along the time dimension, we found that the proposed E3D-LSTM becomes more capable of characterizing and memorizing local or short-term motions. ", + "bbox": [ + 173, + 401, + 825, + 486 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "To capture the long-term frame interactions, we improve the recurrent transition function of the memory states by proposing a new memory RECALL mechanism: ", + "bbox": [ + 173, + 492, + 823, + 521 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/cc8a6dd29230d8644c0af099c67e13e7ab5002774b305cd32d7091f51650b924.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathcal { R } _ { t } = \\sigma \\big ( W _ { x r } * \\mathcal { X } _ { t } + W _ { h r } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { r } \\big ) } \\\\ & { \\mathcal { Z } _ { t } = \\sigma \\big ( W _ { x i } * \\mathcal { X } _ { t } + W _ { h i } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { i } \\big ) } \\\\ & { \\mathcal { G } _ { t } = \\mathrm { t a n h } \\big ( W _ { x g } * \\mathcal { X } _ { t } + W _ { h g } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { g } \\big ) } \\\\ & { \\mathrm { R E C A L L } \\big ( \\mathcal { R } _ { t } , \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) = \\mathrm { s o f t m a x } \\big ( \\mathcal { R } _ { t } \\cdot \\big ( \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) ^ { \\sf T } \\big ) \\cdot \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } } \\\\ & { \\mathcal { C } _ { t } ^ { k } = \\mathcal { T } _ { t } \\odot \\mathcal { G } _ { t } + \\mathrm { L a y e r N o r m } ( \\mathcal { C } _ { t - 1 } ^ { k } + \\mathrm { R E C A L L } \\big ( \\mathcal { R } _ { t } , \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) \\big ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 294, + 521, + 697, + 619 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $\\sigma$ is the sigmoid function, $^ *$ is the 3D-Conv operation, $\\odot$ is the Hadamard product, · is the matrix product after reshaping the recall gate $\\mathcal { R } _ { t }$ and memory states $ { \\mathcal { C } } _ { t - \\tau : t - 1 } ^ { k }$ into $\\mathbb { R } ^ { T H W \\times C }$ and $\\mathbb { R } ^ { \\tau T H W \\times C }$ matrices, respectively, and $\\tau$ is the number of memory states that are concatenated along the temporal dimension. Three terms are involved in computing $\\mathcal { C } _ { t } ^ { k }$ . The first one $\\mathcal { T } _ { t } \\odot \\mathcal { G } _ { t }$ encodes local video appearance and motions, where $\\mathcal { T } _ { t }$ is the input gate and $\\mathcal { G } _ { t }$ is the input modulation gate like standard LSTMs. The second one $\\mathcal { C } _ { t - 1 } ^ { k }$ can be viewed as a short-cut connection from the previous memory state, which captures short-term changes between adjacent time stamps. In this process, the accessible memory field is fixed and limited. Therefore, we introduce the third term of memory transition function, modeling long-term video relations according to local motion and appearance (as encoded in $\\mathcal { X } _ { t }$ and $\\mathcal { H } _ { t - 1 } ^ { k }$ ). The RECALL function is implemented as an attentive module to compute the relationship between the encoded local patterns and the whole memory space. A set of parameterized gates $\\mathcal { R } _ { t }$ , acting as memory access instructions, control where and what to attend in historical memory records. These two terms are respectively designed for shortterm and long-term video modeling. We integrate them in a unified network by applying layer normalization (Ba et al., 2016) to their element-wise sum, in order to mitigate the covariant shift and stabilize the training process, as it has been commonly used in RNNs. The hyper-parameter $\\tau$ in $\\mathcal { C } _ { t - \\tau : t - 1 } ^ { k }$ decides how many historical memory states are attended by the recall gate $\\mathcal { R } _ { t }$ . To involve more long-term relations, in most experiments, we take as the inputs of the RECALL function and do not fix $\\tau$ . Whereas in particular, we enable online recognition by setting $\\tau$ to 5. ", + "bbox": [ + 174, + 619, + 825, + 888 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Unlike the conventional memory transition function, the RECALL function learns the size of temporal interactions. For longer sequences, this allows attending to distant states containing salient information. Our work is partially motivated by self-attention mechanisms (Lin et al., 2017; Vaswani et al., 2017). However, in our model, the attention mechanism is not applied over the output states but during the memory transitions. It is used to evoke past memories from distant time stamps for memorizing and distilling useful information from what has been perceived. We show that learning attention over previous memory states is beneficial in recalling the long-range historical context. The memory tensor $\\mathcal { C } _ { t } ^ { k }$ is named eidetic $3 D$ memory and the entire unit is called E3D-LSTM. We also exploit the same RECALL method to correlate $\\dot { \\mathcal { M } } _ { t } ^ { 1 : k }$ along the vertical memory transition flow, but it turns out to be less helpful. With the updated memory state $\\mathcal { C } _ { t } ^ { k }$ , the output hidden states are: ", + "bbox": [ + 174, + 895, + 820, + 924 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 103, + 826, + 215 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/6ab8c63f91cb312c7867acfaf47c85bd45b3c17901997deb11247fec972aea80.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathcal { Z } _ { t } ^ { \\prime } = \\sigma ( W _ { x t } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m i } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { i } ^ { \\prime } ) } \\\\ & { \\mathcal { G } _ { t } ^ { \\prime } = \\operatorname { t a n h } ( W _ { x g } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m g } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { g } ^ { \\prime } ) } \\\\ & { \\mathcal { F } _ { t } ^ { \\prime } = \\sigma ( W _ { x f } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m f } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { f } ^ { \\prime } ) } \\\\ & { \\mathcal { M } _ { t } ^ { k } = \\mathcal { Z } _ { t } ^ { \\prime } \\odot \\mathcal { G } _ { t } ^ { \\prime } + \\mathcal { F } _ { t } ^ { \\prime } \\odot \\mathcal { M } _ { t } ^ { k - 1 } } \\\\ & { \\mathcal { O } _ { t } = \\sigma ( W _ { x o } \\ast \\mathcal { X } _ { t } + W _ { h o } \\ast \\mathcal { H } _ { t - 1 } ^ { k } + W _ { c o } \\ast \\mathcal { C } _ { t } ^ { k } + W _ { m o } \\ast \\mathcal { M } _ { t } ^ { k } + b _ { o } ) } \\\\ & { \\mathcal { H } _ { t } ^ { k } = \\mathcal { O } _ { t } \\odot \\operatorname { t a n h } ( W _ { 1 \\times 1 \\times 1 } \\ast [ \\mathcal { C } _ { t } ^ { k } , \\mathcal { M } _ { t } ^ { k } ] ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 292, + 223, + 704, + 342 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $W _ { 1 \\times 1 \\times 1 }$ is the $1 \\times 1 \\times 1$ convolutions for the transformation of the channel number. $\\mathcal { T } _ { t } ^ { \\prime } , \\mathcal { G } _ { t } ^ { \\prime }$ , and $\\mathcal { F } _ { t } ^ { \\prime }$ are gate structures of the spatiotemporal memory. ${ \\mathcal { O } } _ { t }$ is the output gate. ", + "bbox": [ + 173, + 349, + 823, + 378 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.3 SELF-SUPERVISED AUXILIARY LEARNING ", + "text_level": 1, + "bbox": [ + 173, + 393, + 508, + 409 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "For many supervised tasks such as video action recognition, there are often not enough supervisions or annotations over time for training a satisfactory RNN. As an auxiliary measure to this problem, future video prediction is considered as a promising representation learning approach that is more densely supervised over time and might extract useful features to assist video understanding. ", + "bbox": [ + 173, + 416, + 825, + 473 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We consider two tasks: the pixel-level future frames prediction and another video-level classification task (early activity recognition in our case). For frames prediction, the objective function is: ", + "bbox": [ + 171, + 479, + 821, + 508 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/2d4ccea0e282e449f34613f89d6e0c059e054bc9d8f0ff8760176a95d65e9773.jpg", + "text": "$$\n\\mathcal { L } _ { \\mathrm { p r e d i c t i o n } } = \\left. \\mathcal { X } - \\widehat { \\mathcal { X } } \\right. _ { F } ^ { 2 } + \\left. \\mathcal { X } - \\widehat { \\mathcal { X } } \\right. _ { 1 } ,\n$$", + "text_format": "latex", + "bbox": [ + 370, + 517, + 625, + 540 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\widehat { \\mathcal X }$ and $\\mathcal { X }$ are respectively predicted and ground truth future frames. $\\| \\cdot \\| _ { F }$ is the Frobenius norm. For early activity recognition, we make the models for these two tasks share the same network backbone in the end-to-end training using a multi-task learning objective: ", + "bbox": [ + 174, + 551, + 825, + 594 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/97393a6b19c8f6c6673661b998ff85dca2c5f3e950dd6f6b130560afe0000ef0.jpg", + "text": "$$\n\\mathcal { L } _ { \\mathrm { r e c o g n i t i o n } } = \\lambda \\| \\mathcal { X } - \\widehat { \\mathcal { X } } \\| _ { F } ^ { 2 } + \\mathcal { L } _ { \\mathrm { c e } } ( \\mathcal { Y } , \\widehat { \\mathcal { Y } } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 366, + 602, + 630, + 626 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\widehat { \\mathcal { V } }$ and $\\mathcal { V }$ are high-level predictions and corresponding ground truth classes. $\\mathcal { L } _ { \\mathrm { c e } }$ is the crossentropy loss for classification, and $\\lambda$ is the weight factor. ", + "bbox": [ + 173, + 637, + 823, + 666 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Although improving both tasks requires proper long short-term contextual representations, there is no guarantee that features learned with pixel-level supervisions will fully align with any highlevel objectives. We thus introduce a scheduled learning strategy where the objective function is gradually inclined from one task to the other in a curriculum learning manner (Bengio et al., 2009). Specifically, we apply a linear decay to $\\lambda$ over the number of iterations $i$ : ", + "bbox": [ + 173, + 672, + 825, + 742 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/bcc72aced1b2b417896806195f79b0152fc53b21baafe890274c448fe18812aa.jpg", + "text": "$$\n\\lambda ( i ) = \\operatorname* { m a x } ( \\eta , \\lambda ( 0 ) - \\epsilon \\cdot i ) ,\n$$", + "text_format": "latex", + "bbox": [ + 401, + 752, + 591, + 768 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\lambda ( 0 )$ and $\\eta$ are respectively maximum and minimum values of $\\lambda ( i )$ , $\\epsilon$ controls the decreasing speed of the role of the auxiliary task. We call this approach the Self-supervised Auxiliary Learning. ", + "bbox": [ + 171, + 777, + 825, + 808 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4 EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 176, + 825, + 326, + 840 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We evaluate the proposed E3D-LSTM model on two tasks: future video prediction and early activity recognition. These two tasks are of great importance with numerous applications that require effective spatiotemporal predictive models. We demonstrate that the E3D-LSTM model performs favorably against the state-of-the-art models on four challenging datasets. The source code and trained models will be made available to the public. ", + "bbox": [ + 174, + 854, + 825, + 925 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/75fe0e2cd497e4e80eecc0ec9f0adce2a5a55b83ce417f48d788ca7b2d8cfb24.jpg", + "image_caption": [ + "Figure 3: Video prediction examples on the Moving MNIST dataset. " + ], + "image_footnote": [], + "bbox": [ + 179, + 88, + 815, + 258 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/b36c04a632d21350cb932109d5719c85a57dff9d56130a6fe44cfa0680abd2e9.jpg", + "table_caption": [ + "Table 1: Results on the Moving MNIST dataset. All models, except DFN and VPN, are trained with a comparable number of parameters. Higher SSIM or lower MSE scores indicate better results. " + ], + "table_footnote": [], + "table_body": "
MODEL10→10COPY
SSIMMSESSIMMSE
CONVLSTM (SHI ET AL., 2015)0.71396.50.539143.2
DFN (DE BRABANDERE ET AL., 2016)0.72689.00.598153.9
CDNA (FINN ET AL., 2016)0.72884.20.671127.1
FRNN (OLIU ET AL., 2018)0.81968.40.694110.5
VPN BASELINE (KALCHBRENNER ET AL., 2017)0.87064.10.73678.0
PREDRNN(WANG ET AL., 2017)0.86956.50.74580.3
PREDRNN++(WANG ET AL.,2018B)0.88546.30.80769.9
E3D-LSTM0.91041.30.85256.8
", + "bbox": [ + 220, + 328, + 771, + 477 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.1 FUTURE VIDEO PREDICTION: MOVING MNIST ", + "text_level": 1, + "bbox": [ + 173, + 492, + 545, + 507 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We first evaluate the E3D-LSTM model against the state-of-the-art video prediction models on a commonly used synthetic benchmark dataset with moving digits. All experiments are conducted using TensorFlow (Abadi et al., 2016) and trained with the ADAM optimizer (Kingma & Ba, 2015) to minimize the $l _ { 1 } + l _ { 2 }$ loss over every pixel in the frame. For fair comparisons, we ensure all models to have comparable numbers of parameters, and apply the same scheduled sampling strategy (Bengio et al., 2015) in order to reduce the difficulty of training recurrent models. ", + "bbox": [ + 173, + 515, + 825, + 598 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Dataset and Setup. The moving MNIST dataset is constructed by randomly sampling two digits from the original MNIST dataset and making them float and bounce at boundaries with a constant velocity and angle inside a black canvas of $6 4 \\times 6 4$ pixels. The whole dataset has a fixed number of entries, 10, 000 sequences for training, $3 , 0 0 0$ for validation and 5, 000 for test. ", + "bbox": [ + 174, + 608, + 825, + 664 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We stack 4 E3D-LSTMs in the architecture illustrated in Figure 1(c), leaving out 3D-CNN encoders for this task. To retain the shape of hidden states over time, the integrated 3D-Conv operators are composed of a $2 \\times 5 \\times 5$ (time $\\times$ height $\\times$ width) convolutions and a corresponding transposed convolution with the same filter size. The number of hidden state channels of each E3D-LSTM is 64. The temporal stride is set to 1 and there is one overlapping frame over consecutive time stamps. A single 3D-Conv layer is used as the decoder to map motion-aware hidden states to output frames. ", + "bbox": [ + 174, + 670, + 825, + 755 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The E3D-LSTM model is evaluated against the state-of-the-art methods including the ConvLSTM network (Shi et al., 2015), DFN (De Brabandere et al., 2016), CDNA (Finn et al., 2016), VPN baseline model with CNN decoders (Kalchbrenner et al., 2017), PredRNN (Wang et al., 2017), PredRNN $^ { + + }$ (Wang et al., 2018b) and FRNN (Oliu et al., 2018). ", + "bbox": [ + 174, + 761, + 825, + 818 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Main Results. Table 1 shows the performance of the evaluated models using a common setting in the literature: generating 10 future frames given the previous 10 observations (denoted as $1 0 1 0$ ). We use the per-frame structural similarity index measure (SSIM) (Wang et al., 2004) and per-frame mean squared error (MSE) for evaluation. The SSIM ranges between $- 1$ and 1, representing the similarity between the generated image and the ground truth. As shown in the second column $1 0 1 0$ ) of Table 1, our model performs well against the state-of-the-art methods in both metrics. The results show that the E3D-LSTM network is effective in modeling spatiotemporal data for video prediction. Figure 3(a) shows the qualitative comparisons in which our model predicts future frames from entangled digits better than other methods. ", + "bbox": [ + 174, + 825, + 825, + 924 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/ede790b7a09d830d8f26501b21e136e414d2008da6a3d89c48ecc554332e378f.jpg", + "table_caption": [ + "Table 2: Ablation study on the Moving MNIST dataset $( 1 0 1 0 )$ ). " + ], + "table_footnote": [], + "table_body": "
MODELSSIMMSE
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))0.85950.6
BASELINE 2: 3D-CNN ON TOP(FIGURE 1(B))0.86253.4
BASELINE 3: ：OURS (W/O 3D CONVOLUTIONS)0.89444.2
BASELINE 4: OURS (W/O MEMORY ATTENTION)0.88045.7
E3D-LSTM0.91041.3
", + "bbox": [ + 271, + 94, + 720, + 193 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 205, + 821, + 233 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Copy Test. We evaluate the proposed model using the Copy Test setting where the task is to memorize useful information in a longer input sequence when the recurrent disturbance is present. The input clip consists of three sub-sequences, as illustrated in Figure 3(b). Seq 1 and Seq 2 are completely irrelevant, and ahead of them, another sub-sequence called prior context is given as the input, which is exactly the same as Seq 2. Frames marked by black arrows are inputs and those marked by red arrows are expected outputs. There are two training objective: a) to predict 10 future frames of Seq 1; and b) to predict 10 future frames of Seq 2. At the test time, we only evaluate the prediction result of Seq 2. The copy test evaluates the modeling capability of long-range video frame relations. A well-designed model should make precise predictions regarding Seq 2, as it has seen all frames of this sequence before. However, this task is difficult for previous LSTM networks. Because Seq 1 is completely irrelevant, the attempt of making predictions of Seq 1 can erase its memory of Seq 2. ", + "bbox": [ + 174, + 243, + 825, + 396 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "The results are presented in the third column (Copy) of Table 1. All baseline models suffer from the influence brought by irrelevant frames in Seq 2 and tend to gradually forget the salient information in the prior context. However, thanks to the eidetic 3D memory, our E3D-LSTM model captures the long-term video frame interactions and performs well in both metrics. A careful inspection of the attention weight shows that the E3D-LSTM model can better attend to useful historical representations across multiple time stamps. The copy test suggests that the E3D-LSTM network is capable of modeling long-range periodical motions effectively. ", + "bbox": [ + 174, + 404, + 825, + 501 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Ablation Study. We conduct a series of ablation studies and summarize the results in Table 2. First, on the first two rows, we show two alternative 3D-LSTM models with 3D-Convs outside the recurrent unit, including 3D-CNN at Bottom (Figure 1(a)) and 3D-CNN on Top (Figure 1(b)). The performance drop validates the integration of 3D-Convs and RNN units via the eidetic 3D memory. Second, the third baseline method is a special case where all 3D convolutional filters in our model are reduced to 2D. The results demonstrate the effect of capturing local spatiotemporal patterns by the 3D memory within an individual recurrent state. Furthermore, the contribution of the memory attention mechanism can be isolated in the fourth baseline method. Note that all evaluated models are trained with a similar number of parameters for fair comparisons, and the performance gain comes from design options rather than increased model parameters. ", + "bbox": [ + 174, + 511, + 825, + 648 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.2 FUTURE VIDEO PREDICTION: KTH ACTION ", + "text_level": 1, + "bbox": [ + 174, + 664, + 521, + 679 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We evaluate the proposed E3D-LSTM model on video prediction of real-world datasets. ", + "bbox": [ + 173, + 686, + 748, + 702 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Dataset and Setup. The KTH action dataset (Schuldt et al., 2004) contains 25 individuals performing 6 types of actions, including walking, jogging, running, boxing, hand waving and hand clapping. On average, each video clip lasts 4 seconds. We follow the experimental setup in (Villegas et al., 2017) by using person 1-16 for training and 17-25 for testing. Each frame is resized to $1 2 8 \\times 1 2 8$ pixels. We employ the same E3D-LSTM network architecture detailed in Section 4.1. Models are trained to predict next 10 frames from the previous 10 observations. The prediction horizon at the test time is extended to 20 or 40 time stamps. ", + "bbox": [ + 174, + 710, + 825, + 809 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Results. Table 3 shows quantitative results of the proposed model and state-of-the-art methods. Same as prior work, we use SSIM and PSNR as metrics. Consistent with the observations on the moving MNIST dataset, the E3D-LSTM model performs favorably against the state-of-the-art methods across three settings of predicting future 10 frames, 20 frames, and copy test. These empirical results demonstrate the effectiveness of the E3D-LSTM model for modeling spatiotemporal data. ", + "bbox": [ + 174, + 819, + 823, + 888 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Figure 4 compares representative generated frames. We select video sequences with relatively complicated spatiotemporal variations (in both moving trajectories and human figure sizes). In the top half (predicting the next 40 frames based on 10 previous frames), E3D-LSTM predicts more accurate motion trajectories into the future, whereas $\\mathrm { P r e d R N N + + }$ and ConvLSTM incorrectly predict the person moving out of the scenes. The lower half shows the copy test providing the expected outputs as prior inputs. We directly apply models, which are trained under the first setting, to this test. Without the prior context, it would be difficult to predict human motions for some cases. With prior inputs, E3D-LSTM benefits the most from its memories and responds well to rapid appearance change. In contrast, PredR $\\mathrm { N N } { + } { + }$ and ConvLSTM are not able to capture useful spatiotemporal patterns from distant observations due to the lack of modeling long-term data relations. ", + "bbox": [ + 173, + 895, + 821, + 924 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/98c7186a972f6792a079875d57cc48e443d0a0c3c46eb79688381a3a07a6d3dc.jpg", + "image_caption": [ + "Figure 4: Comparisons of the generated frames on KTH. (Top) predictions of next 40 frames based on 10 previous observations. (Bottom) the copy test that requires to reproducing prior inputs. " + ], + "image_footnote": [], + "bbox": [ + 205, + 83, + 790, + 313 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/5596cb4faae1b7850e31d5f1d402066dbe1313a867a1a91cfd0ad3f5395ed28c.jpg", + "table_caption": [ + "Table 3: Quantitative evaluation of different methods on the KTH human action test set. The metrics are averaged over the predicted frames. Higher scores indicate better prediction results. " + ], + "table_footnote": [], + "table_body": "
MODEL10→2010→40COPY(→40)
PSNRSSIMPSNRSSIMPSNRSSIM
CONVLSTM(SHI ET AL.,2015)23.580.71222.850.63923.490.670
DFN (DE BRABANDERE ET AL., 2016)27.260.79423.010.65223.370.664
MCNET(VILLEGAS ET AL.,2017)25.950.804---1
FRNN(OLIU ET AL., 2018)26.120.77123.770.67824.000.685
PREDRNN(WANG ET AL., 2017)27.550.83924.160.70324.450.711
PREDRNN++(WANG ET AL., 2018B)28.470.86525.210.74125.900.759
E3D-LSTM29.310.87927.240.81030.590.874
", + "bbox": [ + 184, + 401, + 800, + 537 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 553, + 825, + 665 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.3 A REAL VIDEO PREDICTION APPLICATION: TRAFFIC FLOW PREDICTION ", + "text_level": 1, + "bbox": [ + 178, + 679, + 720, + 694 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We further evaluate our method on the TaxiBJ dataset, which contains real-world traffic flow data in consecutive heat maps. Predicting urban traffic conditions is a complex setting, as the heat maps are very noisy and we do not have any underlying or additional information that can facilitate this task. ", + "bbox": [ + 176, + 702, + 820, + 744 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/b0f35d6c49511747d612440df0457840d2556622aa1e59af1992b77cc1e6b086.jpg", + "table_caption": [ + "Table 4: Experimental results on the TaxiBJ dataset. We report MSE at every time stamp. " + ], + "table_footnote": [], + "table_body": "
MODELFRAME1FRAME 2FRAME 3FRAME 4
ST-RESNET (ZHANG ET AL., 2017)0.6880.9391.1301.288
VPN(KALCHBRENNER ET AL., 2017)0.7441.0311.2511.444
FRNN(OLIU ET AL., 2018)0.6820.8230.9891.183
PREDRNN(WANG ET AL.,2017)0.6340.9341.0471.263
PREDRNN++(WANG ET AL., 2018B)0.6410.8550.9791.158
E3D-LSTM0.6200.7730.8880.984
", + "bbox": [ + 209, + 773, + 777, + 886 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Dataset and Setup. The TaxiBJ dataset is collected from the chaotic real-world environment using GPS monitors of taxicabs Beijing. Each frame is a $3 2 \\times 3 2 \\times 2$ heat map. The last dimension denotes the entering and leaving traffic flow intensities at the same area. We split the whole dataset into a training set and a test set as described in (Zhang et al., 2017). We train the networks to predict 4 frames (the next 2 hours) from 4 observations. We use the same network architecture and training setups as the one on Moving MNIST and KTH datasets. ", + "bbox": [ + 174, + 895, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/1d1f26e3ee7b1adcf8951a828d333cd1d19a35b5578e98204d3503dab5694b12.jpg", + "image_caption": [ + "Figure 5: Prediction results on the TaxiBJ traffic flow dataset. For ease of comparison, we visualize the differences between the generated heat maps and their corresponding ground truth heat maps. " + ], + "image_footnote": [], + "bbox": [ + 173, + 83, + 823, + 196 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 251, + 825, + 306 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Results. We report MSE at every time stamp in Table 4 where lower scores indicate better prediction results. We also show a prediction example in Figure 5. Furthermore, we visualize the differences between the generated heat maps and the ground truth heat maps. Overall, the E3D-LSTM model outperforms the other methods, with the lowest differences intensities in most areas. ", + "bbox": [ + 174, + 315, + 825, + 371 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.4 EARLY ACTIVITY RECOGNITION: SOMETHING-SOMETHING ", + "text_level": 1, + "bbox": [ + 174, + 385, + 629, + 398 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "To validate that the E3D-LSTM model can learn high-level video representations effectively, we carry out experiments on early activity recognition. The task is to predict an activity category in a video after only observing a fraction of frames. We choose not to evaluate on the full-length video for the activity recognition task, because when a model sees the full-length video, it may make decisions solely based on the scene information, e.g. seeing only the last frame is enough to recognize many actions. As a result, the full-length video task may not align well with our previous video prediction tasks, in which the sequential tendency and causality are important. ", + "bbox": [ + 174, + 406, + 825, + 503 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Dataset and Setup. The something-something dataset (Goyal et al., 2017) is a recent benchmark for activity/action recognition (https://20bn.com/datasets/something-something). We use the standard and official subset which contains 56, 769 short videos for the training set and 7, 503 videos for the validation set on 41 action categories. The video length ranges between 2 and 6 seconds with 24 fps. We adopt the early activity recognition setting (Ma et al., 2016; Zeng et al., 2017; Zhou et al., 2018), where a model predicts an action type after observing the first $2 5 \\%$ or $50 \\%$ frames of each video. As these actions appear in diverse scenes and involve interaction with different objects, it is challenging to predict actions even for humans (See Figure 6). There are only subtle differences between some actions in this dataset, such as “Poking a stack of [Something] so that the stack collapses” versus “Poking a stack of [Something] without the stack collapsing”, or “Pouring [Something] into [Something]” versus “Trying to pour [Something] into [Something], but missing so it spills next to it”. To make a correct prediction, a model needs to exploit spatiotemporal cues to understand the subtle differences between actions. Namely, one can evaluate the model effectiveness for high-level video tasks. Recognizing early action accurately requires predictions into future frames, which can only be achieved using an effective model based on historical observations. ", + "bbox": [ + 173, + 512, + 825, + 720 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Hyper-parameters and Baselines. We use the architecture illustrated in Figure 1(c) as our model, which consists of 2 layers of 3D-CNN encoders, 4 layers of E3D-LSTMs, and 2 layers of 3D-CNN decoders. The 3D-CNN encoders take 4 consecutive $2 2 4 \\times 2 2 4$ raw frames, encode them into $2 \\times 5 6 \\times 5 6 \\times 6 4$ feature maps at each time stamp, and then feed them into E3D-LSTM. Each encoder layer has 64 filters (the filter dimensions are $2 \\times 5 \\times 5 )$ . We use the same hyper-parameters for E3D-LSTMs as for video prediction. The decoder layers map the output of E3D-LSTMs back to RGB space, which is an $1 \\times 3$ matrix, predicting the next frame following the inputs. We train the network to predict the next 10 frames using the front $2 5 \\%$ or $5 0 \\%$ frames of the video. Note that we do not extend any predictive states into the future at the test time. For both training and testing, we concatenate hidden representations of the top recurrent units with respect to the last 16 input time stamps (considering the first $2 5 \\%$ video snippets usually have about 20 to 30 frames), and feed them into the classifier for activity recognition. The classifier contains 2 layers of 3D-Convs with 128 filters (filter dimensions: $2 \\times 3 \\times 3$ , filter strides: $2 \\times 2 \\times 2$ ) followed by a $2 \\times 2 \\times 2$ pooling layer. They transform the concatenated recurrent features from $1 6 \\times 5 6 \\times 5 6 \\times 6 4$ to $1 \\times 7 \\times 7 \\times 1 2 8$ , then pass them to a 512-channel fully-connected layer followed by a 41-way classification. We also exploit the self-supervised auxiliary learning approach and train the model with an objective function in Equation 4. We set $\\lambda ( i )$ in Equation 5 to 10 in the beginning $( i = 0$ ), and decrease it with a speed of $2 \\times 1 0 ^ { - 5 }$ per iteration, lower bounded by $\\eta = 0 . 1$ . ", + "bbox": [ + 174, + 729, + 825, + 924 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/35208ce44d8e892975f7b7f7278d7cae6d8d5d530e0543611b9a676083f43948.jpg", + "image_caption": [ + "Figure 6: Early activity recognition results given the first $2 5 \\%$ and $5 0 \\%$ frames of videos on the Something-Something validation set. The blue bars indicate making correct classifications and the red bars are incorrect results. The length of the bar denotes the confidence of the result. " + ], + "image_footnote": [], + "bbox": [ + 236, + 82, + 759, + 388 + ], + "page_idx": 9 + }, + { + "type": "table", + "img_path": "images/9e461803e8c6ed2be0a39e8947a22ad2f9b1a416b85cd6ec06d0277054a13bba.jpg", + "table_caption": [ + "Table 5: Early activity recognition accuracy on the 41-category subset of Something-Something. " + ], + "table_footnote": [], + "table_body": "
MODELFRONT 25%FRONT 50%
3D-CNN9.1110.30
SEPARABLE-CNN:SEPARABLE-CONV AT BOTTOM8.949.62
(2+1)D-CNN:SEPARABLE-CONV ON TOP9.0810.17
E(2+1)D-LSTM: SEPARABLE INSIDE UNITS12.4519.86
E3D-LSTM14.5922.73
", + "bbox": [ + 230, + 477, + 759, + 575 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 588, + 821, + 643 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We evaluate the E3D-LSTM model against the state-of-the-art feed-forward 3D-Conv architectures including C3D/I3D (Diba et al., 2016; Carreira & Zisserman, 2017), Separable 3D-CNN (Xie et al., 2018; Qiu et al., 2017) and (2+1)D-CNN (Tran et al., 2018). These networks achieve the stateof-the-art results on the UCF-101 and Kinetics benchmark datasets for action recognition. For fair comparisons, we train these baseline models using similar backbones to the E3D-LSTM network. ", + "bbox": [ + 173, + 650, + 823, + 720 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Results. Table 5 shows the classification accuracy of the E3D-LSTM network against the stateof-the-art feed-forward 3D-CNNs. The E3D-LSTM model performs favorably against the other methods in two settings of using the first $2 5 \\%$ and $5 0 \\%$ frames, showing its effectiveness in learning high-level spatiotemporal representations. Figure 6 shows two pairs of video activities that are easy to confuse, especially with such limited observations. For instance, our model correctly forecasts the collapse of books, while only a tendency of it has been shown explicitly within the first $2 5 \\%$ frames. This reasoning ability comes from the integrated design of our model to capture both shortterm motions and long-term dependencies. On the other hand, as the feed-forward 3D-CNN models long-term relations by sampling and assembling, it does not perform well in finding the temporal dependencies between cause and effect. We note that Zhou et al. (2018) introduced a feed-forward CNN model and also reported early recognition results on the same dataset. It is not meaningful to compare these two methods in terms of accuracy as our model is trained only using $2 5 \\% { - } 5 0 \\%$ frames of a video instead of the entire video in (Zhou et al., 2018). Moreover, the two methods are trained using different backbone networks and different splits of datasets. ", + "bbox": [ + 174, + 729, + 825, + 924 + ], + "page_idx": 9 + }, + { + "type": "table", + "img_path": "images/88b60c346776649f137d3c2dcfd89d74f5d0f5409dc8aefdf84b527efbdf1393.jpg", + "table_caption": [ + "Table 6: Ablation study of early activity recognition on the Something-Something dataset. " + ], + "table_footnote": [], + "table_body": "
MODELFRONT 25%FRONT 50%
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))10.2816.05
BASELINE 2: 3D-CNN ON TOP (FIGURE 1(B))9.6314.82
BASELINE 3:OURS W/O 3D CONVOLUTIONS9.5813.92
BASELINE 4: OURS W/O MEMORY ATTENTION11.3918.84
E3D-LSTM14.5922.73
", + "bbox": [ + 230, + 98, + 761, + 196 + ], + "page_idx": 10 + }, + { + "type": "table", + "img_path": "images/904c177dfafb176ae79b496bde706bfa003edbd9c6867587e3c84cc82fd32d69.jpg", + "table_caption": [ + "Table 7: Accuracy comparisons of different training strategies on the Something-Something dataset. " + ], + "table_footnote": [], + "table_body": "
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.7820.91
PRE-TRAINED ON THE AUXILIARY TASK14.0022.15
TRAINED ON BOTH TASKS WITH A FIXED LOSS RATIO13.5720.46
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.5922.73
", + "bbox": [ + 194, + 228, + 795, + 314 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "A number of recent studies show that separating temporal and spatial convolution operations in a 3D-CNN model leads to better results (Xie et al., 2018; Qiu et al., 2017; Tran et al., 2018). This observation is validated by our results shown in Table 5. However, it seems counter-intuitive since such separation leads to a pseudo-3D convolution, in which spatial and temporal filters are independent. Interestingly, such separation in our model leads to performance loss, suggesting the 3D convolution in the E3D-LSTM jointly captures the temporal and spatial information. ", + "bbox": [ + 173, + 323, + 825, + 407 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ablation Study. We conduct similar ablation studies as in Section 4.1 and summarize the results in Table 6. The results from the first two rows show that our deeper integration of 3D-Convs inside RNNs is helpful not only for pixel-level video prediction, but also for high-level activity recognition. The results on rows 3 and 4 show the contribution of the two important components in the proposed Eidetic 3D LSTM: a) 3D convolution features, and b) memory attention mechanism. Both components are useful and important for modeling spatiotemporal data effectively. Table 7 shows applying self-supervised training in different settings. The proposed self-supervised auxiliary learning approach performs better than other alternatives, including using video prediction models as network initialization, or training the model under these two tasks with a fixed objective function ratio. ", + "bbox": [ + 174, + 414, + 825, + 540 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "We enable online early activity recognition by making the classifier only depend on a concatenation of the last 5 recurrent output states. Using Equation 1, we fix the length of the attended memory states by setting $\\tau$ to 5. Such settings are applied to both training and testing. Table 8 shows the experimental results. Despite the slight decrease of accuracy, it enables an online prediction. ", + "bbox": [ + 174, + 546, + 823, + 603 + ], + "page_idx": 10 + }, + { + "type": "table", + "img_path": "images/a1031335ebb524ddee38071c7f4e97fb72adf594501edd2998e9ce9c95972e19.jpg", + "table_caption": [ + "Table 8: Online early recognition accuracy: the classifier is built on the last 5 recurrent output states. " + ], + "table_footnote": [], + "table_body": "
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.4918.94
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.3020.85
", + "bbox": [ + 194, + 633, + 795, + 689 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "5 CONCLUSION ", + "text_level": 1, + "bbox": [ + 174, + 704, + 320, + 719 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Spatiotemporal predictive learning has shown significant improvements in a variety of applications, such as weather forecasting, traffic flow prediction, and physical interaction simulation. Although considered to be a promising self-supervised feature learning paradigm, it seldom shows its effectiveness beyond video prediction. In this paper, we presented the E3D-LSTM model based on 3D convolutional recurrent units for this task. In this model, we integrated 3D-Convs into state transitions to perceive short-term motions and designed a memory attentive module controlled by recurrent gates to capture the long-term video frame interaction. Experimental results demonstrate that the E3D-LSTM model performs favorably against the state-of-the-art methods on video prediction and early activity recognition tasks. ", + "bbox": [ + 174, + 731, + 825, + 856 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "ACKNOWLEDGMENTS ", + "text_level": 1, + "bbox": [ + 176, + 872, + 326, + 885 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "We would like to thank anonymous reviewers for useful comments. Mingsheng Long was supported by National Natural Science Foundation of China (61772299, 71690231). ", + "bbox": [ + 174, + 896, + 823, + 922 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 174, + 103, + 287, + 117 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, et al. Tensorflow: Large-scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467, 2016. ", + "bbox": [ + 174, + 126, + 823, + 169 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016. ", + "bbox": [ + 171, + 176, + 823, + 207 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Samy Bengio, Oriol Vinyals, Navdeep Jaitly, and Noam Shazeer. Scheduled sampling for sequence prediction with recurrent neural networks. In NIPS, 2015. ", + "bbox": [ + 171, + 214, + 823, + 244 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. Curriculum learning. In ICML, 2009. ", + "bbox": [ + 173, + 252, + 825, + 281 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Prateep Bhattacharjee and Sukhendu Das. Temporal coherency based criteria for predicting video frames using deep multi-stage generative adversarial networks. In NIPS, 2017. ", + "bbox": [ + 171, + 290, + 823, + 320 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, 2017. ", + "bbox": [ + 171, + 328, + 823, + 357 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Bert De Brabandere, Xu Jia, Tinne Tuytelaars, and Luc Van Gool. Dynamic filter networks. In NIPS, 2016. ", + "bbox": [ + 173, + 366, + 823, + 395 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Emily Denton and Rob Fergus. Stochastic video generation with a learned prior. In ICML, 2018. ", + "bbox": [ + 173, + 404, + 810, + 420 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Ali Diba, Ali Mohammad Pazandeh, and Luc Van Gool. Efficient two-stream motion and appearance 3d cnns for video classification. In ECCV Workshop, 2016. ", + "bbox": [ + 169, + 428, + 823, + 458 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Chelsea Finn, Ian Goodfellow, and Sergey Levine. Unsupervised learning for physical interaction through video prediction. In NIPS, 2016. ", + "bbox": [ + 173, + 465, + 823, + 494 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Raghav Goyal, Samira Ebrahimi Kahou, Vincent Michalski, Joanna Materzynska, Susanne Westphal, Heuna Kim, Valentin Haenel, Ingo Fruend, Peter Yianilos, Moritz Mueller-Freitag, et al. The \"something something\" video database for learning and evaluating visual common sense. In ICCV, 2017. ", + "bbox": [ + 174, + 502, + 825, + 560 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computaiton, 9(8): 1735–1780, 1997. ", + "bbox": [ + 171, + 568, + 823, + 597 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Shuiwang Ji, Wei Xu, Ming Yang, and Kai Yu. 3d convolutional neural networks for human action recognition. TPAMI, 35(1):221–231, 2013. ", + "bbox": [ + 174, + 606, + 821, + 636 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Nal Kalchbrenner, Aaron van den Oord, Karen Simonyan, Ivo Danihelka, Oriol Vinyals, Alex Graves, and Koray Kavukcuoglu. Video pixel networks. In ICML, 2017. ", + "bbox": [ + 174, + 643, + 820, + 674 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015. ", + "bbox": [ + 173, + 683, + 797, + 698 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Zhouhan Lin, Minwei Feng, Cicero Nogueira dos Santos, Mo Yu, Bing Xiang, Bowen Zhou, and Yoshua Bengio. A structured self-attentive sentence embedding. ICLR, 2017. ", + "bbox": [ + 173, + 705, + 820, + 736 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Chaochao Lu, Michael Hirsch, and Bernhard Schölkopf. Flexible spatio-temporal networks for video prediction. In CVPR, 2017. ", + "bbox": [ + 173, + 743, + 821, + 773 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Shugao Ma, Leonid Sigal, and Stan Sclaroff. Learning activity progression in lstms for activity detection and early detection. In CVPR, 2016. ", + "bbox": [ + 173, + 781, + 821, + 810 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Michael Mathieu, Camille Couprie, and Yann LeCun. Deep multi-scale video prediction beyond mean square error. In ICLR, 2016. ", + "bbox": [ + 173, + 819, + 821, + 848 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Marc Oliu, Javier Selva, and Sergio Escalera. Folded recurrent neural networks for future video prediction. In ECCV, 2018. ", + "bbox": [ + 173, + 857, + 821, + 886 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Zhaofan Qiu, Ting Yao, and Tao Mei. Learning spatio-temporal representation with pseudo-3d residual networks. In ICCV, 2017. ", + "bbox": [ + 173, + 895, + 821, + 924 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Christian Schuldt, Ivan Laptev, and Barbara Caputo. Recognizing human actions: a local svm approach. In ICPR, 2004. ", + "bbox": [ + 171, + 103, + 823, + 132 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo. Convolutional lstm network: A machine learning approach for precipitation nowcasting. In NIPS, 2015. ", + "bbox": [ + 178, + 140, + 823, + 184 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Nitish Srivastava, Elman Mansimov, and Ruslan Salakhutdinov. Unsupervised learning of video representations using lstms. In ICML, 2015. ", + "bbox": [ + 173, + 193, + 823, + 222 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Ilya Sutskever, Oriol Vinyals, and Quoc V Le. Sequence to sequence learning with neural networks. In NIPS, 2014. ", + "bbox": [ + 174, + 229, + 823, + 260 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Du Tran, Lubomir Bourdev, Rob Fergus, Lorenzo Torresani, and Manohar Paluri. Learning spatiotemporal features with 3d convolutional networks. In ICCV, 2015. ", + "bbox": [ + 173, + 268, + 823, + 297 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Du Tran, Heng Wang, Lorenzo Torresani, Jamie Ray, Yann LeCun, and Manohar Paluri. A closer look at spatiotemporal convolutions for action recognition. In CVPR, 2018. ", + "bbox": [ + 173, + 305, + 823, + 335 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Sergey Tulyakov, Ming-Yu Liu, Xiaodong Yang, and Jan Kautz. Mocogan: Decomposing motion and content for video generation. In CVPR, 2018. ", + "bbox": [ + 171, + 343, + 823, + 372 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In NIPS, 2017. ", + "bbox": [ + 174, + 381, + 821, + 410 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Ruben Villegas, Jimei Yang, Seunghoon Hong, Xunyu Lin, and Honglak Lee. Decomposing motion and content for natural video sequence prediction. In ICLR, 2017. ", + "bbox": [ + 173, + 419, + 821, + 448 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Carl Vondrick, Hamed Pirsiavash, and Antonio Torralba. Generating videos with scene dynamics. In NIPS, 2016. ", + "bbox": [ + 174, + 457, + 821, + 486 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In CVPR, 2018a. ", + "bbox": [ + 174, + 494, + 820, + 523 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Yunbo Wang, Mingsheng Long, Jianmin Wang, Zhifeng Gao, and S Yu Philip. Predrnn: Recurrent neural networks for predictive learning using spatiotemporal lstms. In NIPS, 2017. ", + "bbox": [ + 173, + 531, + 820, + 561 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Yunbo Wang, Zhifeng Gao, Mingsheng Long, Jianmin Wang, and Philip S Yu. Predrnn $^ { + + }$ : Towards a resolution of the deep-in-time dilemma in spatiotemporal predictive learning. In ICML, 2018b. ", + "bbox": [ + 173, + 569, + 821, + 599 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Zhou Wang, A. C Bovik, H. R Sheikh, and E. P Simoncelli. Image quality assessment: from error visibility to structural similarity. TIP, 13(4):600, 2004. ", + "bbox": [ + 173, + 607, + 820, + 637 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Nevan Wichers, Ruben Villegas, Dumitru Erhan, and Honglak Lee. Hierarchical long-term video prediction without supervision. In ICML, 2018. ", + "bbox": [ + 173, + 645, + 821, + 674 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Saining Xie, Chen Sun, Jonathan Huang, Zhuowen Tu, and Kevin Murphy. Rethinking spatiotemporal feature learning for video understanding. In ECCV, 2018. ", + "bbox": [ + 173, + 683, + 821, + 712 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Jingwei Xu, Bingbing Ni, Zefan Li, Shuo Cheng, and Xiaokang Yang. Structure preserving video prediction. In CVPR, 2018. ", + "bbox": [ + 173, + 720, + 821, + 750 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Kuo-Hao Zeng, William B Shen, De-An Huang, Min Sun, and Juan Carlos Niebles. Visual forecasting by imitating dynamics in natural sequences. In ICCV, 2017. ", + "bbox": [ + 173, + 758, + 821, + 787 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Junbo Zhang, Yu Zheng, and Dekang Qi. Deep spatio-temporal residual networks for citywide crowd flows prediction. In AAAI, 2017. ", + "bbox": [ + 171, + 796, + 821, + 825 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Bolei Zhou, Alex Andonian, and Antonio Torralba. Temporal relational reasoning in videos. In ECCV, 2018. ", + "bbox": [ + 173, + 834, + 825, + 863 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "A KEY EQUATIONS OF SPATIOTEMPORAL LSTM ", + "text_level": 1, + "bbox": [ + 174, + 101, + 596, + 119 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Operations inside a Spatiotemporal LSTM unit at time stamp $t$ and layer $k$ are shown as follows: ", + "bbox": [ + 166, + 132, + 805, + 148 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/5fb1fa23342e02bab61b95df231fb0eb6c3c5c86c246e293837978b0f9bbcd9a.jpg", + "text": "$$\n\\begin{array} { r l } & { i _ { t } = \\sigma ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + { \\bar { W } } _ { i h } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { i } ) } \\\\ & { g _ { t } = \\operatorname { t a n h } ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + W _ { h , p } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { p } ) } \\\\ & { f _ { t } = \\sigma ( W _ { s ^ { t } } + { \\bar { X } } _ { t } + W _ { h , p } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { f } ) } \\\\ & { i _ { t } ^ { * } = \\sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { X } } _ { t } ^ { k - 1 } + b _ { i } ) } \\\\ & { i _ { t } ^ { * } = \\operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { i } ^ { k - 1 } + b _ { i } ^ { * } ) } \\\\ & { f _ { t } ^ { * } = \\operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\\\ & { c _ { t } ^ { * } = \\sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\\\ & { M _ { t } ^ { * } = i _ { t } \\odot g _ { t } + f _ { t } \\odot { \\bar { C } } _ { t - 1 } ^ { k } } \\\\ & { \\boldsymbol { M } _ { t } ^ { * } = i _ { t } ^ { * } \\odot g _ { t } ^ { * } + f _ { t } ^ { * } \\odot { \\bar { M } } _ { t } ^ { k - 1 } } \\\\ & o _ { s } = \\sigma ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + W _ { i s ^ { t } } + { \\bar { M } } _ { t - 1 } ^ { k } + W _ { i s ^ { t } } + \\bar \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 290, + 155, + 700, + 349 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "where $\\sigma$ is the sigmoid function, $^ *$ is the convolution operator, and $\\odot$ denotes the Hadamard product. There are four inputs: $\\mathcal { X } _ { t }$ , the raw frame or hidden states from the previous layer; $\\mathbf { \\mathcal { M } } _ { t } ^ { k - 1 }$ , the previous spatiotemporal memory; $\\mathcal { H } _ { t - 1 } ^ { k }$ and $\\mathcal { C } _ { t - 1 } ^ { k }$ , the previous hidden states and memory states. Two sets of gate structures, including input gate $i _ { t }$ and $i _ { t } ^ { \\prime }$ , forget gate $f _ { t }$ and $f _ { t } ^ { \\prime }$ , as well as the output gate $o _ { t }$ , control the information flow in space-time domain. All of them can be presented by $\\mathbb { R } ^ { H \\times W \\times C }$ dimensional tensors, where the first two dimensions are the width and height of feature maps, and the last one is the number of feature map channels. ", + "bbox": [ + 173, + 353, + 825, + 454 + ], + "page_idx": 13 + } +] \ No newline at end of file diff --git a/parse/train/B1lKS2AqtX/B1lKS2AqtX_middle.json b/parse/train/B1lKS2AqtX/B1lKS2AqtX_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..52677d2f3598e59b3ca5667690d722964331e38c --- /dev/null +++ b/parse/train/B1lKS2AqtX/B1lKS2AqtX_middle.json @@ -0,0 +1,35262 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 471, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 257, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 257, + 96 + ], + "score": 1.0, + "content": "EIDETIC 3D LSTM:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 473, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 473, + 117 + ], + "score": 1.0, + "content": "A MODEL FOR VIDEO PREDICTION AND BEYOND", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 133, + 499, + 158 + ], + "lines": [ + { + "bbox": [ + 114, + 133, + 500, + 148 + ], + "spans": [ + { + "bbox": [ + 114, + 133, + 350, + 148 + ], + "score": 1.0, + "content": "Yunbo Wang1∗, Lu Jiang2, Ming-Hsuan Yang2,3, Li-Jia", + "type": "text" + }, + { + "bbox": [ + 351, + 135, + 366, + 146 + ], + "score": 0.86, + "content": "\\mathbf { L i } ^ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 133, + 500, + 148 + ], + "score": 1.0, + "content": ", Mingsheng Long1, Li Fei-Fei4", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 145, + 479, + 160 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 479, + 160 + ], + "score": 1.0, + "content": "1Tsinghua University, 2Google AI, 3University of California, Merced, 4Stanford University", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 278, + 187, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 276, + 186, + 335, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 186, + 335, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 210, + 468, + 386 + ], + "lines": [ + { + "bbox": [ + 142, + 211, + 469, + 223 + ], + "spans": [ + { + "bbox": [ + 142, + 211, + 469, + 223 + ], + "score": 1.0, + "content": "Spatiotemporal predictive learning, though long considered to be a promising self-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 222, + 469, + 234 + ], + "spans": [ + { + "bbox": [ + 141, + 222, + 469, + 234 + ], + "score": 1.0, + "content": "supervised feature learning method, seldom shows its effectiveness beyond future", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 232, + 469, + 245 + ], + "spans": [ + { + "bbox": [ + 141, + 232, + 469, + 245 + ], + "score": 1.0, + "content": "video prediction. The reason is that it is difficult to learn good representations", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 243, + 470, + 257 + ], + "spans": [ + { + "bbox": [ + 141, + 243, + 470, + 257 + ], + "score": 1.0, + "content": "for both short-term frame dependency and long-term high-level relations. We", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 254, + 469, + 267 + ], + "spans": [ + { + "bbox": [ + 141, + 254, + 469, + 267 + ], + "score": 1.0, + "content": "present a new model, Eidetic 3D LSTM (E3D-LSTM), that integrates 3D convo-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 266, + 469, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 266, + 469, + 277 + ], + "score": 1.0, + "content": "lutions into RNNs. The encapsulated 3D-Conv makes local perceptrons of RNNs", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 277, + 470, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 470, + 288 + ], + "score": 1.0, + "content": "motion-aware and enables the memory cell to store better short-term features. For", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 288, + 469, + 299 + ], + "spans": [ + { + "bbox": [ + 141, + 288, + 469, + 299 + ], + "score": 1.0, + "content": "long-term relations, we make the present memory state interact with its histori-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 298, + 469, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 298, + 469, + 311 + ], + "score": 1.0, + "content": "cal records via a gate-controlled self-attention module. We describe this memory", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 309, + 469, + 321 + ], + "spans": [ + { + "bbox": [ + 142, + 309, + 469, + 321 + ], + "score": 1.0, + "content": "transition mechanism eidetic as it is able to effectively recall the stored memo-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 320, + 469, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 320, + 469, + 332 + ], + "score": 1.0, + "content": "ries across multiple time stamps even after long periods of disturbance. We first", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "score": 1.0, + "content": "evaluate the E3D-LSTM network on widely-used future video prediction datasets", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 342, + 469, + 354 + ], + "spans": [ + { + "bbox": [ + 142, + 342, + 469, + 354 + ], + "score": 1.0, + "content": "and achieve the state-of-the-art performance. Then we show that the E3D-LSTM", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 353, + 469, + 365 + ], + "spans": [ + { + "bbox": [ + 142, + 353, + 469, + 365 + ], + "score": 1.0, + "content": "network also performs well on the early activity recognition to infer what is hap-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 364, + 469, + 376 + ], + "spans": [ + { + "bbox": [ + 141, + 364, + 469, + 376 + ], + "score": 1.0, + "content": "pening or what will happen after observing only limited frames of video. This", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 375, + 469, + 388 + ], + "spans": [ + { + "bbox": [ + 141, + 375, + 469, + 388 + ], + "score": 1.0, + "content": "task aligns well with video prediction in modeling action intentions and tendency.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 108, + 407, + 206, + 420 + ], + "lines": [ + { + "bbox": [ + 105, + 406, + 208, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 208, + 423 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 428, + 505, + 560 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "A fundamental problem in spatiotemporal predictive learning is how to effectively learn good repre-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 104, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "sentations for video inference or reasoning. Currently, recurrent neural networks (RNNs) remain to", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 451, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 463 + ], + "score": 1.0, + "content": "be the most promising models in this field, and have achieved state-of-the-art results on a number of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "score": 1.0, + "content": "future video prediction benchmarks (Wang et al., 2018b; Oliu et al., 2018). However, beyond frames", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 472, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 487 + ], + "score": 1.0, + "content": "prediction, RNN based models are less effective in learning high-level video representations or cap-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 485, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 505, + 496 + ], + "score": 1.0, + "content": "turing long-term relations. On the other hand, recent studies demonstrate that 3D Convolutional", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "Neural networks (3D-CNNs) surpass RNNs in learning better representations for action classifica-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "score": 1.0, + "content": "tion (Carreira & Zisserman, 2017; Tran et al., 2015). For instance, variants of 3D-CNNs, such as", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 517, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 505, + 529 + ], + "score": 1.0, + "content": "Inflated 3D-CNNs, have significantly increased action classification accuracy over the UCF 101 and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 540 + ], + "score": 1.0, + "content": "Kinetics datasets. These 3D-CNN architectures have no recurrent structures but instead employ 3D", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "convolution (3D-Conv) and 3D pooling operations to preserve temporal information of the input", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 550, + 454, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 454, + 562 + ], + "score": 1.0, + "content": "sequences which would be otherwise discarded in classical 2D convolution operations.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 505, + 676 + ], + "lines": [ + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "score": 1.0, + "content": "Motivated by the recent success of 3D-CNNs, in this paper we propose a new model for spatiotem-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 577, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 591 + ], + "score": 1.0, + "content": "poral predictive learning based on both recurrent modeling (for temporal dependency) and feed-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 587, + 505, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 505, + 602 + ], + "score": 1.0, + "content": "forward 3D-Conv modeling (for local dynamics). A plausible approach, of course, is to simply", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 597, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 104, + 597, + 505, + 614 + ], + "score": 1.0, + "content": "stack 3D-Convs and each RNN unit in a feed-forward way using 3D-Convs for either perceiving", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "fine-grained features from raw videos or combining high-level representations. However, as shown", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "in our experiments, these straightforward extensions may not outperform the baseline RNN model.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 631, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 505, + 645 + ], + "score": 1.0, + "content": "We attribute these findings to that RNNs and 3D-CNNs represent two very different mechanisms", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "for the same purpose of spatiotemporal modeling, and connecting them directly fails to exploit their", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "complementary advantages. Therefore, it remains challenging and requires principled approaches", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 666, + 292, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 292, + 677 + ], + "score": 1.0, + "content": "to design an effective spatiotemporal network.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 682, + 504, + 715 + ], + "lines": [ + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "score": 1.0, + "content": "To this end, we propose a new model called Eidetic 3D LSTM (E3D-LSTM) for spatiotemporal", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 693, + 506, + 706 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 506, + 706 + ], + "score": 1.0, + "content": "predictive learning. We introduce an eidetic 3D memory to: a) memorize local appearance and mo-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 703, + 506, + 718 + ], + "spans": [ + { + "bbox": [ + 105, + 703, + 506, + 718 + ], + "score": 1.0, + "content": "tion in a short spatiotemporal volume, and b) recall the long-range historical context by learning", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 116, + 722, + 489, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 720, + 491, + 734 + ], + "spans": [ + { + "bbox": [ + 119, + 720, + 491, + 734 + ], + "score": 1.0, + "content": "∗Corresponding author: wangyb15@mails.tsinghua.edu.cn. Work done in part at Google AI.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 471, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 257, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 257, + 96 + ], + "score": 1.0, + "content": "EIDETIC 3D LSTM:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 473, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 473, + 117 + ], + "score": 1.0, + "content": "A MODEL FOR VIDEO PREDICTION AND BEYOND", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 133, + 499, + 158 + ], + "lines": [ + { + "bbox": [ + 114, + 133, + 500, + 148 + ], + "spans": [ + { + "bbox": [ + 114, + 133, + 350, + 148 + ], + "score": 1.0, + "content": "Yunbo Wang1∗, Lu Jiang2, Ming-Hsuan Yang2,3, Li-Jia", + "type": "text" + }, + { + "bbox": [ + 351, + 135, + 366, + 146 + ], + "score": 0.86, + "content": "\\mathbf { L i } ^ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 133, + 500, + 148 + ], + "score": 1.0, + "content": ", Mingsheng Long1, Li Fei-Fei4", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 145, + 479, + 160 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 479, + 160 + ], + "score": 1.0, + "content": "1Tsinghua University, 2Google AI, 3University of California, Merced, 4Stanford University", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 111, + 133, + 500, + 160 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 187, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 276, + 186, + 335, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 186, + 335, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 210, + 468, + 386 + ], + "lines": [ + { + "bbox": [ + 142, + 211, + 469, + 223 + ], + "spans": [ + { + "bbox": [ + 142, + 211, + 469, + 223 + ], + "score": 1.0, + "content": "Spatiotemporal predictive learning, though long considered to be a promising self-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 222, + 469, + 234 + ], + "spans": [ + { + "bbox": [ + 141, + 222, + 469, + 234 + ], + "score": 1.0, + "content": "supervised feature learning method, seldom shows its effectiveness beyond future", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 232, + 469, + 245 + ], + "spans": [ + { + "bbox": [ + 141, + 232, + 469, + 245 + ], + "score": 1.0, + "content": "video prediction. The reason is that it is difficult to learn good representations", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 243, + 470, + 257 + ], + "spans": [ + { + "bbox": [ + 141, + 243, + 470, + 257 + ], + "score": 1.0, + "content": "for both short-term frame dependency and long-term high-level relations. We", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 254, + 469, + 267 + ], + "spans": [ + { + "bbox": [ + 141, + 254, + 469, + 267 + ], + "score": 1.0, + "content": "present a new model, Eidetic 3D LSTM (E3D-LSTM), that integrates 3D convo-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 266, + 469, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 266, + 469, + 277 + ], + "score": 1.0, + "content": "lutions into RNNs. The encapsulated 3D-Conv makes local perceptrons of RNNs", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 277, + 470, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 470, + 288 + ], + "score": 1.0, + "content": "motion-aware and enables the memory cell to store better short-term features. For", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 288, + 469, + 299 + ], + "spans": [ + { + "bbox": [ + 141, + 288, + 469, + 299 + ], + "score": 1.0, + "content": "long-term relations, we make the present memory state interact with its histori-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 298, + 469, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 298, + 469, + 311 + ], + "score": 1.0, + "content": "cal records via a gate-controlled self-attention module. We describe this memory", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 309, + 469, + 321 + ], + "spans": [ + { + "bbox": [ + 142, + 309, + 469, + 321 + ], + "score": 1.0, + "content": "transition mechanism eidetic as it is able to effectively recall the stored memo-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 320, + 469, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 320, + 469, + 332 + ], + "score": 1.0, + "content": "ries across multiple time stamps even after long periods of disturbance. We first", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "score": 1.0, + "content": "evaluate the E3D-LSTM network on widely-used future video prediction datasets", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 342, + 469, + 354 + ], + "spans": [ + { + "bbox": [ + 142, + 342, + 469, + 354 + ], + "score": 1.0, + "content": "and achieve the state-of-the-art performance. Then we show that the E3D-LSTM", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 353, + 469, + 365 + ], + "spans": [ + { + "bbox": [ + 142, + 353, + 469, + 365 + ], + "score": 1.0, + "content": "network also performs well on the early activity recognition to infer what is hap-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 364, + 469, + 376 + ], + "spans": [ + { + "bbox": [ + 141, + 364, + 469, + 376 + ], + "score": 1.0, + "content": "pening or what will happen after observing only limited frames of video. This", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 375, + 469, + 388 + ], + "spans": [ + { + "bbox": [ + 141, + 375, + 469, + 388 + ], + "score": 1.0, + "content": "task aligns well with video prediction in modeling action intentions and tendency.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 12.5, + "bbox_fs": [ + 141, + 211, + 470, + 388 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 407, + 206, + 420 + ], + "lines": [ + { + "bbox": [ + 105, + 406, + 208, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 208, + 423 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 428, + 505, + 560 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "A fundamental problem in spatiotemporal predictive learning is how to effectively learn good repre-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 104, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "sentations for video inference or reasoning. Currently, recurrent neural networks (RNNs) remain to", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 451, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 463 + ], + "score": 1.0, + "content": "be the most promising models in this field, and have achieved state-of-the-art results on a number of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "score": 1.0, + "content": "future video prediction benchmarks (Wang et al., 2018b; Oliu et al., 2018). However, beyond frames", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 472, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 487 + ], + "score": 1.0, + "content": "prediction, RNN based models are less effective in learning high-level video representations or cap-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 485, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 505, + 496 + ], + "score": 1.0, + "content": "turing long-term relations. On the other hand, recent studies demonstrate that 3D Convolutional", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "Neural networks (3D-CNNs) surpass RNNs in learning better representations for action classifica-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "score": 1.0, + "content": "tion (Carreira & Zisserman, 2017; Tran et al., 2015). For instance, variants of 3D-CNNs, such as", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 517, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 505, + 529 + ], + "score": 1.0, + "content": "Inflated 3D-CNNs, have significantly increased action classification accuracy over the UCF 101 and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 540 + ], + "score": 1.0, + "content": "Kinetics datasets. These 3D-CNN architectures have no recurrent structures but instead employ 3D", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "convolution (3D-Conv) and 3D pooling operations to preserve temporal information of the input", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 550, + 454, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 454, + 562 + ], + "score": 1.0, + "content": "sequences which would be otherwise discarded in classical 2D convolution operations.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 27.5, + "bbox_fs": [ + 104, + 428, + 506, + 562 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 505, + 676 + ], + "lines": [ + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "score": 1.0, + "content": "Motivated by the recent success of 3D-CNNs, in this paper we propose a new model for spatiotem-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 577, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 591 + ], + "score": 1.0, + "content": "poral predictive learning based on both recurrent modeling (for temporal dependency) and feed-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 587, + 505, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 505, + 602 + ], + "score": 1.0, + "content": "forward 3D-Conv modeling (for local dynamics). A plausible approach, of course, is to simply", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 597, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 104, + 597, + 505, + 614 + ], + "score": 1.0, + "content": "stack 3D-Convs and each RNN unit in a feed-forward way using 3D-Convs for either perceiving", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "fine-grained features from raw videos or combining high-level representations. However, as shown", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "in our experiments, these straightforward extensions may not outperform the baseline RNN model.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 631, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 505, + 645 + ], + "score": 1.0, + "content": "We attribute these findings to that RNNs and 3D-CNNs represent two very different mechanisms", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "for the same purpose of spatiotemporal modeling, and connecting them directly fails to exploit their", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "complementary advantages. Therefore, it remains challenging and requires principled approaches", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 666, + 292, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 292, + 677 + ], + "score": 1.0, + "content": "to design an effective spatiotemporal network.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 38.5, + "bbox_fs": [ + 104, + 565, + 505, + 677 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 682, + 504, + 715 + ], + "lines": [ + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "score": 1.0, + "content": "To this end, we propose a new model called Eidetic 3D LSTM (E3D-LSTM) for spatiotemporal", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 693, + 506, + 706 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 506, + 706 + ], + "score": 1.0, + "content": "predictive learning. We introduce an eidetic 3D memory to: a) memorize local appearance and mo-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 703, + 506, + 718 + ], + "spans": [ + { + "bbox": [ + 105, + 703, + 506, + 718 + ], + "score": 1.0, + "content": "tion in a short spatiotemporal volume, and b) recall the long-range historical context by learning", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "to attend to previous memory states. Regarding the short-term dependency, in many cases, spa-", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 107 + ], + "score": 1.0, + "content": "tiotemporal predictive modeling mainly depends on temporally nearby appearances and on-going", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "short-term motions. All the information is encapsulated into the eidetic 3D memory cell with a", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "short time convolution window, and used in recurrent transitions. Our experimental results show", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 505, + 139 + ], + "score": 1.0, + "content": "that integrating 3D-Conv deep into RNNs is effective for modeling local representations in a con-", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "score": 1.0, + "content": "secutive manner. On the other hand, for long-term interactions, which is important for predicting", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 150, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 505, + 160 + ], + "score": 1.0, + "content": "non-stationary or periodical videos as well as learning high-level video representations, we exploit a", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 157, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 157, + 506, + 173 + ], + "score": 1.0, + "content": "self-attention mechanism controlled by revised recurrent gates to recall temporally distant memory.", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 505, + 182 + ], + "score": 1.0, + "content": "The current memory state of E3D-LSTM is learned to attend to all previous relevant moments. Our", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "experimental results verify that this attention mechanism is beneficial for long-term memorization.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 192, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 505, + 205 + ], + "score": 1.0, + "content": "We describe this memory transition mechanism eidetic as it is able to effectively recall the stored", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 203, + 416, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 416, + 216 + ], + "score": 1.0, + "content": "memories across multiple time stamps even after long periods of disturbance.", + "type": "text", + "cross_page": true + } + ], + "index": 11 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 681, + 506, + 718 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 214 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "to attend to previous memory states. Regarding the short-term dependency, in many cases, spa-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 107 + ], + "score": 1.0, + "content": "tiotemporal predictive modeling mainly depends on temporally nearby appearances and on-going", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "short-term motions. All the information is encapsulated into the eidetic 3D memory cell with a", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "short time convolution window, and used in recurrent transitions. Our experimental results show", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 505, + 139 + ], + "score": 1.0, + "content": "that integrating 3D-Conv deep into RNNs is effective for modeling local representations in a con-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "score": 1.0, + "content": "secutive manner. On the other hand, for long-term interactions, which is important for predicting", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 150, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 505, + 160 + ], + "score": 1.0, + "content": "non-stationary or periodical videos as well as learning high-level video representations, we exploit a", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 157, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 157, + 506, + 173 + ], + "score": 1.0, + "content": "self-attention mechanism controlled by revised recurrent gates to recall temporally distant memory.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 505, + 182 + ], + "score": 1.0, + "content": "The current memory state of E3D-LSTM is learned to attend to all previous relevant moments. Our", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "experimental results verify that this attention mechanism is beneficial for long-term memorization.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 192, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 505, + 205 + ], + "score": 1.0, + "content": "We describe this memory transition mechanism eidetic as it is able to effectively recall the stored", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 203, + 416, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 416, + 216 + ], + "score": 1.0, + "content": "memories across multiple time stamps even after long periods of disturbance.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 107, + 220, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "score": 1.0, + "content": "To the best of our knowledge, the proposed E3D-LSTM model is among the first approaches that", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "score": 1.0, + "content": "leverage 3D-Conv in RNNs. We empirically validate it on standard spatiotemporal predictive tasks", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 242, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 505, + 254 + ], + "score": 1.0, + "content": "and an early activity recognition task over four benchmarks: a) on future video prediction, it achieves", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 254, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 505, + 265 + ], + "score": 1.0, + "content": "the best-published accuracy on three classical benchmarks; b) on early activity recognition, it out-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 506, + 277 + ], + "score": 1.0, + "content": "performs the state-of-the-art action recognition methods. In addition, we show that self-supervised", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "learning can further improve the performance of early activity recognition. We present ablation", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 286, + 446, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 446, + 298 + ], + "score": 1.0, + "content": "studies to verify the effectiveness of all modules in the proposed E3D-LSTM model.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 311, + 344, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 309, + 345, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 345, + 325 + ], + "score": 1.0, + "content": "2 RELATED WORK AND PROBLEM CONTEXT", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 333, + 505, + 454 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "score": 1.0, + "content": "Spatiotemporal Predictive Learning Models. In recent years, RNNs have been extensively used", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 505, + 357 + ], + "score": 1.0, + "content": "in sequence prediction and future frame prediction. Srivastava et al. (2015) extended the LSTM-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "score": 1.0, + "content": "based sequence to sequence model (Sutskever et al., 2014) for language modeling to learning video", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 366, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 379 + ], + "score": 1.0, + "content": "representations. Shi et al. (2015) proposed the convolutional LSTM by integrating convolutions", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 378, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 506, + 389 + ], + "score": 1.0, + "content": "into recurrent state transitions for high-dimensional sequence prediction. The convolutional LSTM", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 388, + 506, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 506, + 402 + ], + "score": 1.0, + "content": "model is extended by Finn et al. (2016) to predict future states of robotic environments. Villegas et al.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "(2017) leveraged optical flow to help capture short-term video dynamics for video prediction. Xu", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "et al. (2018) proposed a two-stream RNN that deals with structural video content in separate streams.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "Kalchbrenner et al. (2017) introduced a sophisticated model that extends recurrent structures to", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 433, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 505, + 444 + ], + "score": 1.0, + "content": "estimate local dependencies between adjacent pixels. While this video pixel network (VPN) model", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 443, + 432, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 432, + 456 + ], + "score": 1.0, + "content": "is able to describe image sequences, the computational load is prohibitively high.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 460, + 504, + 526 + ], + "lines": [ + { + "bbox": [ + 106, + 460, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 506, + 473 + ], + "score": 1.0, + "content": "The above-mentioned recurrent models predict future frame mainly based on sequentially updated", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 471, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 483 + ], + "score": 1.0, + "content": "memory states. When the memory cell is refreshed, older memories will be discarded immediately.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 481, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 506, + 495 + ], + "score": 1.0, + "content": "In contrast, the proposed E3D-LSTM model maintains a list of historical memory records and re-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 493, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 506, + 505 + ], + "score": 1.0, + "content": "vokes them when necessary, thereby facilitating long-range video reasoning. While in spirit this", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 502, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 517 + ], + "score": 1.0, + "content": "idea is similar to the self-attention module in feed-forward networks (Vaswani et al., 2017; Wang", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 515, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 505, + 527 + ], + "score": 1.0, + "content": "et al., 2018a), we exploit it to correlate long-term and short-term video representations in this work.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 620 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "score": 1.0, + "content": "Another significant difference between the above-mentioned prior work and the proposed model is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 555 + ], + "score": 1.0, + "content": "that we use 3D-Convs as basic operations inside the E3D-LSTM instead of fully-connected or 2D", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 554, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 566 + ], + "score": 1.0, + "content": "convolution operations. We show using 3D-Convs to model recurrent state-to-state transitions can", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "score": 1.0, + "content": "significantly improve prediction performance. This idea is motivated by recent advances in video", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "classification (a high-level representation learning task) (Ji et al., 2013; Tran et al., 2015; Carreira", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 587, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 598 + ], + "score": 1.0, + "content": "& Zisserman, 2017). We note that Vondrick et al. (2016) and Tulyakov et al. (2018) also introduced", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 597, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 505, + 610 + ], + "score": 1.0, + "content": "3D-CNNs for spatiotemporal predictive learning. However, these networks are feed-forward and do", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 608, + 286, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 286, + 622 + ], + "score": 1.0, + "content": "not capture temporal consistency effectively.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 107, + 625, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 638 + ], + "score": 1.0, + "content": "Future prediction errors of an imperfect model can be categorized by two factors: a) the “systematic", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "errors” caused by a lack of modeling ability to the deterministic variations; b) the stochastic, inherent", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "uncertainty of the future. We aim to minimize the first factor in this work. For the second factor,", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 659, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 505, + 670 + ], + "score": 1.0, + "content": "numerous methods have applied adversarial training or variational auto-encoders to video prediction,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 670, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 681 + ], + "score": 1.0, + "content": "for example (Mathieu et al., 2016; Vondrick et al., 2016; Denton & Fergus, 2018; Bhattacharjee &", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 680, + 393, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 393, + 693 + ], + "score": 1.0, + "content": "Das, 2017; Tulyakov et al., 2018; Lu et al., 2017; Wichers et al., 2018).", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 47.5 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Convolutional Recurrent Networks. Our model is closely related to convolutional recurrent net-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "works. In the ConvLSTM network (Shi et al., 2015), all state transitions are implemented with 2D", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "convolutions. As such, the transition function is no longer permutation invariant and able to better", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 52 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 214 + ], + "lines": [], + "index": 5.5, + "bbox_fs": [ + 105, + 81, + 506, + 216 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 220, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "score": 1.0, + "content": "To the best of our knowledge, the proposed E3D-LSTM model is among the first approaches that", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "score": 1.0, + "content": "leverage 3D-Conv in RNNs. We empirically validate it on standard spatiotemporal predictive tasks", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 242, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 505, + 254 + ], + "score": 1.0, + "content": "and an early activity recognition task over four benchmarks: a) on future video prediction, it achieves", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 254, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 505, + 265 + ], + "score": 1.0, + "content": "the best-published accuracy on three classical benchmarks; b) on early activity recognition, it out-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 506, + 277 + ], + "score": 1.0, + "content": "performs the state-of-the-art action recognition methods. In addition, we show that self-supervised", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "learning can further improve the performance of early activity recognition. We present ablation", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 286, + 446, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 446, + 298 + ], + "score": 1.0, + "content": "studies to verify the effectiveness of all modules in the proposed E3D-LSTM model.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 219, + 506, + 298 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 311, + 344, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 309, + 345, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 345, + 325 + ], + "score": 1.0, + "content": "2 RELATED WORK AND PROBLEM CONTEXT", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 333, + 505, + 454 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "score": 1.0, + "content": "Spatiotemporal Predictive Learning Models. In recent years, RNNs have been extensively used", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 505, + 357 + ], + "score": 1.0, + "content": "in sequence prediction and future frame prediction. Srivastava et al. (2015) extended the LSTM-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "score": 1.0, + "content": "based sequence to sequence model (Sutskever et al., 2014) for language modeling to learning video", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 366, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 379 + ], + "score": 1.0, + "content": "representations. Shi et al. (2015) proposed the convolutional LSTM by integrating convolutions", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 378, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 506, + 389 + ], + "score": 1.0, + "content": "into recurrent state transitions for high-dimensional sequence prediction. The convolutional LSTM", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 388, + 506, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 506, + 402 + ], + "score": 1.0, + "content": "model is extended by Finn et al. (2016) to predict future states of robotic environments. Villegas et al.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "(2017) leveraged optical flow to help capture short-term video dynamics for video prediction. Xu", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "et al. (2018) proposed a two-stream RNN that deals with structural video content in separate streams.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "Kalchbrenner et al. (2017) introduced a sophisticated model that extends recurrent structures to", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 433, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 505, + 444 + ], + "score": 1.0, + "content": "estimate local dependencies between adjacent pixels. While this video pixel network (VPN) model", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 443, + 432, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 432, + 456 + ], + "score": 1.0, + "content": "is able to describe image sequences, the computational load is prohibitively high.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 333, + 506, + 456 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 460, + 504, + 526 + ], + "lines": [ + { + "bbox": [ + 106, + 460, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 506, + 473 + ], + "score": 1.0, + "content": "The above-mentioned recurrent models predict future frame mainly based on sequentially updated", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 471, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 483 + ], + "score": 1.0, + "content": "memory states. When the memory cell is refreshed, older memories will be discarded immediately.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 481, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 506, + 495 + ], + "score": 1.0, + "content": "In contrast, the proposed E3D-LSTM model maintains a list of historical memory records and re-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 493, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 506, + 505 + ], + "score": 1.0, + "content": "vokes them when necessary, thereby facilitating long-range video reasoning. While in spirit this", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 502, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 517 + ], + "score": 1.0, + "content": "idea is similar to the self-attention module in feed-forward networks (Vaswani et al., 2017; Wang", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 515, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 505, + 527 + ], + "score": 1.0, + "content": "et al., 2018a), we exploit it to correlate long-term and short-term video representations in this work.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 460, + 506, + 527 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 620 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "score": 1.0, + "content": "Another significant difference between the above-mentioned prior work and the proposed model is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 555 + ], + "score": 1.0, + "content": "that we use 3D-Convs as basic operations inside the E3D-LSTM instead of fully-connected or 2D", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 554, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 566 + ], + "score": 1.0, + "content": "convolution operations. We show using 3D-Convs to model recurrent state-to-state transitions can", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "score": 1.0, + "content": "significantly improve prediction performance. This idea is motivated by recent advances in video", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "classification (a high-level representation learning task) (Ji et al., 2013; Tran et al., 2015; Carreira", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 587, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 598 + ], + "score": 1.0, + "content": "& Zisserman, 2017). We note that Vondrick et al. (2016) and Tulyakov et al. (2018) also introduced", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 597, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 505, + 610 + ], + "score": 1.0, + "content": "3D-CNNs for spatiotemporal predictive learning. However, these networks are feed-forward and do", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 608, + 286, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 286, + 622 + ], + "score": 1.0, + "content": "not capture temporal consistency effectively.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 532, + 505, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 625, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 638 + ], + "score": 1.0, + "content": "Future prediction errors of an imperfect model can be categorized by two factors: a) the “systematic", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "errors” caused by a lack of modeling ability to the deterministic variations; b) the stochastic, inherent", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "uncertainty of the future. We aim to minimize the first factor in this work. For the second factor,", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 659, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 505, + 670 + ], + "score": 1.0, + "content": "numerous methods have applied adversarial training or variational auto-encoders to video prediction,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 670, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 681 + ], + "score": 1.0, + "content": "for example (Mathieu et al., 2016; Vondrick et al., 2016; Denton & Fergus, 2018; Bhattacharjee &", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 680, + 393, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 393, + 693 + ], + "score": 1.0, + "content": "Das, 2017; Tulyakov et al., 2018; Lu et al., 2017; Wichers et al., 2018).", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 47.5, + "bbox_fs": [ + 105, + 624, + 505, + 693 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Convolutional Recurrent Networks. Our model is closely related to convolutional recurrent net-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "works. In the ConvLSTM network (Shi et al., 2015), all state transitions are implemented with 2D", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "convolutions. As such, the transition function is no longer permutation invariant and able to better", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 505, + 289 + ], + "score": 1.0, + "content": "perceive relations in a spatiotemporal neighborhood. The spatiotemporal LSTM (ST-LSTM) is char-", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 286, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 443, + 301 + ], + "score": 1.0, + "content": "acterized by delivering two memory states separately (Wang et al., 2017): memory", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 444, + 288, + 457, + 297 + ], + "score": 0.84, + "content": "\\mathcal { M }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 457, + 286, + 506, + 301 + ], + "score": 1.0, + "content": "in a zigzag", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 200, + 311 + ], + "score": 1.0, + "content": "direction and memory", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 200, + 299, + 208, + 308 + ], + "score": 0.76, + "content": "\\mathcal { C }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 208, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "being passed horizontally (see Appendix A for details). In this model,", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 308, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 107, + 309, + 120, + 319 + ], + "score": 0.76, + "content": "\\mathcal { M }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 120, + 308, + 366, + 322 + ], + "score": 1.0, + "content": "provides greater capability to model short-term motions, and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 366, + 309, + 374, + 319 + ], + "score": 0.78, + "content": "\\mathcal { C }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 374, + 308, + 506, + 322 + ], + "score": 1.0, + "content": "is adopted from fully-connected", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 319, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 505, + 334 + ], + "score": 1.0, + "content": "LSTMs (Hochreiter & Schmidhuber, 1997) to ease the vanishing gradient problem. Although the", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 330, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 505, + 344 + ], + "score": 1.0, + "content": "ST-LSTM performs well on video prediction benchmarks, it does not capture long-term video re-", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 340, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 302, + 356 + ], + "score": 1.0, + "content": "lations effectively. The forget gates of memory", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 302, + 343, + 309, + 352 + ], + "score": 0.74, + "content": "\\mathcal { C }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 309, + 340, + 506, + 356 + ], + "score": 1.0, + "content": "tend to respond strongly to short-term features,", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "score": 1.0, + "content": "thereby easily falling into a saturated zone (with values between 0 and 0.1) and interrupting long-", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 363, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 406, + 377 + ], + "score": 1.0, + "content": "range information flows. We adopt the zigzag updating route of memory", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 407, + 364, + 420, + 374 + ], + "score": 0.79, + "content": "\\mathcal { M }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 420, + 363, + 505, + 377 + ], + "score": 1.0, + "content": "from the ST-LSTM,", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 375, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 421, + 388 + ], + "score": 1.0, + "content": "while improving the forgetting mechanism in updating the temporal memory", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 421, + 375, + 428, + 385 + ], + "score": 0.6, + "content": "\\mathcal { C }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 429, + 375, + 506, + 388 + ], + "score": 1.0, + "content": ". We also increase", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 385, + 488, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 488, + 398 + ], + "score": 1.0, + "content": "the dimensions of memory states and take 3D-Convs as the basic operators for state transitions.", + "type": "text", + "cross_page": true + } + ], + "index": 17 + } + ], + "index": 52, + "bbox_fs": [ + 106, + 699, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 124, + 70, + 487, + 212 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 124, + 70, + 487, + 212 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 70, + 487, + 212 + ], + "spans": [ + { + "bbox": [ + 124, + 70, + 487, + 212 + ], + "score": 0.972, + "type": "image", + "image_path": "299a5e3b87a1b0eb569a94f56cb3cd52fca6f4a79109947cd0b9f83db0b77962.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 124, + 70, + 487, + 117.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 124, + 117.33333333333334, + 487, + 164.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 124, + 164.66666666666669, + 487, + 212.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 219, + 505, + 263 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "Figure 1: Three approaches to integrate 3D-Convs into recurrent networks. Blue arrows indicate data", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 231, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 242 + ], + "score": 1.0, + "content": "transition paths with 3D-Convs (for feed-forward features or recurrent hidden states). The diagrams", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 254 + ], + "score": 1.0, + "content": "are simplified for illustration, with fewer layers and RNN states than what are actually used in our", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 252, + 461, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 461, + 264 + ], + "score": 1.0, + "content": "experiments. The classifiers are removed when being trained for future video prediction.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 276, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 505, + 289 + ], + "score": 1.0, + "content": "perceive relations in a spatiotemporal neighborhood. The spatiotemporal LSTM (ST-LSTM) is char-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 286, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 443, + 301 + ], + "score": 1.0, + "content": "acterized by delivering two memory states separately (Wang et al., 2017): memory", + "type": "text" + }, + { + "bbox": [ + 444, + 288, + 457, + 297 + ], + "score": 0.84, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 286, + 506, + 301 + ], + "score": 1.0, + "content": "in a zigzag", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 200, + 311 + ], + "score": 1.0, + "content": "direction and memory", + "type": "text" + }, + { + "bbox": [ + 200, + 299, + 208, + 308 + ], + "score": 0.76, + "content": "\\mathcal { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "being passed horizontally (see Appendix A for details). In this model,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 308, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 107, + 309, + 120, + 319 + ], + "score": 0.76, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 308, + 366, + 322 + ], + "score": 1.0, + "content": "provides greater capability to model short-term motions, and", + "type": "text" + }, + { + "bbox": [ + 366, + 309, + 374, + 319 + ], + "score": 0.78, + "content": "\\mathcal { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 308, + 506, + 322 + ], + "score": 1.0, + "content": "is adopted from fully-connected", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 319, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 505, + 334 + ], + "score": 1.0, + "content": "LSTMs (Hochreiter & Schmidhuber, 1997) to ease the vanishing gradient problem. Although the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 330, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 505, + 344 + ], + "score": 1.0, + "content": "ST-LSTM performs well on video prediction benchmarks, it does not capture long-term video re-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 340, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 302, + 356 + ], + "score": 1.0, + "content": "lations effectively. The forget gates of memory", + "type": "text" + }, + { + "bbox": [ + 302, + 343, + 309, + 352 + ], + "score": 0.74, + "content": "\\mathcal { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 340, + 506, + 356 + ], + "score": 1.0, + "content": "tend to respond strongly to short-term features,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "score": 1.0, + "content": "thereby easily falling into a saturated zone (with values between 0 and 0.1) and interrupting long-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 363, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 406, + 377 + ], + "score": 1.0, + "content": "range information flows. We adopt the zigzag updating route of memory", + "type": "text" + }, + { + "bbox": [ + 407, + 364, + 420, + 374 + ], + "score": 0.79, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 363, + 505, + 377 + ], + "score": 1.0, + "content": "from the ST-LSTM,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 375, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 421, + 388 + ], + "score": 1.0, + "content": "while improving the forgetting mechanism in updating the temporal memory", + "type": "text" + }, + { + "bbox": [ + 421, + 375, + 428, + 385 + ], + "score": 0.6, + "content": "\\mathcal { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 375, + 506, + 388 + ], + "score": 1.0, + "content": ". We also increase", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 385, + 488, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 488, + 398 + ], + "score": 1.0, + "content": "the dimensions of memory states and take 3D-Convs as the basic operators for state transitions.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 108, + 414, + 225, + 427 + ], + "lines": [ + { + "bbox": [ + 104, + 412, + 227, + 429 + ], + "spans": [ + { + "bbox": [ + 104, + 412, + 227, + 429 + ], + "score": 1.0, + "content": "3 EIDETIC 3D LSTM", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 108, + 438, + 504, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 438, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 506, + 451 + ], + "score": 1.0, + "content": "This section first presents the Eidetic 3D LSTM for perceiving and memorizing both short-term and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 449, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 506, + 463 + ], + "score": 1.0, + "content": "long-term representations in videos. We then discuss a scheduled multi-task learning strategy that", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 460, + 442, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 442, + 474 + ], + "score": 1.0, + "content": "uses predictive learning as an auxiliary self-supervised task for activity recognition.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 107, + 486, + 331, + 497 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 333, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 333, + 498 + ], + "score": 1.0, + "content": "3.1 3D CONVOLUTIONS IN RECURRENT NETWORK", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "An ideal predictive model relies on effective learning of video representations. RNNs and 3D-CNNs", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "are network architectures of different mechanisms for modeling spatiotemporal data. In this work,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 527, + 504, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 504, + 539 + ], + "score": 1.0, + "content": "we aim to leverage the strength of each one in a unified architecture and start the discussion with", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "two plausible extensions of stacking 3D-Convs and RNN units. Figure 1(a) and 1(b) illustrate two", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 549, + 484, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 484, + 562 + ], + "score": 1.0, + "content": "hybrid baseline networks which add 3D-CNNs before or after stacked spatiotemporal LSTMs.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 565, + 505, + 665 + ], + "lines": [ + { + "bbox": [ + 106, + 566, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 505, + 578 + ], + "score": 1.0, + "content": "However, we find that integrating the 3D-Convs outside the LSTM unit performs noticeably worse", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 576, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 590 + ], + "score": 1.0, + "content": "than the baseline RNN model. To this end, we propose a “deeper” integration of 3D-Convs inside", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 588, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 505, + 600 + ], + "score": 1.0, + "content": "the LSTM unit in order to incorporate the convolutional features into the recurrent state transition", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 598, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 611 + ], + "score": 1.0, + "content": "over time. Figure 1(c) shows the overall encoder-decoder architecture. In this model, a consecutive", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 610, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 117, + 623 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 610, + 126, + 620 + ], + "score": 0.8, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 610, + 506, + 623 + ], + "score": 1.0, + "content": "input frames are first encoded by a few layers of 3D-Convs to obtain high-dimension feature", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 619, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 104, + 619, + 506, + 635 + ], + "score": 1.0, + "content": "maps. The 3D-Conv feature maps are directly fed into a novel E3D-LSTM to model the long-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 632, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 644 + ], + "score": 1.0, + "content": "term spatiotemporal interaction. Finally, the E3D-LSTM hidden states are decoded by a number of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "stacked 3D-Conv layers to get the predicted video frames. For classification tasks, the hidden states", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 654, + 331, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 331, + 666 + ], + "score": 1.0, + "content": "can be directly used as the learned video representation.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32 + }, + { + "type": "title", + "bbox": [ + 108, + 679, + 266, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 268, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 268, + 691 + ], + "score": 1.0, + "content": "3.2 EIDETIC MEMORY TRANSITION", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "The architecture of the proposed Eidetic 3D LSTM is illustrated in Figure 2, where the red arrows", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "indicate short-term information flow and the blue arrows denote long-term information flow. There", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 159, + 733 + ], + "score": 1.0, + "content": "are 4 inputs:", + "type": "text" + }, + { + "bbox": [ + 159, + 721, + 171, + 732 + ], + "score": 0.88, + "content": "\\mathcal { X } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 720, + 505, + 733 + ], + "score": 1.0, + "content": ", the 3D-Conv feature maps from encoders or hidden states from the previous E3D-", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 124, + 70, + 487, + 212 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 124, + 70, + 487, + 212 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 70, + 487, + 212 + ], + "spans": [ + { + "bbox": [ + 124, + 70, + 487, + 212 + ], + "score": 0.972, + "type": "image", + "image_path": "299a5e3b87a1b0eb569a94f56cb3cd52fca6f4a79109947cd0b9f83db0b77962.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 124, + 70, + 487, + 117.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 124, + 117.33333333333334, + 487, + 164.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 124, + 164.66666666666669, + 487, + 212.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 219, + 505, + 263 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "Figure 1: Three approaches to integrate 3D-Convs into recurrent networks. Blue arrows indicate data", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 231, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 242 + ], + "score": 1.0, + "content": "transition paths with 3D-Convs (for feed-forward features or recurrent hidden states). The diagrams", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 254 + ], + "score": 1.0, + "content": "are simplified for illustration, with fewer layers and RNN states than what are actually used in our", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 252, + 461, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 461, + 264 + ], + "score": 1.0, + "content": "experiments. The classifiers are removed when being trained for future video prediction.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 276, + 505, + 397 + ], + "lines": [], + "index": 12, + "bbox_fs": [ + 105, + 276, + 506, + 398 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 414, + 225, + 427 + ], + "lines": [ + { + "bbox": [ + 104, + 412, + 227, + 429 + ], + "spans": [ + { + "bbox": [ + 104, + 412, + 227, + 429 + ], + "score": 1.0, + "content": "3 EIDETIC 3D LSTM", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 108, + 438, + 504, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 438, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 506, + 451 + ], + "score": 1.0, + "content": "This section first presents the Eidetic 3D LSTM for perceiving and memorizing both short-term and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 449, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 506, + 463 + ], + "score": 1.0, + "content": "long-term representations in videos. We then discuss a scheduled multi-task learning strategy that", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 460, + 442, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 442, + 474 + ], + "score": 1.0, + "content": "uses predictive learning as an auxiliary self-supervised task for activity recognition.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 438, + 506, + 474 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 486, + 331, + 497 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 333, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 333, + 498 + ], + "score": 1.0, + "content": "3.1 3D CONVOLUTIONS IN RECURRENT NETWORK", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "An ideal predictive model relies on effective learning of video representations. RNNs and 3D-CNNs", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "are network architectures of different mechanisms for modeling spatiotemporal data. In this work,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 527, + 504, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 504, + 539 + ], + "score": 1.0, + "content": "we aim to leverage the strength of each one in a unified architecture and start the discussion with", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "two plausible extensions of stacking 3D-Convs and RNN units. Figure 1(a) and 1(b) illustrate two", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 549, + 484, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 484, + 562 + ], + "score": 1.0, + "content": "hybrid baseline networks which add 3D-CNNs before or after stacked spatiotemporal LSTMs.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 505, + 506, + 562 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 565, + 505, + 665 + ], + "lines": [ + { + "bbox": [ + 106, + 566, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 505, + 578 + ], + "score": 1.0, + "content": "However, we find that integrating the 3D-Convs outside the LSTM unit performs noticeably worse", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 576, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 590 + ], + "score": 1.0, + "content": "than the baseline RNN model. To this end, we propose a “deeper” integration of 3D-Convs inside", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 588, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 505, + 600 + ], + "score": 1.0, + "content": "the LSTM unit in order to incorporate the convolutional features into the recurrent state transition", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 598, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 611 + ], + "score": 1.0, + "content": "over time. Figure 1(c) shows the overall encoder-decoder architecture. In this model, a consecutive", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 610, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 117, + 623 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 610, + 126, + 620 + ], + "score": 0.8, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 610, + 506, + 623 + ], + "score": 1.0, + "content": "input frames are first encoded by a few layers of 3D-Convs to obtain high-dimension feature", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 619, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 104, + 619, + 506, + 635 + ], + "score": 1.0, + "content": "maps. The 3D-Conv feature maps are directly fed into a novel E3D-LSTM to model the long-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 632, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 644 + ], + "score": 1.0, + "content": "term spatiotemporal interaction. Finally, the E3D-LSTM hidden states are decoded by a number of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "stacked 3D-Conv layers to get the predicted video frames. For classification tasks, the hidden states", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 654, + 331, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 331, + 666 + ], + "score": 1.0, + "content": "can be directly used as the learned video representation.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32, + "bbox_fs": [ + 104, + 566, + 506, + 666 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 679, + 266, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 268, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 268, + 691 + ], + "score": 1.0, + "content": "3.2 EIDETIC MEMORY TRANSITION", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "The architecture of the proposed Eidetic 3D LSTM is illustrated in Figure 2, where the red arrows", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "indicate short-term information flow and the blue arrows denote long-term information flow. There", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 159, + 733 + ], + "score": 1.0, + "content": "are 4 inputs:", + "type": "text" + }, + { + "bbox": [ + 159, + 721, + 171, + 732 + ], + "score": 0.88, + "content": "\\mathcal { X } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 720, + 505, + 733 + ], + "score": 1.0, + "content": ", the 3D-Conv feature maps from encoders or hidden states from the previous E3D-", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 698, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 69, + 502, + 203 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 69, + 502, + 203 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 69, + 502, + 203 + ], + "spans": [ + { + "bbox": [ + 110, + 69, + 502, + 203 + ], + "score": 0.973, + "type": "image", + "image_path": "924cf68bdd48840e3b363ec3c0c9bc6abeab5aaa00eac072e9b283bd4a42e9c3.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 69, + 502, + 113.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 113.66666666666666, + 502, + 158.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 158.33333333333331, + 502, + 202.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 210, + 505, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 210, + 506, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 506, + 223 + ], + "score": 1.0, + "content": "Figure 2: Comparison of (a) the standard memory transition approach in the Spatiotemporal LSTM", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "score": 1.0, + "content": "and (b) the attentive memory transition approach in the Eidetic 3D LSTM. Red arrows indicate the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 232, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 505, + 244 + ], + "score": 1.0, + "content": "short-term information flow. Blue arrows are the attentive memory flow, which potentially enables", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "our model to capture the long-term relations. Cubes denote higher-dimensional hidden states and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 254, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 351, + 266 + ], + "score": 1.0, + "content": "memory states. Cylinders denote higher-dimensional gates.", + "type": "text" + }, + { + "bbox": [ + 352, + 255, + 361, + 265 + ], + "score": 0.79, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 254, + 469, + 266 + ], + "score": 1.0, + "content": "is the Hadamard product.", + "type": "text" + }, + { + "bbox": [ + 469, + 255, + 479, + 265 + ], + "score": 0.81, + "content": "\\otimes", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 254, + 505, + 266 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 266, + 420, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 420, + 277 + ], + "score": 1.0, + "content": "matrix product after reshaping matrices into appropriate 2-dimensional forms.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 106, + 288, + 504, + 314 + ], + "lines": [ + { + "bbox": [ + 104, + 286, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 104, + 286, + 163, + 304 + ], + "score": 1.0, + "content": "LSTM layer;", + "type": "text" + }, + { + "bbox": [ + 164, + 288, + 187, + 302 + ], + "score": 0.92, + "content": "\\mathcal { H } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 286, + 378, + 304 + ], + "score": 1.0, + "content": ", the hidden states from previous time stamp;", + "type": "text" + }, + { + "bbox": [ + 379, + 288, + 398, + 302 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 286, + 506, + 304 + ], + "score": 1.0, + "content": ", the memory states from", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 299, + 486, + 316 + ], + "spans": [ + { + "bbox": [ + 104, + 299, + 209, + 316 + ], + "score": 1.0, + "content": "previous time stamp; and", + "type": "text" + }, + { + "bbox": [ + 209, + 300, + 237, + 314 + ], + "score": 0.93, + "content": "\\mathcal { M } _ { t } ^ { k - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 299, + 486, + 316 + ], + "score": 1.0, + "content": ", the previous spatiotemporal memory states described earlier.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 318, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "We use recurrent 3D-Convs as motion-aware perceptrons to extract short-term appearance and lo-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 329, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 506, + 342 + ], + "score": 1.0, + "content": "cal motions in continuous space-time fields and store them in a small spatiotemporal volume. As", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 103, + 337, + 507, + 355 + ], + "spans": [ + { + "bbox": [ + 103, + 337, + 376, + 355 + ], + "score": 1.0, + "content": "such, video appearance and short-term motions can be encoded in", + "type": "text" + }, + { + "bbox": [ + 377, + 340, + 431, + 351 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { T \\times \\mathbf { \\dot { H } } \\times W \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 337, + 507, + 355 + ], + "score": 1.0, + "content": "tensors, in which", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 352, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 506, + 364 + ], + "score": 1.0, + "content": "each dimension indicates temporal depth, spatial size, and the number of feature map channels, re-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "spectively. By inflating the memory state along the time dimension, we found that the proposed", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 373, + 502, + 387 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 502, + 387 + ], + "score": 1.0, + "content": "E3D-LSTM becomes more capable of characterizing and memorizing local or short-term motions.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 390, + 504, + 413 + ], + "lines": [ + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "score": 1.0, + "content": "To capture the long-term frame interactions, we improve the recurrent transition function of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 402, + 375, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 375, + 414 + ], + "score": 1.0, + "content": "memory states by proposing a new memory RECALL mechanism:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "interline_equation", + "bbox": [ + 180, + 413, + 427, + 491 + ], + "lines": [ + { + "bbox": [ + 180, + 413, + 427, + 491 + ], + "spans": [ + { + "bbox": [ + 180, + 413, + 427, + 491 + ], + "score": 0.89, + "content": "\\begin{array} { r l } & { \\mathcal { R } _ { t } = \\sigma \\big ( W _ { x r } * \\mathcal { X } _ { t } + W _ { h r } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { r } \\big ) } \\\\ & { \\mathcal { Z } _ { t } = \\sigma \\big ( W _ { x i } * \\mathcal { X } _ { t } + W _ { h i } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { i } \\big ) } \\\\ & { \\mathcal { G } _ { t } = \\mathrm { t a n h } \\big ( W _ { x g } * \\mathcal { X } _ { t } + W _ { h g } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { g } \\big ) } \\\\ & { \\mathrm { R E C A L L } \\big ( \\mathcal { R } _ { t } , \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) = \\mathrm { s o f t m a x } \\big ( \\mathcal { R } _ { t } \\cdot \\big ( \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) ^ { \\sf T } \\big ) \\cdot \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } } \\\\ & { \\mathcal { C } _ { t } ^ { k } = \\mathcal { T } _ { t } \\odot \\mathcal { G } _ { t } + \\mathrm { L a y e r N o r m } ( \\mathcal { C } _ { t - 1 } ^ { k } + \\mathrm { R E C A L L } \\big ( \\mathcal { R } _ { t } , \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) \\big ) , } \\end{array}", + "type": "interline_equation", + "image_path": "cc8a6dd29230d8644c0af099c67e13e7ab5002774b305cd32d7091f51650b924.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 180, + 413, + 427, + 439.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 180, + 439.0, + 427, + 465.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 180, + 465.0, + 427, + 491.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 491, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 133, + 504 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 494, + 141, + 502 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 492, + 242, + 504 + ], + "score": 1.0, + "content": "is the sigmoid function,", + "type": "text" + }, + { + "bbox": [ + 242, + 494, + 249, + 502 + ], + "score": 0.76, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 492, + 358, + 504 + ], + "score": 1.0, + "content": "is the 3D-Conv operation,", + "type": "text" + }, + { + "bbox": [ + 358, + 493, + 367, + 502 + ], + "score": 0.82, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 492, + 506, + 504 + ], + "score": 1.0, + "content": "is the Hadamard product, · is the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 103, + 499, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 103, + 499, + 292, + 519 + ], + "score": 1.0, + "content": "matrix product after reshaping the recall gate", + "type": "text" + }, + { + "bbox": [ + 293, + 503, + 306, + 514 + ], + "score": 0.89, + "content": "\\mathcal { R } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 499, + 387, + 519 + ], + "score": 1.0, + "content": "and memory states", + "type": "text" + }, + { + "bbox": [ + 387, + 502, + 423, + 515 + ], + "score": 0.93, + "content": " { \\mathcal { C } } _ { t - \\tau : t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 499, + 443, + 519 + ], + "score": 1.0, + "content": "into", + "type": "text" + }, + { + "bbox": [ + 444, + 502, + 486, + 513 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { T H W \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 499, + 505, + 519 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 512, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 154, + 525 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { \\tau T H W \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 512, + 261, + 531 + ], + "score": 1.0, + "content": "matrices, respectively, and", + "type": "text" + }, + { + "bbox": [ + 262, + 517, + 268, + 525 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 512, + 506, + 531 + ], + "score": 1.0, + "content": "is the number of memory states that are concatenated along", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 524, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 104, + 524, + 366, + 541 + ], + "score": 1.0, + "content": "the temporal dimension. Three terms are involved in computing", + "type": "text" + }, + { + "bbox": [ + 366, + 526, + 378, + 538 + ], + "score": 0.9, + "content": "\\mathcal { C } _ { t } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 524, + 436, + 541 + ], + "score": 1.0, + "content": ". The first one", + "type": "text" + }, + { + "bbox": [ + 437, + 527, + 469, + 538 + ], + "score": 0.92, + "content": "\\mathcal { T } _ { t } \\odot \\mathcal { G } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 524, + 506, + 541 + ], + "score": 1.0, + "content": "encodes", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 537, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 289, + 550 + ], + "score": 1.0, + "content": "local video appearance and motions, where", + "type": "text" + }, + { + "bbox": [ + 290, + 538, + 300, + 548 + ], + "score": 0.88, + "content": "\\mathcal { T } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 537, + 392, + 550 + ], + "score": 1.0, + "content": "is the input gate and", + "type": "text" + }, + { + "bbox": [ + 393, + 538, + 403, + 549 + ], + "score": 0.88, + "content": "\\mathcal { G } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 537, + 505, + 550 + ], + "score": 1.0, + "content": "is the input modulation", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 103, + 545, + 508, + 565 + ], + "spans": [ + { + "bbox": [ + 103, + 545, + 282, + 565 + ], + "score": 1.0, + "content": "gate like standard LSTMs. The second one", + "type": "text" + }, + { + "bbox": [ + 283, + 548, + 303, + 561 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 545, + 508, + 565 + ], + "score": 1.0, + "content": "can be viewed as a short-cut connection from the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "previous memory state, which captures short-term changes between adjacent time stamps. In this", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 570, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 506, + 583 + ], + "score": 1.0, + "content": "process, the accessible memory field is fixed and limited. Therefore, we introduce the third term", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 582, + 505, + 593 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 505, + 593 + ], + "score": 1.0, + "content": "of memory transition function, modeling long-term video relations according to local motion and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 591, + 507, + 606 + ], + "spans": [ + { + "bbox": [ + 104, + 591, + 217, + 606 + ], + "score": 1.0, + "content": "appearance (as encoded in", + "type": "text" + }, + { + "bbox": [ + 217, + 592, + 229, + 604 + ], + "score": 0.87, + "content": "\\mathcal { X } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 591, + 248, + 606 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 249, + 592, + 272, + 605 + ], + "score": 0.91, + "content": "\\mathcal { H } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 591, + 507, + 606 + ], + "score": 1.0, + "content": "). The RECALL function is implemented as an attentive", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 602, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 104, + 602, + 505, + 617 + ], + "score": 1.0, + "content": "module to compute the relationship between the encoded local patterns and the whole memory", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 255, + 627 + ], + "score": 1.0, + "content": "space. A set of parameterized gates", + "type": "text" + }, + { + "bbox": [ + 255, + 614, + 268, + 625 + ], + "score": 0.89, + "content": "\\mathcal { R } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 614, + 505, + 627 + ], + "score": 1.0, + "content": ", acting as memory access instructions, control where and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 624, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 638 + ], + "score": 1.0, + "content": "what to attend in historical memory records. These two terms are respectively designed for short-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "score": 1.0, + "content": "term and long-term video modeling. We integrate them in a unified network by applying layer", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 647, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 506, + 659 + ], + "score": 1.0, + "content": "normalization (Ba et al., 2016) to their element-wise sum, in order to mitigate the covariant shift", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 657, + 507, + 672 + ], + "spans": [ + { + "bbox": [ + 104, + 657, + 507, + 672 + ], + "score": 1.0, + "content": "and stabilize the training process, as it has been commonly used in RNNs. The hyper-parameter", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 663, + 509, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 114, + 679 + ], + "score": 0.71, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 663, + 126, + 690 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 127, + 668, + 163, + 682 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { t - \\tau : t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 663, + 472, + 690 + ], + "score": 1.0, + "content": "decides how many historical memory states are attended by the recall gate", + "type": "text" + }, + { + "bbox": [ + 472, + 669, + 485, + 680 + ], + "score": 0.87, + "content": "\\mathcal { R } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 663, + 509, + 690 + ], + "score": 1.0, + "content": ". To", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 103, + 677, + 508, + 699 + ], + "spans": [ + { + "bbox": [ + 103, + 677, + 359, + 699 + ], + "score": 1.0, + "content": "involve more long-term relations, in most experiments, we take", + "type": "text" + }, + { + "bbox": [ + 386, + 677, + 508, + 699 + ], + "score": 1.0, + "content": "as the inputs of the RECALL", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 693, + 489, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 199, + 705 + ], + "score": 1.0, + "content": "function and do not fix", + "type": "text" + }, + { + "bbox": [ + 200, + 695, + 207, + 703 + ], + "score": 0.67, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 693, + 460, + 705 + ], + "score": 1.0, + "content": ". Whereas in particular, we enable online recognition by setting", + "type": "text" + }, + { + "bbox": [ + 461, + 695, + 468, + 703 + ], + "score": 0.74, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 693, + 489, + 705 + ], + "score": 1.0, + "content": "to 5.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Unlike the conventional memory transition function, the RECALL function learns the size of tem-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 721, + 504, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 504, + 733 + ], + "score": 1.0, + "content": "poral interactions. For longer sequences, this allows attending to distant states containing salient in-", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 69, + 502, + 203 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 69, + 502, + 203 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 69, + 502, + 203 + ], + "spans": [ + { + "bbox": [ + 110, + 69, + 502, + 203 + ], + "score": 0.973, + "type": "image", + "image_path": "924cf68bdd48840e3b363ec3c0c9bc6abeab5aaa00eac072e9b283bd4a42e9c3.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 69, + 502, + 113.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 113.66666666666666, + 502, + 158.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 158.33333333333331, + 502, + 202.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 210, + 505, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 210, + 506, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 506, + 223 + ], + "score": 1.0, + "content": "Figure 2: Comparison of (a) the standard memory transition approach in the Spatiotemporal LSTM", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "score": 1.0, + "content": "and (b) the attentive memory transition approach in the Eidetic 3D LSTM. Red arrows indicate the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 232, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 505, + 244 + ], + "score": 1.0, + "content": "short-term information flow. Blue arrows are the attentive memory flow, which potentially enables", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "our model to capture the long-term relations. Cubes denote higher-dimensional hidden states and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 254, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 351, + 266 + ], + "score": 1.0, + "content": "memory states. Cylinders denote higher-dimensional gates.", + "type": "text" + }, + { + "bbox": [ + 352, + 255, + 361, + 265 + ], + "score": 0.79, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 254, + 469, + 266 + ], + "score": 1.0, + "content": "is the Hadamard product.", + "type": "text" + }, + { + "bbox": [ + 469, + 255, + 479, + 265 + ], + "score": 0.81, + "content": "\\otimes", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 254, + 505, + 266 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 266, + 420, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 420, + 277 + ], + "score": 1.0, + "content": "matrix product after reshaping matrices into appropriate 2-dimensional forms.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 106, + 288, + 504, + 314 + ], + "lines": [ + { + "bbox": [ + 104, + 286, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 104, + 286, + 163, + 304 + ], + "score": 1.0, + "content": "LSTM layer;", + "type": "text" + }, + { + "bbox": [ + 164, + 288, + 187, + 302 + ], + "score": 0.92, + "content": "\\mathcal { H } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 286, + 378, + 304 + ], + "score": 1.0, + "content": ", the hidden states from previous time stamp;", + "type": "text" + }, + { + "bbox": [ + 379, + 288, + 398, + 302 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 286, + 506, + 304 + ], + "score": 1.0, + "content": ", the memory states from", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 299, + 486, + 316 + ], + "spans": [ + { + "bbox": [ + 104, + 299, + 209, + 316 + ], + "score": 1.0, + "content": "previous time stamp; and", + "type": "text" + }, + { + "bbox": [ + 209, + 300, + 237, + 314 + ], + "score": 0.93, + "content": "\\mathcal { M } _ { t } ^ { k - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 299, + 486, + 316 + ], + "score": 1.0, + "content": ", the previous spatiotemporal memory states described earlier.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 104, + 286, + 506, + 316 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 318, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "We use recurrent 3D-Convs as motion-aware perceptrons to extract short-term appearance and lo-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 329, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 506, + 342 + ], + "score": 1.0, + "content": "cal motions in continuous space-time fields and store them in a small spatiotemporal volume. As", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 103, + 337, + 507, + 355 + ], + "spans": [ + { + "bbox": [ + 103, + 337, + 376, + 355 + ], + "score": 1.0, + "content": "such, video appearance and short-term motions can be encoded in", + "type": "text" + }, + { + "bbox": [ + 377, + 340, + 431, + 351 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { T \\times \\mathbf { \\dot { H } } \\times W \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 337, + 507, + 355 + ], + "score": 1.0, + "content": "tensors, in which", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 352, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 506, + 364 + ], + "score": 1.0, + "content": "each dimension indicates temporal depth, spatial size, and the number of feature map channels, re-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "spectively. By inflating the memory state along the time dimension, we found that the proposed", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 373, + 502, + 387 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 502, + 387 + ], + "score": 1.0, + "content": "E3D-LSTM becomes more capable of characterizing and memorizing local or short-term motions.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5, + "bbox_fs": [ + 103, + 318, + 507, + 387 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 390, + 504, + 413 + ], + "lines": [ + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "score": 1.0, + "content": "To capture the long-term frame interactions, we improve the recurrent transition function of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 402, + 375, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 375, + 414 + ], + "score": 1.0, + "content": "memory states by proposing a new memory RECALL mechanism:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 390, + 505, + 414 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 180, + 413, + 427, + 491 + ], + "lines": [ + { + "bbox": [ + 180, + 413, + 427, + 491 + ], + "spans": [ + { + "bbox": [ + 180, + 413, + 427, + 491 + ], + "score": 0.89, + "content": "\\begin{array} { r l } & { \\mathcal { R } _ { t } = \\sigma \\big ( W _ { x r } * \\mathcal { X } _ { t } + W _ { h r } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { r } \\big ) } \\\\ & { \\mathcal { Z } _ { t } = \\sigma \\big ( W _ { x i } * \\mathcal { X } _ { t } + W _ { h i } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { i } \\big ) } \\\\ & { \\mathcal { G } _ { t } = \\mathrm { t a n h } \\big ( W _ { x g } * \\mathcal { X } _ { t } + W _ { h g } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { g } \\big ) } \\\\ & { \\mathrm { R E C A L L } \\big ( \\mathcal { R } _ { t } , \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) = \\mathrm { s o f t m a x } \\big ( \\mathcal { R } _ { t } \\cdot \\big ( \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) ^ { \\sf T } \\big ) \\cdot \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } } \\\\ & { \\mathcal { C } _ { t } ^ { k } = \\mathcal { T } _ { t } \\odot \\mathcal { G } _ { t } + \\mathrm { L a y e r N o r m } ( \\mathcal { C } _ { t - 1 } ^ { k } + \\mathrm { R E C A L L } \\big ( \\mathcal { R } _ { t } , \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) \\big ) , } \\end{array}", + "type": "interline_equation", + "image_path": "cc8a6dd29230d8644c0af099c67e13e7ab5002774b305cd32d7091f51650b924.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 180, + 413, + 427, + 439.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 180, + 439.0, + 427, + 465.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 180, + 465.0, + 427, + 491.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 491, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 133, + 504 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 494, + 141, + 502 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 492, + 242, + 504 + ], + "score": 1.0, + "content": "is the sigmoid function,", + "type": "text" + }, + { + "bbox": [ + 242, + 494, + 249, + 502 + ], + "score": 0.76, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 492, + 358, + 504 + ], + "score": 1.0, + "content": "is the 3D-Conv operation,", + "type": "text" + }, + { + "bbox": [ + 358, + 493, + 367, + 502 + ], + "score": 0.82, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 492, + 506, + 504 + ], + "score": 1.0, + "content": "is the Hadamard product, · is the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 103, + 499, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 103, + 499, + 292, + 519 + ], + "score": 1.0, + "content": "matrix product after reshaping the recall gate", + "type": "text" + }, + { + "bbox": [ + 293, + 503, + 306, + 514 + ], + "score": 0.89, + "content": "\\mathcal { R } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 499, + 387, + 519 + ], + "score": 1.0, + "content": "and memory states", + "type": "text" + }, + { + "bbox": [ + 387, + 502, + 423, + 515 + ], + "score": 0.93, + "content": " { \\mathcal { C } } _ { t - \\tau : t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 499, + 443, + 519 + ], + "score": 1.0, + "content": "into", + "type": "text" + }, + { + "bbox": [ + 444, + 502, + 486, + 513 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { T H W \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 499, + 505, + 519 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 512, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 154, + 525 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { \\tau T H W \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 512, + 261, + 531 + ], + "score": 1.0, + "content": "matrices, respectively, and", + "type": "text" + }, + { + "bbox": [ + 262, + 517, + 268, + 525 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 512, + 506, + 531 + ], + "score": 1.0, + "content": "is the number of memory states that are concatenated along", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 524, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 104, + 524, + 366, + 541 + ], + "score": 1.0, + "content": "the temporal dimension. Three terms are involved in computing", + "type": "text" + }, + { + "bbox": [ + 366, + 526, + 378, + 538 + ], + "score": 0.9, + "content": "\\mathcal { C } _ { t } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 524, + 436, + 541 + ], + "score": 1.0, + "content": ". The first one", + "type": "text" + }, + { + "bbox": [ + 437, + 527, + 469, + 538 + ], + "score": 0.92, + "content": "\\mathcal { T } _ { t } \\odot \\mathcal { G } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 524, + 506, + 541 + ], + "score": 1.0, + "content": "encodes", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 537, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 289, + 550 + ], + "score": 1.0, + "content": "local video appearance and motions, where", + "type": "text" + }, + { + "bbox": [ + 290, + 538, + 300, + 548 + ], + "score": 0.88, + "content": "\\mathcal { T } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 537, + 392, + 550 + ], + "score": 1.0, + "content": "is the input gate and", + "type": "text" + }, + { + "bbox": [ + 393, + 538, + 403, + 549 + ], + "score": 0.88, + "content": "\\mathcal { G } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 537, + 505, + 550 + ], + "score": 1.0, + "content": "is the input modulation", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 103, + 545, + 508, + 565 + ], + "spans": [ + { + "bbox": [ + 103, + 545, + 282, + 565 + ], + "score": 1.0, + "content": "gate like standard LSTMs. The second one", + "type": "text" + }, + { + "bbox": [ + 283, + 548, + 303, + 561 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 545, + 508, + 565 + ], + "score": 1.0, + "content": "can be viewed as a short-cut connection from the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "previous memory state, which captures short-term changes between adjacent time stamps. In this", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 570, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 506, + 583 + ], + "score": 1.0, + "content": "process, the accessible memory field is fixed and limited. Therefore, we introduce the third term", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 582, + 505, + 593 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 505, + 593 + ], + "score": 1.0, + "content": "of memory transition function, modeling long-term video relations according to local motion and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 591, + 507, + 606 + ], + "spans": [ + { + "bbox": [ + 104, + 591, + 217, + 606 + ], + "score": 1.0, + "content": "appearance (as encoded in", + "type": "text" + }, + { + "bbox": [ + 217, + 592, + 229, + 604 + ], + "score": 0.87, + "content": "\\mathcal { X } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 591, + 248, + 606 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 249, + 592, + 272, + 605 + ], + "score": 0.91, + "content": "\\mathcal { H } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 591, + 507, + 606 + ], + "score": 1.0, + "content": "). The RECALL function is implemented as an attentive", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 602, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 104, + 602, + 505, + 617 + ], + "score": 1.0, + "content": "module to compute the relationship between the encoded local patterns and the whole memory", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 255, + 627 + ], + "score": 1.0, + "content": "space. A set of parameterized gates", + "type": "text" + }, + { + "bbox": [ + 255, + 614, + 268, + 625 + ], + "score": 0.89, + "content": "\\mathcal { R } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 614, + 505, + 627 + ], + "score": 1.0, + "content": ", acting as memory access instructions, control where and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 624, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 638 + ], + "score": 1.0, + "content": "what to attend in historical memory records. These two terms are respectively designed for short-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "score": 1.0, + "content": "term and long-term video modeling. We integrate them in a unified network by applying layer", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 647, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 506, + 659 + ], + "score": 1.0, + "content": "normalization (Ba et al., 2016) to their element-wise sum, in order to mitigate the covariant shift", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 657, + 507, + 672 + ], + "spans": [ + { + "bbox": [ + 104, + 657, + 507, + 672 + ], + "score": 1.0, + "content": "and stabilize the training process, as it has been commonly used in RNNs. The hyper-parameter", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 663, + 509, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 114, + 679 + ], + "score": 0.71, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 663, + 126, + 690 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 127, + 668, + 163, + 682 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { t - \\tau : t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 663, + 472, + 690 + ], + "score": 1.0, + "content": "decides how many historical memory states are attended by the recall gate", + "type": "text" + }, + { + "bbox": [ + 472, + 669, + 485, + 680 + ], + "score": 0.87, + "content": "\\mathcal { R } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 663, + 509, + 690 + ], + "score": 1.0, + "content": ". To", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 103, + 677, + 508, + 699 + ], + "spans": [ + { + "bbox": [ + 103, + 677, + 359, + 699 + ], + "score": 1.0, + "content": "involve more long-term relations, in most experiments, we take", + "type": "text" + }, + { + "bbox": [ + 386, + 677, + 508, + 699 + ], + "score": 1.0, + "content": "as the inputs of the RECALL", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 693, + 489, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 199, + 705 + ], + "score": 1.0, + "content": "function and do not fix", + "type": "text" + }, + { + "bbox": [ + 200, + 695, + 207, + 703 + ], + "score": 0.67, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 693, + 460, + 705 + ], + "score": 1.0, + "content": ". Whereas in particular, we enable online recognition by setting", + "type": "text" + }, + { + "bbox": [ + 461, + 695, + 468, + 703 + ], + "score": 0.74, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 693, + 489, + 705 + ], + "score": 1.0, + "content": "to 5.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 31, + "bbox_fs": [ + 103, + 492, + 509, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Unlike the conventional memory transition function, the RECALL function learns the size of tem-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 721, + 504, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 504, + 733 + ], + "score": 1.0, + "content": "poral interactions. For longer sequences, this allows attending to distant states containing salient in-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "formation. Our work is partially motivated by self-attention mechanisms (Lin et al., 2017; Vaswani", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "et al., 2017). However, in our model, the attention mechanism is not applied over the output states", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "but during the memory transitions. It is used to evoke past memories from distant time stamps for", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 114, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 104, + 114, + 506, + 129 + ], + "score": 1.0, + "content": "memorizing and distilling useful information from what has been perceived. We show that learning", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 505, + 138 + ], + "score": 1.0, + "content": "attention over previous memory states is beneficial in recalling the long-range historical context.", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 189, + 150 + ], + "score": 1.0, + "content": "The memory tensor", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 189, + 137, + 201, + 149 + ], + "score": 0.9, + "content": "\\mathcal { C } _ { t } ^ { k }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 201, + 137, + 271, + 150 + ], + "score": 1.0, + "content": "is named eidetic", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 272, + 138, + 285, + 147 + ], + "score": 0.65, + "content": "3 D", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 286, + 137, + 506, + 150 + ], + "score": 1.0, + "content": "memory and the entire unit is called E3D-LSTM. We", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 147, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 104, + 147, + 312, + 162 + ], + "score": 1.0, + "content": "also exploit the same RECALL method to correlate", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 312, + 148, + 336, + 160 + ], + "score": 0.92, + "content": "\\dot { \\mathcal { M } } _ { t } ^ { 1 : k }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 336, + 147, + 506, + 162 + ], + "score": 1.0, + "content": "along the vertical memory transition flow,", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 157, + 500, + 173 + ], + "spans": [ + { + "bbox": [ + 104, + 157, + 367, + 173 + ], + "score": 1.0, + "content": "but it turns out to be less helpful. With the updated memory state", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 368, + 159, + 379, + 171 + ], + "score": 0.89, + "content": "\\mathcal { C } _ { t } ^ { k }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 380, + 157, + 500, + 173 + ], + "score": 1.0, + "content": ", the output hidden states are:", + "type": "text", + "cross_page": true + } + ], + "index": 7 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 709, + 504, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 506, + 171 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "formation. Our work is partially motivated by self-attention mechanisms (Lin et al., 2017; Vaswani", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "et al., 2017). However, in our model, the attention mechanism is not applied over the output states", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "but during the memory transitions. It is used to evoke past memories from distant time stamps for", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 114, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 104, + 114, + 506, + 129 + ], + "score": 1.0, + "content": "memorizing and distilling useful information from what has been perceived. We show that learning", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 505, + 138 + ], + "score": 1.0, + "content": "attention over previous memory states is beneficial in recalling the long-range historical context.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 189, + 150 + ], + "score": 1.0, + "content": "The memory tensor", + "type": "text" + }, + { + "bbox": [ + 189, + 137, + 201, + 149 + ], + "score": 0.9, + "content": "\\mathcal { C } _ { t } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 137, + 271, + 150 + ], + "score": 1.0, + "content": "is named eidetic", + "type": "text" + }, + { + "bbox": [ + 272, + 138, + 285, + 147 + ], + "score": 0.65, + "content": "3 D", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 137, + 506, + 150 + ], + "score": 1.0, + "content": "memory and the entire unit is called E3D-LSTM. We", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 147, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 104, + 147, + 312, + 162 + ], + "score": 1.0, + "content": "also exploit the same RECALL method to correlate", + "type": "text" + }, + { + "bbox": [ + 312, + 148, + 336, + 160 + ], + "score": 0.92, + "content": "\\dot { \\mathcal { M } } _ { t } ^ { 1 : k }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 147, + 506, + 162 + ], + "score": 1.0, + "content": "along the vertical memory transition flow,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 157, + 500, + 173 + ], + "spans": [ + { + "bbox": [ + 104, + 157, + 367, + 173 + ], + "score": 1.0, + "content": "but it turns out to be less helpful. With the updated memory state", + "type": "text" + }, + { + "bbox": [ + 368, + 159, + 379, + 171 + ], + "score": 0.89, + "content": "\\mathcal { C } _ { t } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 157, + 500, + 173 + ], + "score": 1.0, + "content": ", the output hidden states are:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 3.5 + }, + { + "type": "interline_equation", + "bbox": [ + 179, + 177, + 431, + 271 + ], + "lines": [ + { + "bbox": [ + 179, + 177, + 431, + 271 + ], + "spans": [ + { + "bbox": [ + 179, + 177, + 431, + 271 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\mathcal { Z } _ { t } ^ { \\prime } = \\sigma ( W _ { x t } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m i } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { i } ^ { \\prime } ) } \\\\ & { \\mathcal { G } _ { t } ^ { \\prime } = \\operatorname { t a n h } ( W _ { x g } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m g } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { g } ^ { \\prime } ) } \\\\ & { \\mathcal { F } _ { t } ^ { \\prime } = \\sigma ( W _ { x f } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m f } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { f } ^ { \\prime } ) } \\\\ & { \\mathcal { M } _ { t } ^ { k } = \\mathcal { Z } _ { t } ^ { \\prime } \\odot \\mathcal { G } _ { t } ^ { \\prime } + \\mathcal { F } _ { t } ^ { \\prime } \\odot \\mathcal { M } _ { t } ^ { k - 1 } } \\\\ & { \\mathcal { O } _ { t } = \\sigma ( W _ { x o } \\ast \\mathcal { X } _ { t } + W _ { h o } \\ast \\mathcal { H } _ { t - 1 } ^ { k } + W _ { c o } \\ast \\mathcal { C } _ { t } ^ { k } + W _ { m o } \\ast \\mathcal { M } _ { t } ^ { k } + b _ { o } ) } \\\\ & { \\mathcal { H } _ { t } ^ { k } = \\mathcal { O } _ { t } \\odot \\operatorname { t a n h } ( W _ { 1 \\times 1 \\times 1 } \\ast [ \\mathcal { C } _ { t } ^ { k } , \\mathcal { M } _ { t } ^ { k } ] ) , } \\end{array}", + "type": "interline_equation", + "image_path": "6ab8c63f91cb312c7867acfaf47c85bd45b3c17901997deb11247fec972aea80.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 179, + 177, + 431, + 208.33333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 179, + 208.33333333333334, + 431, + 239.66666666666669 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 179, + 239.66666666666669, + 431, + 271.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 277, + 504, + 300 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 133, + 290 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 277, + 169, + 289 + ], + "score": 0.92, + "content": "W _ { 1 \\times 1 \\times 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 276, + 194, + 290 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 195, + 277, + 236, + 288 + ], + "score": 0.91, + "content": "1 \\times 1 \\times 1", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 276, + 476, + 290 + ], + "score": 1.0, + "content": "convolutions for the transformation of the channel number.", + "type": "text" + }, + { + "bbox": [ + 476, + 277, + 501, + 289 + ], + "score": 0.9, + "content": "\\mathcal { T } _ { t } ^ { \\prime } , \\mathcal { G } _ { t } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 276, + 505, + 290 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 288, + 424, + 301 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 123, + 301 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 289, + 136, + 300 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { t } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 288, + 337, + 301 + ], + "score": 1.0, + "content": "are gate structures of the spatiotemporal memory.", + "type": "text" + }, + { + "bbox": [ + 337, + 289, + 349, + 299 + ], + "score": 0.88, + "content": "{ \\mathcal { O } } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 288, + 424, + 301 + ], + "score": 1.0, + "content": "is the output gate.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 106, + 312, + 311, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 311, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 311, + 324 + ], + "score": 1.0, + "content": "3.3 SELF-SUPERVISED AUXILIARY LEARNING", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 330, + 505, + 375 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "score": 1.0, + "content": "For many supervised tasks such as video action recognition, there are often not enough supervisions", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "score": 1.0, + "content": "or annotations over time for training a satisfactory RNN. As an auxiliary measure to this problem,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 352, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 366 + ], + "score": 1.0, + "content": "future video prediction is considered as a promising representation learning approach that is more", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 363, + 478, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 478, + 377 + ], + "score": 1.0, + "content": "densely supervised over time and might extract useful features to assist video understanding.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 105, + 380, + 503, + 403 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "We consider two tasks: the pixel-level future frames prediction and another video-level classification", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 392, + 475, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 475, + 405 + ], + "score": 1.0, + "content": "task (early activity recognition in our case). For frames prediction, the objective function is:", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 410, + 383, + 428 + ], + "lines": [ + { + "bbox": [ + 227, + 410, + 383, + 428 + ], + "spans": [ + { + "bbox": [ + 227, + 410, + 383, + 428 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { \\mathrm { p r e d i c t i o n } } = \\left. \\mathcal { X } - \\widehat { \\mathcal { X } } \\right. _ { F } ^ { 2 } + \\left. \\mathcal { X } - \\widehat { \\mathcal { X } } \\right. _ { 1 } ,", + "type": "interline_equation", + "image_path": "2d4ccea0e282e449f34613f89d6e0c059e054bc9d8f0ff8760176a95d65e9773.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 227, + 410, + 383, + 428 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 437, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 133, + 450 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 435, + 144, + 447 + ], + "score": 0.86, + "content": "\\widehat { \\mathcal X }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 436, + 163, + 450 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 163, + 438, + 173, + 447 + ], + "score": 0.8, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 436, + 411, + 450 + ], + "score": 1.0, + "content": "are respectively predicted and ground truth future frames.", + "type": "text" + }, + { + "bbox": [ + 411, + 437, + 436, + 450 + ], + "score": 0.92, + "content": "\\| \\cdot \\| _ { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 436, + 505, + 450 + ], + "score": 1.0, + "content": "is the Frobenius", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "norm. For early activity recognition, we make the models for these two tasks share the same network", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 459, + 402, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 402, + 472 + ], + "score": 1.0, + "content": "backbone in the end-to-end training using a multi-task learning objective:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 477, + 386, + 496 + ], + "lines": [ + { + "bbox": [ + 224, + 477, + 386, + 496 + ], + "spans": [ + { + "bbox": [ + 224, + 477, + 386, + 496 + ], + "score": 0.93, + "content": "\\mathcal { L } _ { \\mathrm { r e c o g n i t i o n } } = \\lambda \\| \\mathcal { X } - \\widehat { \\mathcal { X } } \\| _ { F } ^ { 2 } + \\mathcal { L } _ { \\mathrm { c e } } ( \\mathcal { Y } , \\widehat { \\mathcal { Y } } ) ,", + "type": "interline_equation", + "image_path": "97393a6b19c8f6c6673661b998ff85dca2c5f3e950dd6f6b130560afe0000ef0.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 224, + 477, + 386, + 496 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 505, + 504, + 528 + ], + "lines": [ + { + "bbox": [ + 106, + 503, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 503, + 133, + 519 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 503, + 143, + 516 + ], + "score": 0.86, + "content": "\\widehat { \\mathcal { V } }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 503, + 160, + 519 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 161, + 505, + 170, + 516 + ], + "score": 0.82, + "content": "\\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 503, + 438, + 519 + ], + "score": 1.0, + "content": "are high-level predictions and corresponding ground truth classes.", + "type": "text" + }, + { + "bbox": [ + 438, + 506, + 453, + 516 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { c e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 503, + 506, + 519 + ], + "score": 1.0, + "content": "is the cross-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 516, + 335, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 245, + 528 + ], + "score": 1.0, + "content": "entropy loss for classification, and", + "type": "text" + }, + { + "bbox": [ + 245, + 517, + 252, + 526 + ], + "score": 0.82, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 516, + 335, + 528 + ], + "score": 1.0, + "content": "is the weight factor.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 533, + 505, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "Although improving both tasks requires proper long short-term contextual representations, there", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 543, + 504, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 504, + 557 + ], + "score": 1.0, + "content": "is no guarantee that features learned with pixel-level supervisions will fully align with any high-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "score": 1.0, + "content": "level objectives. We thus introduce a scheduled learning strategy where the objective function is", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "gradually inclined from one task to the other in a curriculum learning manner (Bengio et al., 2009).", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 577, + 399, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 264, + 589 + ], + "score": 1.0, + "content": "Specifically, we apply a linear decay to", + "type": "text" + }, + { + "bbox": [ + 265, + 578, + 272, + 587 + ], + "score": 0.82, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 577, + 391, + 589 + ], + "score": 1.0, + "content": "over the number of iterations", + "type": "text" + }, + { + "bbox": [ + 391, + 578, + 396, + 587 + ], + "score": 0.77, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 577, + 399, + 589 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29 + }, + { + "type": "interline_equation", + "bbox": [ + 246, + 596, + 362, + 609 + ], + "lines": [ + { + "bbox": [ + 246, + 596, + 362, + 609 + ], + "spans": [ + { + "bbox": [ + 246, + 596, + 362, + 609 + ], + "score": 0.91, + "content": "\\lambda ( i ) = \\operatorname* { m a x } ( \\eta , \\lambda ( 0 ) - \\epsilon \\cdot i ) ,", + "type": "interline_equation", + "image_path": "bcc72aced1b2b417896806195f79b0152fc53b21baafe890274c448fe18812aa.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 246, + 596, + 362, + 609 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 616, + 505, + 640 + ], + "lines": [ + { + "bbox": [ + 105, + 615, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 133, + 630 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 616, + 153, + 628 + ], + "score": 0.91, + "content": "\\lambda ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 615, + 171, + 630 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 171, + 618, + 178, + 628 + ], + "score": 0.8, + "content": "\\eta", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 615, + 383, + 630 + ], + "score": 1.0, + "content": "are respectively maximum and minimum values of", + "type": "text" + }, + { + "bbox": [ + 383, + 616, + 400, + 628 + ], + "score": 0.89, + "content": "\\lambda ( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 615, + 404, + 630 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 405, + 618, + 410, + 626 + ], + "score": 0.54, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 615, + 505, + 630 + ], + "score": 1.0, + "content": "controls the decreasing", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 626, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 641 + ], + "score": 1.0, + "content": "speed of the role of the auxiliary task. We call this approach the Self-supervised Auxiliary Learning.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 108, + 654, + 200, + 666 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 201, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 201, + 668 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 733 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 504, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 504, + 689 + ], + "score": 1.0, + "content": "We evaluate the proposed E3D-LSTM model on two tasks: future video prediction and early activ-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "score": 1.0, + "content": "ity recognition. These two tasks are of great importance with numerous applications that require", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "effective spatiotemporal predictive models. We demonstrate that the E3D-LSTM model performs", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "favorably against the state-of-the-art models on four challenging datasets. The source code and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 721, + 314, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 314, + 733 + ], + "score": 1.0, + "content": "trained models will be made available to the public.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 506, + 171 + ], + "lines": [], + "index": 3.5, + "bbox_fs": [ + 104, + 82, + 506, + 173 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 179, + 177, + 431, + 271 + ], + "lines": [ + { + "bbox": [ + 179, + 177, + 431, + 271 + ], + "spans": [ + { + "bbox": [ + 179, + 177, + 431, + 271 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\mathcal { Z } _ { t } ^ { \\prime } = \\sigma ( W _ { x t } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m i } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { i } ^ { \\prime } ) } \\\\ & { \\mathcal { G } _ { t } ^ { \\prime } = \\operatorname { t a n h } ( W _ { x g } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m g } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { g } ^ { \\prime } ) } \\\\ & { \\mathcal { F } _ { t } ^ { \\prime } = \\sigma ( W _ { x f } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m f } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { f } ^ { \\prime } ) } \\\\ & { \\mathcal { M } _ { t } ^ { k } = \\mathcal { Z } _ { t } ^ { \\prime } \\odot \\mathcal { G } _ { t } ^ { \\prime } + \\mathcal { F } _ { t } ^ { \\prime } \\odot \\mathcal { M } _ { t } ^ { k - 1 } } \\\\ & { \\mathcal { O } _ { t } = \\sigma ( W _ { x o } \\ast \\mathcal { X } _ { t } + W _ { h o } \\ast \\mathcal { H } _ { t - 1 } ^ { k } + W _ { c o } \\ast \\mathcal { C } _ { t } ^ { k } + W _ { m o } \\ast \\mathcal { M } _ { t } ^ { k } + b _ { o } ) } \\\\ & { \\mathcal { H } _ { t } ^ { k } = \\mathcal { O } _ { t } \\odot \\operatorname { t a n h } ( W _ { 1 \\times 1 \\times 1 } \\ast [ \\mathcal { C } _ { t } ^ { k } , \\mathcal { M } _ { t } ^ { k } ] ) , } \\end{array}", + "type": "interline_equation", + "image_path": "6ab8c63f91cb312c7867acfaf47c85bd45b3c17901997deb11247fec972aea80.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 179, + 177, + 431, + 208.33333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 179, + 208.33333333333334, + 431, + 239.66666666666669 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 179, + 239.66666666666669, + 431, + 271.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 277, + 504, + 300 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 133, + 290 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 277, + 169, + 289 + ], + "score": 0.92, + "content": "W _ { 1 \\times 1 \\times 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 276, + 194, + 290 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 195, + 277, + 236, + 288 + ], + "score": 0.91, + "content": "1 \\times 1 \\times 1", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 276, + 476, + 290 + ], + "score": 1.0, + "content": "convolutions for the transformation of the channel number.", + "type": "text" + }, + { + "bbox": [ + 476, + 277, + 501, + 289 + ], + "score": 0.9, + "content": "\\mathcal { T } _ { t } ^ { \\prime } , \\mathcal { G } _ { t } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 276, + 505, + 290 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 288, + 424, + 301 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 123, + 301 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 289, + 136, + 300 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { t } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 288, + 337, + 301 + ], + "score": 1.0, + "content": "are gate structures of the spatiotemporal memory.", + "type": "text" + }, + { + "bbox": [ + 337, + 289, + 349, + 299 + ], + "score": 0.88, + "content": "{ \\mathcal { O } } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 288, + 424, + 301 + ], + "score": 1.0, + "content": "is the output gate.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 106, + 276, + 505, + 301 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 312, + 311, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 311, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 311, + 324 + ], + "score": 1.0, + "content": "3.3 SELF-SUPERVISED AUXILIARY LEARNING", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 330, + 505, + 375 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "score": 1.0, + "content": "For many supervised tasks such as video action recognition, there are often not enough supervisions", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "score": 1.0, + "content": "or annotations over time for training a satisfactory RNN. As an auxiliary measure to this problem,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 352, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 366 + ], + "score": 1.0, + "content": "future video prediction is considered as a promising representation learning approach that is more", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 363, + 478, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 478, + 377 + ], + "score": 1.0, + "content": "densely supervised over time and might extract useful features to assist video understanding.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 331, + 506, + 377 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 380, + 503, + 403 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "We consider two tasks: the pixel-level future frames prediction and another video-level classification", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 392, + 475, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 475, + 405 + ], + "score": 1.0, + "content": "task (early activity recognition in our case). For frames prediction, the objective function is:", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 106, + 380, + 505, + 405 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 410, + 383, + 428 + ], + "lines": [ + { + "bbox": [ + 227, + 410, + 383, + 428 + ], + "spans": [ + { + "bbox": [ + 227, + 410, + 383, + 428 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { \\mathrm { p r e d i c t i o n } } = \\left. \\mathcal { X } - \\widehat { \\mathcal { X } } \\right. _ { F } ^ { 2 } + \\left. \\mathcal { X } - \\widehat { \\mathcal { X } } \\right. _ { 1 } ,", + "type": "interline_equation", + "image_path": "2d4ccea0e282e449f34613f89d6e0c059e054bc9d8f0ff8760176a95d65e9773.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 227, + 410, + 383, + 428 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 437, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 133, + 450 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 435, + 144, + 447 + ], + "score": 0.86, + "content": "\\widehat { \\mathcal X }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 436, + 163, + 450 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 163, + 438, + 173, + 447 + ], + "score": 0.8, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 436, + 411, + 450 + ], + "score": 1.0, + "content": "are respectively predicted and ground truth future frames.", + "type": "text" + }, + { + "bbox": [ + 411, + 437, + 436, + 450 + ], + "score": 0.92, + "content": "\\| \\cdot \\| _ { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 436, + 505, + 450 + ], + "score": 1.0, + "content": "is the Frobenius", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "norm. For early activity recognition, we make the models for these two tasks share the same network", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 459, + 402, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 402, + 472 + ], + "score": 1.0, + "content": "backbone in the end-to-end training using a multi-task learning objective:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 435, + 505, + 472 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 477, + 386, + 496 + ], + "lines": [ + { + "bbox": [ + 224, + 477, + 386, + 496 + ], + "spans": [ + { + "bbox": [ + 224, + 477, + 386, + 496 + ], + "score": 0.93, + "content": "\\mathcal { L } _ { \\mathrm { r e c o g n i t i o n } } = \\lambda \\| \\mathcal { X } - \\widehat { \\mathcal { X } } \\| _ { F } ^ { 2 } + \\mathcal { L } _ { \\mathrm { c e } } ( \\mathcal { Y } , \\widehat { \\mathcal { Y } } ) ,", + "type": "interline_equation", + "image_path": "97393a6b19c8f6c6673661b998ff85dca2c5f3e950dd6f6b130560afe0000ef0.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 224, + 477, + 386, + 496 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 505, + 504, + 528 + ], + "lines": [ + { + "bbox": [ + 106, + 503, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 503, + 133, + 519 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 503, + 143, + 516 + ], + "score": 0.86, + "content": "\\widehat { \\mathcal { V } }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 503, + 160, + 519 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 161, + 505, + 170, + 516 + ], + "score": 0.82, + "content": "\\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 503, + 438, + 519 + ], + "score": 1.0, + "content": "are high-level predictions and corresponding ground truth classes.", + "type": "text" + }, + { + "bbox": [ + 438, + 506, + 453, + 516 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { c e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 503, + 506, + 519 + ], + "score": 1.0, + "content": "is the cross-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 516, + 335, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 245, + 528 + ], + "score": 1.0, + "content": "entropy loss for classification, and", + "type": "text" + }, + { + "bbox": [ + 245, + 517, + 252, + 526 + ], + "score": 0.82, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 516, + 335, + 528 + ], + "score": 1.0, + "content": "is the weight factor.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 503, + 506, + 528 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 533, + 505, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "Although improving both tasks requires proper long short-term contextual representations, there", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 543, + 504, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 504, + 557 + ], + "score": 1.0, + "content": "is no guarantee that features learned with pixel-level supervisions will fully align with any high-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "score": 1.0, + "content": "level objectives. We thus introduce a scheduled learning strategy where the objective function is", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "gradually inclined from one task to the other in a curriculum learning manner (Bengio et al., 2009).", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 577, + 399, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 264, + 589 + ], + "score": 1.0, + "content": "Specifically, we apply a linear decay to", + "type": "text" + }, + { + "bbox": [ + 265, + 578, + 272, + 587 + ], + "score": 0.82, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 577, + 391, + 589 + ], + "score": 1.0, + "content": "over the number of iterations", + "type": "text" + }, + { + "bbox": [ + 391, + 578, + 396, + 587 + ], + "score": 0.77, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 577, + 399, + 589 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 533, + 505, + 589 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 246, + 596, + 362, + 609 + ], + "lines": [ + { + "bbox": [ + 246, + 596, + 362, + 609 + ], + "spans": [ + { + "bbox": [ + 246, + 596, + 362, + 609 + ], + "score": 0.91, + "content": "\\lambda ( i ) = \\operatorname* { m a x } ( \\eta , \\lambda ( 0 ) - \\epsilon \\cdot i ) ,", + "type": "interline_equation", + "image_path": "bcc72aced1b2b417896806195f79b0152fc53b21baafe890274c448fe18812aa.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 246, + 596, + 362, + 609 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 616, + 505, + 640 + ], + "lines": [ + { + "bbox": [ + 105, + 615, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 133, + 630 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 616, + 153, + 628 + ], + "score": 0.91, + "content": "\\lambda ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 615, + 171, + 630 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 171, + 618, + 178, + 628 + ], + "score": 0.8, + "content": "\\eta", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 615, + 383, + 630 + ], + "score": 1.0, + "content": "are respectively maximum and minimum values of", + "type": "text" + }, + { + "bbox": [ + 383, + 616, + 400, + 628 + ], + "score": 0.89, + "content": "\\lambda ( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 615, + 404, + 630 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 405, + 618, + 410, + 626 + ], + "score": 0.54, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 615, + 505, + 630 + ], + "score": 1.0, + "content": "controls the decreasing", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 626, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 641 + ], + "score": 1.0, + "content": "speed of the role of the auxiliary task. We call this approach the Self-supervised Auxiliary Learning.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 615, + 505, + 641 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 654, + 200, + 666 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 201, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 201, + 668 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 733 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 504, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 504, + 689 + ], + "score": 1.0, + "content": "We evaluate the proposed E3D-LSTM model on two tasks: future video prediction and early activ-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "score": 1.0, + "content": "ity recognition. These two tasks are of great importance with numerous applications that require", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "effective spatiotemporal predictive models. We demonstrate that the E3D-LSTM model performs", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "favorably against the state-of-the-art models on four challenging datasets. The source code and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 721, + 314, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 314, + 733 + ], + "score": 1.0, + "content": "trained models will be made available to the public.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 677, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 70, + 499, + 205 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 70, + 499, + 205 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 70, + 499, + 205 + ], + "spans": [ + { + "bbox": [ + 110, + 70, + 499, + 205 + ], + "score": 0.97, + "type": "image", + "image_path": "75fe0e2cd497e4e80eecc0ec9f0adce2a5a55b83ce417f48d788ca7b2d8cfb24.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 70, + 499, + 115.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 115.0, + 499, + 160.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 160.0, + 499, + 205.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 167, + 211, + 443, + 223 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 167, + 210, + 444, + 225 + ], + "spans": [ + { + "bbox": [ + 167, + 210, + 444, + 225 + ], + "score": 1.0, + "content": "Figure 3: Video prediction examples on the Moving MNIST dataset.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "table", + "bbox": [ + 135, + 260, + 472, + 378 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 231, + 505, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 232, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 505, + 243 + ], + "score": 1.0, + "content": "Table 1: Results on the Moving MNIST dataset. All models, except DFN and VPN, are trained with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 242, + 487, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 487, + 255 + ], + "score": 1.0, + "content": "a comparable number of parameters. Higher SSIM or lower MSE scores indicate better results.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "table_body", + "bbox": [ + 135, + 260, + 472, + 378 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 135, + 260, + 472, + 378 + ], + "spans": [ + { + "bbox": [ + 135, + 260, + 472, + 378 + ], + "score": 0.972, + "html": "
MODEL10→10COPY
SSIMMSESSIMMSE
CONVLSTM (SHI ET AL., 2015)0.71396.50.539143.2
DFN (DE BRABANDERE ET AL., 2016)0.72689.00.598153.9
CDNA (FINN ET AL., 2016)0.72884.20.671127.1
FRNN (OLIU ET AL., 2018)0.81968.40.694110.5
VPN BASELINE (KALCHBRENNER ET AL., 2017)0.87064.10.73678.0
PREDRNN(WANG ET AL., 2017)0.86956.50.74580.3
PREDRNN++(WANG ET AL.,2018B)0.88546.30.80769.9
E3D-LSTM0.91041.30.85256.8
", + "type": "table", + "image_path": "b36c04a632d21350cb932109d5719c85a57dff9d56130a6fe44cfa0680abd2e9.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 135, + 260, + 472, + 299.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 135, + 299.3333333333333, + 472, + 338.66666666666663 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 135, + 338.66666666666663, + 472, + 377.99999999999994 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "title", + "bbox": [ + 106, + 390, + 334, + 402 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 334, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 334, + 402 + ], + "score": 1.0, + "content": "4.1 FUTURE VIDEO PREDICTION: MOVING MNIST", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 408, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 407, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 506, + 421 + ], + "score": 1.0, + "content": "We first evaluate the E3D-LSTM model against the state-of-the-art video prediction models on a", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 420, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 432 + ], + "score": 1.0, + "content": "commonly used synthetic benchmark dataset with moving digits. All experiments are conducted", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "score": 1.0, + "content": "using TensorFlow (Abadi et al., 2016) and trained with the ADAM optimizer (Kingma & Ba, 2015)", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 442, + 504, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 170, + 453 + ], + "score": 1.0, + "content": "to minimize the", + "type": "text" + }, + { + "bbox": [ + 170, + 442, + 196, + 453 + ], + "score": 0.92, + "content": "l _ { 1 } + l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 442, + 504, + 453 + ], + "score": 1.0, + "content": "loss over every pixel in the frame. For fair comparisons, we ensure all models", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 451, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 465 + ], + "score": 1.0, + "content": "to have comparable numbers of parameters, and apply the same scheduled sampling strategy (Bengio", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 463, + 399, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 399, + 476 + ], + "score": 1.0, + "content": "et al., 2015) in order to reduce the difficulty of training recurrent models.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 482, + 505, + 526 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 505, + 495 + ], + "score": 1.0, + "content": "Dataset and Setup. The moving MNIST dataset is constructed by randomly sampling two digits", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "score": 1.0, + "content": "from the original MNIST dataset and making them float and bounce at boundaries with a constant", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 278, + 516 + ], + "score": 1.0, + "content": "velocity and angle inside a black canvas of", + "type": "text" + }, + { + "bbox": [ + 279, + 504, + 312, + 514 + ], + "score": 0.9, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "pixels. The whole dataset has a fixed number of", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 515, + 423, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 263, + 527 + ], + "score": 1.0, + "content": "entries, 10, 000 sequences for training,", + "type": "text" + }, + { + "bbox": [ + 263, + 515, + 289, + 526 + ], + "score": 0.32, + "content": "3 , 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 515, + 423, + 527 + ], + "score": 1.0, + "content": "for validation and 5, 000 for test.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 544 + ], + "score": 1.0, + "content": "We stack 4 E3D-LSTMs in the architecture illustrated in Figure 1(c), leaving out 3D-CNN encoders", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "score": 1.0, + "content": "for this task. To retain the shape of hidden states over time, the integrated 3D-Conv operators are", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 553, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 170, + 566 + ], + "score": 1.0, + "content": "composed of a", + "type": "text" + }, + { + "bbox": [ + 171, + 554, + 214, + 564 + ], + "score": 0.91, + "content": "2 \\times 5 \\times 5", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 553, + 239, + 566 + ], + "score": 1.0, + "content": "(time", + "type": "text" + }, + { + "bbox": [ + 240, + 554, + 250, + 564 + ], + "score": 0.6, + "content": "\\times", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 553, + 278, + 566 + ], + "score": 1.0, + "content": "height", + "type": "text" + }, + { + "bbox": [ + 279, + 554, + 288, + 564 + ], + "score": 0.73, + "content": "\\times", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 553, + 506, + 566 + ], + "score": 1.0, + "content": "width) convolutions and a corresponding transposed", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 565, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 576 + ], + "score": 1.0, + "content": "convolution with the same filter size. The number of hidden state channels of each E3D-LSTM is", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 575, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 505, + 588 + ], + "score": 1.0, + "content": "64. The temporal stride is set to 1 and there is one overlapping frame over consecutive time stamps.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "score": 1.0, + "content": "A single 3D-Conv layer is used as the decoder to map motion-aware hidden states to output frames.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 603, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 105, + 603, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 506, + 616 + ], + "score": 1.0, + "content": "The E3D-LSTM model is evaluated against the state-of-the-art methods including the ConvLSTM", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 614, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 506, + 627 + ], + "score": 1.0, + "content": "network (Shi et al., 2015), DFN (De Brabandere et al., 2016), CDNA (Finn et al., 2016), VPN", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "score": 1.0, + "content": "baseline model with CNN decoders (Kalchbrenner et al., 2017), PredRNN (Wang et al., 2017),", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 637, + 366, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 146, + 648 + ], + "score": 1.0, + "content": "PredRNN", + "type": "text" + }, + { + "bbox": [ + 146, + 637, + 159, + 646 + ], + "score": 0.28, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 637, + 366, + 648 + ], + "score": 1.0, + "content": "(Wang et al., 2018b) and FRNN (Oliu et al., 2018).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "Main Results. Table 1 shows the performance of the evaluated models using a common setting in", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 665, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 461, + 679 + ], + "score": 1.0, + "content": "the literature: generating 10 future frames given the previous 10 observations (denoted as", + "type": "text" + }, + { + "bbox": [ + 461, + 666, + 498, + 676 + ], + "score": 0.87, + "content": "1 0 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 665, + 505, + 679 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "We use the per-frame structural similarity index measure (SSIM) (Wang et al., 2004) and per-frame", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 393, + 701 + ], + "score": 1.0, + "content": "mean squared error (MSE) for evaluation. The SSIM ranges between", + "type": "text" + }, + { + "bbox": [ + 393, + 688, + 407, + 698 + ], + "score": 0.67, + "content": "- 1", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "and 1, representing the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "similarity between the generated image and the ground truth. As shown in the second column", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 110, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 110, + 710, + 147, + 720 + ], + "score": 0.86, + "content": "1 0 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 709, + 505, + 722 + ], + "score": 1.0, + "content": ") of Table 1, our model performs well against the state-of-the-art methods in both metrics.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "The results show that the E3D-LSTM network is effective in modeling spatiotemporal data for video", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 70, + 499, + 205 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 70, + 499, + 205 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 70, + 499, + 205 + ], + "spans": [ + { + "bbox": [ + 110, + 70, + 499, + 205 + ], + "score": 0.97, + "type": "image", + "image_path": "75fe0e2cd497e4e80eecc0ec9f0adce2a5a55b83ce417f48d788ca7b2d8cfb24.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 70, + 499, + 115.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 115.0, + 499, + 160.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 160.0, + 499, + 205.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 167, + 211, + 443, + 223 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 167, + 210, + 444, + 225 + ], + "spans": [ + { + "bbox": [ + 167, + 210, + 444, + 225 + ], + "score": 1.0, + "content": "Figure 3: Video prediction examples on the Moving MNIST dataset.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "table", + "bbox": [ + 135, + 260, + 472, + 378 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 231, + 505, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 232, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 505, + 243 + ], + "score": 1.0, + "content": "Table 1: Results on the Moving MNIST dataset. All models, except DFN and VPN, are trained with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 242, + 487, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 487, + 255 + ], + "score": 1.0, + "content": "a comparable number of parameters. Higher SSIM or lower MSE scores indicate better results.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "table_body", + "bbox": [ + 135, + 260, + 472, + 378 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 135, + 260, + 472, + 378 + ], + "spans": [ + { + "bbox": [ + 135, + 260, + 472, + 378 + ], + "score": 0.972, + "html": "
MODEL10→10COPY
SSIMMSESSIMMSE
CONVLSTM (SHI ET AL., 2015)0.71396.50.539143.2
DFN (DE BRABANDERE ET AL., 2016)0.72689.00.598153.9
CDNA (FINN ET AL., 2016)0.72884.20.671127.1
FRNN (OLIU ET AL., 2018)0.81968.40.694110.5
VPN BASELINE (KALCHBRENNER ET AL., 2017)0.87064.10.73678.0
PREDRNN(WANG ET AL., 2017)0.86956.50.74580.3
PREDRNN++(WANG ET AL.,2018B)0.88546.30.80769.9
E3D-LSTM0.91041.30.85256.8
", + "type": "table", + "image_path": "b36c04a632d21350cb932109d5719c85a57dff9d56130a6fe44cfa0680abd2e9.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 135, + 260, + 472, + 299.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 135, + 299.3333333333333, + 472, + 338.66666666666663 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 135, + 338.66666666666663, + 472, + 377.99999999999994 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "title", + "bbox": [ + 106, + 390, + 334, + 402 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 334, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 334, + 402 + ], + "score": 1.0, + "content": "4.1 FUTURE VIDEO PREDICTION: MOVING MNIST", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 408, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 407, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 506, + 421 + ], + "score": 1.0, + "content": "We first evaluate the E3D-LSTM model against the state-of-the-art video prediction models on a", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 420, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 432 + ], + "score": 1.0, + "content": "commonly used synthetic benchmark dataset with moving digits. All experiments are conducted", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "score": 1.0, + "content": "using TensorFlow (Abadi et al., 2016) and trained with the ADAM optimizer (Kingma & Ba, 2015)", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 442, + 504, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 170, + 453 + ], + "score": 1.0, + "content": "to minimize the", + "type": "text" + }, + { + "bbox": [ + 170, + 442, + 196, + 453 + ], + "score": 0.92, + "content": "l _ { 1 } + l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 442, + 504, + 453 + ], + "score": 1.0, + "content": "loss over every pixel in the frame. For fair comparisons, we ensure all models", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 451, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 465 + ], + "score": 1.0, + "content": "to have comparable numbers of parameters, and apply the same scheduled sampling strategy (Bengio", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 463, + 399, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 399, + 476 + ], + "score": 1.0, + "content": "et al., 2015) in order to reduce the difficulty of training recurrent models.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 407, + 506, + 476 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 482, + 505, + 526 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 505, + 495 + ], + "score": 1.0, + "content": "Dataset and Setup. The moving MNIST dataset is constructed by randomly sampling two digits", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "score": 1.0, + "content": "from the original MNIST dataset and making them float and bounce at boundaries with a constant", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 278, + 516 + ], + "score": 1.0, + "content": "velocity and angle inside a black canvas of", + "type": "text" + }, + { + "bbox": [ + 279, + 504, + 312, + 514 + ], + "score": 0.9, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "pixels. The whole dataset has a fixed number of", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 515, + 423, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 263, + 527 + ], + "score": 1.0, + "content": "entries, 10, 000 sequences for training,", + "type": "text" + }, + { + "bbox": [ + 263, + 515, + 289, + 526 + ], + "score": 0.32, + "content": "3 , 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 515, + 423, + 527 + ], + "score": 1.0, + "content": "for validation and 5, 000 for test.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 481, + 506, + 527 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 544 + ], + "score": 1.0, + "content": "We stack 4 E3D-LSTMs in the architecture illustrated in Figure 1(c), leaving out 3D-CNN encoders", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "score": 1.0, + "content": "for this task. To retain the shape of hidden states over time, the integrated 3D-Conv operators are", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 553, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 170, + 566 + ], + "score": 1.0, + "content": "composed of a", + "type": "text" + }, + { + "bbox": [ + 171, + 554, + 214, + 564 + ], + "score": 0.91, + "content": "2 \\times 5 \\times 5", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 553, + 239, + 566 + ], + "score": 1.0, + "content": "(time", + "type": "text" + }, + { + "bbox": [ + 240, + 554, + 250, + 564 + ], + "score": 0.6, + "content": "\\times", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 553, + 278, + 566 + ], + "score": 1.0, + "content": "height", + "type": "text" + }, + { + "bbox": [ + 279, + 554, + 288, + 564 + ], + "score": 0.73, + "content": "\\times", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 553, + 506, + 566 + ], + "score": 1.0, + "content": "width) convolutions and a corresponding transposed", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 565, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 576 + ], + "score": 1.0, + "content": "convolution with the same filter size. The number of hidden state channels of each E3D-LSTM is", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 575, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 505, + 588 + ], + "score": 1.0, + "content": "64. The temporal stride is set to 1 and there is one overlapping frame over consecutive time stamps.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "score": 1.0, + "content": "A single 3D-Conv layer is used as the decoder to map motion-aware hidden states to output frames.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 531, + 506, + 599 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 603, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 105, + 603, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 506, + 616 + ], + "score": 1.0, + "content": "The E3D-LSTM model is evaluated against the state-of-the-art methods including the ConvLSTM", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 614, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 506, + 627 + ], + "score": 1.0, + "content": "network (Shi et al., 2015), DFN (De Brabandere et al., 2016), CDNA (Finn et al., 2016), VPN", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "score": 1.0, + "content": "baseline model with CNN decoders (Kalchbrenner et al., 2017), PredRNN (Wang et al., 2017),", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 637, + 366, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 146, + 648 + ], + "score": 1.0, + "content": "PredRNN", + "type": "text" + }, + { + "bbox": [ + 146, + 637, + 159, + 646 + ], + "score": 0.28, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 637, + 366, + 648 + ], + "score": 1.0, + "content": "(Wang et al., 2018b) and FRNN (Oliu et al., 2018).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 603, + 506, + 648 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "Main Results. Table 1 shows the performance of the evaluated models using a common setting in", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 665, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 461, + 679 + ], + "score": 1.0, + "content": "the literature: generating 10 future frames given the previous 10 observations (denoted as", + "type": "text" + }, + { + "bbox": [ + 461, + 666, + 498, + 676 + ], + "score": 0.87, + "content": "1 0 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 665, + 505, + 679 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "We use the per-frame structural similarity index measure (SSIM) (Wang et al., 2004) and per-frame", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 393, + 701 + ], + "score": 1.0, + "content": "mean squared error (MSE) for evaluation. The SSIM ranges between", + "type": "text" + }, + { + "bbox": [ + 393, + 688, + 407, + 698 + ], + "score": 0.67, + "content": "- 1", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "and 1, representing the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "similarity between the generated image and the ground truth. As shown in the second column", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 110, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 110, + 710, + 147, + 720 + ], + "score": 0.86, + "content": "1 0 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 709, + 505, + 722 + ], + "score": 1.0, + "content": ") of Table 1, our model performs well against the state-of-the-art methods in both metrics.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "The results show that the E3D-LSTM network is effective in modeling spatiotemporal data for video", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 162, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 505, + 176 + ], + "score": 1.0, + "content": "prediction. Figure 3(a) shows the qualitative comparisons in which our model predicts future frames", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 174, + 301, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 301, + 185 + ], + "score": 1.0, + "content": "from entangled digits better than other methods.", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 654, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 166, + 75, + 441, + 153 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 171, + 61, + 439, + 73 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 171, + 60, + 439, + 74 + ], + "spans": [ + { + "bbox": [ + 171, + 60, + 395, + 74 + ], + "score": 1.0, + "content": "Table 2: Ablation study on the Moving MNIST dataset", + "type": "text" + }, + { + "bbox": [ + 395, + 61, + 434, + 72 + ], + "score": 0.72, + "content": "( 1 0 1 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 60, + 439, + 74 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 166, + 75, + 441, + 153 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 166, + 75, + 441, + 153 + ], + "spans": [ + { + "bbox": [ + 166, + 75, + 441, + 153 + ], + "score": 0.979, + "html": "
MODELSSIMMSE
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))0.85950.6
BASELINE 2: 3D-CNN ON TOP(FIGURE 1(B))0.86253.4
BASELINE 3: ：OURS (W/O 3D CONVOLUTIONS)0.89444.2
BASELINE 4: OURS (W/O MEMORY ATTENTION)0.88045.7
E3D-LSTM0.91041.3
", + "type": "table", + "image_path": "ede790b7a09d830d8f26501b21e136e414d2008da6a3d89c48ecc554332e378f.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 166, + 75, + 441, + 101.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 166, + 101.0, + 441, + 127.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 166, + 127.0, + 441, + 153.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 105, + 163, + 503, + 185 + ], + "lines": [ + { + "bbox": [ + 105, + 162, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 505, + 176 + ], + "score": 1.0, + "content": "prediction. Figure 3(a) shows the qualitative comparisons in which our model predicts future frames", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 174, + 301, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 301, + 185 + ], + "score": 1.0, + "content": "from entangled digits better than other methods.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 193, + 505, + 314 + ], + "lines": [ + { + "bbox": [ + 106, + 192, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 505, + 207 + ], + "score": 1.0, + "content": "Copy Test. We evaluate the proposed model using the Copy Test setting where the task is to mem-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 204, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 505, + 217 + ], + "score": 1.0, + "content": "orize useful information in a longer input sequence when the recurrent disturbance is present. The", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "score": 1.0, + "content": "input clip consists of three sub-sequences, as illustrated in Figure 3(b). Seq 1 and Seq 2 are com-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "score": 1.0, + "content": "pletely irrelevant, and ahead of them, another sub-sequence called prior context is given as the input,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "which is exactly the same as Seq 2. Frames marked by black arrows are inputs and those marked by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 248, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 506, + 261 + ], + "score": 1.0, + "content": "red arrows are expected outputs. There are two training objective: a) to predict 10 future frames of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 505, + 272 + ], + "score": 1.0, + "content": "Seq 1; and b) to predict 10 future frames of Seq 2. At the test time, we only evaluate the prediction", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 270, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 505, + 283 + ], + "score": 1.0, + "content": "result of Seq 2. The copy test evaluates the modeling capability of long-range video frame relations.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "score": 1.0, + "content": "A well-designed model should make precise predictions regarding Seq 2, as it has seen all frames of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "this sequence before. However, this task is difficult for previous LSTM networks. Because Seq 1 is", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 303, + 494, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 494, + 316 + ], + "score": 1.0, + "content": "completely irrelevant, the attempt of making predictions of Seq 1 can erase its memory of Seq 2.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 320, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 332 + ], + "score": 1.0, + "content": "The results are presented in the third column (Copy) of Table 1. All baseline models suffer from the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 332, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 343 + ], + "score": 1.0, + "content": "influence brought by irrelevant frames in Seq 2 and tend to gradually forget the salient information", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "in the prior context. However, thanks to the eidetic 3D memory, our E3D-LSTM model captures the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 353, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 365 + ], + "score": 1.0, + "content": "long-term video frame interactions and performs well in both metrics. A careful inspection of the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 363, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 376 + ], + "score": 1.0, + "content": "attention weight shows that the E3D-LSTM model can better attend to useful historical representa-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 375, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 506, + 387 + ], + "score": 1.0, + "content": "tions across multiple time stamps. The copy test suggests that the E3D-LSTM network is capable of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 386, + 315, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 315, + 398 + ], + "score": 1.0, + "content": "modeling long-range periodical motions effectively.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 405, + 505, + 514 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 505, + 416 + ], + "score": 1.0, + "content": "Ablation Study. We conduct a series of ablation studies and summarize the results in Table 2.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "First, on the first two rows, we show two alternative 3D-LSTM models with 3D-Convs outside the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "recurrent unit, including 3D-CNN at Bottom (Figure 1(a)) and 3D-CNN on Top (Figure 1(b)). The", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "score": 1.0, + "content": "performance drop validates the integration of 3D-Convs and RNN units via the eidetic 3D memory.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 449, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 505, + 460 + ], + "score": 1.0, + "content": "Second, the third baseline method is a special case where all 3D convolutional filters in our model", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 459, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 472 + ], + "score": 1.0, + "content": "are reduced to 2D. The results demonstrate the effect of capturing local spatiotemporal patterns by", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "score": 1.0, + "content": "the 3D memory within an individual recurrent state. Furthermore, the contribution of the memory", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "score": 1.0, + "content": "attention mechanism can be isolated in the fourth baseline method. Note that all evaluated models", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 492, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 505 + ], + "score": 1.0, + "content": "are trained with a similar number of parameters for fair comparisons, and the performance gain", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 504, + 378, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 378, + 515 + ], + "score": 1.0, + "content": "comes from design options rather than increased model parameters.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 28.5 + }, + { + "type": "title", + "bbox": [ + 107, + 526, + 319, + 538 + ], + "lines": [ + { + "bbox": [ + 106, + 526, + 320, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 320, + 539 + ], + "score": 1.0, + "content": "4.2 FUTURE VIDEO PREDICTION: KTH ACTION", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 544, + 458, + 556 + ], + "lines": [ + { + "bbox": [ + 106, + 544, + 460, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 460, + 558 + ], + "score": 1.0, + "content": "We evaluate the proposed E3D-LSTM model on video prediction of real-world datasets.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "Dataset and Setup. The KTH action dataset (Schuldt et al., 2004) contains 25 individuals per-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 574, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 589 + ], + "score": 1.0, + "content": "forming 6 types of actions, including walking, jogging, running, boxing, hand waving and hand", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 586, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 599 + ], + "score": 1.0, + "content": "clapping. On average, each video clip lasts 4 seconds. We follow the experimental setup in (Ville-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 597, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 609 + ], + "score": 1.0, + "content": "gas et al., 2017) by using person 1-16 for training and 17-25 for testing. Each frame is resized to", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 151, + 618 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "pixels. We employ the same E3D-LSTM network architecture detailed in Section 4.1.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 619, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 505, + 631 + ], + "score": 1.0, + "content": "Models are trained to predict next 10 frames from the previous 10 observations. The prediction", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 629, + 347, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 347, + 642 + ], + "score": 1.0, + "content": "horizon at the test time is extended to 20 or 40 time stamps.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 649, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "score": 1.0, + "content": "Results. Table 3 shows quantitative results of the proposed model and state-of-the-art methods.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "Same as prior work, we use SSIM and PSNR as metrics. Consistent with the observations on the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "moving MNIST dataset, the E3D-LSTM model performs favorably against the state-of-the-art meth-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 680, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 694 + ], + "score": 1.0, + "content": "ods across three settings of predicting future 10 frames, 20 frames, and copy test. These empirical", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 692, + 495, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 495, + 705 + ], + "score": 1.0, + "content": "results demonstrate the effectiveness of the E3D-LSTM model for modeling spatiotemporal data.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Figure 4 compares representative generated frames. We select video sequences with relatively com-", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "plicated spatiotemporal variations (in both moving trajectories and human figure sizes). In the top", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 166, + 75, + 441, + 153 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 171, + 61, + 439, + 73 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 171, + 60, + 439, + 74 + ], + "spans": [ + { + "bbox": [ + 171, + 60, + 395, + 74 + ], + "score": 1.0, + "content": "Table 2: Ablation study on the Moving MNIST dataset", + "type": "text" + }, + { + "bbox": [ + 395, + 61, + 434, + 72 + ], + "score": 0.72, + "content": "( 1 0 1 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 60, + 439, + 74 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 166, + 75, + 441, + 153 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 166, + 75, + 441, + 153 + ], + "spans": [ + { + "bbox": [ + 166, + 75, + 441, + 153 + ], + "score": 0.979, + "html": "
MODELSSIMMSE
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))0.85950.6
BASELINE 2: 3D-CNN ON TOP(FIGURE 1(B))0.86253.4
BASELINE 3: ：OURS (W/O 3D CONVOLUTIONS)0.89444.2
BASELINE 4: OURS (W/O MEMORY ATTENTION)0.88045.7
E3D-LSTM0.91041.3
", + "type": "table", + "image_path": "ede790b7a09d830d8f26501b21e136e414d2008da6a3d89c48ecc554332e378f.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 166, + 75, + 441, + 101.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 166, + 101.0, + 441, + 127.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 166, + 127.0, + 441, + 153.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 105, + 163, + 503, + 185 + ], + "lines": [], + "index": 4.5, + "bbox_fs": [ + 105, + 162, + 505, + 185 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 193, + 505, + 314 + ], + "lines": [ + { + "bbox": [ + 106, + 192, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 505, + 207 + ], + "score": 1.0, + "content": "Copy Test. We evaluate the proposed model using the Copy Test setting where the task is to mem-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 204, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 505, + 217 + ], + "score": 1.0, + "content": "orize useful information in a longer input sequence when the recurrent disturbance is present. The", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "score": 1.0, + "content": "input clip consists of three sub-sequences, as illustrated in Figure 3(b). Seq 1 and Seq 2 are com-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "score": 1.0, + "content": "pletely irrelevant, and ahead of them, another sub-sequence called prior context is given as the input,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "which is exactly the same as Seq 2. Frames marked by black arrows are inputs and those marked by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 248, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 506, + 261 + ], + "score": 1.0, + "content": "red arrows are expected outputs. There are two training objective: a) to predict 10 future frames of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 505, + 272 + ], + "score": 1.0, + "content": "Seq 1; and b) to predict 10 future frames of Seq 2. At the test time, we only evaluate the prediction", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 270, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 505, + 283 + ], + "score": 1.0, + "content": "result of Seq 2. The copy test evaluates the modeling capability of long-range video frame relations.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "score": 1.0, + "content": "A well-designed model should make precise predictions regarding Seq 2, as it has seen all frames of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "this sequence before. However, this task is difficult for previous LSTM networks. Because Seq 1 is", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 303, + 494, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 494, + 316 + ], + "score": 1.0, + "content": "completely irrelevant, the attempt of making predictions of Seq 1 can erase its memory of Seq 2.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 192, + 506, + 316 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 320, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 332 + ], + "score": 1.0, + "content": "The results are presented in the third column (Copy) of Table 1. All baseline models suffer from the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 332, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 343 + ], + "score": 1.0, + "content": "influence brought by irrelevant frames in Seq 2 and tend to gradually forget the salient information", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "in the prior context. However, thanks to the eidetic 3D memory, our E3D-LSTM model captures the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 353, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 365 + ], + "score": 1.0, + "content": "long-term video frame interactions and performs well in both metrics. A careful inspection of the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 363, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 376 + ], + "score": 1.0, + "content": "attention weight shows that the E3D-LSTM model can better attend to useful historical representa-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 375, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 506, + 387 + ], + "score": 1.0, + "content": "tions across multiple time stamps. The copy test suggests that the E3D-LSTM network is capable of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 386, + 315, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 315, + 398 + ], + "score": 1.0, + "content": "modeling long-range periodical motions effectively.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 320, + 506, + 398 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 405, + 505, + 514 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 505, + 416 + ], + "score": 1.0, + "content": "Ablation Study. We conduct a series of ablation studies and summarize the results in Table 2.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "First, on the first two rows, we show two alternative 3D-LSTM models with 3D-Convs outside the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "recurrent unit, including 3D-CNN at Bottom (Figure 1(a)) and 3D-CNN on Top (Figure 1(b)). The", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "score": 1.0, + "content": "performance drop validates the integration of 3D-Convs and RNN units via the eidetic 3D memory.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 449, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 505, + 460 + ], + "score": 1.0, + "content": "Second, the third baseline method is a special case where all 3D convolutional filters in our model", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 459, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 472 + ], + "score": 1.0, + "content": "are reduced to 2D. The results demonstrate the effect of capturing local spatiotemporal patterns by", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "score": 1.0, + "content": "the 3D memory within an individual recurrent state. Furthermore, the contribution of the memory", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "score": 1.0, + "content": "attention mechanism can be isolated in the fourth baseline method. Note that all evaluated models", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 492, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 505 + ], + "score": 1.0, + "content": "are trained with a similar number of parameters for fair comparisons, and the performance gain", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 504, + 378, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 378, + 515 + ], + "score": 1.0, + "content": "comes from design options rather than increased model parameters.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 405, + 506, + 515 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 526, + 319, + 538 + ], + "lines": [ + { + "bbox": [ + 106, + 526, + 320, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 320, + 539 + ], + "score": 1.0, + "content": "4.2 FUTURE VIDEO PREDICTION: KTH ACTION", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 544, + 458, + 556 + ], + "lines": [ + { + "bbox": [ + 106, + 544, + 460, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 460, + 558 + ], + "score": 1.0, + "content": "We evaluate the proposed E3D-LSTM model on video prediction of real-world datasets.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 106, + 544, + 460, + 558 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "Dataset and Setup. The KTH action dataset (Schuldt et al., 2004) contains 25 individuals per-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 574, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 589 + ], + "score": 1.0, + "content": "forming 6 types of actions, including walking, jogging, running, boxing, hand waving and hand", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 586, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 599 + ], + "score": 1.0, + "content": "clapping. On average, each video clip lasts 4 seconds. We follow the experimental setup in (Ville-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 597, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 609 + ], + "score": 1.0, + "content": "gas et al., 2017) by using person 1-16 for training and 17-25 for testing. Each frame is resized to", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 151, + 618 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "pixels. We employ the same E3D-LSTM network architecture detailed in Section 4.1.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 619, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 505, + 631 + ], + "score": 1.0, + "content": "Models are trained to predict next 10 frames from the previous 10 observations. The prediction", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 629, + 347, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 347, + 642 + ], + "score": 1.0, + "content": "horizon at the test time is extended to 20 or 40 time stamps.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 563, + 506, + 642 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 649, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "score": 1.0, + "content": "Results. Table 3 shows quantitative results of the proposed model and state-of-the-art methods.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "Same as prior work, we use SSIM and PSNR as metrics. Consistent with the observations on the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "moving MNIST dataset, the E3D-LSTM model performs favorably against the state-of-the-art meth-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 680, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 694 + ], + "score": 1.0, + "content": "ods across three settings of predicting future 10 frames, 20 frames, and copy test. These empirical", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 692, + 495, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 495, + 705 + ], + "score": 1.0, + "content": "results demonstrate the effectiveness of the E3D-LSTM model for modeling spatiotemporal data.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 648, + 505, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Figure 4 compares representative generated frames. We select video sequences with relatively com-", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "plicated spatiotemporal variations (in both moving trajectories and human figure sizes). In the top", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 438, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 452 + ], + "score": 1.0, + "content": "half (predicting the next 40 frames based on 10 previous frames), E3D-LSTM predicts more ac-", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 450, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 308, + 462 + ], + "score": 1.0, + "content": "curate motion trajectories into the future, whereas", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 309, + 450, + 361, + 461 + ], + "score": 0.45, + "content": "\\mathrm { P r e d R N N + + }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 361, + 450, + 506, + 462 + ], + "score": 1.0, + "content": "and ConvLSTM incorrectly predict", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "score": 1.0, + "content": "the person moving out of the scenes. The lower half shows the copy test providing the expected", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 472, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 484 + ], + "score": 1.0, + "content": "outputs as prior inputs. We directly apply models, which are trained under the first setting, to this", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "test. Without the prior context, it would be difficult to predict human motions for some cases. With", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 493, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 505, + 507 + ], + "score": 1.0, + "content": "prior inputs, E3D-LSTM benefits the most from its memories and responds well to rapid appear-", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 505, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 233, + 517 + ], + "score": 1.0, + "content": "ance change. In contrast, PredR", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 234, + 505, + 260, + 515 + ], + "score": 0.25, + "content": "\\mathrm { N N } { + } { + }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 260, + 505, + 506, + 517 + ], + "score": 1.0, + "content": "and ConvLSTM are not able to capture useful spatiotemporal", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 516, + 456, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 456, + 528 + ], + "score": 1.0, + "content": "patterns from distant observations due to the lack of modeling long-term data relations.", + "type": "text", + "cross_page": true + } + ], + "index": 17 + } + ], + "index": 48.5, + "bbox_fs": [ + 105, + 709, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 126, + 66, + 484, + 248 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 66, + 484, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 66, + 484, + 248 + ], + "spans": [ + { + "bbox": [ + 126, + 66, + 484, + 248 + ], + "score": 0.973, + "type": "image", + "image_path": "98c7186a972f6792a079875d57cc48e443d0a0c3c46eb79688381a3a07a6d3dc.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 126, + 66, + 484, + 126.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 126, + 126.66666666666666, + 484, + 187.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 126, + 187.33333333333331, + 484, + 247.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 255, + 505, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 255, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 268 + ], + "score": 1.0, + "content": "Figure 4: Comparisons of the generated frames on KTH. (Top) predictions of next 40 frames based", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 267, + 479, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 479, + 280 + ], + "score": 1.0, + "content": "on 10 previous observations. (Bottom) the copy test that requires to reproducing prior inputs.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "table", + "bbox": [ + 113, + 318, + 490, + 426 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 289, + 504, + 313 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 290, + 504, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 504, + 302 + ], + "score": 1.0, + "content": "Table 3: Quantitative evaluation of different methods on the KTH human action test set. The metrics", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 300, + 456, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 456, + 313 + ], + "score": 1.0, + "content": "are averaged over the predicted frames. Higher scores indicate better prediction results.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 113, + 318, + 490, + 426 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 318, + 490, + 426 + ], + "spans": [ + { + "bbox": [ + 113, + 318, + 490, + 426 + ], + "score": 0.981, + "html": "
MODEL10→2010→40COPY(→40)
PSNRSSIMPSNRSSIMPSNRSSIM
CONVLSTM(SHI ET AL.,2015)23.580.71222.850.63923.490.670
DFN (DE BRABANDERE ET AL., 2016)27.260.79423.010.65223.370.664
MCNET(VILLEGAS ET AL.,2017)25.950.804---1
FRNN(OLIU ET AL., 2018)26.120.77123.770.67824.000.685
PREDRNN(WANG ET AL., 2017)27.550.83924.160.70324.450.711
PREDRNN++(WANG ET AL., 2018B)28.470.86525.210.74125.900.759
E3D-LSTM29.310.87927.240.81030.590.874
", + "type": "table", + "image_path": "5596cb4faae1b7850e31d5f1d402066dbe1313a867a1a91cfd0ad3f5395ed28c.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 113, + 318, + 490, + 354.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 354.0, + 490, + 390.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 113, + 390.0, + 490, + 426.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "text", + "bbox": [ + 106, + 438, + 505, + 527 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 452 + ], + "score": 1.0, + "content": "half (predicting the next 40 frames based on 10 previous frames), E3D-LSTM predicts more ac-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 450, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 308, + 462 + ], + "score": 1.0, + "content": "curate motion trajectories into the future, whereas", + "type": "text" + }, + { + "bbox": [ + 309, + 450, + 361, + 461 + ], + "score": 0.45, + "content": "\\mathrm { P r e d R N N + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 450, + 506, + 462 + ], + "score": 1.0, + "content": "and ConvLSTM incorrectly predict", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "score": 1.0, + "content": "the person moving out of the scenes. The lower half shows the copy test providing the expected", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 472, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 484 + ], + "score": 1.0, + "content": "outputs as prior inputs. We directly apply models, which are trained under the first setting, to this", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "test. Without the prior context, it would be difficult to predict human motions for some cases. With", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 493, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 505, + 507 + ], + "score": 1.0, + "content": "prior inputs, E3D-LSTM benefits the most from its memories and responds well to rapid appear-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 505, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 233, + 517 + ], + "score": 1.0, + "content": "ance change. In contrast, PredR", + "type": "text" + }, + { + "bbox": [ + 234, + 505, + 260, + 515 + ], + "score": 0.25, + "content": "\\mathrm { N N } { + } { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 505, + 506, + 517 + ], + "score": 1.0, + "content": "and ConvLSTM are not able to capture useful spatiotemporal", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 516, + 456, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 456, + 528 + ], + "score": 1.0, + "content": "patterns from distant observations due to the lack of modeling long-term data relations.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 13.5 + }, + { + "type": "title", + "bbox": [ + 109, + 538, + 441, + 550 + ], + "lines": [ + { + "bbox": [ + 107, + 538, + 441, + 551 + ], + "spans": [ + { + "bbox": [ + 107, + 538, + 441, + 551 + ], + "score": 1.0, + "content": "4.3 A REAL VIDEO PREDICTION APPLICATION: TRAFFIC FLOW PREDICTION", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 108, + 556, + 502, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "We further evaluate our method on the TaxiBJ dataset, which contains real-world traffic flow data in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 568, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 505, + 579 + ], + "score": 1.0, + "content": "consecutive heat maps. Predicting urban traffic conditions is a complex setting, as the heat maps are", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 579, + 504, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 504, + 590 + ], + "score": 1.0, + "content": "very noisy and we do not have any underlying or additional information that can facilitate this task.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "table", + "bbox": [ + 128, + 613, + 476, + 702 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 124, + 596, + 484, + 608 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 124, + 594, + 485, + 610 + ], + "spans": [ + { + "bbox": [ + 124, + 594, + 485, + 610 + ], + "score": 1.0, + "content": "Table 4: Experimental results on the TaxiBJ dataset. We report MSE at every time stamp.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "table_body", + "bbox": [ + 128, + 613, + 476, + 702 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 128, + 613, + 476, + 702 + ], + "spans": [ + { + "bbox": [ + 128, + 613, + 476, + 702 + ], + "score": 0.982, + "html": "
MODELFRAME1FRAME 2FRAME 3FRAME 4
ST-RESNET (ZHANG ET AL., 2017)0.6880.9391.1301.288
VPN(KALCHBRENNER ET AL., 2017)0.7441.0311.2511.444
FRNN(OLIU ET AL., 2018)0.6820.8230.9891.183
PREDRNN(WANG ET AL.,2017)0.6340.9341.0471.263
PREDRNN++(WANG ET AL., 2018B)0.6410.8550.9791.158
E3D-LSTM0.6200.7730.8880.984
", + "type": "table", + "image_path": "b0f35d6c49511747d612440df0457840d2556622aa1e59af1992b77cc1e6b086.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 128, + 613, + 476, + 642.6666666666666 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 128, + 642.6666666666666, + 476, + 672.3333333333333 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 128, + 672.3333333333333, + 476, + 701.9999999999999 + ], + "spans": [], + "index": 25 + } + ] + } + ], + "index": 23.0 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "Dataset and Setup. The TaxiBJ dataset is collected from the chaotic real-world environment using", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 307, + 732 + ], + "score": 1.0, + "content": "GPS monitors of taxicabs Beijing. Each frame is a", + "type": "text" + }, + { + "bbox": [ + 307, + 721, + 353, + 731 + ], + "score": 0.92, + "content": "3 2 \\times 3 2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "heat map. The last dimension denotes", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 126, + 66, + 484, + 248 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 66, + 484, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 66, + 484, + 248 + ], + "spans": [ + { + "bbox": [ + 126, + 66, + 484, + 248 + ], + "score": 0.973, + "type": "image", + "image_path": "98c7186a972f6792a079875d57cc48e443d0a0c3c46eb79688381a3a07a6d3dc.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 126, + 66, + 484, + 126.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 126, + 126.66666666666666, + 484, + 187.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 126, + 187.33333333333331, + 484, + 247.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 255, + 505, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 255, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 268 + ], + "score": 1.0, + "content": "Figure 4: Comparisons of the generated frames on KTH. (Top) predictions of next 40 frames based", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 267, + 479, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 479, + 280 + ], + "score": 1.0, + "content": "on 10 previous observations. (Bottom) the copy test that requires to reproducing prior inputs.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "table", + "bbox": [ + 113, + 318, + 490, + 426 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 289, + 504, + 313 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 290, + 504, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 504, + 302 + ], + "score": 1.0, + "content": "Table 3: Quantitative evaluation of different methods on the KTH human action test set. The metrics", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 300, + 456, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 456, + 313 + ], + "score": 1.0, + "content": "are averaged over the predicted frames. Higher scores indicate better prediction results.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 113, + 318, + 490, + 426 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 318, + 490, + 426 + ], + "spans": [ + { + "bbox": [ + 113, + 318, + 490, + 426 + ], + "score": 0.981, + "html": "
MODEL10→2010→40COPY(→40)
PSNRSSIMPSNRSSIMPSNRSSIM
CONVLSTM(SHI ET AL.,2015)23.580.71222.850.63923.490.670
DFN (DE BRABANDERE ET AL., 2016)27.260.79423.010.65223.370.664
MCNET(VILLEGAS ET AL.,2017)25.950.804---1
FRNN(OLIU ET AL., 2018)26.120.77123.770.67824.000.685
PREDRNN(WANG ET AL., 2017)27.550.83924.160.70324.450.711
PREDRNN++(WANG ET AL., 2018B)28.470.86525.210.74125.900.759
E3D-LSTM29.310.87927.240.81030.590.874
", + "type": "table", + "image_path": "5596cb4faae1b7850e31d5f1d402066dbe1313a867a1a91cfd0ad3f5395ed28c.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 113, + 318, + 490, + 354.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 354.0, + 490, + 390.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 113, + 390.0, + 490, + 426.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "text", + "bbox": [ + 106, + 438, + 505, + 527 + ], + "lines": [], + "index": 13.5, + "bbox_fs": [ + 105, + 438, + 506, + 528 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 109, + 538, + 441, + 550 + ], + "lines": [ + { + "bbox": [ + 107, + 538, + 441, + 551 + ], + "spans": [ + { + "bbox": [ + 107, + 538, + 441, + 551 + ], + "score": 1.0, + "content": "4.3 A REAL VIDEO PREDICTION APPLICATION: TRAFFIC FLOW PREDICTION", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 108, + 556, + 502, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "We further evaluate our method on the TaxiBJ dataset, which contains real-world traffic flow data in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 568, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 505, + 579 + ], + "score": 1.0, + "content": "consecutive heat maps. Predicting urban traffic conditions is a complex setting, as the heat maps are", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 579, + 504, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 504, + 590 + ], + "score": 1.0, + "content": "very noisy and we do not have any underlying or additional information that can facilitate this task.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 106, + 556, + 505, + 590 + ] + }, + { + "type": "table", + "bbox": [ + 128, + 613, + 476, + 702 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 124, + 596, + 484, + 608 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 124, + 594, + 485, + 610 + ], + "spans": [ + { + "bbox": [ + 124, + 594, + 485, + 610 + ], + "score": 1.0, + "content": "Table 4: Experimental results on the TaxiBJ dataset. We report MSE at every time stamp.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "table_body", + "bbox": [ + 128, + 613, + 476, + 702 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 128, + 613, + 476, + 702 + ], + "spans": [ + { + "bbox": [ + 128, + 613, + 476, + 702 + ], + "score": 0.982, + "html": "
MODELFRAME1FRAME 2FRAME 3FRAME 4
ST-RESNET (ZHANG ET AL., 2017)0.6880.9391.1301.288
VPN(KALCHBRENNER ET AL., 2017)0.7441.0311.2511.444
FRNN(OLIU ET AL., 2018)0.6820.8230.9891.183
PREDRNN(WANG ET AL.,2017)0.6340.9341.0471.263
PREDRNN++(WANG ET AL., 2018B)0.6410.8550.9791.158
E3D-LSTM0.6200.7730.8880.984
", + "type": "table", + "image_path": "b0f35d6c49511747d612440df0457840d2556622aa1e59af1992b77cc1e6b086.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 128, + 613, + 476, + 642.6666666666666 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 128, + 642.6666666666666, + 476, + 672.3333333333333 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 128, + 672.3333333333333, + 476, + 701.9999999999999 + ], + "spans": [], + "index": 25 + } + ] + } + ], + "index": 23.0 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "Dataset and Setup. The TaxiBJ dataset is collected from the chaotic real-world environment using", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 307, + 732 + ], + "score": 1.0, + "content": "GPS monitors of taxicabs Beijing. Each frame is a", + "type": "text" + }, + { + "bbox": [ + 307, + 721, + 353, + 731 + ], + "score": 0.92, + "content": "3 2 \\times 3 2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "heat map. The last dimension denotes", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 199, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 506, + 211 + ], + "score": 1.0, + "content": "the entering and leaving traffic flow intensities at the same area. We split the whole dataset into a", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 210, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 506, + 222 + ], + "score": 1.0, + "content": "training set and a test set as described in (Zhang et al., 2017). We train the networks to predict 4", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 220, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 234 + ], + "score": 1.0, + "content": "frames (the next 2 hours) from 4 observations. We use the same network architecture and training", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 232, + 333, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 333, + 244 + ], + "score": 1.0, + "content": "setups as the one on Moving MNIST and KTH datasets.", + "type": "text", + "cross_page": true + } + ], + "index": 8 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 708, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 66, + 504, + 156 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 66, + 504, + 156 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 66, + 504, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 66, + 504, + 156 + ], + "score": 0.966, + "type": "image", + "image_path": "1d1f26e3ee7b1adcf8951a828d333cd1d19a35b5578e98204d3503dab5694b12.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 66, + 504, + 96.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 96.0, + 504, + 126.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 126.0, + 504, + 156.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 167, + 503, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 167, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 505, + 180 + ], + "score": 1.0, + "content": "Figure 5: Prediction results on the TaxiBJ traffic flow dataset. For ease of comparison, we visualize", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 178, + 496, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 496, + 192 + ], + "score": 1.0, + "content": "the differences between the generated heat maps and their corresponding ground truth heat maps.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 199, + 505, + 243 + ], + "lines": [ + { + "bbox": [ + 106, + 199, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 506, + 211 + ], + "score": 1.0, + "content": "the entering and leaving traffic flow intensities at the same area. We split the whole dataset into a", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 210, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 506, + 222 + ], + "score": 1.0, + "content": "training set and a test set as described in (Zhang et al., 2017). We train the networks to predict 4", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 220, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 234 + ], + "score": 1.0, + "content": "frames (the next 2 hours) from 4 observations. We use the same network architecture and training", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 232, + 333, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 333, + 244 + ], + "score": 1.0, + "content": "setups as the one on Moving MNIST and KTH datasets.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 250, + 505, + 294 + ], + "lines": [ + { + "bbox": [ + 106, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "Results. We report MSE at every time stamp in Table 4 where lower scores indicate better predic-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "tion results. We also show a prediction example in Figure 5. Furthermore, we visualize the differ-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 272, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 505, + 284 + ], + "score": 1.0, + "content": "ences between the generated heat maps and the ground truth heat maps. Overall, the E3D-LSTM", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 283, + 471, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 471, + 295 + ], + "score": 1.0, + "content": "model outperforms the other methods, with the lowest differences intensities in most areas.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 107, + 305, + 385, + 316 + ], + "lines": [ + { + "bbox": [ + 106, + 305, + 386, + 317 + ], + "spans": [ + { + "bbox": [ + 106, + 305, + 386, + 317 + ], + "score": 1.0, + "content": "4.4 EARLY ACTIVITY RECOGNITION: SOMETHING-SOMETHING", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 322, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 106, + 322, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 505, + 336 + ], + "score": 1.0, + "content": "To validate that the E3D-LSTM model can learn high-level video representations effectively, we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 333, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 506, + 347 + ], + "score": 1.0, + "content": "carry out experiments on early activity recognition. The task is to predict an activity category in", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 345, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 505, + 357 + ], + "score": 1.0, + "content": "a video after only observing a fraction of frames. We choose not to evaluate on the full-length", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "score": 1.0, + "content": "video for the activity recognition task, because when a model sees the full-length video, it may", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "make decisions solely based on the scene information, e.g. seeing only the last frame is enough to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "recognize many actions. As a result, the full-length video task may not align well with our previous", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 388, + 445, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 445, + 401 + ], + "score": 1.0, + "content": "video prediction tasks, in which the sequential tendency and causality are important.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 406, + 505, + 571 + ], + "lines": [ + { + "bbox": [ + 106, + 406, + 504, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 504, + 419 + ], + "score": 1.0, + "content": "Dataset and Setup. The something-something dataset (Goyal et al., 2017) is a recent benchmark", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "score": 1.0, + "content": "for activity/action recognition (https://20bn.com/datasets/something-something). We", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 428, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 506, + 441 + ], + "score": 1.0, + "content": "use the standard and official subset which contains 56, 769 short videos for the training set and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "7, 503 videos for the validation set on 41 action categories. The video length ranges between 2 and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 449, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 104, + 449, + 505, + 464 + ], + "score": 1.0, + "content": "6 seconds with 24 fps. We adopt the early activity recognition setting (Ma et al., 2016; Zeng et al.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 461, + 504, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 453, + 475 + ], + "score": 1.0, + "content": "2017; Zhou et al., 2018), where a model predicts an action type after observing the first", + "type": "text" + }, + { + "bbox": [ + 453, + 462, + 473, + 472 + ], + "score": 0.88, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 461, + 484, + 475 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 484, + 462, + 504, + 472 + ], + "score": 0.86, + "content": "50 \\%", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "frames of each video. As these actions appear in diverse scenes and involve interaction with differ-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "score": 1.0, + "content": "ent objects, it is challenging to predict actions even for humans (See Figure 6). There are only subtle", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 494, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 505, + 507 + ], + "score": 1.0, + "content": "differences between some actions in this dataset, such as “Poking a stack of [Something] so that the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 503, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 104, + 503, + 506, + 519 + ], + "score": 1.0, + "content": "stack collapses” versus “Poking a stack of [Something] without the stack collapsing”, or “Pouring", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 515, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 506, + 530 + ], + "score": 1.0, + "content": "[Something] into [Something]” versus “Trying to pour [Something] into [Something], but missing", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 527, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 104, + 527, + 505, + 541 + ], + "score": 1.0, + "content": "so it spills next to it”. To make a correct prediction, a model needs to exploit spatiotemporal cues", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 538, + 504, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 504, + 550 + ], + "score": 1.0, + "content": "to understand the subtle differences between actions. Namely, one can evaluate the model effective-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 505, + 562 + ], + "score": 1.0, + "content": "ness for high-level video tasks. Recognizing early action accurately requires predictions into future", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 561, + 484, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 484, + 572 + ], + "score": 1.0, + "content": "frames, which can only be achieved using an effective model based on historical observations.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 578, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "Hyper-parameters and Baselines. We use the architecture illustrated in Figure 1(c) as our model,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 589, + 504, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 504, + 600 + ], + "score": 1.0, + "content": "which consists of 2 layers of 3D-CNN encoders, 4 layers of E3D-LSTMs, and 2 layers of 3D-CNN", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 330, + 612 + ], + "score": 1.0, + "content": "decoders. The 3D-CNN encoders take 4 consecutive", + "type": "text" + }, + { + "bbox": [ + 331, + 600, + 376, + 611 + ], + "score": 0.9, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "raw frames, encode them into", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 185, + 622 + ], + "score": 0.92, + "content": "2 \\times 5 6 \\times 5 6 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "feature maps at each time stamp, and then feed them into E3D-LSTM. Each", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 622, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 317, + 634 + ], + "score": 1.0, + "content": "encoder layer has 64 filters (the filter dimensions are", + "type": "text" + }, + { + "bbox": [ + 317, + 622, + 358, + 633 + ], + "score": 0.9, + "content": "2 \\times 5 \\times 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 622, + 505, + 634 + ], + "score": 1.0, + "content": ". We use the same hyper-parameters", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 633, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 506, + 646 + ], + "score": 1.0, + "content": "for E3D-LSTMs as for video prediction. The decoder layers map the output of E3D-LSTMs back to", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 205, + 656 + ], + "score": 1.0, + "content": "RGB space, which is an", + "type": "text" + }, + { + "bbox": [ + 206, + 644, + 230, + 654 + ], + "score": 0.89, + "content": "1 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "matrix, predicting the next frame following the inputs. We train the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 317, + 668 + ], + "score": 1.0, + "content": "network to predict the next 10 frames using the front", + "type": "text" + }, + { + "bbox": [ + 318, + 655, + 337, + 666 + ], + "score": 0.88, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 654, + 348, + 668 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 349, + 655, + 369, + 666 + ], + "score": 0.89, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "frames of the video. Note that we", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "do not extend any predictive states into the future at the test time. For both training and testing, we", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "concatenate hidden representations of the top recurrent units with respect to the last 16 input time", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 221, + 700 + ], + "score": 1.0, + "content": "stamps (considering the first", + "type": "text" + }, + { + "bbox": [ + 221, + 687, + 240, + 698 + ], + "score": 0.89, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "video snippets usually have about 20 to 30 frames), and feed them", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "into the classifier for activity recognition. The classifier contains 2 layers of 3D-Convs with 128", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 207, + 722 + ], + "score": 1.0, + "content": "filters (filter dimensions:", + "type": "text" + }, + { + "bbox": [ + 208, + 710, + 248, + 720 + ], + "score": 0.91, + "content": "2 \\times 3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 709, + 304, + 722 + ], + "score": 1.0, + "content": ", filter strides:", + "type": "text" + }, + { + "bbox": [ + 305, + 710, + 345, + 720 + ], + "score": 0.9, + "content": "2 \\times 2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 709, + 406, + 722 + ], + "score": 1.0, + "content": ") followed by a", + "type": "text" + }, + { + "bbox": [ + 407, + 710, + 447, + 720 + ], + "score": 0.9, + "content": "2 \\times 2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "pooling layer.", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 338, + 732 + ], + "score": 1.0, + "content": "They transform the concatenated recurrent features from", + "type": "text" + }, + { + "bbox": [ + 339, + 721, + 419, + 731 + ], + "score": 0.91, + "content": "1 6 \\times 5 6 \\times 5 6 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 720, + 431, + 732 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 431, + 721, + 501, + 731 + ], + "score": 0.93, + "content": "1 \\times 7 \\times 7 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 720, + 505, + 732 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 42.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 66, + 504, + 156 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 66, + 504, + 156 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 66, + 504, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 66, + 504, + 156 + ], + "score": 0.966, + "type": "image", + "image_path": "1d1f26e3ee7b1adcf8951a828d333cd1d19a35b5578e98204d3503dab5694b12.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 66, + 504, + 96.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 96.0, + 504, + 126.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 126.0, + 504, + 156.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 167, + 503, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 167, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 505, + 180 + ], + "score": 1.0, + "content": "Figure 5: Prediction results on the TaxiBJ traffic flow dataset. For ease of comparison, we visualize", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 178, + 496, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 496, + 192 + ], + "score": 1.0, + "content": "the differences between the generated heat maps and their corresponding ground truth heat maps.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 199, + 505, + 243 + ], + "lines": [], + "index": 6.5, + "bbox_fs": [ + 105, + 199, + 506, + 244 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 250, + 505, + 294 + ], + "lines": [ + { + "bbox": [ + 106, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "Results. We report MSE at every time stamp in Table 4 where lower scores indicate better predic-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "tion results. We also show a prediction example in Figure 5. Furthermore, we visualize the differ-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 272, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 505, + 284 + ], + "score": 1.0, + "content": "ences between the generated heat maps and the ground truth heat maps. Overall, the E3D-LSTM", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 283, + 471, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 471, + 295 + ], + "score": 1.0, + "content": "model outperforms the other methods, with the lowest differences intensities in most areas.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 250, + 505, + 295 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 305, + 385, + 316 + ], + "lines": [ + { + "bbox": [ + 106, + 305, + 386, + 317 + ], + "spans": [ + { + "bbox": [ + 106, + 305, + 386, + 317 + ], + "score": 1.0, + "content": "4.4 EARLY ACTIVITY RECOGNITION: SOMETHING-SOMETHING", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 322, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 106, + 322, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 505, + 336 + ], + "score": 1.0, + "content": "To validate that the E3D-LSTM model can learn high-level video representations effectively, we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 333, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 506, + 347 + ], + "score": 1.0, + "content": "carry out experiments on early activity recognition. The task is to predict an activity category in", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 345, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 505, + 357 + ], + "score": 1.0, + "content": "a video after only observing a fraction of frames. We choose not to evaluate on the full-length", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 505, + 369 + ], + "score": 1.0, + "content": "video for the activity recognition task, because when a model sees the full-length video, it may", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "make decisions solely based on the scene information, e.g. seeing only the last frame is enough to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "recognize many actions. As a result, the full-length video task may not align well with our previous", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 388, + 445, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 445, + 401 + ], + "score": 1.0, + "content": "video prediction tasks, in which the sequential tendency and causality are important.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 322, + 506, + 401 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 406, + 505, + 571 + ], + "lines": [ + { + "bbox": [ + 106, + 406, + 504, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 504, + 419 + ], + "score": 1.0, + "content": "Dataset and Setup. The something-something dataset (Goyal et al., 2017) is a recent benchmark", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "score": 1.0, + "content": "for activity/action recognition (https://20bn.com/datasets/something-something). We", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 428, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 506, + 441 + ], + "score": 1.0, + "content": "use the standard and official subset which contains 56, 769 short videos for the training set and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "7, 503 videos for the validation set on 41 action categories. The video length ranges between 2 and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 449, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 104, + 449, + 505, + 464 + ], + "score": 1.0, + "content": "6 seconds with 24 fps. We adopt the early activity recognition setting (Ma et al., 2016; Zeng et al.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 461, + 504, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 453, + 475 + ], + "score": 1.0, + "content": "2017; Zhou et al., 2018), where a model predicts an action type after observing the first", + "type": "text" + }, + { + "bbox": [ + 453, + 462, + 473, + 472 + ], + "score": 0.88, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 461, + 484, + 475 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 484, + 462, + 504, + 472 + ], + "score": 0.86, + "content": "50 \\%", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "frames of each video. As these actions appear in diverse scenes and involve interaction with differ-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "score": 1.0, + "content": "ent objects, it is challenging to predict actions even for humans (See Figure 6). There are only subtle", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 494, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 505, + 507 + ], + "score": 1.0, + "content": "differences between some actions in this dataset, such as “Poking a stack of [Something] so that the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 503, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 104, + 503, + 506, + 519 + ], + "score": 1.0, + "content": "stack collapses” versus “Poking a stack of [Something] without the stack collapsing”, or “Pouring", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 515, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 506, + 530 + ], + "score": 1.0, + "content": "[Something] into [Something]” versus “Trying to pour [Something] into [Something], but missing", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 527, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 104, + 527, + 505, + 541 + ], + "score": 1.0, + "content": "so it spills next to it”. To make a correct prediction, a model needs to exploit spatiotemporal cues", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 538, + 504, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 504, + 550 + ], + "score": 1.0, + "content": "to understand the subtle differences between actions. Namely, one can evaluate the model effective-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 505, + 562 + ], + "score": 1.0, + "content": "ness for high-level video tasks. Recognizing early action accurately requires predictions into future", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 561, + 484, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 484, + 572 + ], + "score": 1.0, + "content": "frames, which can only be achieved using an effective model based on historical observations.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 28, + "bbox_fs": [ + 104, + 406, + 506, + 572 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 578, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "Hyper-parameters and Baselines. We use the architecture illustrated in Figure 1(c) as our model,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 589, + 504, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 504, + 600 + ], + "score": 1.0, + "content": "which consists of 2 layers of 3D-CNN encoders, 4 layers of E3D-LSTMs, and 2 layers of 3D-CNN", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 330, + 612 + ], + "score": 1.0, + "content": "decoders. The 3D-CNN encoders take 4 consecutive", + "type": "text" + }, + { + "bbox": [ + 331, + 600, + 376, + 611 + ], + "score": 0.9, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "raw frames, encode them into", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 185, + 622 + ], + "score": 0.92, + "content": "2 \\times 5 6 \\times 5 6 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "feature maps at each time stamp, and then feed them into E3D-LSTM. Each", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 622, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 317, + 634 + ], + "score": 1.0, + "content": "encoder layer has 64 filters (the filter dimensions are", + "type": "text" + }, + { + "bbox": [ + 317, + 622, + 358, + 633 + ], + "score": 0.9, + "content": "2 \\times 5 \\times 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 622, + 505, + 634 + ], + "score": 1.0, + "content": ". We use the same hyper-parameters", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 633, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 506, + 646 + ], + "score": 1.0, + "content": "for E3D-LSTMs as for video prediction. The decoder layers map the output of E3D-LSTMs back to", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 205, + 656 + ], + "score": 1.0, + "content": "RGB space, which is an", + "type": "text" + }, + { + "bbox": [ + 206, + 644, + 230, + 654 + ], + "score": 0.89, + "content": "1 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "matrix, predicting the next frame following the inputs. We train the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 317, + 668 + ], + "score": 1.0, + "content": "network to predict the next 10 frames using the front", + "type": "text" + }, + { + "bbox": [ + 318, + 655, + 337, + 666 + ], + "score": 0.88, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 654, + 348, + 668 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 349, + 655, + 369, + 666 + ], + "score": 0.89, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "frames of the video. Note that we", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "do not extend any predictive states into the future at the test time. For both training and testing, we", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "concatenate hidden representations of the top recurrent units with respect to the last 16 input time", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 221, + 700 + ], + "score": 1.0, + "content": "stamps (considering the first", + "type": "text" + }, + { + "bbox": [ + 221, + 687, + 240, + 698 + ], + "score": 0.89, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "video snippets usually have about 20 to 30 frames), and feed them", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "into the classifier for activity recognition. The classifier contains 2 layers of 3D-Convs with 128", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 207, + 722 + ], + "score": 1.0, + "content": "filters (filter dimensions:", + "type": "text" + }, + { + "bbox": [ + 208, + 710, + 248, + 720 + ], + "score": 0.91, + "content": "2 \\times 3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 709, + 304, + 722 + ], + "score": 1.0, + "content": ", filter strides:", + "type": "text" + }, + { + "bbox": [ + 305, + 710, + 345, + 720 + ], + "score": 0.9, + "content": "2 \\times 2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 709, + 406, + 722 + ], + "score": 1.0, + "content": ") followed by a", + "type": "text" + }, + { + "bbox": [ + 407, + 710, + 447, + 720 + ], + "score": 0.9, + "content": "2 \\times 2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "pooling layer.", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 338, + 732 + ], + "score": 1.0, + "content": "They transform the concatenated recurrent features from", + "type": "text" + }, + { + "bbox": [ + 339, + 721, + 419, + 731 + ], + "score": 0.91, + "content": "1 6 \\times 5 6 \\times 5 6 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 720, + 431, + 732 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 431, + 721, + 501, + 731 + ], + "score": 0.93, + "content": "1 \\times 7 \\times 7 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 720, + 505, + 732 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "then pass them to a 512-channel fully-connected layer followed by a 41-way classification. We", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 476, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 489 + ], + "score": 1.0, + "content": "also exploit the self-supervised auxiliary learning approach and train the model with an objective", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 487, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 235, + 500 + ], + "score": 1.0, + "content": "function in Equation 4. We set", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 235, + 488, + 253, + 500 + ], + "score": 0.91, + "content": "\\lambda ( i )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 254, + 487, + 408, + 500 + ], + "score": 1.0, + "content": "in Equation 5 to 10 in the beginning", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 408, + 488, + 435, + 499 + ], + "score": 0.85, + "content": "( i = 0", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 435, + 487, + 506, + 500 + ], + "score": 1.0, + "content": "), and decrease it", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 498, + 374, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 169, + 511 + ], + "score": 1.0, + "content": "with a speed of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 170, + 498, + 208, + 509 + ], + "score": 0.91, + "content": "2 \\times 1 0 ^ { - 5 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 209, + 498, + 337, + 511 + ], + "score": 1.0, + "content": "per iteration, lower bounded by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 337, + 499, + 370, + 510 + ], + "score": 0.9, + "content": "\\eta = 0 . 1", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 370, + 498, + 374, + 511 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 13 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 578, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 145, + 65, + 465, + 308 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 145, + 65, + 465, + 308 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 145, + 65, + 465, + 308 + ], + "spans": [ + { + "bbox": [ + 145, + 65, + 465, + 308 + ], + "score": 0.963, + "type": "image", + "image_path": "35208ce44d8e892975f7b7f7278d7cae6d8d5d530e0543611b9a676083f43948.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 145, + 65, + 465, + 146.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 145, + 146.0, + 465, + 227.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 145, + 227.0, + 465, + 308.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 318, + 503, + 352 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 344, + 331 + ], + "score": 1.0, + "content": "Figure 6: Early activity recognition results given the first", + "type": "text" + }, + { + "bbox": [ + 344, + 318, + 364, + 330 + ], + "score": 0.88, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 318, + 383, + 331 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 384, + 318, + 403, + 329 + ], + "score": 0.88, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "frames of videos on the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 330, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 505, + 342 + ], + "score": 1.0, + "content": "Something-Something validation set. The blue bars indicate making correct classifications and the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 341, + 456, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 456, + 352 + ], + "score": 1.0, + "content": "red bars are incorrect results. The length of the bar denotes the confidence of the result.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "table", + "bbox": [ + 141, + 378, + 465, + 456 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 360, + 496, + 372 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 358, + 498, + 375 + ], + "spans": [ + { + "bbox": [ + 109, + 358, + 498, + 375 + ], + "score": 1.0, + "content": "Table 5: Early activity recognition accuracy on the 41-category subset of Something-Something.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "table_body", + "bbox": [ + 141, + 378, + 465, + 456 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 141, + 378, + 465, + 456 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 465, + 456 + ], + "score": 0.975, + "html": "
MODELFRONT 25%FRONT 50%
3D-CNN9.1110.30
SEPARABLE-CNN:SEPARABLE-CONV AT BOTTOM8.949.62
(2+1)D-CNN:SEPARABLE-CONV ON TOP9.0810.17
E(2+1)D-LSTM: SEPARABLE INSIDE UNITS12.4519.86
E3D-LSTM14.5922.73
", + "type": "table", + "image_path": "9e461803e8c6ed2be0a39e8947a22ad2f9b1a416b85cd6ec06d0277054a13bba.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 141, + 378, + 465, + 404.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 141, + 404.0, + 465, + 430.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 141, + 430.0, + 465, + 456.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 7.0 + }, + { + "type": "text", + "bbox": [ + 108, + 466, + 503, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "then pass them to a 512-channel fully-connected layer followed by a 41-way classification. We", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 476, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 489 + ], + "score": 1.0, + "content": "also exploit the self-supervised auxiliary learning approach and train the model with an objective", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 487, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 235, + 500 + ], + "score": 1.0, + "content": "function in Equation 4. We set", + "type": "text" + }, + { + "bbox": [ + 235, + 488, + 253, + 500 + ], + "score": 0.91, + "content": "\\lambda ( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 487, + 408, + 500 + ], + "score": 1.0, + "content": "in Equation 5 to 10 in the beginning", + "type": "text" + }, + { + "bbox": [ + 408, + 488, + 435, + 499 + ], + "score": 0.85, + "content": "( i = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 487, + 506, + 500 + ], + "score": 1.0, + "content": "), and decrease it", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 498, + 374, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 169, + 511 + ], + "score": 1.0, + "content": "with a speed of", + "type": "text" + }, + { + "bbox": [ + 170, + 498, + 208, + 509 + ], + "score": 0.91, + "content": "2 \\times 1 0 ^ { - 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 498, + 337, + 511 + ], + "score": 1.0, + "content": "per iteration, lower bounded by", + "type": "text" + }, + { + "bbox": [ + 337, + 499, + 370, + 510 + ], + "score": 0.9, + "content": "\\eta = 0 . 1", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 498, + 374, + 511 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 515, + 504, + 571 + ], + "lines": [ + { + "bbox": [ + 106, + 516, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 506, + 528 + ], + "score": 1.0, + "content": "We evaluate the E3D-LSTM model against the state-of-the-art feed-forward 3D-Conv architectures", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 527, + 505, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 505, + 539 + ], + "score": 1.0, + "content": "including C3D/I3D (Diba et al., 2016; Carreira & Zisserman, 2017), Separable 3D-CNN (Xie et al.,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 536, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 506, + 551 + ], + "score": 1.0, + "content": "2018; Qiu et al., 2017) and (2+1)D-CNN (Tran et al., 2018). These networks achieve the state-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 549, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 505, + 561 + ], + "score": 1.0, + "content": "of-the-art results on the UCF-101 and Kinetics benchmark datasets for action recognition. For fair", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 559, + 498, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 498, + 572 + ], + "score": 1.0, + "content": "comparisons, we train these baseline models using similar backbones to the E3D-LSTM network.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 578, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 577, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 591 + ], + "score": 1.0, + "content": "Results. Table 5 shows the classification accuracy of the E3D-LSTM network against the state-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 588, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 601 + ], + "score": 1.0, + "content": "of-the-art feed-forward 3D-CNNs. The E3D-LSTM model performs favorably against the other", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 598, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 104, + 598, + 269, + 614 + ], + "score": 1.0, + "content": "methods in two settings of using the first", + "type": "text" + }, + { + "bbox": [ + 269, + 600, + 289, + 611 + ], + "score": 0.88, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 598, + 306, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 306, + 600, + 326, + 611 + ], + "score": 0.89, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 598, + 506, + 614 + ], + "score": 1.0, + "content": "frames, showing its effectiveness in learning", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 610, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 506, + 625 + ], + "score": 1.0, + "content": "high-level spatiotemporal representations. Figure 6 shows two pairs of video activities that are easy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "to confuse, especially with such limited observations. For instance, our model correctly forecasts", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 632, + 504, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 484, + 646 + ], + "score": 1.0, + "content": "the collapse of books, while only a tendency of it has been shown explicitly within the first", + "type": "text" + }, + { + "bbox": [ + 484, + 632, + 504, + 644 + ], + "score": 0.87, + "content": "2 5 \\%", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "frames. This reasoning ability comes from the integrated design of our model to capture both short-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "term motions and long-term dependencies. On the other hand, as the feed-forward 3D-CNN models", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "long-term relations by sampling and assembling, it does not perform well in finding the temporal", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "dependencies between cause and effect. We note that Zhou et al. (2018) introduced a feed-forward", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "CNN model and also reported early recognition results on the same dataset. It is not meaningful", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 698, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 462, + 711 + ], + "score": 1.0, + "content": "to compare these two methods in terms of accuracy as our model is trained only using", + "type": "text" + }, + { + "bbox": [ + 462, + 698, + 504, + 710 + ], + "score": 0.86, + "content": "2 5 \\% { - } 5 0 \\%", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "frames of a video instead of the entire video in (Zhou et al., 2018). Moreover, the two methods are", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 721, + 401, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 401, + 732 + ], + "score": 1.0, + "content": "trained using different backbone networks and different splits of datasets.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 25.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 145, + 65, + 465, + 308 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 145, + 65, + 465, + 308 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 145, + 65, + 465, + 308 + ], + "spans": [ + { + "bbox": [ + 145, + 65, + 465, + 308 + ], + "score": 0.963, + "type": "image", + "image_path": "35208ce44d8e892975f7b7f7278d7cae6d8d5d530e0543611b9a676083f43948.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 145, + 65, + 465, + 146.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 145, + 146.0, + 465, + 227.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 145, + 227.0, + 465, + 308.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 318, + 503, + 352 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 344, + 331 + ], + "score": 1.0, + "content": "Figure 6: Early activity recognition results given the first", + "type": "text" + }, + { + "bbox": [ + 344, + 318, + 364, + 330 + ], + "score": 0.88, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 318, + 383, + 331 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 384, + 318, + 403, + 329 + ], + "score": 0.88, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "frames of videos on the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 330, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 505, + 342 + ], + "score": 1.0, + "content": "Something-Something validation set. The blue bars indicate making correct classifications and the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 341, + 456, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 456, + 352 + ], + "score": 1.0, + "content": "red bars are incorrect results. The length of the bar denotes the confidence of the result.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "table", + "bbox": [ + 141, + 378, + 465, + 456 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 360, + 496, + 372 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 358, + 498, + 375 + ], + "spans": [ + { + "bbox": [ + 109, + 358, + 498, + 375 + ], + "score": 1.0, + "content": "Table 5: Early activity recognition accuracy on the 41-category subset of Something-Something.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "table_body", + "bbox": [ + 141, + 378, + 465, + 456 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 141, + 378, + 465, + 456 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 465, + 456 + ], + "score": 0.975, + "html": "
MODELFRONT 25%FRONT 50%
3D-CNN9.1110.30
SEPARABLE-CNN:SEPARABLE-CONV AT BOTTOM8.949.62
(2+1)D-CNN:SEPARABLE-CONV ON TOP9.0810.17
E(2+1)D-LSTM: SEPARABLE INSIDE UNITS12.4519.86
E3D-LSTM14.5922.73
", + "type": "table", + "image_path": "9e461803e8c6ed2be0a39e8947a22ad2f9b1a416b85cd6ec06d0277054a13bba.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 141, + 378, + 465, + 404.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 141, + 404.0, + 465, + 430.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 141, + 430.0, + 465, + 456.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 7.0 + }, + { + "type": "text", + "bbox": [ + 108, + 466, + 503, + 510 + ], + "lines": [], + "index": 11.5, + "bbox_fs": [ + 105, + 465, + 506, + 511 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 515, + 504, + 571 + ], + "lines": [ + { + "bbox": [ + 106, + 516, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 506, + 528 + ], + "score": 1.0, + "content": "We evaluate the E3D-LSTM model against the state-of-the-art feed-forward 3D-Conv architectures", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 527, + 505, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 505, + 539 + ], + "score": 1.0, + "content": "including C3D/I3D (Diba et al., 2016; Carreira & Zisserman, 2017), Separable 3D-CNN (Xie et al.,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 536, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 506, + 551 + ], + "score": 1.0, + "content": "2018; Qiu et al., 2017) and (2+1)D-CNN (Tran et al., 2018). These networks achieve the state-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 549, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 505, + 561 + ], + "score": 1.0, + "content": "of-the-art results on the UCF-101 and Kinetics benchmark datasets for action recognition. For fair", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 559, + 498, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 498, + 572 + ], + "score": 1.0, + "content": "comparisons, we train these baseline models using similar backbones to the E3D-LSTM network.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 516, + 506, + 572 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 578, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 577, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 591 + ], + "score": 1.0, + "content": "Results. Table 5 shows the classification accuracy of the E3D-LSTM network against the state-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 588, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 601 + ], + "score": 1.0, + "content": "of-the-art feed-forward 3D-CNNs. The E3D-LSTM model performs favorably against the other", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 598, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 104, + 598, + 269, + 614 + ], + "score": 1.0, + "content": "methods in two settings of using the first", + "type": "text" + }, + { + "bbox": [ + 269, + 600, + 289, + 611 + ], + "score": 0.88, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 598, + 306, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 306, + 600, + 326, + 611 + ], + "score": 0.89, + "content": "5 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 598, + 506, + 614 + ], + "score": 1.0, + "content": "frames, showing its effectiveness in learning", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 610, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 506, + 625 + ], + "score": 1.0, + "content": "high-level spatiotemporal representations. Figure 6 shows two pairs of video activities that are easy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "to confuse, especially with such limited observations. For instance, our model correctly forecasts", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 632, + 504, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 484, + 646 + ], + "score": 1.0, + "content": "the collapse of books, while only a tendency of it has been shown explicitly within the first", + "type": "text" + }, + { + "bbox": [ + 484, + 632, + 504, + 644 + ], + "score": 0.87, + "content": "2 5 \\%", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "frames. This reasoning ability comes from the integrated design of our model to capture both short-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "term motions and long-term dependencies. On the other hand, as the feed-forward 3D-CNN models", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "long-term relations by sampling and assembling, it does not perform well in finding the temporal", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "dependencies between cause and effect. We note that Zhou et al. (2018) introduced a feed-forward", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "CNN model and also reported early recognition results on the same dataset. It is not meaningful", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 698, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 462, + 711 + ], + "score": 1.0, + "content": "to compare these two methods in terms of accuracy as our model is trained only using", + "type": "text" + }, + { + "bbox": [ + 462, + 698, + 504, + 710 + ], + "score": 0.86, + "content": "2 5 \\% { - } 5 0 \\%", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "frames of a video instead of the entire video in (Zhou et al., 2018). Moreover, the two methods are", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 721, + 401, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 401, + 732 + ], + "score": 1.0, + "content": "trained using different backbone networks and different splits of datasets.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 25.5, + "bbox_fs": [ + 104, + 577, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 141, + 78, + 466, + 156 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 124, + 61, + 486, + 73 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 59, + 487, + 74 + ], + "spans": [ + { + "bbox": [ + 123, + 59, + 487, + 74 + ], + "score": 1.0, + "content": "Table 6: Ablation study of early activity recognition on the Something-Something dataset.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 141, + 78, + 466, + 156 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 141, + 78, + 466, + 156 + ], + "spans": [ + { + "bbox": [ + 141, + 78, + 466, + 156 + ], + "score": 0.977, + "html": "
MODELFRONT 25%FRONT 50%
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))10.2816.05
BASELINE 2: 3D-CNN ON TOP (FIGURE 1(B))9.6314.82
BASELINE 3:OURS W/O 3D CONVOLUTIONS9.5813.92
BASELINE 4: OURS W/O MEMORY ATTENTION11.3918.84
E3D-LSTM14.5922.73
", + "type": "table", + "image_path": "88b60c346776649f137d3c2dcfd89d74f5d0f5409dc8aefdf84b527efbdf1393.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 141, + 78, + 466, + 104.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 141, + 104.0, + 466, + 130.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 141, + 130.0, + 466, + 156.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "table", + "bbox": [ + 119, + 181, + 487, + 249 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 163, + 504, + 175 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 161, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 161, + 505, + 178 + ], + "score": 1.0, + "content": "Table 7: Accuracy comparisons of different training strategies on the Something-Something dataset.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "table_body", + "bbox": [ + 119, + 181, + 487, + 249 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 119, + 181, + 487, + 249 + ], + "spans": [ + { + "bbox": [ + 119, + 181, + 487, + 249 + ], + "score": 0.98, + "html": "
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.7820.91
PRE-TRAINED ON THE AUXILIARY TASK14.0022.15
TRAINED ON BOTH TASKS WITH A FIXED LOSS RATIO13.5720.46
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.5922.73
", + "type": "table", + "image_path": "904c177dfafb176ae79b496bde706bfa003edbd9c6867587e3c84cc82fd32d69.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 119, + 181, + 487, + 203.66666666666666 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 119, + 203.66666666666666, + 487, + 226.33333333333331 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 119, + 226.33333333333331, + 487, + 248.99999999999997 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 106, + 256, + 505, + 323 + ], + "lines": [ + { + "bbox": [ + 106, + 256, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 506, + 268 + ], + "score": 1.0, + "content": "A number of recent studies show that separating temporal and spatial convolution operations in a", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "3D-CNN model leads to better results (Xie et al., 2018; Qiu et al., 2017; Tran et al., 2018). This", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "score": 1.0, + "content": "observation is validated by our results shown in Table 5. However, it seems counter-intuitive since", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 289, + 504, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 504, + 302 + ], + "score": 1.0, + "content": "such separation leads to a pseudo-3D convolution, in which spatial and temporal filters are inde-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "score": 1.0, + "content": "pendent. Interestingly, such separation in our model leads to performance loss, suggesting the 3D", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 311, + 446, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 446, + 325 + ], + "score": 1.0, + "content": "convolution in the E3D-LSTM jointly captures the temporal and spatial information.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 107, + 328, + 505, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "score": 1.0, + "content": "Ablation Study. We conduct similar ablation studies as in Section 4.1 and summarize the results", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "in Table 6. The results from the first two rows show that our deeper integration of 3D-Convs inside", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 350, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 505, + 364 + ], + "score": 1.0, + "content": "RNNs is helpful not only for pixel-level video prediction, but also for high-level activity recognition.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 361, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 505, + 375 + ], + "score": 1.0, + "content": "The results on rows 3 and 4 show the contribution of the two important components in the proposed", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 372, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 386 + ], + "score": 1.0, + "content": "Eidetic 3D LSTM: a) 3D convolution features, and b) memory attention mechanism. Both compo-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 382, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 104, + 382, + 505, + 398 + ], + "score": 1.0, + "content": "nents are useful and important for modeling spatiotemporal data effectively. Table 7 shows applying", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 393, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 409 + ], + "score": 1.0, + "content": "self-supervised training in different settings. The proposed self-supervised auxiliary learning ap-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "score": 1.0, + "content": "proach performs better than other alternatives, including using video prediction models as network", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 416, + 483, + 429 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 483, + 429 + ], + "score": 1.0, + "content": "initialization, or training the model under these two tasks with a fixed objective function ratio.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 433, + 504, + 478 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 505, + 446 + ], + "score": 1.0, + "content": "We enable online early activity recognition by making the classifier only depend on a concatenation", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 443, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 505, + 457 + ], + "score": 1.0, + "content": "of the last 5 recurrent output states. Using Equation 1, we fix the length of the attended memory", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 174, + 469 + ], + "score": 1.0, + "content": "states by setting", + "type": "text" + }, + { + "bbox": [ + 174, + 457, + 182, + 466 + ], + "score": 0.72, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "to 5. Such settings are applied to both training and testing. Table 8 shows the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 466, + 476, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 476, + 479 + ], + "score": 1.0, + "content": "experimental results. Despite the slight decrease of accuracy, it enables an online prediction.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "table", + "bbox": [ + 119, + 502, + 487, + 546 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 485, + 503, + 497 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 483, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 505, + 498 + ], + "score": 1.0, + "content": "Table 8: Online early recognition accuracy: the classifier is built on the last 5 recurrent output states.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "table_body", + "bbox": [ + 119, + 502, + 487, + 546 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 119, + 502, + 487, + 546 + ], + "spans": [ + { + "bbox": [ + 119, + 502, + 487, + 546 + ], + "score": 0.976, + "html": "
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.4918.94
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.3020.85
", + "type": "table", + "image_path": "a1031335ebb524ddee38071c7f4e97fb72adf594501edd2998e9ce9c95972e19.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 119, + 502, + 487, + 516.6666666666666 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 119, + 516.6666666666666, + 487, + 531.3333333333333 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 119, + 531.3333333333333, + 487, + 545.9999999999999 + ], + "spans": [], + "index": 30 + } + ] + } + ], + "index": 28.0 + }, + { + "type": "title", + "bbox": [ + 107, + 558, + 196, + 570 + ], + "lines": [ + { + "bbox": [ + 104, + 555, + 198, + 574 + ], + "spans": [ + { + "bbox": [ + 104, + 555, + 198, + 574 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 579, + 505, + 678 + ], + "lines": [ + { + "bbox": [ + 106, + 580, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 505, + 592 + ], + "score": 1.0, + "content": "Spatiotemporal predictive learning has shown significant improvements in a variety of applications,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 591, + 504, + 602 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 504, + 602 + ], + "score": 1.0, + "content": "such as weather forecasting, traffic flow prediction, and physical interaction simulation. Although", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "score": 1.0, + "content": "considered to be a promising self-supervised feature learning paradigm, it seldom shows its effec-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 612, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 506, + 626 + ], + "score": 1.0, + "content": "tiveness beyond video prediction. In this paper, we presented the E3D-LSTM model based on 3D", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "convolutional recurrent units for this task. In this model, we integrated 3D-Convs into state transi-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 635, + 504, + 647 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 504, + 647 + ], + "score": 1.0, + "content": "tions to perceive short-term motions and designed a memory attentive module controlled by recur-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "score": 1.0, + "content": "rent gates to capture the long-term video frame interaction. Experimental results demonstrate that", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 655, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 670 + ], + "score": 1.0, + "content": "the E3D-LSTM model performs favorably against the state-of-the-art methods on video prediction", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 668, + 250, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 668, + 250, + 680 + ], + "score": 1.0, + "content": "and early activity recognition tasks.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36 + }, + { + "type": "title", + "bbox": [ + 108, + 691, + 200, + 701 + ], + "lines": [ + { + "bbox": [ + 107, + 692, + 200, + 702 + ], + "spans": [ + { + "bbox": [ + 107, + 692, + 200, + 702 + ], + "score": 1.0, + "content": "ACKNOWLEDGMENTS", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 710, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "We would like to thank anonymous reviewers for useful comments. Mingsheng Long was supported", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 720, + 402, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 402, + 733 + ], + "score": 1.0, + "content": "by National Natural Science Foundation of China (61772299, 71690231).", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 141, + 78, + 466, + 156 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 124, + 61, + 486, + 73 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 59, + 487, + 74 + ], + "spans": [ + { + "bbox": [ + 123, + 59, + 487, + 74 + ], + "score": 1.0, + "content": "Table 6: Ablation study of early activity recognition on the Something-Something dataset.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 141, + 78, + 466, + 156 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 141, + 78, + 466, + 156 + ], + "spans": [ + { + "bbox": [ + 141, + 78, + 466, + 156 + ], + "score": 0.977, + "html": "
MODELFRONT 25%FRONT 50%
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))10.2816.05
BASELINE 2: 3D-CNN ON TOP (FIGURE 1(B))9.6314.82
BASELINE 3:OURS W/O 3D CONVOLUTIONS9.5813.92
BASELINE 4: OURS W/O MEMORY ATTENTION11.3918.84
E3D-LSTM14.5922.73
", + "type": "table", + "image_path": "88b60c346776649f137d3c2dcfd89d74f5d0f5409dc8aefdf84b527efbdf1393.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 141, + 78, + 466, + 104.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 141, + 104.0, + 466, + 130.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 141, + 130.0, + 466, + 156.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "table", + "bbox": [ + 119, + 181, + 487, + 249 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 163, + 504, + 175 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 161, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 161, + 505, + 178 + ], + "score": 1.0, + "content": "Table 7: Accuracy comparisons of different training strategies on the Something-Something dataset.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "table_body", + "bbox": [ + 119, + 181, + 487, + 249 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 119, + 181, + 487, + 249 + ], + "spans": [ + { + "bbox": [ + 119, + 181, + 487, + 249 + ], + "score": 0.98, + "html": "
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.7820.91
PRE-TRAINED ON THE AUXILIARY TASK14.0022.15
TRAINED ON BOTH TASKS WITH A FIXED LOSS RATIO13.5720.46
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.5922.73
", + "type": "table", + "image_path": "904c177dfafb176ae79b496bde706bfa003edbd9c6867587e3c84cc82fd32d69.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 119, + 181, + 487, + 203.66666666666666 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 119, + 203.66666666666666, + 487, + 226.33333333333331 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 119, + 226.33333333333331, + 487, + 248.99999999999997 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 106, + 256, + 505, + 323 + ], + "lines": [ + { + "bbox": [ + 106, + 256, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 506, + 268 + ], + "score": 1.0, + "content": "A number of recent studies show that separating temporal and spatial convolution operations in a", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "3D-CNN model leads to better results (Xie et al., 2018; Qiu et al., 2017; Tran et al., 2018). This", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "score": 1.0, + "content": "observation is validated by our results shown in Table 5. However, it seems counter-intuitive since", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 289, + 504, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 504, + 302 + ], + "score": 1.0, + "content": "such separation leads to a pseudo-3D convolution, in which spatial and temporal filters are inde-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "score": 1.0, + "content": "pendent. Interestingly, such separation in our model leads to performance loss, suggesting the 3D", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 311, + 446, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 446, + 325 + ], + "score": 1.0, + "content": "convolution in the E3D-LSTM jointly captures the temporal and spatial information.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 256, + 506, + 325 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 328, + 505, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "score": 1.0, + "content": "Ablation Study. We conduct similar ablation studies as in Section 4.1 and summarize the results", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "in Table 6. The results from the first two rows show that our deeper integration of 3D-Convs inside", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 350, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 505, + 364 + ], + "score": 1.0, + "content": "RNNs is helpful not only for pixel-level video prediction, but also for high-level activity recognition.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 361, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 505, + 375 + ], + "score": 1.0, + "content": "The results on rows 3 and 4 show the contribution of the two important components in the proposed", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 372, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 386 + ], + "score": 1.0, + "content": "Eidetic 3D LSTM: a) 3D convolution features, and b) memory attention mechanism. Both compo-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 382, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 104, + 382, + 505, + 398 + ], + "score": 1.0, + "content": "nents are useful and important for modeling spatiotemporal data effectively. Table 7 shows applying", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 393, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 409 + ], + "score": 1.0, + "content": "self-supervised training in different settings. The proposed self-supervised auxiliary learning ap-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "score": 1.0, + "content": "proach performs better than other alternatives, including using video prediction models as network", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 416, + 483, + 429 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 483, + 429 + ], + "score": 1.0, + "content": "initialization, or training the model under these two tasks with a fixed objective function ratio.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18, + "bbox_fs": [ + 104, + 329, + 506, + 429 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 433, + 504, + 478 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 505, + 446 + ], + "score": 1.0, + "content": "We enable online early activity recognition by making the classifier only depend on a concatenation", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 443, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 505, + 457 + ], + "score": 1.0, + "content": "of the last 5 recurrent output states. Using Equation 1, we fix the length of the attended memory", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 174, + 469 + ], + "score": 1.0, + "content": "states by setting", + "type": "text" + }, + { + "bbox": [ + 174, + 457, + 182, + 466 + ], + "score": 0.72, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "to 5. Such settings are applied to both training and testing. Table 8 shows the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 466, + 476, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 476, + 479 + ], + "score": 1.0, + "content": "experimental results. Despite the slight decrease of accuracy, it enables an online prediction.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 434, + 505, + 479 + ] + }, + { + "type": "table", + "bbox": [ + 119, + 502, + 487, + 546 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 485, + 503, + 497 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 483, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 505, + 498 + ], + "score": 1.0, + "content": "Table 8: Online early recognition accuracy: the classifier is built on the last 5 recurrent output states.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "table_body", + "bbox": [ + 119, + 502, + 487, + 546 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 119, + 502, + 487, + 546 + ], + "spans": [ + { + "bbox": [ + 119, + 502, + 487, + 546 + ], + "score": 0.976, + "html": "
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.4918.94
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.3020.85
", + "type": "table", + "image_path": "a1031335ebb524ddee38071c7f4e97fb72adf594501edd2998e9ce9c95972e19.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 119, + 502, + 487, + 516.6666666666666 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 119, + 516.6666666666666, + 487, + 531.3333333333333 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 119, + 531.3333333333333, + 487, + 545.9999999999999 + ], + "spans": [], + "index": 30 + } + ] + } + ], + "index": 28.0 + }, + { + "type": "title", + "bbox": [ + 107, + 558, + 196, + 570 + ], + "lines": [ + { + "bbox": [ + 104, + 555, + 198, + 574 + ], + "spans": [ + { + "bbox": [ + 104, + 555, + 198, + 574 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 579, + 505, + 678 + ], + "lines": [ + { + "bbox": [ + 106, + 580, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 505, + 592 + ], + "score": 1.0, + "content": "Spatiotemporal predictive learning has shown significant improvements in a variety of applications,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 591, + 504, + 602 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 504, + 602 + ], + "score": 1.0, + "content": "such as weather forecasting, traffic flow prediction, and physical interaction simulation. Although", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "score": 1.0, + "content": "considered to be a promising self-supervised feature learning paradigm, it seldom shows its effec-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 612, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 506, + 626 + ], + "score": 1.0, + "content": "tiveness beyond video prediction. In this paper, we presented the E3D-LSTM model based on 3D", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "convolutional recurrent units for this task. In this model, we integrated 3D-Convs into state transi-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 635, + 504, + 647 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 504, + 647 + ], + "score": 1.0, + "content": "tions to perceive short-term motions and designed a memory attentive module controlled by recur-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "score": 1.0, + "content": "rent gates to capture the long-term video frame interaction. Experimental results demonstrate that", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 655, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 670 + ], + "score": 1.0, + "content": "the E3D-LSTM model performs favorably against the state-of-the-art methods on video prediction", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 668, + 250, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 668, + 250, + 680 + ], + "score": 1.0, + "content": "and early activity recognition tasks.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 580, + 506, + 680 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 691, + 200, + 701 + ], + "lines": [ + { + "bbox": [ + 107, + 692, + 200, + 702 + ], + "spans": [ + { + "bbox": [ + 107, + 692, + 200, + 702 + ], + "score": 1.0, + "content": "ACKNOWLEDGMENTS", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 710, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "We would like to thank anonymous reviewers for useful comments. Mingsheng Long was supported", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 720, + 402, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 402, + 733 + ], + "score": 1.0, + "content": "by National Natural Science Foundation of China (61772299, 71690231).", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 708, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 504, + 134 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 506, + 114 + ], + "score": 1.0, + "content": "Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 111, + 506, + 124 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 506, + 124 + ], + "score": 1.0, + "content": "Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, et al. Tensorflow: Large-scale machine", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 122, + 469, + 136 + ], + "spans": [ + { + "bbox": [ + 116, + 122, + 469, + 136 + ], + "score": 1.0, + "content": "learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467, 2016.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 105, + 140, + 504, + 164 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 506, + 154 + ], + "score": 1.0, + "content": "Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 152, + 219, + 163 + ], + "spans": [ + { + "bbox": [ + 115, + 152, + 219, + 163 + ], + "score": 1.0, + "content": "arXiv:1607.06450, 2016.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 105, + 170, + 504, + 194 + ], + "lines": [ + { + "bbox": [ + 105, + 170, + 505, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 505, + 184 + ], + "score": 1.0, + "content": "Samy Bengio, Oriol Vinyals, Navdeep Jaitly, and Noam Shazeer. Scheduled sampling for sequence", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 182, + 350, + 194 + ], + "spans": [ + { + "bbox": [ + 115, + 182, + 350, + 194 + ], + "score": 1.0, + "content": "prediction with recurrent neural networks. In NIPS, 2015.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 505, + 223 + ], + "lines": [ + { + "bbox": [ + 106, + 200, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 505, + 213 + ], + "score": 1.0, + "content": "Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. Curriculum learning. In", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 211, + 171, + 223 + ], + "spans": [ + { + "bbox": [ + 115, + 211, + 171, + 223 + ], + "score": 1.0, + "content": "ICML, 2009.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 105, + 230, + 504, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "Prateep Bhattacharjee and Sukhendu Das. Temporal coherency based criteria for predicting video", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 241, + 433, + 254 + ], + "spans": [ + { + "bbox": [ + 115, + 241, + 433, + 254 + ], + "score": 1.0, + "content": "frames using deep multi-stage generative adversarial networks. In NIPS, 2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 105, + 260, + 504, + 283 + ], + "lines": [ + { + "bbox": [ + 105, + 260, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 505, + 273 + ], + "score": 1.0, + "content": "Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 272, + 216, + 282 + ], + "spans": [ + { + "bbox": [ + 116, + 272, + 216, + 282 + ], + "score": 1.0, + "content": "dataset. In CVPR, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 290, + 504, + 313 + ], + "lines": [ + { + "bbox": [ + 106, + 290, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 505, + 303 + ], + "score": 1.0, + "content": "Bert De Brabandere, Xu Jia, Tinne Tuytelaars, and Luc Van Gool. Dynamic filter networks. In", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 301, + 168, + 313 + ], + "spans": [ + { + "bbox": [ + 115, + 301, + 168, + 313 + ], + "score": 1.0, + "content": "NIPS, 2016.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 320, + 496, + 333 + ], + "lines": [ + { + "bbox": [ + 106, + 320, + 495, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 495, + 334 + ], + "score": 1.0, + "content": "Emily Denton and Rob Fergus. Stochastic video generation with a learned prior. In ICML, 2018.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 104, + 339, + 504, + 363 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 353 + ], + "score": 1.0, + "content": "Ali Diba, Ali Mohammad Pazandeh, and Luc Van Gool. Efficient two-stream motion and appearance", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 351, + 354, + 362 + ], + "spans": [ + { + "bbox": [ + 116, + 351, + 354, + 362 + ], + "score": 1.0, + "content": "3d cnns for video classification. In ECCV Workshop, 2016.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 369, + 504, + 392 + ], + "lines": [ + { + "bbox": [ + 106, + 368, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 368, + 505, + 383 + ], + "score": 1.0, + "content": "Chelsea Finn, Ian Goodfellow, and Sergey Levine. Unsupervised learning for physical interaction", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 380, + 282, + 392 + ], + "spans": [ + { + "bbox": [ + 116, + 380, + 282, + 392 + ], + "score": 1.0, + "content": "through video prediction. In NIPS, 2016.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 398, + 505, + 444 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "Raghav Goyal, Samira Ebrahimi Kahou, Vincent Michalski, Joanna Materzynska, Susanne West-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 115, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "phal, Heuna Kim, Valentin Haenel, Ingo Fruend, Peter Yianilos, Moritz Mueller-Freitag, et al.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 115, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "The \"something something\" video database for learning and evaluating visual common sense. In", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 432, + 170, + 444 + ], + "spans": [ + { + "bbox": [ + 115, + 432, + 170, + 444 + ], + "score": 1.0, + "content": "ICCV, 2017.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 105, + 450, + 504, + 473 + ], + "lines": [ + { + "bbox": [ + 105, + 449, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 505, + 465 + ], + "score": 1.0, + "content": "Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computaiton, 9(8):", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 462, + 191, + 474 + ], + "spans": [ + { + "bbox": [ + 116, + 462, + 191, + 474 + ], + "score": 1.0, + "content": "1735–1780, 1997.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 480, + 503, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "score": 1.0, + "content": "Shuiwang Ji, Wei Xu, Ming Yang, and Kai Yu. 3d convolutional neural networks for human action", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 492, + 291, + 504 + ], + "spans": [ + { + "bbox": [ + 115, + 492, + 291, + 504 + ], + "score": 1.0, + "content": "recognition. TPAMI, 35(1):221–231, 2013.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 502, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 511, + 504, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 504, + 524 + ], + "score": 1.0, + "content": "Nal Kalchbrenner, Aaron van den Oord, Karen Simonyan, Ivo Danihelka, Oriol Vinyals, Alex", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 522, + 408, + 534 + ], + "spans": [ + { + "bbox": [ + 116, + 522, + 408, + 534 + ], + "score": 1.0, + "content": "Graves, and Koray Kavukcuoglu. Video pixel networks. In ICML, 2017.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 488, + 553 + ], + "lines": [ + { + "bbox": [ + 106, + 540, + 489, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 489, + 554 + ], + "score": 1.0, + "content": "Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 502, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 558, + 504, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 504, + 573 + ], + "score": 1.0, + "content": "Zhouhan Lin, Minwei Feng, Cicero Nogueira dos Santos, Mo Yu, Bing Xiang, Bowen Zhou, and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 570, + 429, + 583 + ], + "spans": [ + { + "bbox": [ + 116, + 570, + 429, + 583 + ], + "score": 1.0, + "content": "Yoshua Bengio. A structured self-attentive sentence embedding. ICLR, 2017.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 106, + 589, + 503, + 613 + ], + "lines": [ + { + "bbox": [ + 106, + 588, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 505, + 603 + ], + "score": 1.0, + "content": "Chaochao Lu, Michael Hirsch, and Bernhard Schölkopf. Flexible spatio-temporal networks for", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 117, + 601, + 253, + 612 + ], + "spans": [ + { + "bbox": [ + 117, + 601, + 253, + 612 + ], + "score": 1.0, + "content": "video prediction. In CVPR, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 106, + 619, + 503, + 642 + ], + "lines": [ + { + "bbox": [ + 106, + 620, + 504, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 504, + 633 + ], + "score": 1.0, + "content": "Shugao Ma, Leonid Sigal, and Stan Sclaroff. Learning activity progression in lstms for activity", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 630, + 302, + 642 + ], + "spans": [ + { + "bbox": [ + 116, + 630, + 302, + 642 + ], + "score": 1.0, + "content": "detection and early detection. In CVPR, 2016.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 106, + 649, + 503, + 672 + ], + "lines": [ + { + "bbox": [ + 106, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "Michael Mathieu, Camille Couprie, and Yann LeCun. Deep multi-scale video prediction beyond", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 661, + 255, + 672 + ], + "spans": [ + { + "bbox": [ + 116, + 661, + 255, + 672 + ], + "score": 1.0, + "content": "mean square error. In ICLR, 2016.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 106, + 679, + 503, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 505, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 505, + 692 + ], + "score": 1.0, + "content": "Marc Oliu, Javier Selva, and Sergio Escalera. Folded recurrent neural networks for future video", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 690, + 228, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 228, + 702 + ], + "score": 1.0, + "content": "prediction. In ECCV, 2018.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "Zhaofan Qiu, Ting Yao, and Tao Mei. Learning spatio-temporal representation with pseudo-3d", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 721, + 257, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 257, + 732 + ], + "score": 1.0, + "content": "residual networks. In ICCV, 2017.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 504, + 134 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 506, + 114 + ], + "score": 1.0, + "content": "Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 111, + 506, + 124 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 506, + 124 + ], + "score": 1.0, + "content": "Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, et al. Tensorflow: Large-scale machine", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 122, + 469, + 136 + ], + "spans": [ + { + "bbox": [ + 116, + 122, + 469, + 136 + ], + "score": 1.0, + "content": "learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467, 2016.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 99, + 506, + 136 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 140, + 504, + 164 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 506, + 154 + ], + "score": 1.0, + "content": "Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 152, + 219, + 163 + ], + "spans": [ + { + "bbox": [ + 115, + 152, + 219, + 163 + ], + "score": 1.0, + "content": "arXiv:1607.06450, 2016.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 140, + 506, + 163 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 170, + 504, + 194 + ], + "lines": [ + { + "bbox": [ + 105, + 170, + 505, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 505, + 184 + ], + "score": 1.0, + "content": "Samy Bengio, Oriol Vinyals, Navdeep Jaitly, and Noam Shazeer. Scheduled sampling for sequence", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 182, + 350, + 194 + ], + "spans": [ + { + "bbox": [ + 115, + 182, + 350, + 194 + ], + "score": 1.0, + "content": "prediction with recurrent neural networks. In NIPS, 2015.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 170, + 505, + 194 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 505, + 223 + ], + "lines": [ + { + "bbox": [ + 106, + 200, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 505, + 213 + ], + "score": 1.0, + "content": "Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. Curriculum learning. In", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 211, + 171, + 223 + ], + "spans": [ + { + "bbox": [ + 115, + 211, + 171, + 223 + ], + "score": 1.0, + "content": "ICML, 2009.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 106, + 200, + 505, + 223 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 230, + 504, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "Prateep Bhattacharjee and Sukhendu Das. Temporal coherency based criteria for predicting video", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 241, + 433, + 254 + ], + "spans": [ + { + "bbox": [ + 115, + 241, + 433, + 254 + ], + "score": 1.0, + "content": "frames using deep multi-stage generative adversarial networks. In NIPS, 2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 230, + 505, + 254 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 260, + 504, + 283 + ], + "lines": [ + { + "bbox": [ + 105, + 260, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 505, + 273 + ], + "score": 1.0, + "content": "Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 272, + 216, + 282 + ], + "spans": [ + { + "bbox": [ + 116, + 272, + 216, + 282 + ], + "score": 1.0, + "content": "dataset. In CVPR, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 260, + 505, + 282 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 290, + 504, + 313 + ], + "lines": [ + { + "bbox": [ + 106, + 290, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 505, + 303 + ], + "score": 1.0, + "content": "Bert De Brabandere, Xu Jia, Tinne Tuytelaars, and Luc Van Gool. Dynamic filter networks. In", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 301, + 168, + 313 + ], + "spans": [ + { + "bbox": [ + 115, + 301, + 168, + 313 + ], + "score": 1.0, + "content": "NIPS, 2016.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 106, + 290, + 505, + 313 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 320, + 496, + 333 + ], + "lines": [ + { + "bbox": [ + 106, + 320, + 495, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 495, + 334 + ], + "score": 1.0, + "content": "Emily Denton and Rob Fergus. Stochastic video generation with a learned prior. In ICML, 2018.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 320, + 495, + 334 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 339, + 504, + 363 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 353 + ], + "score": 1.0, + "content": "Ali Diba, Ali Mohammad Pazandeh, and Luc Van Gool. Efficient two-stream motion and appearance", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 351, + 354, + 362 + ], + "spans": [ + { + "bbox": [ + 116, + 351, + 354, + 362 + ], + "score": 1.0, + "content": "3d cnns for video classification. In ECCV Workshop, 2016.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 338, + 505, + 362 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 369, + 504, + 392 + ], + "lines": [ + { + "bbox": [ + 106, + 368, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 368, + 505, + 383 + ], + "score": 1.0, + "content": "Chelsea Finn, Ian Goodfellow, and Sergey Levine. Unsupervised learning for physical interaction", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 380, + 282, + 392 + ], + "spans": [ + { + "bbox": [ + 116, + 380, + 282, + 392 + ], + "score": 1.0, + "content": "through video prediction. In NIPS, 2016.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 368, + 505, + 392 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 398, + 505, + 444 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "Raghav Goyal, Samira Ebrahimi Kahou, Vincent Michalski, Joanna Materzynska, Susanne West-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 115, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "phal, Heuna Kim, Valentin Haenel, Ingo Fruend, Peter Yianilos, Moritz Mueller-Freitag, et al.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 115, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "The \"something something\" video database for learning and evaluating visual common sense. In", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 432, + 170, + 444 + ], + "spans": [ + { + "bbox": [ + 115, + 432, + 170, + 444 + ], + "score": 1.0, + "content": "ICCV, 2017.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 399, + 506, + 444 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 450, + 504, + 473 + ], + "lines": [ + { + "bbox": [ + 105, + 449, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 505, + 465 + ], + "score": 1.0, + "content": "Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computaiton, 9(8):", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 462, + 191, + 474 + ], + "spans": [ + { + "bbox": [ + 116, + 462, + 191, + 474 + ], + "score": 1.0, + "content": "1735–1780, 1997.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 449, + 505, + 474 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 480, + 503, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "score": 1.0, + "content": "Shuiwang Ji, Wei Xu, Ming Yang, and Kai Yu. 3d convolutional neural networks for human action", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 492, + 291, + 504 + ], + "spans": [ + { + "bbox": [ + 115, + 492, + 291, + 504 + ], + "score": 1.0, + "content": "recognition. TPAMI, 35(1):221–231, 2013.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 480, + 505, + 504 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 502, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 511, + 504, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 504, + 524 + ], + "score": 1.0, + "content": "Nal Kalchbrenner, Aaron van den Oord, Karen Simonyan, Ivo Danihelka, Oriol Vinyals, Alex", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 522, + 408, + 534 + ], + "spans": [ + { + "bbox": [ + 116, + 522, + 408, + 534 + ], + "score": 1.0, + "content": "Graves, and Koray Kavukcuoglu. Video pixel networks. In ICML, 2017.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 511, + 504, + 534 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 488, + 553 + ], + "lines": [ + { + "bbox": [ + 106, + 540, + 489, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 489, + 554 + ], + "score": 1.0, + "content": "Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 540, + 489, + 554 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 502, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 558, + 504, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 504, + 573 + ], + "score": 1.0, + "content": "Zhouhan Lin, Minwei Feng, Cicero Nogueira dos Santos, Mo Yu, Bing Xiang, Bowen Zhou, and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 570, + 429, + 583 + ], + "spans": [ + { + "bbox": [ + 116, + 570, + 429, + 583 + ], + "score": 1.0, + "content": "Yoshua Bengio. A structured self-attentive sentence embedding. ICLR, 2017.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 558, + 504, + 583 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 589, + 503, + 613 + ], + "lines": [ + { + "bbox": [ + 106, + 588, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 505, + 603 + ], + "score": 1.0, + "content": "Chaochao Lu, Michael Hirsch, and Bernhard Schölkopf. Flexible spatio-temporal networks for", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 117, + 601, + 253, + 612 + ], + "spans": [ + { + "bbox": [ + 117, + 601, + 253, + 612 + ], + "score": 1.0, + "content": "video prediction. In CVPR, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 106, + 588, + 505, + 612 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 619, + 503, + 642 + ], + "lines": [ + { + "bbox": [ + 106, + 620, + 504, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 504, + 633 + ], + "score": 1.0, + "content": "Shugao Ma, Leonid Sigal, and Stan Sclaroff. Learning activity progression in lstms for activity", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 630, + 302, + 642 + ], + "spans": [ + { + "bbox": [ + 116, + 630, + 302, + 642 + ], + "score": 1.0, + "content": "detection and early detection. In CVPR, 2016.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 620, + 504, + 642 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 649, + 503, + 672 + ], + "lines": [ + { + "bbox": [ + 106, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "Michael Mathieu, Camille Couprie, and Yann LeCun. Deep multi-scale video prediction beyond", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 661, + 255, + 672 + ], + "spans": [ + { + "bbox": [ + 116, + 661, + 255, + 672 + ], + "score": 1.0, + "content": "mean square error. In ICLR, 2016.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 106, + 649, + 505, + 672 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 679, + 503, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 505, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 505, + 692 + ], + "score": 1.0, + "content": "Marc Oliu, Javier Selva, and Sergio Escalera. Folded recurrent neural networks for future video", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 690, + 228, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 228, + 702 + ], + "score": 1.0, + "content": "prediction. In ECCV, 2018.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 678, + 505, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "Zhaofan Qiu, Ting Yao, and Tao Mei. Learning spatio-temporal representation with pseudo-3d", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 721, + 257, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 257, + 732 + ], + "score": 1.0, + "content": "residual networks. In ICCV, 2017.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 708, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "Christian Schuldt, Ivan Laptev, and Barbara Caputo. Recognizing human actions: a local svm", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 94, + 222, + 105 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 222, + 105 + ], + "score": 1.0, + "content": "approach. In ICPR, 2004.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 109, + 111, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 106, + 111, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 111, + 505, + 125 + ], + "score": 1.0, + "content": "Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 505, + 136 + ], + "score": 1.0, + "content": "Convolutional lstm network: A machine learning approach for precipitation nowcasting. In NIPS,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 133, + 143, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 133, + 143, + 146 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 153, + 504, + 176 + ], + "lines": [ + { + "bbox": [ + 105, + 151, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 167 + ], + "score": 1.0, + "content": "Nitish Srivastava, Elman Mansimov, and Ruslan Salakhutdinov. Unsupervised learning of video", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 165, + 294, + 176 + ], + "spans": [ + { + "bbox": [ + 116, + 165, + 294, + 176 + ], + "score": 1.0, + "content": "representations using lstms. In ICML, 2015.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 107, + 182, + 504, + 206 + ], + "lines": [ + { + "bbox": [ + 105, + 183, + 504, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 504, + 196 + ], + "score": 1.0, + "content": "Ilya Sutskever, Oriol Vinyals, and Quoc V Le. Sequence to sequence learning with neural networks.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 194, + 178, + 206 + ], + "spans": [ + { + "bbox": [ + 115, + 194, + 178, + 206 + ], + "score": 1.0, + "content": "In NIPS, 2014.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 504, + 236 + ], + "lines": [ + { + "bbox": [ + 105, + 211, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 505, + 227 + ], + "score": 1.0, + "content": "Du Tran, Lubomir Bourdev, Rob Fergus, Lorenzo Torresani, and Manohar Paluri. Learning spa-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 224, + 394, + 236 + ], + "spans": [ + { + "bbox": [ + 117, + 224, + 394, + 236 + ], + "score": 1.0, + "content": "tiotemporal features with 3d convolutional networks. In ICCV, 2015.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 242, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "Du Tran, Heng Wang, Lorenzo Torresani, Jamie Ray, Yann LeCun, and Manohar Paluri. A closer", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 253, + 419, + 266 + ], + "spans": [ + { + "bbox": [ + 115, + 253, + 419, + 266 + ], + "score": 1.0, + "content": "look at spatiotemporal convolutions for action recognition. In CVPR, 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 105, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 272, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 505, + 286 + ], + "score": 1.0, + "content": "Sergey Tulyakov, Ming-Yu Liu, Xiaodong Yang, and Jan Kautz. Mocogan: Decomposing motion", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 284, + 317, + 295 + ], + "spans": [ + { + "bbox": [ + 115, + 284, + 317, + 295 + ], + "score": 1.0, + "content": "and content for video generation. In CVPR, 2018.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 302, + 503, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 300, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 316 + ], + "score": 1.0, + "content": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 313, + 430, + 325 + ], + "spans": [ + { + "bbox": [ + 116, + 313, + 430, + 325 + ], + "score": 1.0, + "content": "Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In NIPS, 2017.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 332, + 503, + 355 + ], + "lines": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "score": 1.0, + "content": "Ruben Villegas, Jimei Yang, Seunghoon Hong, Xunyu Lin, and Honglak Lee. Decomposing motion", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 343, + 382, + 356 + ], + "spans": [ + { + "bbox": [ + 115, + 343, + 382, + 356 + ], + "score": 1.0, + "content": "and content for natural video sequence prediction. In ICLR, 2017.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 107, + 362, + 503, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 375 + ], + "score": 1.0, + "content": "Carl Vondrick, Hamed Pirsiavash, and Antonio Torralba. Generating videos with scene dynamics.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 373, + 178, + 384 + ], + "spans": [ + { + "bbox": [ + 115, + 373, + 178, + 384 + ], + "score": 1.0, + "content": "In NIPS, 2016.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 502, + 415 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "score": 1.0, + "content": "Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 117, + 403, + 176, + 415 + ], + "spans": [ + { + "bbox": [ + 117, + 403, + 176, + 415 + ], + "score": 1.0, + "content": "CVPR, 2018a.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 421, + 502, + 445 + ], + "lines": [ + { + "bbox": [ + 106, + 422, + 504, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 504, + 435 + ], + "score": 1.0, + "content": "Yunbo Wang, Mingsheng Long, Jianmin Wang, Zhifeng Gao, and S Yu Philip. Predrnn: Recurrent", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 432, + 449, + 446 + ], + "spans": [ + { + "bbox": [ + 115, + 432, + 449, + 446 + ], + "score": 1.0, + "content": "neural networks for predictive learning using spatiotemporal lstms. In NIPS, 2017.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 451, + 503, + 475 + ], + "lines": [ + { + "bbox": [ + 106, + 452, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 452, + 465 + ], + "score": 1.0, + "content": "Yunbo Wang, Zhifeng Gao, Mingsheng Long, Jianmin Wang, and Philip S Yu. Predrnn", + "type": "text" + }, + { + "bbox": [ + 452, + 454, + 464, + 462 + ], + "score": 0.57, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 452, + 505, + 465 + ], + "score": 1.0, + "content": ": Towards", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 462, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 462, + 505, + 476 + ], + "score": 1.0, + "content": "a resolution of the deep-in-time dilemma in spatiotemporal predictive learning. In ICML, 2018b.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 481, + 502, + 505 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 504, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 504, + 495 + ], + "score": 1.0, + "content": "Zhou Wang, A. C Bovik, H. R Sheikh, and E. P Simoncelli. Image quality assessment: from error", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 117, + 493, + 337, + 505 + ], + "spans": [ + { + "bbox": [ + 117, + 493, + 337, + 505 + ], + "score": 1.0, + "content": "visibility to structural similarity. TIP, 13(4):600, 2004.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 511, + 503, + 534 + ], + "lines": [ + { + "bbox": [ + 106, + 512, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 525 + ], + "score": 1.0, + "content": "Nevan Wichers, Ruben Villegas, Dumitru Erhan, and Honglak Lee. Hierarchical long-term video", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 523, + 309, + 535 + ], + "spans": [ + { + "bbox": [ + 115, + 523, + 309, + 535 + ], + "score": 1.0, + "content": "prediction without supervision. In ICML, 2018.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 503, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 540, + 504, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 504, + 555 + ], + "score": 1.0, + "content": "Saining Xie, Chen Sun, Jonathan Huang, Zhuowen Tu, and Kevin Murphy. Rethinking spatiotem-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 553, + 372, + 565 + ], + "spans": [ + { + "bbox": [ + 115, + 553, + 372, + 565 + ], + "score": 1.0, + "content": "poral feature learning for video understanding. In ECCV, 2018.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 571, + 503, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "Jingwei Xu, Bingbing Ni, Zefan Li, Shuo Cheng, and Xiaokang Yang. Structure preserving video", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 582, + 228, + 594 + ], + "spans": [ + { + "bbox": [ + 115, + 582, + 228, + 594 + ], + "score": 1.0, + "content": "prediction. In CVPR, 2018.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 106, + 601, + 503, + 624 + ], + "lines": [ + { + "bbox": [ + 106, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 505, + 614 + ], + "score": 1.0, + "content": "Kuo-Hao Zeng, William B Shen, De-An Huang, Min Sun, and Juan Carlos Niebles. Visual forecast-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 613, + 374, + 624 + ], + "spans": [ + { + "bbox": [ + 115, + 613, + 374, + 624 + ], + "score": 1.0, + "content": "ing by imitating dynamics in natural sequences. In ICCV, 2017.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 105, + 631, + 503, + 654 + ], + "lines": [ + { + "bbox": [ + 106, + 631, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 644 + ], + "score": 1.0, + "content": "Junbo Zhang, Yu Zheng, and Dekang Qi. Deep spatio-temporal residual networks for citywide", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 642, + 277, + 653 + ], + "spans": [ + { + "bbox": [ + 116, + 642, + 277, + 653 + ], + "score": 1.0, + "content": "crowd flows prediction. In AAAI, 2017.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 661, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 105, + 660, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 506, + 675 + ], + "score": 1.0, + "content": "Bolei Zhou, Alex Andonian, and Antonio Torralba. Temporal relational reasoning in videos. In", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 672, + 172, + 684 + ], + "spans": [ + { + "bbox": [ + 116, + 672, + 172, + 684 + ], + "score": 1.0, + "content": "ECCV, 2018.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "Christian Schuldt, Ivan Laptev, and Barbara Caputo. Recognizing human actions: a local svm", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 94, + 222, + 105 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 222, + 105 + ], + "score": 1.0, + "content": "approach. In ICPR, 2004.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 111, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 106, + 111, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 111, + 505, + 125 + ], + "score": 1.0, + "content": "Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 505, + 136 + ], + "score": 1.0, + "content": "Convolutional lstm network: A machine learning approach for precipitation nowcasting. In NIPS,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 133, + 143, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 133, + 143, + 146 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 106, + 111, + 505, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 153, + 504, + 176 + ], + "lines": [ + { + "bbox": [ + 105, + 151, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 167 + ], + "score": 1.0, + "content": "Nitish Srivastava, Elman Mansimov, and Ruslan Salakhutdinov. Unsupervised learning of video", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 165, + 294, + 176 + ], + "spans": [ + { + "bbox": [ + 116, + 165, + 294, + 176 + ], + "score": 1.0, + "content": "representations using lstms. In ICML, 2015.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 105, + 151, + 506, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 182, + 504, + 206 + ], + "lines": [ + { + "bbox": [ + 105, + 183, + 504, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 504, + 196 + ], + "score": 1.0, + "content": "Ilya Sutskever, Oriol Vinyals, and Quoc V Le. Sequence to sequence learning with neural networks.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 194, + 178, + 206 + ], + "spans": [ + { + "bbox": [ + 115, + 194, + 178, + 206 + ], + "score": 1.0, + "content": "In NIPS, 2014.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 183, + 504, + 206 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 504, + 236 + ], + "lines": [ + { + "bbox": [ + 105, + 211, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 505, + 227 + ], + "score": 1.0, + "content": "Du Tran, Lubomir Bourdev, Rob Fergus, Lorenzo Torresani, and Manohar Paluri. Learning spa-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 224, + 394, + 236 + ], + "spans": [ + { + "bbox": [ + 117, + 224, + 394, + 236 + ], + "score": 1.0, + "content": "tiotemporal features with 3d convolutional networks. In ICCV, 2015.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 211, + 505, + 236 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 242, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "Du Tran, Heng Wang, Lorenzo Torresani, Jamie Ray, Yann LeCun, and Manohar Paluri. A closer", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 253, + 419, + 266 + ], + "spans": [ + { + "bbox": [ + 115, + 253, + 419, + 266 + ], + "score": 1.0, + "content": "look at spatiotemporal convolutions for action recognition. In CVPR, 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 106, + 243, + 505, + 266 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 272, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 505, + 286 + ], + "score": 1.0, + "content": "Sergey Tulyakov, Ming-Yu Liu, Xiaodong Yang, and Jan Kautz. Mocogan: Decomposing motion", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 284, + 317, + 295 + ], + "spans": [ + { + "bbox": [ + 115, + 284, + 317, + 295 + ], + "score": 1.0, + "content": "and content for video generation. In CVPR, 2018.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 272, + 505, + 295 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 302, + 503, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 300, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 316 + ], + "score": 1.0, + "content": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 313, + 430, + 325 + ], + "spans": [ + { + "bbox": [ + 116, + 313, + 430, + 325 + ], + "score": 1.0, + "content": "Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In NIPS, 2017.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 300, + 506, + 325 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 332, + 503, + 355 + ], + "lines": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 505, + 345 + ], + "score": 1.0, + "content": "Ruben Villegas, Jimei Yang, Seunghoon Hong, Xunyu Lin, and Honglak Lee. Decomposing motion", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 343, + 382, + 356 + ], + "spans": [ + { + "bbox": [ + 115, + 343, + 382, + 356 + ], + "score": 1.0, + "content": "and content for natural video sequence prediction. In ICLR, 2017.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 332, + 505, + 356 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 362, + 503, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 375 + ], + "score": 1.0, + "content": "Carl Vondrick, Hamed Pirsiavash, and Antonio Torralba. Generating videos with scene dynamics.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 373, + 178, + 384 + ], + "spans": [ + { + "bbox": [ + 115, + 373, + 178, + 384 + ], + "score": 1.0, + "content": "In NIPS, 2016.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 360, + 505, + 384 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 502, + 415 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "score": 1.0, + "content": "Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 117, + 403, + 176, + 415 + ], + "spans": [ + { + "bbox": [ + 117, + 403, + 176, + 415 + ], + "score": 1.0, + "content": "CVPR, 2018a.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 391, + 505, + 415 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 421, + 502, + 445 + ], + "lines": [ + { + "bbox": [ + 106, + 422, + 504, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 504, + 435 + ], + "score": 1.0, + "content": "Yunbo Wang, Mingsheng Long, Jianmin Wang, Zhifeng Gao, and S Yu Philip. Predrnn: Recurrent", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 432, + 449, + 446 + ], + "spans": [ + { + "bbox": [ + 115, + 432, + 449, + 446 + ], + "score": 1.0, + "content": "neural networks for predictive learning using spatiotemporal lstms. In NIPS, 2017.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 422, + 504, + 446 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 451, + 503, + 475 + ], + "lines": [ + { + "bbox": [ + 106, + 452, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 452, + 465 + ], + "score": 1.0, + "content": "Yunbo Wang, Zhifeng Gao, Mingsheng Long, Jianmin Wang, and Philip S Yu. Predrnn", + "type": "text" + }, + { + "bbox": [ + 452, + 454, + 464, + 462 + ], + "score": 0.57, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 452, + 505, + 465 + ], + "score": 1.0, + "content": ": Towards", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 462, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 462, + 505, + 476 + ], + "score": 1.0, + "content": "a resolution of the deep-in-time dilemma in spatiotemporal predictive learning. In ICML, 2018b.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 452, + 505, + 476 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 481, + 502, + 505 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 504, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 504, + 495 + ], + "score": 1.0, + "content": "Zhou Wang, A. C Bovik, H. R Sheikh, and E. P Simoncelli. Image quality assessment: from error", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 117, + 493, + 337, + 505 + ], + "spans": [ + { + "bbox": [ + 117, + 493, + 337, + 505 + ], + "score": 1.0, + "content": "visibility to structural similarity. TIP, 13(4):600, 2004.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 481, + 504, + 505 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 511, + 503, + 534 + ], + "lines": [ + { + "bbox": [ + 106, + 512, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 525 + ], + "score": 1.0, + "content": "Nevan Wichers, Ruben Villegas, Dumitru Erhan, and Honglak Lee. Hierarchical long-term video", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 523, + 309, + 535 + ], + "spans": [ + { + "bbox": [ + 115, + 523, + 309, + 535 + ], + "score": 1.0, + "content": "prediction without supervision. In ICML, 2018.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 106, + 512, + 505, + 535 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 503, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 540, + 504, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 504, + 555 + ], + "score": 1.0, + "content": "Saining Xie, Chen Sun, Jonathan Huang, Zhuowen Tu, and Kevin Murphy. Rethinking spatiotem-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 553, + 372, + 565 + ], + "spans": [ + { + "bbox": [ + 115, + 553, + 372, + 565 + ], + "score": 1.0, + "content": "poral feature learning for video understanding. In ECCV, 2018.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 540, + 504, + 565 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 571, + 503, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "Jingwei Xu, Bingbing Ni, Zefan Li, Shuo Cheng, and Xiaokang Yang. Structure preserving video", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 582, + 228, + 594 + ], + "spans": [ + { + "bbox": [ + 115, + 582, + 228, + 594 + ], + "score": 1.0, + "content": "prediction. In CVPR, 2018.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 106, + 572, + 505, + 594 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 601, + 503, + 624 + ], + "lines": [ + { + "bbox": [ + 106, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 505, + 614 + ], + "score": 1.0, + "content": "Kuo-Hao Zeng, William B Shen, De-An Huang, Min Sun, and Juan Carlos Niebles. Visual forecast-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 613, + 374, + 624 + ], + "spans": [ + { + "bbox": [ + 115, + 613, + 374, + 624 + ], + "score": 1.0, + "content": "ing by imitating dynamics in natural sequences. In ICCV, 2017.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 106, + 601, + 505, + 624 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 631, + 503, + 654 + ], + "lines": [ + { + "bbox": [ + 106, + 631, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 644 + ], + "score": 1.0, + "content": "Junbo Zhang, Yu Zheng, and Dekang Qi. Deep spatio-temporal residual networks for citywide", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 642, + 277, + 653 + ], + "spans": [ + { + "bbox": [ + 116, + 642, + 277, + 653 + ], + "score": 1.0, + "content": "crowd flows prediction. In AAAI, 2017.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 106, + 631, + 505, + 653 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 661, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 105, + 660, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 506, + 675 + ], + "score": 1.0, + "content": "Bolei Zhou, Alex Andonian, and Antonio Torralba. Temporal relational reasoning in videos. In", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 672, + 172, + 684 + ], + "spans": [ + { + "bbox": [ + 116, + 672, + 172, + 684 + ], + "score": 1.0, + "content": "ECCV, 2018.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 660, + 506, + 684 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 80, + 365, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 365, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 365, + 96 + ], + "score": 1.0, + "content": "A KEY EQUATIONS OF SPATIOTEMPORAL LSTM", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 102, + 105, + 493, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 495, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 352, + 120 + ], + "score": 1.0, + "content": "Operations inside a Spatiotemporal LSTM unit at time stamp", + "type": "text" + }, + { + "bbox": [ + 353, + 108, + 357, + 116 + ], + "score": 0.77, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 105, + 398, + 120 + ], + "score": 1.0, + "content": "and layer", + "type": "text" + }, + { + "bbox": [ + 398, + 107, + 405, + 116 + ], + "score": 0.83, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 105, + 495, + 120 + ], + "score": 1.0, + "content": "are shown as follows:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 123, + 429, + 277 + ], + "lines": [ + { + "bbox": [ + 178, + 123, + 429, + 277 + ], + "spans": [ + { + "bbox": [ + 178, + 123, + 429, + 277 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { i _ { t } = \\sigma ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + { \\bar { W } } _ { i h } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { i } ) } \\\\ & { g _ { t } = \\operatorname { t a n h } ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + W _ { h , p } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { p } ) } \\\\ & { f _ { t } = \\sigma ( W _ { s ^ { t } } + { \\bar { X } } _ { t } + W _ { h , p } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { f } ) } \\\\ & { i _ { t } ^ { * } = \\sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { X } } _ { t } ^ { k - 1 } + b _ { i } ) } \\\\ & { i _ { t } ^ { * } = \\operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { i } ^ { k - 1 } + b _ { i } ^ { * } ) } \\\\ & { f _ { t } ^ { * } = \\operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\\\ & { c _ { t } ^ { * } = \\sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\\\ & { M _ { t } ^ { * } = i _ { t } \\odot g _ { t } + f _ { t } \\odot { \\bar { C } } _ { t - 1 } ^ { k } } \\\\ & { \\boldsymbol { M } _ { t } ^ { * } = i _ { t } ^ { * } \\odot g _ { t } ^ { * } + f _ { t } ^ { * } \\odot { \\bar { M } } _ { t } ^ { k - 1 } } \\\\ & o _ { s } = \\sigma ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + W _ { i s ^ { t } } + { \\bar { M } } _ { t - 1 } ^ { k } + W _ { i s ^ { t } } + \\bar \\end{array}", + "type": "interline_equation", + "image_path": "5fb1fa23342e02bab61b95df231fb0eb6c3c5c86c246e293837978b0f9bbcd9a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 178, + 123, + 429, + 174.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 178, + 174.33333333333334, + 429, + 225.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 178, + 225.66666666666669, + 429, + 277.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 280, + 505, + 360 + ], + "lines": [ + { + "bbox": [ + 107, + 282, + 504, + 293 + ], + "spans": [ + { + "bbox": [ + 107, + 282, + 133, + 293 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 284, + 140, + 291 + ], + "score": 0.78, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 282, + 236, + 293 + ], + "score": 1.0, + "content": "is the sigmoid function,", + "type": "text" + }, + { + "bbox": [ + 236, + 283, + 243, + 291 + ], + "score": 0.77, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 282, + 369, + 293 + ], + "score": 1.0, + "content": "is the convolution operator, and", + "type": "text" + }, + { + "bbox": [ + 370, + 282, + 379, + 292 + ], + "score": 0.84, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 282, + 504, + 293 + ], + "score": 1.0, + "content": "denotes the Hadamard product.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 290, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 104, + 290, + 203, + 307 + ], + "score": 1.0, + "content": "There are four inputs:", + "type": "text" + }, + { + "bbox": [ + 203, + 294, + 215, + 305 + ], + "score": 0.86, + "content": "\\mathcal { X } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 290, + 456, + 307 + ], + "score": 1.0, + "content": ", the raw frame or hidden states from the previous layer;", + "type": "text" + }, + { + "bbox": [ + 457, + 292, + 484, + 306 + ], + "score": 0.92, + "content": "\\mathbf { \\mathcal { M } } _ { t } ^ { k - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 290, + 506, + 307 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 103, + 301, + 507, + 321 + ], + "spans": [ + { + "bbox": [ + 103, + 301, + 246, + 321 + ], + "score": 1.0, + "content": "previous spatiotemporal memory;", + "type": "text" + }, + { + "bbox": [ + 246, + 304, + 270, + 317 + ], + "score": 0.92, + "content": "\\mathcal { H } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 301, + 289, + 321 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 289, + 304, + 309, + 317 + ], + "score": 0.91, + "content": "\\mathcal { C } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 301, + 507, + 321 + ], + "score": 1.0, + "content": ", the previous hidden states and memory states.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 315, + 506, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 309, + 329 + ], + "score": 1.0, + "content": "Two sets of gate structures, including input gate", + "type": "text" + }, + { + "bbox": [ + 310, + 317, + 318, + 327 + ], + "score": 0.86, + "content": "i _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 315, + 338, + 329 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 339, + 316, + 347, + 327 + ], + "score": 0.86, + "content": "i _ { t } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 315, + 399, + 329 + ], + "score": 1.0, + "content": ", forget gate", + "type": "text" + }, + { + "bbox": [ + 400, + 316, + 410, + 327 + ], + "score": 0.87, + "content": "f _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 315, + 430, + 329 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 430, + 316, + 440, + 327 + ], + "score": 0.88, + "content": "f _ { t } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 315, + 506, + 329 + ], + "score": 1.0, + "content": ", as well as the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 326, + 504, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 153, + 339 + ], + "score": 1.0, + "content": "output gate", + "type": "text" + }, + { + "bbox": [ + 154, + 328, + 163, + 338 + ], + "score": 0.82, + "content": "o _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 326, + 504, + 339 + ], + "score": 1.0, + "content": ", control the information flow in space-time domain. All of them can be presented by", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 333, + 507, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 149, + 347 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { H \\times W \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 333, + 507, + 353 + ], + "score": 1.0, + "content": "dimensional tensors, where the first two dimensions are the width and height of feature", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 349, + 354, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 354, + 361 + ], + "score": 1.0, + "content": "maps, and the last one is the number of feature map channels.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 80, + 365, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 365, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 365, + 96 + ], + "score": 1.0, + "content": "A KEY EQUATIONS OF SPATIOTEMPORAL LSTM", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 102, + 105, + 493, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 495, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 352, + 120 + ], + "score": 1.0, + "content": "Operations inside a Spatiotemporal LSTM unit at time stamp", + "type": "text" + }, + { + "bbox": [ + 353, + 108, + 357, + 116 + ], + "score": 0.77, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 105, + 398, + 120 + ], + "score": 1.0, + "content": "and layer", + "type": "text" + }, + { + "bbox": [ + 398, + 107, + 405, + 116 + ], + "score": 0.83, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 105, + 495, + 120 + ], + "score": 1.0, + "content": "are shown as follows:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 105, + 495, + 120 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 123, + 429, + 277 + ], + "lines": [ + { + "bbox": [ + 178, + 123, + 429, + 277 + ], + "spans": [ + { + "bbox": [ + 178, + 123, + 429, + 277 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { i _ { t } = \\sigma ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + { \\bar { W } } _ { i h } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { i } ) } \\\\ & { g _ { t } = \\operatorname { t a n h } ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + W _ { h , p } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { p } ) } \\\\ & { f _ { t } = \\sigma ( W _ { s ^ { t } } + { \\bar { X } } _ { t } + W _ { h , p } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { f } ) } \\\\ & { i _ { t } ^ { * } = \\sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { X } } _ { t } ^ { k - 1 } + b _ { i } ) } \\\\ & { i _ { t } ^ { * } = \\operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { i } ^ { k - 1 } + b _ { i } ^ { * } ) } \\\\ & { f _ { t } ^ { * } = \\operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\\\ & { c _ { t } ^ { * } = \\sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\\\ & { M _ { t } ^ { * } = i _ { t } \\odot g _ { t } + f _ { t } \\odot { \\bar { C } } _ { t - 1 } ^ { k } } \\\\ & { \\boldsymbol { M } _ { t } ^ { * } = i _ { t } ^ { * } \\odot g _ { t } ^ { * } + f _ { t } ^ { * } \\odot { \\bar { M } } _ { t } ^ { k - 1 } } \\\\ & o _ { s } = \\sigma ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + W _ { i s ^ { t } } + { \\bar { M } } _ { t - 1 } ^ { k } + W _ { i s ^ { t } } + \\bar \\end{array}", + "type": "interline_equation", + "image_path": "5fb1fa23342e02bab61b95df231fb0eb6c3c5c86c246e293837978b0f9bbcd9a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 178, + 123, + 429, + 174.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 178, + 174.33333333333334, + 429, + 225.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 178, + 225.66666666666669, + 429, + 277.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 280, + 505, + 360 + ], + "lines": [ + { + "bbox": [ + 107, + 282, + 504, + 293 + ], + "spans": [ + { + "bbox": [ + 107, + 282, + 133, + 293 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 284, + 140, + 291 + ], + "score": 0.78, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 282, + 236, + 293 + ], + "score": 1.0, + "content": "is the sigmoid function,", + "type": "text" + }, + { + "bbox": [ + 236, + 283, + 243, + 291 + ], + "score": 0.77, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 282, + 369, + 293 + ], + "score": 1.0, + "content": "is the convolution operator, and", + "type": "text" + }, + { + "bbox": [ + 370, + 282, + 379, + 292 + ], + "score": 0.84, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 282, + 504, + 293 + ], + "score": 1.0, + "content": "denotes the Hadamard product.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 290, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 104, + 290, + 203, + 307 + ], + "score": 1.0, + "content": "There are four inputs:", + "type": "text" + }, + { + "bbox": [ + 203, + 294, + 215, + 305 + ], + "score": 0.86, + "content": "\\mathcal { X } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 290, + 456, + 307 + ], + "score": 1.0, + "content": ", the raw frame or hidden states from the previous layer;", + "type": "text" + }, + { + "bbox": [ + 457, + 292, + 484, + 306 + ], + "score": 0.92, + "content": "\\mathbf { \\mathcal { M } } _ { t } ^ { k - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 290, + 506, + 307 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 103, + 301, + 507, + 321 + ], + "spans": [ + { + "bbox": [ + 103, + 301, + 246, + 321 + ], + "score": 1.0, + "content": "previous spatiotemporal memory;", + "type": "text" + }, + { + "bbox": [ + 246, + 304, + 270, + 317 + ], + "score": 0.92, + "content": "\\mathcal { H } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 301, + 289, + 321 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 289, + 304, + 309, + 317 + ], + "score": 0.91, + "content": "\\mathcal { C } _ { t - 1 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 301, + 507, + 321 + ], + "score": 1.0, + "content": ", the previous hidden states and memory states.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 315, + 506, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 309, + 329 + ], + "score": 1.0, + "content": "Two sets of gate structures, including input gate", + "type": "text" + }, + { + "bbox": [ + 310, + 317, + 318, + 327 + ], + "score": 0.86, + "content": "i _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 315, + 338, + 329 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 339, + 316, + 347, + 327 + ], + "score": 0.86, + "content": "i _ { t } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 315, + 399, + 329 + ], + "score": 1.0, + "content": ", forget gate", + "type": "text" + }, + { + "bbox": [ + 400, + 316, + 410, + 327 + ], + "score": 0.87, + "content": "f _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 315, + 430, + 329 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 430, + 316, + 440, + 327 + ], + "score": 0.88, + "content": "f _ { t } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 315, + 506, + 329 + ], + "score": 1.0, + "content": ", as well as the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 326, + 504, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 153, + 339 + ], + "score": 1.0, + "content": "output gate", + "type": "text" + }, + { + "bbox": [ + 154, + 328, + 163, + 338 + ], + "score": 0.82, + "content": "o _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 326, + 504, + 339 + ], + "score": 1.0, + "content": ", control the information flow in space-time domain. All of them can be presented by", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 333, + 507, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 149, + 347 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { H \\times W \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 333, + 507, + 353 + ], + "score": 1.0, + "content": "dimensional tensors, where the first two dimensions are the width and height of feature", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 349, + 354, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 354, + 361 + ], + "score": 1.0, + "content": "maps, and the last one is the number of feature map channels.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8, + "bbox_fs": [ + 103, + 282, + 507, + 361 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/B1lKS2AqtX/B1lKS2AqtX_model.json b/parse/train/B1lKS2AqtX/B1lKS2AqtX_model.json new file mode 100644 index 0000000000000000000000000000000000000000..6884ac7f53e0d7e866ad6415d66c80ace3604a94 --- /dev/null +++ b/parse/train/B1lKS2AqtX/B1lKS2AqtX_model.json @@ -0,0 +1,19346 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 586, + 1301, + 586, + 1301, + 1074, + 398, + 1074 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1191, + 1403, + 1191, + 1403, + 1558, + 298, + 1558 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1573, + 1403, + 1573, + 1403, + 1879, + 298, + 1879 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 301, + 1895, + 1402, + 1895, + 1402, + 1988, + 301, + 1988 + ], + "score": 0.963 + }, + { + "category_id": 0, + "poly": [ + 301, + 218, + 1311, + 218, + 1311, + 324, + 301, + 324 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 316, + 372, + 1387, + 372, + 1387, + 440, + 316, + 440 + ], + "score": 0.93 + }, + { + "category_id": 2, + "poly": [ + 323, + 2006, + 1361, + 2006, + 1361, + 2034, + 323, + 2034 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 302, + 1133, + 573, + 1133, + 573, + 1167, + 302, + 1167 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 816, + 76, + 816, + 104, + 299, + 104 + ], + "score": 0.879 + }, + { + "category_id": 0, + "poly": [ + 773, + 520, + 926, + 520, + 926, + 553, + 773, + 553 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 842, + 2089, + 857, + 2089, + 857, + 2112, + 842, + 2112 + ], + "score": 0.677 + }, + { + "category_id": 13, + "poly": [ + 975, + 375, + 1018, + 375, + 1018, + 406, + 975, + 406 + ], + "score": 0.86, + "latex": "\\mathbf { L i } ^ { 4 }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 218.0, + 716.0, + 218.0, + 716.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 272.0, + 1315.0, + 272.0, + 1315.0, + 327.0, + 294.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 2001.0, + 1366.0, + 2001.0, + 1366.0, + 2039.0, + 331.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1129.0, + 579.0, + 1129.0, + 579.0, + 1175.0, + 294.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 519.0, + 932.0, + 519.0, + 932.0, + 556.0, + 769.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 2088.0, + 860.0, + 2088.0, + 860.0, + 2118.0, + 841.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 587.0, + 1303.0, + 587.0, + 1303.0, + 622.0, + 396.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 617.0, + 1305.0, + 617.0, + 1305.0, + 651.0, + 394.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 645.0, + 1305.0, + 645.0, + 1305.0, + 682.0, + 393.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 676.0, + 1306.0, + 676.0, + 1306.0, + 714.0, + 393.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 707.0, + 1305.0, + 707.0, + 1305.0, + 742.0, + 393.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 739.0, + 1305.0, + 739.0, + 1305.0, + 771.0, + 393.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 770.0, + 1306.0, + 770.0, + 1306.0, + 801.0, + 394.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 800.0, + 1305.0, + 800.0, + 1305.0, + 832.0, + 394.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 829.0, + 1304.0, + 829.0, + 1304.0, + 864.0, + 394.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 861.0, + 1304.0, + 861.0, + 1304.0, + 892.0, + 395.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 891.0, + 1305.0, + 891.0, + 1305.0, + 923.0, + 393.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 921.0, + 1305.0, + 921.0, + 1305.0, + 955.0, + 394.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 952.0, + 1305.0, + 952.0, + 1305.0, + 984.0, + 395.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 983.0, + 1304.0, + 983.0, + 1304.0, + 1015.0, + 395.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1013.0, + 1305.0, + 1013.0, + 1305.0, + 1046.0, + 393.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1042.0, + 1305.0, + 1042.0, + 1305.0, + 1078.0, + 393.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1190.0, + 1404.0, + 1190.0, + 1404.0, + 1229.0, + 292.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1222.0, + 1408.0, + 1222.0, + 1408.0, + 1259.0, + 291.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1254.0, + 1407.0, + 1254.0, + 1407.0, + 1288.0, + 294.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1283.0, + 1405.0, + 1283.0, + 1405.0, + 1320.0, + 293.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1312.0, + 1404.0, + 1312.0, + 1404.0, + 1353.0, + 292.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1348.0, + 1404.0, + 1348.0, + 1404.0, + 1378.0, + 294.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1376.0, + 1404.0, + 1376.0, + 1404.0, + 1409.0, + 292.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1405.0, + 1405.0, + 1405.0, + 1405.0, + 1439.0, + 294.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1437.0, + 1405.0, + 1437.0, + 1405.0, + 1471.0, + 294.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1467.0, + 1404.0, + 1467.0, + 1404.0, + 1501.0, + 294.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1499.0, + 1405.0, + 1499.0, + 1405.0, + 1533.0, + 294.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1529.0, + 1262.0, + 1529.0, + 1262.0, + 1563.0, + 294.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1572.0, + 1404.0, + 1572.0, + 1404.0, + 1611.0, + 292.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1604.0, + 1405.0, + 1604.0, + 1405.0, + 1642.0, + 292.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1633.0, + 1403.0, + 1633.0, + 1403.0, + 1673.0, + 292.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1661.0, + 1405.0, + 1661.0, + 1405.0, + 1706.0, + 291.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1697.0, + 1405.0, + 1697.0, + 1405.0, + 1732.0, + 294.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1727.0, + 1405.0, + 1727.0, + 1405.0, + 1763.0, + 293.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1755.0, + 1405.0, + 1755.0, + 1405.0, + 1793.0, + 293.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1786.0, + 1404.0, + 1786.0, + 1404.0, + 1822.0, + 293.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1819.0, + 1405.0, + 1819.0, + 1405.0, + 1854.0, + 294.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1850.0, + 813.0, + 1850.0, + 813.0, + 1882.0, + 296.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1892.0, + 1406.0, + 1892.0, + 1406.0, + 1932.0, + 294.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1926.0, + 1407.0, + 1926.0, + 1407.0, + 1963.0, + 295.0, + 1963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1955.0, + 1406.0, + 1955.0, + 1406.0, + 1995.0, + 294.0, + 1995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 372.0, + 974.0, + 372.0, + 974.0, + 413.0, + 317.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 372.0, + 1389.0, + 372.0, + 1389.0, + 413.0, + 1019.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 405.0, + 1333.0, + 405.0, + 1333.0, + 445.0, + 311.0, + 445.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 926, + 1404, + 926, + 1404, + 1263, + 298, + 1263 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 230, + 1403, + 230, + 1403, + 596, + 298, + 596 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 299, + 612, + 1403, + 612, + 1403, + 827, + 299, + 827 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1738, + 1403, + 1738, + 1403, + 1922, + 299, + 1922 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1477, + 1403, + 1477, + 1403, + 1723, + 298, + 1723 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1279, + 1402, + 1279, + 1402, + 1463, + 299, + 1463 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 301, + 1942, + 1401, + 1942, + 1401, + 2034, + 301, + 2034 + ], + "score": 0.97 + }, + { + "category_id": 0, + "poly": [ + 299, + 864, + 956, + 864, + 956, + 900, + 299, + 900 + ], + "score": 0.918 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2112, + 841, + 2112 + ], + "score": 0.686 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.204 + }, + { + "category_id": 15, + "poly": [ + 292.0, + 861.0, + 959.0, + 861.0, + 959.0, + 904.0, + 292.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2121.0, + 839.0, + 2121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 926.0, + 1406.0, + 926.0, + 1406.0, + 963.0, + 293.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 958.0, + 1404.0, + 958.0, + 1404.0, + 993.0, + 293.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 988.0, + 1405.0, + 988.0, + 1405.0, + 1026.0, + 293.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1019.0, + 1405.0, + 1019.0, + 1405.0, + 1054.0, + 293.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1050.0, + 1406.0, + 1050.0, + 1406.0, + 1083.0, + 293.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1078.0, + 1406.0, + 1078.0, + 1406.0, + 1117.0, + 292.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1111.0, + 1404.0, + 1111.0, + 1404.0, + 1145.0, + 296.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1139.0, + 1406.0, + 1139.0, + 1406.0, + 1176.0, + 292.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1171.0, + 1406.0, + 1171.0, + 1406.0, + 1206.0, + 294.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1204.0, + 1404.0, + 1204.0, + 1404.0, + 1235.0, + 296.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1232.0, + 1200.0, + 1232.0, + 1200.0, + 1267.0, + 293.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 227.0, + 1405.0, + 227.0, + 1405.0, + 269.0, + 292.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 260.0, + 1405.0, + 260.0, + 1405.0, + 298.0, + 293.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 292.0, + 1405.0, + 292.0, + 1405.0, + 326.0, + 294.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 321.0, + 1404.0, + 321.0, + 1404.0, + 356.0, + 293.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 354.0, + 1404.0, + 354.0, + 1404.0, + 388.0, + 294.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 380.0, + 1405.0, + 380.0, + 1405.0, + 420.0, + 292.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 417.0, + 1404.0, + 417.0, + 1404.0, + 447.0, + 294.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 438.0, + 1407.0, + 438.0, + 1407.0, + 482.0, + 292.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 476.0, + 1404.0, + 476.0, + 1404.0, + 506.0, + 297.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 504.0, + 1404.0, + 504.0, + 1404.0, + 539.0, + 292.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 536.0, + 1405.0, + 536.0, + 1405.0, + 570.0, + 296.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 566.0, + 1158.0, + 566.0, + 1158.0, + 600.0, + 294.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 609.0, + 1407.0, + 609.0, + 1407.0, + 649.0, + 293.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 642.0, + 1407.0, + 642.0, + 1407.0, + 680.0, + 293.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 674.0, + 1404.0, + 674.0, + 1404.0, + 708.0, + 295.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 706.0, + 1403.0, + 706.0, + 1403.0, + 737.0, + 297.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 733.0, + 1407.0, + 733.0, + 1407.0, + 770.0, + 293.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 765.0, + 1405.0, + 765.0, + 1405.0, + 799.0, + 295.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 796.0, + 1240.0, + 796.0, + 1240.0, + 830.0, + 294.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1736.0, + 1404.0, + 1736.0, + 1404.0, + 1773.0, + 294.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1769.0, + 1403.0, + 1769.0, + 1403.0, + 1801.0, + 295.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1800.0, + 1405.0, + 1800.0, + 1405.0, + 1835.0, + 295.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1831.0, + 1404.0, + 1831.0, + 1404.0, + 1863.0, + 295.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1862.0, + 1404.0, + 1862.0, + 1404.0, + 1893.0, + 295.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1890.0, + 1092.0, + 1890.0, + 1092.0, + 1925.0, + 295.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1479.0, + 1405.0, + 1479.0, + 1405.0, + 1512.0, + 296.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1509.0, + 1404.0, + 1509.0, + 1404.0, + 1542.0, + 294.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1540.0, + 1405.0, + 1540.0, + 1405.0, + 1574.0, + 294.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1570.0, + 1405.0, + 1570.0, + 1405.0, + 1604.0, + 293.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1601.0, + 1405.0, + 1601.0, + 1405.0, + 1635.0, + 294.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1632.0, + 1404.0, + 1632.0, + 1404.0, + 1662.0, + 296.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1661.0, + 1405.0, + 1661.0, + 1405.0, + 1695.0, + 294.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1691.0, + 797.0, + 1691.0, + 797.0, + 1729.0, + 293.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1278.0, + 1406.0, + 1278.0, + 1406.0, + 1314.0, + 297.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1309.0, + 1404.0, + 1309.0, + 1404.0, + 1344.0, + 294.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1338.0, + 1406.0, + 1338.0, + 1406.0, + 1375.0, + 293.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1370.0, + 1406.0, + 1370.0, + 1406.0, + 1405.0, + 295.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1395.0, + 1407.0, + 1395.0, + 1407.0, + 1438.0, + 293.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1433.0, + 1403.0, + 1433.0, + 1403.0, + 1465.0, + 295.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1942.0, + 1403.0, + 1942.0, + 1403.0, + 1976.0, + 297.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2006.0, + 296.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2004.0, + 1405.0, + 2004.0, + 1405.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1572, + 1403, + 1572, + 1403, + 1849, + 298, + 1849 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 767, + 1404, + 767, + 1404, + 1103, + 298, + 1103 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1403, + 1402, + 1403, + 1402, + 1557, + 299, + 1557 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 301, + 1942, + 1401, + 1942, + 1401, + 2034, + 301, + 2034 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 347, + 197, + 1354, + 197, + 1354, + 591, + 347, + 591 + ], + "score": 0.972 + }, + { + "category_id": 4, + "poly": [ + 297, + 610, + 1404, + 610, + 1404, + 732, + 297, + 732 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 300, + 1218, + 1402, + 1218, + 1402, + 1312, + 300, + 1312 + ], + "score": 0.962 + }, + { + "category_id": 0, + "poly": [ + 301, + 1150, + 626, + 1150, + 626, + 1187, + 301, + 1187 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 299, + 1351, + 921, + 1351, + 921, + 1382, + 299, + 1382 + ], + "score": 0.893 + }, + { + "category_id": 0, + "poly": [ + 302, + 1888, + 739, + 1888, + 739, + 1919, + 302, + 1919 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 814, + 76, + 814, + 104, + 300, + 104 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 858, + 2088, + 858, + 2112, + 842, + 2112 + ], + "score": 0.589 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 859, + 2088, + 859, + 2112, + 842, + 2112 + ], + "score": 0.406 + }, + { + "category_id": 13, + "poly": [ + 443, + 2003, + 475, + 2003, + 475, + 2034, + 443, + 2034 + ], + "score": 0.88, + "latex": "\\mathcal { X } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1234, + 800, + 1271, + 800, + 1271, + 827, + 1234, + 827 + ], + "score": 0.84, + "latex": "\\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 328, + 1697, + 352, + 1697, + 352, + 1723, + 328, + 1723 + ], + "score": 0.8, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 1131, + 1013, + 1167, + 1013, + 1167, + 1040, + 1131, + 1040 + ], + "score": 0.79, + "latex": "\\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 1018, + 861, + 1039, + 861, + 1039, + 887, + 1018, + 887 + ], + "score": 0.78, + "latex": "\\mathcal { C }" + }, + { + "category_id": 13, + "poly": [ + 298, + 860, + 334, + 860, + 334, + 888, + 298, + 888 + ], + "score": 0.76, + "latex": "\\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 558, + 831, + 578, + 831, + 578, + 858, + 558, + 858 + ], + "score": 0.76, + "latex": "\\mathcal { C }" + }, + { + "category_id": 13, + "poly": [ + 840, + 953, + 860, + 953, + 860, + 980, + 840, + 980 + ], + "score": 0.74, + "latex": "\\mathcal { C }" + }, + { + "category_id": 13, + "poly": [ + 1171, + 1044, + 1191, + 1044, + 1191, + 1070, + 1171, + 1070 + ], + "score": 0.6, + "latex": "\\mathcal { C }" + }, + { + "category_id": 13, + "poly": [ + 1303, + 217, + 1314, + 217, + 1314, + 227, + 1303, + 227 + ], + "score": 0.27, + "latex": "\\ndot" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 210.0, + 623.0, + 210.0, + 623.0, + 229.0, + 546.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 210.0, + 954.0, + 210.0, + 954.0, + 229.0, + 877.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 210.0, + 1285.0, + 210.0, + 1285.0, + 233.0, + 1206.0, + 233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 236.0, + 563.0, + 236.0, + 563.0, + 247.0, + 553.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 235.0, + 893.0, + 235.0, + 893.0, + 249.0, + 879.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 236.0, + 1246.0, + 236.0, + 1246.0, + 247.0, + 1237.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 293.0, + 446.0, + 293.0, + 446.0, + 326.0, + 382.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 293.0, + 542.0, + 293.0, + 542.0, + 327.0, + 477.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 293.0, + 637.0, + 293.0, + 637.0, + 326.0, + 574.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 291.0, + 771.0, + 291.0, + 771.0, + 323.0, + 706.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 291.0, + 868.0, + 291.0, + 868.0, + 324.0, + 804.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 291.0, + 962.0, + 291.0, + 962.0, + 323.0, + 899.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 286.0, + 1108.0, + 286.0, + 1108.0, + 319.0, + 1044.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 286.0, + 1217.0, + 286.0, + 1217.0, + 319.0, + 1154.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 286.0, + 1328.0, + 286.0, + 1328.0, + 319.0, + 1263.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 331.0, + 419.0, + 331.0, + 419.0, + 343.0, + 410.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 331.0, + 516.0, + 331.0, + 516.0, + 343.0, + 506.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 331.0, + 612.0, + 331.0, + 612.0, + 343.0, + 601.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 329.0, + 749.0, + 329.0, + 749.0, + 341.0, + 740.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 330.0, + 846.0, + 330.0, + 846.0, + 341.0, + 837.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 327.0, + 944.0, + 327.0, + 944.0, + 344.0, + 930.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 325.0, + 1305.0, + 325.0, + 1305.0, + 335.0, + 1296.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 345.0, + 739.0, + 345.0, + 739.0, + 360.0, + 724.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 352.0, + 751.0, + 352.0, + 751.0, + 368.0, + 737.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 352.0, + 943.0, + 352.0, + 943.0, + 369.0, + 930.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 361.0, + 446.0, + 361.0, + 446.0, + 381.0, + 380.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 370.0, + 469.0, + 370.0, + 469.0, + 376.0, + 463.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 362.0, + 543.0, + 362.0, + 543.0, + 382.0, + 477.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 362.0, + 638.0, + 362.0, + 638.0, + 382.0, + 573.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 368.0, + 1037.0, + 368.0, + 1037.0, + 376.0, + 1028.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 364.0, + 1109.0, + 364.0, + 1109.0, + 383.0, + 1044.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 364.0, + 1218.0, + 364.0, + 1218.0, + 383.0, + 1154.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 363.0, + 1328.0, + 363.0, + 1328.0, + 383.0, + 1263.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 381.0, + 777.0, + 381.0, + 777.0, + 401.0, + 711.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 381.0, + 872.0, + 381.0, + 872.0, + 401.0, + 808.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 381.0, + 968.0, + 381.0, + 968.0, + 400.0, + 904.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 392.0, + 422.0, + 392.0, + 422.0, + 407.0, + 408.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 392.0, + 517.0, + 392.0, + 517.0, + 407.0, + 503.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 392.0, + 613.0, + 392.0, + 613.0, + 409.0, + 599.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 396.0, + 1086.0, + 396.0, + 1086.0, + 408.0, + 1077.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 411.0, + 751.0, + 411.0, + 751.0, + 426.0, + 737.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 411.0, + 847.0, + 411.0, + 847.0, + 426.0, + 834.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 411.0, + 944.0, + 411.0, + 944.0, + 426.0, + 930.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 420.0, + 435.0, + 420.0, + 435.0, + 439.0, + 385.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 420.0, + 532.0, + 420.0, + 532.0, + 439.0, + 480.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 420.0, + 628.0, + 420.0, + 628.0, + 439.0, + 577.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 422.0, + 1101.0, + 422.0, + 1101.0, + 443.0, + 1050.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 422.0, + 1211.0, + 422.0, + 1211.0, + 443.0, + 1160.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 422.0, + 1321.0, + 422.0, + 1321.0, + 443.0, + 1269.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 433.0, + 441.0, + 433.0, + 441.0, + 452.0, + 378.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 433.0, + 538.0, + 433.0, + 538.0, + 452.0, + 473.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 433.0, + 635.0, + 433.0, + 635.0, + 452.0, + 569.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 434.0, + 770.0, + 434.0, + 770.0, + 455.0, + 718.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 434.0, + 867.0, + 434.0, + 867.0, + 455.0, + 815.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 436.0, + 962.0, + 436.0, + 962.0, + 455.0, + 910.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 437.0, + 1108.0, + 437.0, + 1108.0, + 456.0, + 1044.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 437.0, + 1217.0, + 437.0, + 1217.0, + 456.0, + 1154.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 437.0, + 1327.0, + 437.0, + 1327.0, + 456.0, + 1263.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 456.0, + 419.0, + 456.0, + 419.0, + 468.0, + 410.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 456.0, + 516.0, + 456.0, + 516.0, + 466.0, + 505.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 449.0, + 776.0, + 449.0, + 776.0, + 468.0, + 712.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 449.0, + 872.0, + 449.0, + 872.0, + 468.0, + 808.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 449.0, + 968.0, + 449.0, + 968.0, + 468.0, + 905.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 463.0, + 1086.0, + 463.0, + 1086.0, + 472.0, + 1075.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 462.0, + 1195.0, + 462.0, + 1195.0, + 474.0, + 1185.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 463.0, + 1305.0, + 463.0, + 1305.0, + 472.0, + 1296.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 471.0, + 751.0, + 471.0, + 751.0, + 488.0, + 737.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 471.0, + 848.0, + 471.0, + 848.0, + 488.0, + 833.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 471.0, + 944.0, + 471.0, + 944.0, + 488.0, + 930.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 559.0, + 626.0, + 559.0, + 626.0, + 592.0, + 378.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 555.0, + 944.0, + 555.0, + 944.0, + 596.0, + 724.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 559.0, + 1313.0, + 559.0, + 1313.0, + 590.0, + 1051.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 249.5, + 1118.0, + 249.5, + 1118.0, + 277.0, + 1042.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.75, + 250.5, + 1234.75, + 250.5, + 1234.75, + 277.5, + 1145.75, + 277.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.25, + 257.5, + 452.25, + 257.5, + 452.25, + 288.0, + 374.25, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.75, + 258.0, + 551.75, + 258.0, + 551.75, + 289.0, + 470.75, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.75, + 258.0, + 647.75, + 258.0, + 647.75, + 289.0, + 567.75, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 253.0, + 783.0, + 253.0, + 783.0, + 284.0, + 704.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 253.0, + 881.0, + 253.0, + 881.0, + 284.0, + 801.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 254.5, + 975.0, + 254.5, + 975.0, + 282.0, + 898.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 250.5, + 1347.0, + 250.5, + 1347.0, + 277.5, + 1254.0, + 277.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 453.0, + 617.0, + 453.0, + 617.0, + 468.0, + 595.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 523.0, + 452.0, + 523.0, + 452.0, + 554.0, + 376.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 523.5, + 553.0, + 523.5, + 553.0, + 554.5, + 465.0, + 554.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 524.0, + 650.0, + 524.0, + 650.0, + 555.0, + 563.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 523.0, + 779.0, + 523.0, + 779.0, + 554.0, + 711.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 525.0, + 884.0, + 525.0, + 884.0, + 553.5, + 806.0, + 553.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.25, + 525.5, + 983.25, + 525.5, + 983.25, + 554.5, + 905.25, + 554.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.25, + 523.5, + 1118.25, + 523.5, + 1118.25, + 554.5, + 1043.25, + 554.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.25, + 524.0, + 1233.25, + 524.0, + 1233.25, + 555.0, + 1144.25, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 525.0, + 1351.0, + 525.0, + 1351.0, + 556.0, + 1250.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 611.0, + 1404.0, + 611.0, + 1404.0, + 643.0, + 296.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 642.0, + 1404.0, + 642.0, + 1404.0, + 674.0, + 295.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 671.0, + 1405.0, + 671.0, + 1405.0, + 707.0, + 294.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 702.0, + 1282.0, + 702.0, + 1282.0, + 736.0, + 293.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1147.0, + 632.0, + 1147.0, + 632.0, + 1194.0, + 291.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1349.0, + 925.0, + 1349.0, + 925.0, + 1386.0, + 294.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1885.0, + 745.0, + 1885.0, + 745.0, + 1922.0, + 292.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 818.0, + 73.0, + 818.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1574.0, + 1404.0, + 1574.0, + 1404.0, + 1607.0, + 296.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1602.0, + 1405.0, + 1602.0, + 1405.0, + 1640.0, + 293.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1635.0, + 1404.0, + 1635.0, + 1404.0, + 1667.0, + 296.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1663.0, + 1405.0, + 1663.0, + 1405.0, + 1699.0, + 292.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1695.0, + 327.0, + 1695.0, + 327.0, + 1731.0, + 293.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1695.0, + 1407.0, + 1695.0, + 1407.0, + 1731.0, + 353.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1722.0, + 1407.0, + 1722.0, + 1407.0, + 1764.0, + 291.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1756.0, + 1406.0, + 1756.0, + 1406.0, + 1789.0, + 293.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1785.0, + 1405.0, + 1785.0, + 1405.0, + 1822.0, + 294.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1819.0, + 921.0, + 1819.0, + 921.0, + 1852.0, + 296.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 768.0, + 1405.0, + 768.0, + 1405.0, + 804.0, + 292.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 796.0, + 1233.0, + 796.0, + 1233.0, + 837.0, + 292.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 796.0, + 1406.0, + 796.0, + 1406.0, + 837.0, + 1272.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 830.0, + 557.0, + 830.0, + 557.0, + 864.0, + 294.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 830.0, + 1404.0, + 830.0, + 1404.0, + 864.0, + 579.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 858.0, + 297.0, + 858.0, + 297.0, + 897.0, + 293.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 858.0, + 1017.0, + 858.0, + 1017.0, + 897.0, + 335.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 858.0, + 1406.0, + 858.0, + 1406.0, + 897.0, + 1040.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 887.0, + 1405.0, + 887.0, + 1405.0, + 928.0, + 292.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 918.0, + 1405.0, + 918.0, + 1405.0, + 957.0, + 292.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 947.0, + 839.0, + 947.0, + 839.0, + 990.0, + 292.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 947.0, + 1406.0, + 947.0, + 1406.0, + 990.0, + 861.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 978.0, + 1404.0, + 978.0, + 1404.0, + 1020.0, + 292.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1011.0, + 1130.0, + 1011.0, + 1130.0, + 1049.0, + 293.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 1011.0, + 1404.0, + 1011.0, + 1404.0, + 1049.0, + 1168.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1042.0, + 1170.0, + 1042.0, + 1170.0, + 1079.0, + 293.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1042.0, + 1406.0, + 1042.0, + 1406.0, + 1079.0, + 1192.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1071.0, + 1357.0, + 1071.0, + 1357.0, + 1107.0, + 293.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1403.0, + 1406.0, + 1403.0, + 1406.0, + 1440.0, + 295.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1436.0, + 1403.0, + 1436.0, + 1403.0, + 1469.0, + 295.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1465.0, + 1402.0, + 1465.0, + 1402.0, + 1499.0, + 295.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1495.0, + 1406.0, + 1495.0, + 1406.0, + 1533.0, + 294.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1525.0, + 1347.0, + 1525.0, + 1347.0, + 1562.0, + 294.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1939.0, + 1406.0, + 1939.0, + 1406.0, + 1978.0, + 295.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2006.0, + 296.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 442.0, + 2001.0, + 442.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 2001.0, + 1405.0, + 2001.0, + 1405.0, + 2037.0, + 476.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1218.0, + 1406.0, + 1218.0, + 1406.0, + 1255.0, + 295.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1248.0, + 1406.0, + 1248.0, + 1406.0, + 1287.0, + 293.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1280.0, + 1228.0, + 1280.0, + 1228.0, + 1317.0, + 295.0, + 1317.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1366, + 1405, + 1366, + 1405, + 1957, + 298, + 1957 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 886, + 1404, + 886, + 1404, + 1072, + 297, + 1072 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 307, + 192, + 1396, + 192, + 1396, + 565, + 307, + 565 + ], + "score": 0.973 + }, + { + "category_id": 4, + "poly": [ + 296, + 585, + 1405, + 585, + 1405, + 770, + 296, + 770 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 297, + 1086, + 1400, + 1086, + 1400, + 1148, + 297, + 1148 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 507, + 1150, + 1187, + 1150, + 1187, + 1363, + 507, + 1363 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 295, + 801, + 1401, + 801, + 1401, + 873, + 295, + 873 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 299, + 1972, + 1397, + 1972, + 1397, + 2034, + 299, + 2034 + ], + "score": 0.94 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1369, + 1241, + 1401, + 1241, + 1401, + 1269, + 1369, + 1269 + ], + "score": 0.871 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.765 + }, + { + "category_id": 13, + "poly": [ + 1076, + 1397, + 1176, + 1397, + 1176, + 1433, + 1076, + 1433 + ], + "score": 0.93, + "latex": " { \\mathcal { C } } _ { t - \\tau : t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 583, + 835, + 660, + 835, + 660, + 874, + 583, + 874 + ], + "score": 0.93, + "latex": "\\mathcal { M } _ { t } ^ { k - 1 }" + }, + { + "category_id": 13, + "poly": [ + 1000, + 1891, + 1073, + 1891, + 1073, + 1929, + 1000, + 1929 + ], + "score": 0.92, + "latex": "\\mathcal { C } _ { 1 : t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 456, + 801, + 520, + 801, + 520, + 839, + 456, + 839 + ], + "score": 0.92, + "latex": "\\mathcal { H } _ { t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 1214, + 1464, + 1304, + 1464, + 1304, + 1495, + 1214, + 1495 + ], + "score": 0.92, + "latex": "\\mathcal { T } _ { t } \\odot \\mathcal { G } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 353, + 1858, + 453, + 1858, + 453, + 1895, + 353, + 1895 + ], + "score": 0.92, + "latex": "\\mathcal { C } _ { t - \\tau : t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 1053, + 801, + 1108, + 801, + 1108, + 839, + 1053, + 839 + ], + "score": 0.92, + "latex": "\\mathcal { C } _ { t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 787, + 1524, + 844, + 1524, + 844, + 1559, + 787, + 1559 + ], + "score": 0.92, + "latex": "\\mathcal { C } _ { t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 1048, + 945, + 1199, + 945, + 1199, + 976, + 1048, + 976 + ], + "score": 0.91, + "latex": "\\mathbb { R } ^ { T \\times \\mathbf { \\dot { H } } \\times W \\times C }" + }, + { + "category_id": 13, + "poly": [ + 1234, + 1396, + 1351, + 1396, + 1351, + 1426, + 1234, + 1426 + ], + "score": 0.91, + "latex": "\\mathbb { R } ^ { T H W \\times C }" + }, + { + "category_id": 13, + "poly": [ + 692, + 1645, + 758, + 1645, + 758, + 1681, + 692, + 1681 + ], + "score": 0.91, + "latex": "\\mathcal { H } _ { t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1429, + 428, + 1429, + 428, + 1461, + 297, + 1461 + ], + "score": 0.9, + "latex": "\\mathbb { R } ^ { \\tau T H W \\times C }" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1462, + 1051, + 1462, + 1051, + 1497, + 1018, + 1497 + ], + "score": 0.9, + "latex": "\\mathcal { C } _ { t } ^ { k }" + }, + { + "category_id": 14, + "poly": [ + 502, + 1149, + 1188, + 1149, + 1188, + 1364, + 502, + 1364 + ], + "score": 0.89, + "latex": "\\begin{array} { r l } & { \\mathcal { R } _ { t } = \\sigma \\big ( W _ { x r } * \\mathcal { X } _ { t } + W _ { h r } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { r } \\big ) } \\\\ & { \\mathcal { Z } _ { t } = \\sigma \\big ( W _ { x i } * \\mathcal { X } _ { t } + W _ { h i } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { i } \\big ) } \\\\ & { \\mathcal { G } _ { t } = \\mathrm { t a n h } \\big ( W _ { x g } * \\mathcal { X } _ { t } + W _ { h g } * \\mathcal { H } _ { t - 1 } ^ { k } + b _ { g } \\big ) } \\\\ & { \\mathrm { R E C A L L } \\big ( \\mathcal { R } _ { t } , \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) = \\mathrm { s o f t m a x } \\big ( \\mathcal { R } _ { t } \\cdot \\big ( \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) ^ { \\sf T } \\big ) \\cdot \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } } \\\\ & { \\mathcal { C } _ { t } ^ { k } = \\mathcal { T } _ { t } \\odot \\mathcal { G } _ { t } + \\mathrm { L a y e r N o r m } ( \\mathcal { C } _ { t - 1 } ^ { k } + \\mathrm { R E C A L L } \\big ( \\mathcal { R } _ { t } , \\mathcal { C } _ { t - \\tau : t - 1 } ^ { k } \\big ) \\big ) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 711, + 1708, + 746, + 1708, + 746, + 1738, + 711, + 1738 + ], + "score": 0.89, + "latex": "\\mathcal { R } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 814, + 1399, + 851, + 1399, + 851, + 1429, + 814, + 1429 + ], + "score": 0.89, + "latex": "\\mathcal { R } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 806, + 1496, + 835, + 1496, + 835, + 1524, + 806, + 1524 + ], + "score": 0.88, + "latex": "\\mathcal { T } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1092, + 1496, + 1122, + 1496, + 1122, + 1525, + 1092, + 1525 + ], + "score": 0.88, + "latex": "\\mathcal { G } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 604, + 1647, + 637, + 1647, + 637, + 1678, + 604, + 1678 + ], + "score": 0.87, + "latex": "\\mathcal { X } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1313, + 1860, + 1349, + 1860, + 1349, + 1891, + 1313, + 1891 + ], + "score": 0.87, + "latex": "\\mathcal { R } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 996, + 1371, + 1022, + 1371, + 1022, + 1396, + 996, + 1396 + ], + "score": 0.82, + "latex": "\\odot" + }, + { + "category_id": 13, + "poly": [ + 1305, + 710, + 1332, + 710, + 1332, + 737, + 1305, + 737 + ], + "score": 0.81, + "latex": "\\otimes" + }, + { + "category_id": 13, + "poly": [ + 978, + 710, + 1004, + 710, + 1004, + 737, + 978, + 737 + ], + "score": 0.79, + "latex": "\\odot" + }, + { + "category_id": 13, + "poly": [ + 728, + 1438, + 747, + 1438, + 747, + 1461, + 728, + 1461 + ], + "score": 0.77, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 373, + 1374, + 394, + 1374, + 394, + 1395, + 373, + 1395 + ], + "score": 0.76, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 674, + 1373, + 693, + 1373, + 693, + 1395, + 674, + 1395 + ], + "score": 0.76, + "latex": "^ *" + }, + { + "category_id": 13, + "poly": [ + 1281, + 1931, + 1300, + 1931, + 1300, + 1953, + 1281, + 1953 + ], + "score": 0.74, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 297, + 1864, + 317, + 1864, + 317, + 1888, + 297, + 1888 + ], + "score": 0.71, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 556, + 1931, + 575, + 1931, + 575, + 1953, + 556, + 1953 + ], + "score": 0.67, + "latex": "\\tau" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 245.0, + 821.0, + 245.0, + 821.0, + 265.0, + 795.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 247.0, + 855.0, + 247.0, + 855.0, + 261.0, + 834.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 243.0, + 963.0, + 243.0, + 963.0, + 268.0, + 933.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 232.0, + 1305.0, + 232.0, + 1305.0, + 268.0, + 1270.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 242.0, + 1392.0, + 242.0, + 1392.0, + 273.0, + 1359.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 257.0, + 876.0, + 257.0, + 876.0, + 274.0, + 847.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 259.0, + 923.0, + 259.0, + 923.0, + 273.0, + 907.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 257.0, + 964.0, + 257.0, + 964.0, + 273.0, + 947.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 273.0, + 526.0, + 273.0, + 526.0, + 297.0, + 500.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 282.0, + 765.0, + 282.0, + 765.0, + 291.0, + 755.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 277.0, + 890.0, + 277.0, + 890.0, + 291.0, + 870.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 276.0, + 1129.0, + 276.0, + 1129.0, + 296.0, + 1109.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 270.0, + 1191.0, + 270.0, + 1191.0, + 298.0, + 1156.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1358.0, + 270.0, + 1394.0, + 270.0, + 1394.0, + 300.0, + 1358.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 291.0, + 528.0, + 291.0, + 528.0, + 354.0, + 494.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 307.0, + 1128.0, + 307.0, + 1128.0, + 328.0, + 1087.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 299.0, + 1261.0, + 299.0, + 1261.0, + 345.0, + 1181.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 317.0, + 351.0, + 317.0, + 351.0, + 359.0, + 308.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 327.0, + 1130.0, + 327.0, + 1130.0, + 350.0, + 1109.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 360.0, + 1211.0, + 360.0, + 1211.0, + 371.0, + 1201.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 370.0, + 1131.0, + 370.0, + 1131.0, + 388.0, + 1110.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 396.0, + 1157.0, + 396.0, + 1157.0, + 408.0, + 1125.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 407.0, + 350.0, + 407.0, + 350.0, + 447.0, + 308.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 407.0, + 532.0, + 407.0, + 532.0, + 441.0, + 496.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 414.0, + 937.0, + 414.0, + 937.0, + 449.0, + 897.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 415.0, + 1103.0, + 415.0, + 1103.0, + 434.0, + 1079.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 431.0, + 505.0, + 431.0, + 505.0, + 455.0, + 422.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 444.0, + 518.0, + 444.0, + 518.0, + 455.0, + 508.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 430.0, + 1092.0, + 430.0, + 1092.0, + 454.0, + 1011.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 427.0, + 1180.0, + 427.0, + 1180.0, + 443.0, + 1165.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 450.0, + 346.0, + 450.0, + 346.0, + 480.0, + 314.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 452.0, + 926.0, + 452.0, + 926.0, + 473.0, + 905.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 454.0, + 957.0, + 454.0, + 957.0, + 472.0, + 936.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 449.0, + 1135.0, + 449.0, + 1135.0, + 460.0, + 1126.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 493.0, + 1270.0, + 493.0, + 1270.0, + 532.0, + 1179.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 535.0, + 678.0, + 535.0, + 678.0, + 567.0, + 405.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 535.0, + 1208.0, + 535.0, + 1208.0, + 567.0, + 980.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 239.0, + 1184.0, + 239.0, + 1184.0, + 271.5, + 1162.0, + 271.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 244.0, + 924.0, + 244.0, + 924.0, + 263.0, + 889.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.25, + 354.0, + 868.25, + 354.0, + 868.25, + 385.5, + 838.25, + 385.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 585.0, + 1406.0, + 585.0, + 1406.0, + 620.0, + 295.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 616.0, + 1403.0, + 616.0, + 1403.0, + 648.0, + 295.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 647.0, + 1403.0, + 647.0, + 1403.0, + 679.0, + 295.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 677.0, + 1403.0, + 677.0, + 1403.0, + 709.0, + 295.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 708.0, + 977.0, + 708.0, + 977.0, + 740.0, + 294.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 708.0, + 1304.0, + 708.0, + 1304.0, + 740.0, + 1005.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 708.0, + 1405.0, + 708.0, + 1405.0, + 740.0, + 1333.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 739.0, + 1167.0, + 739.0, + 1167.0, + 770.0, + 294.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1367.0, + 372.0, + 1367.0, + 372.0, + 1402.0, + 296.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1367.0, + 673.0, + 1367.0, + 673.0, + 1402.0, + 395.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1367.0, + 995.0, + 1367.0, + 995.0, + 1402.0, + 694.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1367.0, + 1406.0, + 1367.0, + 1406.0, + 1402.0, + 1023.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1387.0, + 813.0, + 1387.0, + 813.0, + 1443.0, + 288.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1387.0, + 1075.0, + 1387.0, + 1075.0, + 1443.0, + 852.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 1387.0, + 1233.0, + 1387.0, + 1233.0, + 1443.0, + 1177.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1387.0, + 1405.0, + 1387.0, + 1405.0, + 1443.0, + 1352.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1424.0, + 296.0, + 1424.0, + 296.0, + 1475.0, + 291.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1424.0, + 727.0, + 1424.0, + 727.0, + 1475.0, + 429.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1424.0, + 1407.0, + 1424.0, + 1407.0, + 1475.0, + 748.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1456.0, + 1017.0, + 1456.0, + 1017.0, + 1504.0, + 291.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1456.0, + 1213.0, + 1456.0, + 1213.0, + 1504.0, + 1052.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 1456.0, + 1408.0, + 1456.0, + 1408.0, + 1504.0, + 1305.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1493.0, + 805.0, + 1493.0, + 805.0, + 1528.0, + 293.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1493.0, + 1091.0, + 1493.0, + 1091.0, + 1528.0, + 836.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1493.0, + 1403.0, + 1493.0, + 1403.0, + 1528.0, + 1123.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1514.0, + 786.0, + 1514.0, + 786.0, + 1570.0, + 287.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1514.0, + 1412.0, + 1514.0, + 1412.0, + 1570.0, + 845.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1556.0, + 1405.0, + 1556.0, + 1405.0, + 1590.0, + 292.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1585.0, + 1406.0, + 1585.0, + 1406.0, + 1620.0, + 292.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1617.0, + 1405.0, + 1617.0, + 1405.0, + 1649.0, + 295.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1642.0, + 603.0, + 1642.0, + 603.0, + 1684.0, + 291.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1642.0, + 691.0, + 1642.0, + 691.0, + 1684.0, + 638.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1642.0, + 1410.0, + 1642.0, + 1410.0, + 1684.0, + 759.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1673.0, + 1405.0, + 1673.0, + 1405.0, + 1715.0, + 291.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1708.0, + 710.0, + 1708.0, + 710.0, + 1744.0, + 295.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1708.0, + 1405.0, + 1708.0, + 1405.0, + 1744.0, + 747.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1735.0, + 1405.0, + 1735.0, + 1405.0, + 1773.0, + 293.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1767.0, + 1407.0, + 1767.0, + 1407.0, + 1805.0, + 292.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1799.0, + 1406.0, + 1799.0, + 1406.0, + 1832.0, + 292.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1825.0, + 1409.0, + 1825.0, + 1409.0, + 1867.0, + 291.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 1842.0, + 296.0, + 1842.0, + 296.0, + 1917.0, + 281.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1842.0, + 352.0, + 1842.0, + 352.0, + 1917.0, + 318.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 1842.0, + 1312.0, + 1842.0, + 1312.0, + 1917.0, + 454.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 1842.0, + 1415.0, + 1842.0, + 1415.0, + 1917.0, + 1350.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1883.0, + 999.0, + 1883.0, + 999.0, + 1942.0, + 287.0, + 1942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1883.0, + 1412.0, + 1883.0, + 1412.0, + 1942.0, + 1074.0, + 1942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1926.0, + 555.0, + 1926.0, + 555.0, + 1961.0, + 295.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1926.0, + 1280.0, + 1926.0, + 1280.0, + 1961.0, + 576.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1926.0, + 1359.0, + 1926.0, + 1359.0, + 1961.0, + 1301.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 885.0, + 1404.0, + 885.0, + 1404.0, + 922.0, + 294.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 916.0, + 1406.0, + 916.0, + 1406.0, + 952.0, + 295.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 937.0, + 1047.0, + 937.0, + 1047.0, + 988.0, + 287.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 937.0, + 1411.0, + 937.0, + 1411.0, + 988.0, + 1200.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 978.0, + 1406.0, + 978.0, + 1406.0, + 1013.0, + 295.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1010.0, + 1405.0, + 1010.0, + 1405.0, + 1044.0, + 292.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1038.0, + 1396.0, + 1038.0, + 1396.0, + 1075.0, + 295.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1085.0, + 1405.0, + 1085.0, + 1405.0, + 1121.0, + 296.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1117.0, + 1043.0, + 1117.0, + 1043.0, + 1152.0, + 293.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 795.0, + 455.0, + 795.0, + 455.0, + 845.0, + 290.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 795.0, + 1052.0, + 795.0, + 1052.0, + 845.0, + 521.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 795.0, + 1408.0, + 795.0, + 1408.0, + 845.0, + 1109.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 832.0, + 582.0, + 832.0, + 582.0, + 880.0, + 290.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 832.0, + 1352.0, + 832.0, + 1352.0, + 880.0, + 661.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1971.0, + 1401.0, + 1971.0, + 1401.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 1401.0, + 2003.0, + 1401.0, + 2037.0, + 294.0, + 2037.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 228, + 1406, + 228, + 1406, + 476, + 296, + 476 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1481, + 1403, + 1481, + 1403, + 1636, + 297, + 1636 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1881, + 1403, + 1881, + 1403, + 2037, + 298, + 2037 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1214, + 1403, + 1214, + 1403, + 1310, + 298, + 1310 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 918, + 1405, + 918, + 1405, + 1044, + 297, + 1044 + ], + "score": 0.978 + }, + { + "category_id": 8, + "poly": [ + 497, + 494, + 1201, + 494, + 1201, + 752, + 497, + 752 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 296, + 770, + 1401, + 770, + 1401, + 834, + 296, + 834 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 292, + 1057, + 1399, + 1057, + 1399, + 1121, + 292, + 1121 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 293, + 1712, + 1403, + 1712, + 1403, + 1778, + 293, + 1778 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 626, + 1330, + 1074, + 1330, + 1074, + 1382, + 626, + 1382 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 297, + 1404, + 1402, + 1404, + 1402, + 1468, + 297, + 1468 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 635, + 1141, + 1065, + 1141, + 1065, + 1193, + 635, + 1193 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 687, + 1656, + 1008, + 1656, + 1008, + 1696, + 687, + 1696 + ], + "score": 0.946 + }, + { + "category_id": 0, + "poly": [ + 297, + 867, + 865, + 867, + 865, + 900, + 297, + 900 + ], + "score": 0.917 + }, + { + "category_id": 0, + "poly": [ + 300, + 1817, + 558, + 1817, + 558, + 1852, + 300, + 1852 + ], + "score": 0.913 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1345, + 1400, + 1345, + 1400, + 1374, + 1366, + 1374 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1155, + 1399, + 1155, + 1399, + 1185, + 1366, + 1185 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1661, + 1400, + 1661, + 1400, + 1691, + 1366, + 1691 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1369, + 609, + 1401, + 609, + 1401, + 637, + 1369, + 637 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.743 + }, + { + "category_id": 14, + "poly": [ + 498, + 493, + 1199, + 493, + 1199, + 755, + 498, + 755 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\mathcal { Z } _ { t } ^ { \\prime } = \\sigma ( W _ { x t } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m i } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { i } ^ { \\prime } ) } \\\\ & { \\mathcal { G } _ { t } ^ { \\prime } = \\operatorname { t a n h } ( W _ { x g } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m g } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { g } ^ { \\prime } ) } \\\\ & { \\mathcal { F } _ { t } ^ { \\prime } = \\sigma ( W _ { x f } ^ { \\prime } \\ast \\mathcal { X } _ { t } + W _ { m f } \\ast \\mathcal { M } _ { t } ^ { k - 1 } + b _ { f } ^ { \\prime } ) } \\\\ & { \\mathcal { M } _ { t } ^ { k } = \\mathcal { Z } _ { t } ^ { \\prime } \\odot \\mathcal { G } _ { t } ^ { \\prime } + \\mathcal { F } _ { t } ^ { \\prime } \\odot \\mathcal { M } _ { t } ^ { k - 1 } } \\\\ & { \\mathcal { O } _ { t } = \\sigma ( W _ { x o } \\ast \\mathcal { X } _ { t } + W _ { h o } \\ast \\mathcal { H } _ { t - 1 } ^ { k } + W _ { c o } \\ast \\mathcal { C } _ { t } ^ { k } + W _ { m o } \\ast \\mathcal { M } _ { t } ^ { k } + b _ { o } ) } \\\\ & { \\mathcal { H } _ { t } ^ { k } = \\mathcal { O } _ { t } \\odot \\operatorname { t a n h } ( W _ { 1 \\times 1 \\times 1 } \\ast [ \\mathcal { C } _ { t } ^ { k } , \\mathcal { M } _ { t } ^ { k } ] ) , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 624, + 1327, + 1074, + 1327, + 1074, + 1379, + 624, + 1379 + ], + "score": 0.93, + "latex": "\\mathcal { L } _ { \\mathrm { r e c o g n i t i o n } } = \\lambda \\| \\mathcal { X } - \\widehat { \\mathcal { X } } \\| _ { F } ^ { 2 } + \\mathcal { L } _ { \\mathrm { c e } } ( \\mathcal { Y } , \\widehat { \\mathcal { Y } } ) ," + }, + { + "category_id": 14, + "poly": [ + 632, + 1139, + 1065, + 1139, + 1065, + 1191, + 632, + 1191 + ], + "score": 0.92, + "latex": "\\mathcal { L } _ { \\mathrm { p r e d i c t i o n } } = \\left. \\mathcal { X } - \\widehat { \\mathcal { X } } \\right. _ { F } ^ { 2 } + \\left. \\mathcal { X } - \\widehat { \\mathcal { X } } \\right. _ { 1 } ," + }, + { + "category_id": 13, + "poly": [ + 372, + 772, + 470, + 772, + 470, + 804, + 372, + 804 + ], + "score": 0.92, + "latex": "W _ { 1 \\times 1 \\times 1 }" + }, + { + "category_id": 13, + "poly": [ + 868, + 412, + 934, + 412, + 934, + 447, + 868, + 447 + ], + "score": 0.92, + "latex": "\\dot { \\mathcal { M } } _ { t } ^ { 1 : k }" + }, + { + "category_id": 13, + "poly": [ + 1144, + 1216, + 1212, + 1216, + 1212, + 1251, + 1144, + 1251 + ], + "score": 0.92, + "latex": "\\| \\cdot \\| _ { F }" + }, + { + "category_id": 13, + "poly": [ + 542, + 772, + 656, + 772, + 656, + 801, + 542, + 801 + ], + "score": 0.91, + "latex": "1 \\times 1 \\times 1" + }, + { + "category_id": 14, + "poly": [ + 684, + 1656, + 1008, + 1656, + 1008, + 1694, + 684, + 1694 + ], + "score": 0.91, + "latex": "\\lambda ( i ) = \\operatorname* { m a x } ( \\eta , \\lambda ( 0 ) - \\epsilon \\cdot i ) ," + }, + { + "category_id": 13, + "poly": [ + 372, + 1713, + 426, + 1713, + 426, + 1747, + 372, + 1747 + ], + "score": 0.91, + "latex": "\\lambda ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 1324, + 771, + 1393, + 771, + 1393, + 805, + 1324, + 805 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { t } ^ { \\prime } , \\mathcal { G } _ { t } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 526, + 381, + 559, + 381, + 559, + 415, + 526, + 415 + ], + "score": 0.9, + "latex": "\\mathcal { C } _ { t } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 344, + 803, + 378, + 803, + 378, + 835, + 344, + 835 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { t } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1023, + 442, + 1055, + 442, + 1055, + 477, + 1023, + 477 + ], + "score": 0.89, + "latex": "\\mathcal { C } _ { t } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 1219, + 1407, + 1260, + 1407, + 1260, + 1436, + 1219, + 1436 + ], + "score": 0.89, + "latex": "\\mathcal { L } _ { \\mathrm { c e } }" + }, + { + "category_id": 13, + "poly": [ + 1065, + 1713, + 1113, + 1713, + 1113, + 1747, + 1065, + 1747 + ], + "score": 0.89, + "latex": "\\lambda ( i )" + }, + { + "category_id": 13, + "poly": [ + 938, + 803, + 972, + 803, + 972, + 832, + 938, + 832 + ], + "score": 0.88, + "latex": "{ \\mathcal { O } } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 372, + 1399, + 398, + 1399, + 398, + 1436, + 372, + 1436 + ], + "score": 0.86, + "latex": "\\widehat { \\mathcal { V } }" + }, + { + "category_id": 13, + "poly": [ + 373, + 1210, + 401, + 1210, + 401, + 1244, + 373, + 1244 + ], + "score": 0.86, + "latex": "\\widehat { \\mathcal X }" + }, + { + "category_id": 13, + "poly": [ + 448, + 1405, + 473, + 1405, + 473, + 1436, + 448, + 1436 + ], + "score": 0.82, + "latex": "\\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 737, + 1606, + 757, + 1606, + 757, + 1632, + 737, + 1632 + ], + "score": 0.82, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 683, + 1437, + 702, + 1437, + 702, + 1463, + 683, + 1463 + ], + "score": 0.82, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 454, + 1217, + 481, + 1217, + 481, + 1244, + 454, + 1244 + ], + "score": 0.8, + "latex": "\\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 477, + 1719, + 495, + 1719, + 495, + 1746, + 477, + 1746 + ], + "score": 0.8, + "latex": "\\eta" + }, + { + "category_id": 13, + "poly": [ + 1088, + 1607, + 1100, + 1607, + 1100, + 1631, + 1088, + 1631 + ], + "score": 0.77, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 756, + 384, + 794, + 384, + 794, + 411, + 756, + 411 + ], + "score": 0.65, + "latex": "3 D" + }, + { + "category_id": 13, + "poly": [ + 1125, + 1719, + 1141, + 1719, + 1141, + 1741, + 1125, + 1741 + ], + "score": 0.54, + "latex": "\\epsilon" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 867.0, + 864.0, + 867.0, + 864.0, + 902.0, + 294.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1817.0, + 561.0, + 1817.0, + 561.0, + 1857.0, + 293.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 230.0, + 1404.0, + 230.0, + 1404.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 260.0, + 1406.0, + 260.0, + 1406.0, + 295.0, + 293.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 290.0, + 1406.0, + 290.0, + 1406.0, + 327.0, + 294.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 318.0, + 1407.0, + 318.0, + 1407.0, + 359.0, + 291.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 354.0, + 1404.0, + 354.0, + 1404.0, + 386.0, + 293.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 381.0, + 525.0, + 381.0, + 525.0, + 418.0, + 294.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 381.0, + 755.0, + 381.0, + 755.0, + 418.0, + 560.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 381.0, + 1407.0, + 381.0, + 1407.0, + 418.0, + 795.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 409.0, + 867.0, + 409.0, + 867.0, + 450.0, + 291.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 409.0, + 1408.0, + 409.0, + 1408.0, + 450.0, + 935.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 438.0, + 1022.0, + 438.0, + 1022.0, + 482.0, + 290.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 438.0, + 1389.0, + 438.0, + 1389.0, + 482.0, + 1056.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1481.0, + 1405.0, + 1481.0, + 1405.0, + 1519.0, + 295.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1511.0, + 1401.0, + 1511.0, + 1401.0, + 1548.0, + 293.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1541.0, + 1405.0, + 1541.0, + 1405.0, + 1578.0, + 293.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1573.0, + 1404.0, + 1573.0, + 1404.0, + 1610.0, + 293.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1605.0, + 736.0, + 1605.0, + 736.0, + 1638.0, + 297.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1605.0, + 1087.0, + 1605.0, + 1087.0, + 1638.0, + 758.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 1605.0, + 1110.0, + 1605.0, + 1110.0, + 1638.0, + 1101.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1883.0, + 1402.0, + 1883.0, + 1402.0, + 1916.0, + 297.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1948.0, + 293.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 873.0, + 2003.0, + 873.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1212.0, + 372.0, + 1212.0, + 372.0, + 1252.0, + 294.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1212.0, + 453.0, + 1212.0, + 453.0, + 1252.0, + 402.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1212.0, + 1143.0, + 1212.0, + 1143.0, + 1252.0, + 482.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1212.0, + 1404.0, + 1212.0, + 1404.0, + 1252.0, + 1213.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1247.0, + 1404.0, + 1247.0, + 1404.0, + 1280.0, + 293.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1275.0, + 1117.0, + 1275.0, + 1117.0, + 1312.0, + 294.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 921.0, + 1403.0, + 921.0, + 1403.0, + 954.0, + 295.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 950.0, + 1406.0, + 950.0, + 1406.0, + 987.0, + 294.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 979.0, + 1406.0, + 979.0, + 1406.0, + 1018.0, + 292.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1010.0, + 1329.0, + 1010.0, + 1329.0, + 1048.0, + 292.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 767.0, + 371.0, + 767.0, + 371.0, + 807.0, + 295.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 767.0, + 541.0, + 767.0, + 541.0, + 807.0, + 471.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 767.0, + 1323.0, + 767.0, + 1323.0, + 807.0, + 657.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 767.0, + 1403.0, + 767.0, + 1403.0, + 807.0, + 1394.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 801.0, + 343.0, + 801.0, + 343.0, + 837.0, + 296.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 801.0, + 937.0, + 801.0, + 937.0, + 837.0, + 379.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 801.0, + 1179.0, + 801.0, + 1179.0, + 837.0, + 973.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1057.0, + 1404.0, + 1057.0, + 1404.0, + 1093.0, + 296.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1089.0, + 1321.0, + 1089.0, + 1321.0, + 1125.0, + 295.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1709.0, + 371.0, + 1709.0, + 371.0, + 1752.0, + 293.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 1709.0, + 476.0, + 1709.0, + 476.0, + 1752.0, + 427.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1709.0, + 1064.0, + 1709.0, + 1064.0, + 1752.0, + 496.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 1709.0, + 1124.0, + 1709.0, + 1124.0, + 1752.0, + 1114.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1709.0, + 1405.0, + 1709.0, + 1405.0, + 1752.0, + 1142.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1741.0, + 1405.0, + 1741.0, + 1405.0, + 1783.0, + 295.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1399.0, + 371.0, + 1399.0, + 371.0, + 1442.0, + 295.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1399.0, + 447.0, + 1399.0, + 447.0, + 1442.0, + 399.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 1399.0, + 1218.0, + 1399.0, + 1218.0, + 1442.0, + 474.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1399.0, + 1406.0, + 1399.0, + 1406.0, + 1442.0, + 1261.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1435.0, + 682.0, + 1435.0, + 682.0, + 1469.0, + 295.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1435.0, + 931.0, + 1435.0, + 931.0, + 1469.0, + 703.0, + 1469.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1134, + 1404, + 1134, + 1404, + 1319, + 297, + 1319 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1819, + 1403, + 1819, + 1403, + 2034, + 298, + 2034 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1339, + 1403, + 1339, + 1403, + 1462, + 299, + 1462 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1476, + 1404, + 1476, + 1404, + 1662, + 298, + 1662 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1677, + 1403, + 1677, + 1403, + 1800, + 299, + 1800 + ], + "score": 0.975 + }, + { + "category_id": 5, + "poly": [ + 376, + 724, + 1312, + 724, + 1312, + 1052, + 376, + 1052 + ], + "score": 0.972, + "html": "
MODEL10→10COPY
SSIMMSESSIMMSE
CONVLSTM (SHI ET AL., 2015)0.71396.50.539143.2
DFN (DE BRABANDERE ET AL., 2016)0.72689.00.598153.9
CDNA (FINN ET AL., 2016)0.72884.20.671127.1
FRNN (OLIU ET AL., 2018)0.81968.40.694110.5
VPN BASELINE (KALCHBRENNER ET AL., 2017)0.87064.10.73678.0
PREDRNN(WANG ET AL., 2017)0.86956.50.74580.3
PREDRNN++(WANG ET AL.,2018B)0.88546.30.80769.9
E3D-LSTM0.91041.30.85256.8
" + }, + { + "category_id": 3, + "poly": [ + 306, + 196, + 1387, + 196, + 1387, + 570, + 306, + 570 + ], + "score": 0.97 + }, + { + "category_id": 0, + "poly": [ + 297, + 1084, + 928, + 1084, + 928, + 1117, + 297, + 1117 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.888 + }, + { + "category_id": 4, + "poly": [ + 465, + 587, + 1232, + 587, + 1232, + 621, + 465, + 621 + ], + "score": 0.789 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.784 + }, + { + "category_id": 6, + "poly": [ + 294, + 644, + 1403, + 644, + 1403, + 707, + 294, + 707 + ], + "score": 0.751 + }, + { + "category_id": 1, + "poly": [ + 294, + 644, + 1403, + 644, + 1403, + 707, + 294, + 707 + ], + "score": 0.154 + }, + { + "category_id": 13, + "poly": [ + 474, + 1228, + 546, + 1228, + 546, + 1259, + 474, + 1259 + ], + "score": 0.92, + "latex": "l _ { 1 } + l _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 475, + 1539, + 597, + 1539, + 597, + 1568, + 475, + 1568 + ], + "score": 0.91, + "latex": "2 \\times 5 \\times 5" + }, + { + "category_id": 13, + "poly": [ + 775, + 1401, + 867, + 1401, + 867, + 1430, + 775, + 1430 + ], + "score": 0.9, + "latex": "6 4 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 1283, + 1852, + 1385, + 1852, + 1385, + 1880, + 1283, + 1880 + ], + "score": 0.87, + "latex": "1 0 1 0" + }, + { + "category_id": 13, + "poly": [ + 306, + 1973, + 409, + 1973, + 409, + 2002, + 306, + 2002 + ], + "score": 0.86, + "latex": "1 0 1 0" + }, + { + "category_id": 13, + "poly": [ + 775, + 1541, + 802, + 1541, + 802, + 1568, + 775, + 1568 + ], + "score": 0.73, + "latex": "\\times" + }, + { + "category_id": 13, + "poly": [ + 1094, + 1913, + 1133, + 1913, + 1133, + 1941, + 1094, + 1941 + ], + "score": 0.67, + "latex": "- 1" + }, + { + "category_id": 13, + "poly": [ + 667, + 1541, + 695, + 1541, + 695, + 1568, + 667, + 1568 + ], + "score": 0.6, + "latex": "\\times" + }, + { + "category_id": 13, + "poly": [ + 496, + 540, + 592, + 540, + 592, + 567, + 496, + 567 + ], + "score": 0.34, + "latex": "1 0 1 0" + }, + { + "category_id": 13, + "poly": [ + 732, + 1432, + 803, + 1432, + 803, + 1462, + 732, + 1462 + ], + "score": 0.32, + "latex": "3 , 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 408, + 1770, + 443, + 1770, + 443, + 1797, + 408, + 1797 + ], + "score": 0.28, + "latex": "^ { + + }" + }, + { + "category_id": 13, + "poly": [ + 473, + 971, + 546, + 971, + 546, + 998, + 473, + 998 + ], + "score": 0.27, + "latex": "\\mathrm { N N } { + } { + }" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 213.0, + 389.0, + 213.0, + 389.0, + 284.0, + 305.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 203.0, + 432.0, + 203.0, + 432.0, + 292.0, + 399.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 202.0, + 477.0, + 202.0, + 477.0, + 290.0, + 439.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 257.0, + 520.0, + 257.0, + 520.0, + 290.0, + 491.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 255.0, + 568.0, + 255.0, + 568.0, + 293.0, + 534.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 205.0, + 620.0, + 205.0, + 620.0, + 299.0, + 582.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 205.0, + 712.0, + 205.0, + 712.0, + 297.0, + 624.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 209.0, + 806.0, + 209.0, + 806.0, + 302.0, + 721.0, + 302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 209.0, + 848.0, + 209.0, + 848.0, + 248.0, + 813.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 235.0, + 952.0, + 235.0, + 952.0, + 290.0, + 877.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 237.0, + 1033.0, + 237.0, + 1033.0, + 260.0, + 1013.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 199.0, + 1250.0, + 199.0, + 1250.0, + 290.0, + 1048.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 231.0, + 1291.0, + 231.0, + 1291.0, + 268.0, + 1257.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 234.0, + 1334.0, + 234.0, + 1334.0, + 269.0, + 1299.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 236.0, + 1379.0, + 236.0, + 1379.0, + 263.0, + 1356.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 308.0, + 348.0, + 308.0, + 348.0, + 335.0, + 305.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 304.0, + 433.0, + 304.0, + 433.0, + 342.0, + 400.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 305.0, + 476.0, + 305.0, + 476.0, + 337.0, + 445.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 304.0, + 519.0, + 304.0, + 519.0, + 337.0, + 491.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 299.0, + 568.0, + 299.0, + 568.0, + 341.0, + 534.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 297.0, + 615.0, + 297.0, + 615.0, + 343.0, + 577.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 308.0, + 844.0, + 308.0, + 844.0, + 343.0, + 802.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 293.0, + 1380.0, + 293.0, + 1380.0, + 336.0, + 867.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 358.0, + 383.0, + 358.0, + 383.0, + 383.0, + 305.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 356.0, + 432.0, + 356.0, + 432.0, + 389.0, + 401.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 356.0, + 476.0, + 356.0, + 476.0, + 387.0, + 445.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 353.0, + 520.0, + 353.0, + 520.0, + 387.0, + 491.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 350.0, + 568.0, + 350.0, + 568.0, + 392.0, + 535.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 343.0, + 617.0, + 343.0, + 617.0, + 396.0, + 585.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 357.0, + 796.0, + 357.0, + 796.0, + 394.0, + 758.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 361.0, + 842.0, + 361.0, + 842.0, + 394.0, + 802.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 336.0, + 958.0, + 336.0, + 958.0, + 404.0, + 878.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 368.0, + 995.0, + 368.0, + 995.0, + 400.0, + 967.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 336.0, + 1334.0, + 336.0, + 1334.0, + 404.0, + 1010.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 373.0, + 1378.0, + 373.0, + 1378.0, + 406.0, + 1337.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 403.0, + 370.0, + 403.0, + 370.0, + 427.0, + 305.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 402.0, + 431.0, + 402.0, + 431.0, + 435.0, + 401.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 404.0, + 475.0, + 404.0, + 475.0, + 430.0, + 448.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 399.0, + 519.0, + 399.0, + 519.0, + 431.0, + 491.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 395.0, + 567.0, + 395.0, + 567.0, + 436.0, + 535.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 393.0, + 658.0, + 393.0, + 658.0, + 434.0, + 625.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 395.0, + 703.0, + 395.0, + 703.0, + 434.0, + 672.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 399.0, + 749.0, + 399.0, + 749.0, + 437.0, + 715.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 404.0, + 794.0, + 404.0, + 794.0, + 436.0, + 758.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 406.0, + 843.0, + 406.0, + 843.0, + 437.0, + 803.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 411.0, + 933.0, + 411.0, + 933.0, + 437.0, + 893.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 411.0, + 997.0, + 411.0, + 997.0, + 443.0, + 968.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 415.0, + 1035.0, + 415.0, + 1035.0, + 439.0, + 1011.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 411.0, + 1077.0, + 411.0, + 1077.0, + 439.0, + 1052.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 408.0, + 1121.0, + 408.0, + 1121.0, + 442.0, + 1091.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 402.0, + 1206.0, + 402.0, + 1206.0, + 443.0, + 1175.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 404.0, + 1249.0, + 404.0, + 1249.0, + 446.0, + 1217.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 409.0, + 1291.0, + 409.0, + 1291.0, + 445.0, + 1257.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 413.0, + 1332.0, + 413.0, + 1332.0, + 446.0, + 1298.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 415.0, + 1377.0, + 415.0, + 1377.0, + 447.0, + 1339.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 446.0, + 392.0, + 446.0, + 392.0, + 475.0, + 306.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 448.0, + 431.0, + 448.0, + 431.0, + 481.0, + 401.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 451.0, + 475.0, + 451.0, + 475.0, + 477.0, + 447.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 448.0, + 520.0, + 448.0, + 520.0, + 479.0, + 491.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 445.0, + 566.0, + 445.0, + 566.0, + 481.0, + 536.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 440.0, + 658.0, + 440.0, + 658.0, + 482.0, + 626.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 458.0, + 696.0, + 458.0, + 696.0, + 484.0, + 671.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 445.0, + 751.0, + 445.0, + 751.0, + 483.0, + 716.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 448.0, + 797.0, + 448.0, + 797.0, + 484.0, + 762.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 452.0, + 844.0, + 452.0, + 844.0, + 485.0, + 806.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 456.0, + 949.0, + 456.0, + 949.0, + 477.0, + 878.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 455.0, + 995.0, + 455.0, + 995.0, + 483.0, + 970.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 456.0, + 1036.0, + 456.0, + 1036.0, + 482.0, + 1009.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 452.0, + 1079.0, + 452.0, + 1079.0, + 484.0, + 1049.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 448.0, + 1123.0, + 448.0, + 1123.0, + 485.0, + 1089.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 443.0, + 1206.0, + 443.0, + 1206.0, + 489.0, + 1173.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 443.0, + 1249.0, + 443.0, + 1249.0, + 488.0, + 1214.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 450.0, + 1291.0, + 450.0, + 1291.0, + 489.0, + 1256.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 453.0, + 1332.0, + 453.0, + 1332.0, + 490.0, + 1298.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 460.0, + 1374.0, + 460.0, + 1374.0, + 490.0, + 1339.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 494.0, + 378.0, + 494.0, + 378.0, + 519.0, + 306.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 494.0, + 432.0, + 494.0, + 432.0, + 527.0, + 401.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 494.0, + 476.0, + 494.0, + 476.0, + 524.0, + 445.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 492.0, + 520.0, + 492.0, + 520.0, + 524.0, + 490.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 489.0, + 566.0, + 489.0, + 566.0, + 524.0, + 535.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 484.0, + 704.0, + 484.0, + 704.0, + 527.0, + 671.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 488.0, + 753.0, + 488.0, + 753.0, + 527.0, + 715.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 493.0, + 797.0, + 493.0, + 797.0, + 530.0, + 759.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 497.0, + 845.0, + 497.0, + 845.0, + 530.0, + 805.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 498.0, + 945.0, + 498.0, + 945.0, + 519.0, + 880.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 498.0, + 995.0, + 498.0, + 995.0, + 525.0, + 970.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 499.0, + 1035.0, + 499.0, + 1035.0, + 522.0, + 1011.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 536.0, + 495.0, + 536.0, + 495.0, + 571.0, + 455.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 536.0, + 705.0, + 536.0, + 705.0, + 571.0, + 593.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 532.0, + 1211.0, + 532.0, + 1211.0, + 577.0, + 1053.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 200.5, + 534.0, + 200.5, + 534.0, + 232.5, + 483.0, + 232.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 205.0, + 790.0, + 205.0, + 790.0, + 250.0, + 778.0, + 250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 203.0, + 585.0, + 203.0, + 585.0, + 236.0, + 526.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 231.5, + 947.0, + 231.5, + 947.0, + 268.0, + 933.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 230.0, + 1115.0, + 230.0, + 1115.0, + 265.0, + 1098.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 224.5, + 1155.0, + 224.5, + 1155.0, + 267.5, + 1142.0, + 267.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.75, + 233.5, + 988.75, + 233.5, + 988.75, + 264.5, + 972.75, + 264.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.75, + 243.5, + 1356.75, + 243.5, + 1356.75, + 271.0, + 1343.75, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 259.0, + 857.0, + 259.0, + 857.0, + 296.0, + 789.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.75, + 294.5, + 647.75, + 294.5, + 647.75, + 343.5, + 634.75, + 343.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.25, + 295.0, + 689.25, + 295.0, + 689.25, + 344.0, + 685.25, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.75, + 261.0, + 766.75, + 261.0, + 766.75, + 289.0, + 697.75, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 343.0, + 647.0, + 343.0, + 647.0, + 392.5, + 636.0, + 392.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 310.0, + 812.0, + 310.0, + 812.0, + 339.5, + 743.0, + 339.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 346.5, + 688.0, + 346.5, + 688.0, + 393.0, + 684.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 363.0, + 1115.0, + 363.0, + 1115.0, + 401.0, + 1097.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 360.5, + 1155.0, + 360.5, + 1155.0, + 401.5, + 1141.0, + 401.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.75, + 392.5, + 603.75, + 392.5, + 603.75, + 435.0, + 589.75, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.25, + 362.5, + 760.25, + 362.5, + 760.25, + 381.5, + 704.25, + 381.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 374.0, + 928.0, + 374.0, + 928.0, + 396.0, + 871.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.25, + 404.0, + 1154.25, + 404.0, + 1154.25, + 445.0, + 1142.25, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 441.0, + 602.0, + 441.0, + 602.0, + 480.5, + 590.0, + 480.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 437.5, + 708.0, + 437.5, + 708.0, + 463.5, + 675.0, + 463.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 445.0, + 1156.0, + 445.0, + 1156.0, + 486.5, + 1140.0, + 486.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 483.0, + 607.0, + 483.0, + 607.0, + 527.5, + 587.0, + 527.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 481.0, + 647.0, + 481.0, + 647.0, + 529.5, + 635.0, + 529.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1083.0, + 930.0, + 1083.0, + 930.0, + 1119.0, + 295.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 584.0, + 1234.0, + 584.0, + 1234.0, + 627.0, + 465.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 645.0, + 1403.0, + 645.0, + 1403.0, + 677.0, + 296.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 674.0, + 1355.0, + 674.0, + 1355.0, + 709.0, + 293.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1132.0, + 1408.0, + 1132.0, + 1408.0, + 1171.0, + 295.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1167.0, + 1406.0, + 1167.0, + 1406.0, + 1200.0, + 293.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1196.0, + 1405.0, + 1196.0, + 1405.0, + 1230.0, + 294.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1229.0, + 473.0, + 1229.0, + 473.0, + 1261.0, + 295.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1229.0, + 1402.0, + 1229.0, + 1402.0, + 1261.0, + 547.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1254.0, + 1405.0, + 1254.0, + 1405.0, + 1294.0, + 292.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1288.0, + 1111.0, + 1288.0, + 1111.0, + 1323.0, + 294.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1818.0, + 1405.0, + 1818.0, + 1405.0, + 1855.0, + 293.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1849.0, + 1282.0, + 1849.0, + 1282.0, + 1887.0, + 293.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1386.0, + 1849.0, + 1404.0, + 1849.0, + 1404.0, + 1887.0, + 1386.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1880.0, + 1405.0, + 1880.0, + 1405.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 1093.0, + 1909.0, + 1093.0, + 1949.0, + 292.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 1909.0, + 1405.0, + 1909.0, + 1405.0, + 1949.0, + 1134.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1971.0, + 305.0, + 1971.0, + 305.0, + 2006.0, + 293.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2006.0, + 410.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1998.0, + 1404.0, + 1998.0, + 1404.0, + 2040.0, + 293.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1337.0, + 1405.0, + 1337.0, + 1405.0, + 1375.0, + 293.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1371.0, + 1405.0, + 1371.0, + 1405.0, + 1404.0, + 295.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1402.0, + 774.0, + 1402.0, + 774.0, + 1434.0, + 297.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1402.0, + 1407.0, + 1402.0, + 1407.0, + 1434.0, + 868.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1431.0, + 731.0, + 1431.0, + 731.0, + 1466.0, + 293.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1431.0, + 1176.0, + 1431.0, + 1176.0, + 1466.0, + 804.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1475.0, + 1404.0, + 1475.0, + 1404.0, + 1512.0, + 295.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1508.0, + 1406.0, + 1508.0, + 1406.0, + 1543.0, + 294.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1538.0, + 474.0, + 1538.0, + 474.0, + 1573.0, + 293.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1538.0, + 666.0, + 1538.0, + 666.0, + 1573.0, + 598.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1538.0, + 774.0, + 1538.0, + 774.0, + 1573.0, + 696.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 1538.0, + 1406.0, + 1538.0, + 1406.0, + 1573.0, + 803.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1570.0, + 1405.0, + 1570.0, + 1405.0, + 1602.0, + 294.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1598.0, + 1403.0, + 1598.0, + 1403.0, + 1634.0, + 293.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1629.0, + 1406.0, + 1629.0, + 1406.0, + 1666.0, + 293.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1675.0, + 1407.0, + 1675.0, + 1407.0, + 1712.0, + 293.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1707.0, + 1407.0, + 1707.0, + 1407.0, + 1743.0, + 293.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1740.0, + 1405.0, + 1740.0, + 1405.0, + 1772.0, + 295.0, + 1772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1770.0, + 407.0, + 1770.0, + 407.0, + 1802.0, + 295.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1770.0, + 1017.0, + 1770.0, + 1017.0, + 1802.0, + 444.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 645.0, + 1403.0, + 645.0, + 1403.0, + 677.0, + 296.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 674.0, + 1355.0, + 674.0, + 1355.0, + 709.0, + 293.0, + 709.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1125, + 1403, + 1125, + 1403, + 1430, + 298, + 1430 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 538, + 1404, + 538, + 1404, + 874, + 298, + 874 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1566, + 1403, + 1566, + 1403, + 1782, + 298, + 1782 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 462, + 209, + 1226, + 209, + 1226, + 427, + 462, + 427 + ], + "score": 0.979, + "html": "
MODELSSIMMSE
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))0.85950.6
BASELINE 2: 3D-CNN ON TOP(FIGURE 1(B))0.86253.4
BASELINE 3: ：OURS (W/O 3D CONVOLUTIONS)0.89444.2
BASELINE 4: OURS (W/O MEMORY ATTENTION)0.88045.7
E3D-LSTM0.91041.3
" + }, + { + "category_id": 1, + "poly": [ + 298, + 889, + 1403, + 889, + 1403, + 1104, + 298, + 1104 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1803, + 1402, + 1803, + 1402, + 1956, + 299, + 1956 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 294, + 454, + 1399, + 454, + 1399, + 515, + 294, + 515 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 297, + 1972, + 1398, + 1972, + 1398, + 2034, + 297, + 2034 + ], + "score": 0.935 + }, + { + "category_id": 6, + "poly": [ + 476, + 170, + 1222, + 170, + 1222, + 204, + 476, + 204 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 104, + 299, + 104 + ], + "score": 0.893 + }, + { + "category_id": 0, + "poly": [ + 298, + 1463, + 888, + 1463, + 888, + 1495, + 298, + 1495 + ], + "score": 0.887 + }, + { + "category_id": 1, + "poly": [ + 295, + 1513, + 1274, + 1513, + 1274, + 1546, + 295, + 1546 + ], + "score": 0.84 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 858, + 2088, + 858, + 2111, + 842, + 2111 + ], + "score": 0.62 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 859, + 2088, + 859, + 2111, + 842, + 2111 + ], + "score": 0.269 + }, + { + "category_id": 13, + "poly": [ + 297, + 1690, + 421, + 1690, + 421, + 1719, + 297, + 1719 + ], + "score": 0.89, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 1099, + 171, + 1206, + 171, + 1206, + 201, + 1099, + 201 + ], + "score": 0.72, + "latex": "( 1 0 1 0 )" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 169.0, + 1098.0, + 169.0, + 1098.0, + 206.0, + 477.0, + 206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 169.0, + 1221.0, + 169.0, + 1221.0, + 206.0, + 1207.0, + 206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1463.0, + 890.0, + 1463.0, + 890.0, + 1499.0, + 295.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2119.0, + 840.0, + 2119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2119.0, + 840.0, + 2119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1126.0, + 1403.0, + 1126.0, + 1403.0, + 1157.0, + 297.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1153.0, + 1405.0, + 1153.0, + 1405.0, + 1190.0, + 293.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1186.0, + 1405.0, + 1186.0, + 1405.0, + 1221.0, + 292.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1215.0, + 1407.0, + 1215.0, + 1407.0, + 1253.0, + 292.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1248.0, + 1404.0, + 1248.0, + 1404.0, + 1280.0, + 296.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1276.0, + 1404.0, + 1276.0, + 1404.0, + 1313.0, + 292.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1307.0, + 1403.0, + 1307.0, + 1403.0, + 1345.0, + 293.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1340.0, + 1404.0, + 1340.0, + 1404.0, + 1371.0, + 296.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1369.0, + 1404.0, + 1369.0, + 1404.0, + 1403.0, + 293.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1402.0, + 1051.0, + 1402.0, + 1051.0, + 1433.0, + 296.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 535.0, + 1404.0, + 535.0, + 1404.0, + 575.0, + 296.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 569.0, + 1404.0, + 569.0, + 1404.0, + 603.0, + 293.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 598.0, + 1405.0, + 598.0, + 1405.0, + 637.0, + 293.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 628.0, + 1405.0, + 628.0, + 1405.0, + 668.0, + 292.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 659.0, + 1404.0, + 659.0, + 1404.0, + 695.0, + 293.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 691.0, + 1406.0, + 691.0, + 1406.0, + 726.0, + 292.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 721.0, + 1404.0, + 721.0, + 1404.0, + 756.0, + 296.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 752.0, + 1405.0, + 752.0, + 1405.0, + 787.0, + 293.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 780.0, + 1408.0, + 780.0, + 1408.0, + 819.0, + 293.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 811.0, + 1404.0, + 811.0, + 1404.0, + 846.0, + 294.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 843.0, + 1374.0, + 843.0, + 1374.0, + 878.0, + 294.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1565.0, + 1403.0, + 1565.0, + 1403.0, + 1602.0, + 293.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1596.0, + 1408.0, + 1596.0, + 1408.0, + 1638.0, + 292.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1630.0, + 1405.0, + 1630.0, + 1405.0, + 1664.0, + 294.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1659.0, + 1406.0, + 1659.0, + 1406.0, + 1694.0, + 292.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 1689.0, + 1405.0, + 1689.0, + 1405.0, + 1723.0, + 422.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1721.0, + 1405.0, + 1721.0, + 1405.0, + 1755.0, + 294.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1749.0, + 964.0, + 1749.0, + 964.0, + 1786.0, + 292.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 889.0, + 1404.0, + 889.0, + 1404.0, + 924.0, + 294.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 923.0, + 1404.0, + 923.0, + 1404.0, + 953.0, + 296.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 950.0, + 1405.0, + 950.0, + 1405.0, + 984.0, + 294.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 981.0, + 1404.0, + 981.0, + 1404.0, + 1015.0, + 294.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1009.0, + 1407.0, + 1009.0, + 1407.0, + 1047.0, + 293.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1042.0, + 1407.0, + 1042.0, + 1407.0, + 1076.0, + 294.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1074.0, + 877.0, + 1074.0, + 877.0, + 1108.0, + 294.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1802.0, + 1404.0, + 1802.0, + 1404.0, + 1839.0, + 294.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1835.0, + 1404.0, + 1835.0, + 1404.0, + 1868.0, + 297.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1864.0, + 1404.0, + 1864.0, + 1404.0, + 1898.0, + 293.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1891.0, + 1404.0, + 1891.0, + 1404.0, + 1930.0, + 293.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1924.0, + 1377.0, + 1924.0, + 1377.0, + 1961.0, + 293.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 452.0, + 1405.0, + 452.0, + 1405.0, + 491.0, + 293.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 485.0, + 838.0, + 485.0, + 838.0, + 516.0, + 297.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1970.0, + 1400.0, + 1970.0, + 1400.0, + 2006.0, + 296.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2000.0, + 1404.0, + 2000.0, + 1404.0, + 2041.0, + 292.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1512.0, + 1280.0, + 1512.0, + 1280.0, + 1551.0, + 296.0, + 1551.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 357, + 1705, + 1324, + 1705, + 1324, + 1951, + 357, + 1951 + ], + "score": 0.982, + "html": "
MODELFRAME1FRAME 2FRAME 3FRAME 4
ST-RESNET (ZHANG ET AL., 2017)0.6880.9391.1301.288
VPN(KALCHBRENNER ET AL., 2017)0.7441.0311.2511.444
FRNN(OLIU ET AL., 2018)0.6820.8230.9891.183
PREDRNN(WANG ET AL.,2017)0.6340.9341.0471.263
PREDRNN++(WANG ET AL., 2018B)0.6410.8550.9791.158
E3D-LSTM0.6200.7730.8880.984
" + }, + { + "category_id": 5, + "poly": [ + 316, + 885, + 1363, + 885, + 1363, + 1186, + 316, + 1186 + ], + "score": 0.981, + "html": "
MODEL10→2010→40COPY(→40)
PSNRSSIMPSNRSSIMPSNRSSIM
CONVLSTM(SHI ET AL.,2015)23.580.71222.850.63923.490.670
DFN (DE BRABANDERE ET AL., 2016)27.260.79423.010.65223.370.664
MCNET(VILLEGAS ET AL.,2017)25.950.804---1
FRNN(OLIU ET AL., 2018)26.120.77123.770.67824.000.685
PREDRNN(WANG ET AL., 2017)27.550.83924.160.70324.450.711
PREDRNN++(WANG ET AL., 2018B)28.470.86525.210.74125.900.759
E3D-LSTM29.310.87927.240.81030.590.874
" + }, + { + "category_id": 1, + "poly": [ + 296, + 1219, + 1404, + 1219, + 1404, + 1465, + 296, + 1465 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 352, + 186, + 1346, + 186, + 1346, + 689, + 352, + 689 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 301, + 1546, + 1397, + 1546, + 1397, + 1640, + 301, + 1640 + ], + "score": 0.957 + }, + { + "category_id": 4, + "poly": [ + 297, + 711, + 1403, + 711, + 1403, + 775, + 297, + 775 + ], + "score": 0.943 + }, + { + "category_id": 6, + "poly": [ + 297, + 805, + 1400, + 805, + 1400, + 870, + 297, + 870 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.886 + }, + { + "category_id": 1, + "poly": [ + 299, + 1972, + 1405, + 1972, + 1405, + 2035, + 299, + 2035 + ], + "score": 0.884 + }, + { + "category_id": 6, + "poly": [ + 345, + 1656, + 1345, + 1656, + 1345, + 1689, + 345, + 1689 + ], + "score": 0.87 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2112, + 840, + 2112 + ], + "score": 0.814 + }, + { + "category_id": 0, + "poly": [ + 305, + 1496, + 1225, + 1496, + 1225, + 1529, + 305, + 1529 + ], + "score": 0.804 + }, + { + "category_id": 1, + "poly": [ + 305, + 1496, + 1225, + 1496, + 1225, + 1529, + 305, + 1529 + ], + "score": 0.111 + }, + { + "category_id": 13, + "poly": [ + 855, + 2003, + 982, + 2003, + 982, + 2033, + 855, + 2033 + ], + "score": 0.92, + "latex": "3 2 \\times 3 2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 859, + 1252, + 1004, + 1252, + 1004, + 1281, + 859, + 1281 + ], + "score": 0.45, + "latex": "\\mathrm { P r e d R N N + + }" + }, + { + "category_id": 13, + "poly": [ + 436, + 1870, + 525, + 1870, + 525, + 1897, + 436, + 1897 + ], + "score": 0.34, + "latex": "{ \\mathrm { R N N } } + +" + }, + { + "category_id": 13, + "poly": [ + 397, + 1105, + 486, + 1105, + 486, + 1132, + 397, + 1132 + ], + "score": 0.33, + "latex": "{ \\mathrm { R N N } } + +" + }, + { + "category_id": 13, + "poly": [ + 650, + 1404, + 723, + 1404, + 723, + 1433, + 650, + 1433 + ], + "score": 0.25, + "latex": "\\mathrm { N N } { + } { + }" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 186.0, + 487.0, + 186.0, + 487.0, + 217.0, + 425.0, + 217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 186.0, + 1045.0, + 186.0, + 1045.0, + 215.0, + 852.0, + 215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 218.0, + 1344.0, + 218.0, + 1344.0, + 242.0, + 360.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 246.0, + 900.0, + 246.0, + 900.0, + 278.0, + 494.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 285.0, + 551.0, + 285.0, + 551.0, + 319.0, + 496.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 328.0, + 552.0, + 328.0, + 552.0, + 360.0, + 442.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 367.0, + 553.0, + 367.0, + 553.0, + 403.0, + 449.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 377.0, + 613.0, + 377.0, + 613.0, + 400.0, + 566.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 408.0, + 901.0, + 408.0, + 901.0, + 440.0, + 799.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 485.0, + 488.0, + 485.0, + 488.0, + 516.0, + 424.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 482.0, + 1047.0, + 482.0, + 1047.0, + 516.0, + 849.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 564.0, + 553.0, + 564.0, + 553.0, + 603.0, + 494.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 608.0, + 552.0, + 608.0, + 552.0, + 641.0, + 442.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 647.0, + 555.0, + 647.0, + 555.0, + 687.0, + 448.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 711.0, + 1405.0, + 711.0, + 1405.0, + 747.0, + 296.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 743.0, + 1332.0, + 743.0, + 1332.0, + 779.0, + 295.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 807.0, + 1402.0, + 807.0, + 1402.0, + 839.0, + 297.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 836.0, + 1269.0, + 836.0, + 1269.0, + 871.0, + 293.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1651.0, + 1349.0, + 1651.0, + 1349.0, + 1696.0, + 347.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1496.0, + 1227.0, + 1496.0, + 1227.0, + 1531.0, + 299.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1218.0, + 1405.0, + 1218.0, + 1405.0, + 1257.0, + 293.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1252.0, + 858.0, + 1252.0, + 858.0, + 1286.0, + 294.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1252.0, + 1406.0, + 1252.0, + 1406.0, + 1286.0, + 1005.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1281.0, + 1406.0, + 1281.0, + 1406.0, + 1318.0, + 294.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1312.0, + 1404.0, + 1312.0, + 1404.0, + 1346.0, + 293.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1343.0, + 1404.0, + 1343.0, + 1404.0, + 1377.0, + 294.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1371.0, + 1405.0, + 1371.0, + 1405.0, + 1410.0, + 293.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1404.0, + 649.0, + 1404.0, + 649.0, + 1438.0, + 294.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1404.0, + 1406.0, + 1404.0, + 1406.0, + 1438.0, + 724.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1435.0, + 1268.0, + 1435.0, + 1268.0, + 1468.0, + 294.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1547.0, + 1405.0, + 1547.0, + 1405.0, + 1581.0, + 297.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1580.0, + 1403.0, + 1580.0, + 1403.0, + 1610.0, + 297.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1610.0, + 1401.0, + 1610.0, + 1401.0, + 1640.0, + 296.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1967.0, + 1406.0, + 1967.0, + 1406.0, + 2011.0, + 293.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 2004.0, + 854.0, + 2004.0, + 854.0, + 2035.0, + 297.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 2004.0, + 1403.0, + 2004.0, + 1403.0, + 2035.0, + 983.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1496.0, + 1227.0, + 1496.0, + 1227.0, + 1531.0, + 299.0, + 1531.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1607, + 1404, + 1607, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 1130, + 1404, + 1130, + 1404, + 1588, + 297, + 1588 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 897, + 1404, + 897, + 1404, + 1111, + 298, + 1111 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 554, + 1403, + 554, + 1403, + 677, + 298, + 677 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 696, + 1403, + 696, + 1403, + 818, + 298, + 818 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 297, + 184, + 1400, + 184, + 1400, + 436, + 297, + 436 + ], + "score": 0.966 + }, + { + "category_id": 4, + "poly": [ + 298, + 466, + 1399, + 466, + 1399, + 530, + 298, + 530 + ], + "score": 0.932 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 815, + 76, + 815, + 104, + 298, + 104 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.791 + }, + { + "category_id": 0, + "poly": [ + 299, + 848, + 1071, + 848, + 1071, + 880, + 299, + 880 + ], + "score": 0.741 + }, + { + "category_id": 1, + "poly": [ + 299, + 848, + 1071, + 848, + 1071, + 880, + 299, + 880 + ], + "score": 0.153 + }, + { + "category_id": 13, + "poly": [ + 1199, + 2003, + 1394, + 2003, + 1394, + 2033, + 1199, + 2033 + ], + "score": 0.93, + "latex": "1 \\times 7 \\times 7 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 297, + 1699, + 514, + 1699, + 514, + 1728, + 297, + 1728 + ], + "score": 0.92, + "latex": "2 \\times 5 6 \\times 5 6 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 578, + 1974, + 689, + 1974, + 689, + 2002, + 578, + 2002 + ], + "score": 0.91, + "latex": "2 \\times 3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 942, + 2003, + 1164, + 2003, + 1164, + 2033, + 942, + 2033 + ], + "score": 0.91, + "latex": "1 6 \\times 5 6 \\times 5 6 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 1131, + 1973, + 1242, + 1973, + 1242, + 2002, + 1131, + 2002 + ], + "score": 0.9, + "latex": "2 \\times 2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 883, + 1729, + 997, + 1729, + 997, + 1759, + 883, + 1759 + ], + "score": 0.9, + "latex": "2 \\times 5 \\times 5 )" + }, + { + "category_id": 13, + "poly": [ + 848, + 1973, + 961, + 1973, + 961, + 2002, + 848, + 2002 + ], + "score": 0.9, + "latex": "2 \\times 2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 920, + 1669, + 1047, + 1669, + 1047, + 1698, + 920, + 1698 + ], + "score": 0.9, + "latex": "2 2 4 \\times 2 2 4" + }, + { + "category_id": 13, + "poly": [ + 573, + 1791, + 640, + 1791, + 640, + 1819, + 573, + 1819 + ], + "score": 0.89, + "latex": "1 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 615, + 1911, + 669, + 1911, + 669, + 1941, + 615, + 1941 + ], + "score": 0.89, + "latex": "2 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 970, + 1820, + 1025, + 1820, + 1025, + 1850, + 970, + 1850 + ], + "score": 0.89, + "latex": "5 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 884, + 1820, + 938, + 1820, + 938, + 1850, + 884, + 1850 + ], + "score": 0.88, + "latex": "2 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1260, + 1284, + 1314, + 1284, + 1314, + 1313, + 1260, + 1313 + ], + "score": 0.88, + "latex": "2 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1346, + 1284, + 1401, + 1284, + 1401, + 1313, + 1346, + 1313 + ], + "score": 0.86, + "latex": "50 \\%" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 185.0, + 499.0, + 185.0, + 499.0, + 210.0, + 450.0, + 210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 185.0, + 768.0, + 185.0, + 768.0, + 210.0, + 711.0, + 210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 199.0, + 879.0, + 199.0, + 879.0, + 290.0, + 335.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 255.0, + 1441.0, + 255.0, + 1441.0, + 480.0, + 404.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 330.0, + 333.0, + 330.0, + 333.0, + 342.0, + 313.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 312.0, + 481.0, + 312.0, + 481.0, + 371.0, + 365.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 312.0, + 1006.0, + 312.0, + 1006.0, + 414.0, + 510.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 465.0, + 1404.0, + 465.0, + 1404.0, + 501.0, + 296.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 495.0, + 1379.0, + 495.0, + 1379.0, + 534.0, + 293.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 848.0, + 1074.0, + 848.0, + 1074.0, + 883.0, + 295.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1606.0, + 1405.0, + 1606.0, + 1405.0, + 1642.0, + 294.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1637.0, + 1402.0, + 1637.0, + 1402.0, + 1668.0, + 296.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1667.0, + 919.0, + 1667.0, + 919.0, + 1701.0, + 295.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 1667.0, + 1405.0, + 1667.0, + 1405.0, + 1701.0, + 1048.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1698.0, + 296.0, + 1698.0, + 296.0, + 1733.0, + 293.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1698.0, + 1404.0, + 1698.0, + 1404.0, + 1733.0, + 515.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1728.0, + 882.0, + 1728.0, + 882.0, + 1762.0, + 295.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1728.0, + 1405.0, + 1728.0, + 1405.0, + 1762.0, + 998.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1760.0, + 1406.0, + 1760.0, + 1406.0, + 1795.0, + 294.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1788.0, + 572.0, + 1788.0, + 572.0, + 1823.0, + 293.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1788.0, + 1404.0, + 1788.0, + 1404.0, + 1823.0, + 641.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1819.0, + 883.0, + 1819.0, + 883.0, + 1856.0, + 292.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1819.0, + 969.0, + 1819.0, + 969.0, + 1856.0, + 939.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 1819.0, + 1405.0, + 1819.0, + 1405.0, + 1856.0, + 1026.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 1406.0, + 1851.0, + 1406.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1879.0, + 1405.0, + 1879.0, + 1405.0, + 1919.0, + 292.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1910.0, + 614.0, + 1910.0, + 614.0, + 1946.0, + 293.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1910.0, + 1405.0, + 1910.0, + 1405.0, + 1946.0, + 670.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1978.0, + 294.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1970.0, + 577.0, + 1970.0, + 577.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 1970.0, + 847.0, + 1970.0, + 847.0, + 2008.0, + 690.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1970.0, + 1130.0, + 1970.0, + 1130.0, + 2008.0, + 962.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 1970.0, + 1405.0, + 1970.0, + 1405.0, + 2008.0, + 1243.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2000.0, + 941.0, + 2000.0, + 941.0, + 2036.0, + 296.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 2000.0, + 1198.0, + 2000.0, + 1198.0, + 2036.0, + 1165.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 2000.0, + 1405.0, + 2000.0, + 1405.0, + 2036.0, + 1395.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1130.0, + 1402.0, + 1130.0, + 1402.0, + 1165.0, + 295.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1163.0, + 1404.0, + 1163.0, + 1404.0, + 1195.0, + 296.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1191.0, + 1406.0, + 1191.0, + 1406.0, + 1226.0, + 295.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1224.0, + 1404.0, + 1224.0, + 1404.0, + 1256.0, + 296.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1248.0, + 1405.0, + 1248.0, + 1405.0, + 1289.0, + 291.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1282.0, + 1259.0, + 1282.0, + 1259.0, + 1320.0, + 294.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 1282.0, + 1345.0, + 1282.0, + 1345.0, + 1320.0, + 1315.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1282.0, + 1405.0, + 1282.0, + 1405.0, + 1320.0, + 1402.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1313.0, + 1404.0, + 1313.0, + 1404.0, + 1348.0, + 295.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1347.0, + 1404.0, + 1347.0, + 1404.0, + 1378.0, + 296.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1374.0, + 1405.0, + 1374.0, + 1405.0, + 1409.0, + 295.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1399.0, + 1406.0, + 1399.0, + 1406.0, + 1444.0, + 291.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1432.0, + 1408.0, + 1432.0, + 1408.0, + 1474.0, + 292.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1464.0, + 1405.0, + 1464.0, + 1405.0, + 1503.0, + 291.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1495.0, + 1402.0, + 1495.0, + 1402.0, + 1530.0, + 294.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1527.0, + 1405.0, + 1527.0, + 1405.0, + 1562.0, + 294.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1560.0, + 1345.0, + 1560.0, + 1345.0, + 1591.0, + 296.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 896.0, + 1404.0, + 896.0, + 1404.0, + 934.0, + 295.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 927.0, + 1406.0, + 927.0, + 1406.0, + 966.0, + 292.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 960.0, + 1405.0, + 960.0, + 1405.0, + 994.0, + 293.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 987.0, + 1405.0, + 987.0, + 1405.0, + 1027.0, + 293.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1021.0, + 1404.0, + 1021.0, + 1404.0, + 1056.0, + 294.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1052.0, + 1405.0, + 1052.0, + 1405.0, + 1086.0, + 294.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1080.0, + 1238.0, + 1080.0, + 1238.0, + 1116.0, + 295.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 555.0, + 1407.0, + 555.0, + 1407.0, + 588.0, + 296.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 585.0, + 1407.0, + 585.0, + 1407.0, + 617.0, + 293.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 613.0, + 1405.0, + 613.0, + 1405.0, + 650.0, + 293.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 647.0, + 925.0, + 647.0, + 925.0, + 679.0, + 294.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 697.0, + 1403.0, + 697.0, + 1403.0, + 729.0, + 296.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 727.0, + 1404.0, + 727.0, + 1404.0, + 759.0, + 296.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 758.0, + 1405.0, + 758.0, + 1405.0, + 791.0, + 293.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 787.0, + 1311.0, + 787.0, + 1311.0, + 822.0, + 294.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 848.0, + 1074.0, + 848.0, + 1074.0, + 883.0, + 295.0, + 883.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1607, + 1403, + 1607, + 1403, + 2034, + 298, + 2034 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 297, + 1432, + 1402, + 1432, + 1402, + 1587, + 297, + 1587 + ], + "score": 0.977 + }, + { + "category_id": 5, + "poly": [ + 394, + 1051, + 1292, + 1051, + 1292, + 1269, + 394, + 1269 + ], + "score": 0.975, + "html": "
MODELFRONT 25%FRONT 50%
3D-CNN9.1110.30
SEPARABLE-CNN:SEPARABLE-CONV AT BOTTOM8.949.62
(2+1)D-CNN:SEPARABLE-CONV ON TOP9.0810.17
E(2+1)D-LSTM: SEPARABLE INSIDE UNITS12.4519.86
E3D-LSTM14.5922.73
" + }, + { + "category_id": 1, + "poly": [ + 301, + 1295, + 1399, + 1295, + 1399, + 1418, + 301, + 1418 + ], + "score": 0.969 + }, + { + "category_id": 3, + "poly": [ + 405, + 183, + 1292, + 183, + 1292, + 856, + 405, + 856 + ], + "score": 0.963 + }, + { + "category_id": 4, + "poly": [ + 299, + 886, + 1399, + 886, + 1399, + 980, + 299, + 980 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 815, + 76, + 815, + 104, + 298, + 104 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 837, + 2088, + 865, + 2088, + 865, + 2112, + 837, + 2112 + ], + "score": 0.856 + }, + { + "category_id": 6, + "poly": [ + 302, + 1002, + 1379, + 1002, + 1379, + 1035, + 302, + 1035 + ], + "score": 0.521 + }, + { + "category_id": 1, + "poly": [ + 302, + 1002, + 1379, + 1002, + 1379, + 1035, + 302, + 1035 + ], + "score": 0.47 + }, + { + "category_id": 4, + "poly": [ + 1018, + 816, + 1264, + 816, + 1264, + 854, + 1018, + 854 + ], + "score": 0.264 + }, + { + "category_id": 13, + "poly": [ + 473, + 1385, + 580, + 1385, + 580, + 1416, + 473, + 1416 + ], + "score": 0.91, + "latex": "2 \\times 1 0 ^ { - 5 }" + }, + { + "category_id": 13, + "poly": [ + 655, + 1356, + 705, + 1356, + 705, + 1389, + 655, + 1389 + ], + "score": 0.91, + "latex": "\\lambda ( i )" + }, + { + "category_id": 13, + "poly": [ + 938, + 1388, + 1029, + 1388, + 1029, + 1419, + 938, + 1419 + ], + "score": 0.9, + "latex": "\\eta = 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 852, + 1668, + 907, + 1668, + 907, + 1698, + 852, + 1698 + ], + "score": 0.89, + "latex": "5 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 749, + 1668, + 804, + 1668, + 804, + 1698, + 749, + 1698 + ], + "score": 0.88, + "latex": "2 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1067, + 886, + 1122, + 886, + 1122, + 916, + 1067, + 916 + ], + "score": 0.88, + "latex": "5 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 958, + 886, + 1013, + 886, + 1013, + 917, + 958, + 917 + ], + "score": 0.88, + "latex": "2 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1346, + 1758, + 1402, + 1758, + 1402, + 1790, + 1346, + 1790 + ], + "score": 0.87, + "latex": "2 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1286, + 1941, + 1401, + 1941, + 1401, + 1973, + 1286, + 1973 + ], + "score": 0.86, + "latex": "2 5 \\% { - } 5 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1136, + 1357, + 1210, + 1357, + 1210, + 1387, + 1136, + 1387 + ], + "score": 0.85, + "latex": "( i = 0" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 190.0, + 532.0, + 190.0, + 532.0, + 220.0, + 485.0, + 220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 186.0, + 737.0, + 186.0, + 737.0, + 222.0, + 671.0, + 222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 186.0, + 934.0, + 186.0, + 934.0, + 222.0, + 868.0, + 222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 187.0, + 1233.0, + 187.0, + 1233.0, + 222.0, + 1153.0, + 222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 332.0, + 477.0, + 332.0, + 477.0, + 359.0, + 406.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 328.0, + 1000.0, + 328.0, + 1000.0, + 361.0, + 602.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 332.0, + 1280.0, + 332.0, + 1280.0, + 361.0, + 1015.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 353.0, + 496.0, + 353.0, + 496.0, + 380.0, + 407.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 350.0, + 1001.0, + 350.0, + 1001.0, + 384.0, + 602.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 354.0, + 1091.0, + 354.0, + 1091.0, + 378.0, + 1016.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 391.0, + 601.0, + 391.0, + 601.0, + 410.0, + 585.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 491.0, + 476.0, + 491.0, + 476.0, + 515.0, + 407.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 489.0, + 779.0, + 489.0, + 779.0, + 519.0, + 602.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 489.0, + 1001.0, + 489.0, + 1001.0, + 519.0, + 799.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 492.0, + 1271.0, + 492.0, + 1271.0, + 518.0, + 1017.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 510.0, + 497.0, + 510.0, + 497.0, + 541.0, + 405.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 507.0, + 771.0, + 507.0, + 771.0, + 544.0, + 600.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 509.0, + 1003.0, + 509.0, + 1003.0, + 541.0, + 798.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 510.0, + 1141.0, + 510.0, + 1141.0, + 539.0, + 1015.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 650.0, + 476.0, + 650.0, + 476.0, + 674.0, + 406.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 650.0, + 771.0, + 650.0, + 771.0, + 677.0, + 603.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 650.0, + 969.0, + 650.0, + 969.0, + 675.0, + 774.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 651.0, + 1240.0, + 651.0, + 1240.0, + 680.0, + 1016.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 669.0, + 497.0, + 669.0, + 497.0, + 699.0, + 405.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 668.0, + 987.0, + 668.0, + 987.0, + 702.0, + 601.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 671.0, + 1094.0, + 671.0, + 1094.0, + 695.0, + 1017.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 808.0, + 476.0, + 808.0, + 476.0, + 832.0, + 407.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 808.0, + 991.0, + 808.0, + 991.0, + 839.0, + 602.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 811.0, + 1268.0, + 811.0, + 1268.0, + 841.0, + 1016.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 829.0, + 496.0, + 829.0, + 496.0, + 857.0, + 407.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 830.0, + 779.0, + 830.0, + 779.0, + 858.0, + 604.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 831.0, + 987.0, + 831.0, + 987.0, + 858.0, + 800.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 831.0, + 1221.0, + 831.0, + 1221.0, + 858.0, + 1016.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 884.0, + 957.0, + 884.0, + 957.0, + 922.0, + 295.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 884.0, + 1066.0, + 884.0, + 1066.0, + 922.0, + 1014.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 884.0, + 1405.0, + 884.0, + 1405.0, + 922.0, + 1123.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 917.0, + 1404.0, + 917.0, + 1404.0, + 951.0, + 296.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 949.0, + 1268.0, + 949.0, + 1268.0, + 979.0, + 296.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 995.0, + 1384.0, + 995.0, + 1384.0, + 1042.0, + 305.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 813.0, + 1266.0, + 813.0, + 1266.0, + 837.0, + 1019.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 832.0, + 1221.0, + 832.0, + 1221.0, + 854.0, + 1016.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1605.0, + 1405.0, + 1605.0, + 1405.0, + 1642.0, + 293.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1636.0, + 1407.0, + 1636.0, + 1407.0, + 1672.0, + 293.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1663.0, + 748.0, + 1663.0, + 748.0, + 1708.0, + 291.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 1663.0, + 851.0, + 1663.0, + 851.0, + 1708.0, + 805.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1663.0, + 1407.0, + 1663.0, + 1407.0, + 1708.0, + 908.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1696.0, + 1407.0, + 1696.0, + 1407.0, + 1737.0, + 292.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1729.0, + 1405.0, + 1729.0, + 1405.0, + 1765.0, + 293.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1757.0, + 1345.0, + 1757.0, + 1345.0, + 1795.0, + 293.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1788.0, + 1405.0, + 1788.0, + 1405.0, + 1824.0, + 292.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1853.0, + 293.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1851.0, + 1405.0, + 1851.0, + 1405.0, + 1886.0, + 293.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1883.0, + 1405.0, + 1883.0, + 1405.0, + 1914.0, + 296.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1909.0, + 1405.0, + 1909.0, + 1405.0, + 1946.0, + 293.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1942.0, + 1285.0, + 1942.0, + 1285.0, + 1976.0, + 292.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1976.0, + 1402.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1974.0, + 1404.0, + 1974.0, + 1404.0, + 2005.0, + 294.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2004.0, + 1115.0, + 2004.0, + 1115.0, + 2036.0, + 296.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1434.0, + 1406.0, + 1434.0, + 1406.0, + 1467.0, + 297.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1464.0, + 1404.0, + 1464.0, + 1404.0, + 1498.0, + 295.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1491.0, + 1406.0, + 1491.0, + 1406.0, + 1531.0, + 293.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1525.0, + 1404.0, + 1525.0, + 1404.0, + 1559.0, + 296.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1555.0, + 1385.0, + 1555.0, + 1385.0, + 1589.0, + 293.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1293.0, + 1405.0, + 1293.0, + 1405.0, + 1330.0, + 293.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1324.0, + 1404.0, + 1324.0, + 1404.0, + 1360.0, + 295.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1355.0, + 654.0, + 1355.0, + 654.0, + 1391.0, + 293.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1355.0, + 1135.0, + 1355.0, + 1135.0, + 1391.0, + 706.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1355.0, + 1407.0, + 1355.0, + 1407.0, + 1391.0, + 1211.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1386.0, + 472.0, + 1386.0, + 472.0, + 1422.0, + 296.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1386.0, + 937.0, + 1386.0, + 937.0, + 1422.0, + 581.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1386.0, + 1040.0, + 1386.0, + 1040.0, + 1422.0, + 1030.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 995.0, + 1384.0, + 995.0, + 1384.0, + 1042.0, + 305.0, + 1042.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 913, + 1403, + 913, + 1403, + 1190, + 298, + 1190 + ], + "score": 0.981 + }, + { + "category_id": 5, + "poly": [ + 331, + 505, + 1354, + 505, + 1354, + 694, + 331, + 694 + ], + "score": 0.98, + "html": "
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.7820.91
PRE-TRAINED ON THE AUXILIARY TASK14.0022.15
TRAINED ON BOTH TASKS WITH A FIXED LOSS RATIO13.5720.46
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.5922.73
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1611, + 1404, + 1611, + 1404, + 1886, + 298, + 1886 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 394, + 219, + 1295, + 219, + 1295, + 436, + 394, + 436 + ], + "score": 0.977, + "html": "
MODELFRONT 25%FRONT 50%
BASELINE 1: 3D-CNN AT BOTTOM (FIGURE 1(A))10.2816.05
BASELINE 2: 3D-CNN ON TOP (FIGURE 1(B))9.6314.82
BASELINE 3:OURS W/O 3D CONVOLUTIONS9.5813.92
BASELINE 4: OURS W/O MEMORY ATTENTION11.3918.84
E3D-LSTM14.5922.73
" + }, + { + "category_id": 5, + "poly": [ + 332, + 1397, + 1354, + 1397, + 1354, + 1518, + 332, + 1518 + ], + "score": 0.976, + "html": "
MODELFRONT 25%FRONT 50%
TRAINED ONLY ON THE PRIMARY CLASSIFICATION TASK13.4918.94
E3D-LSTM(WITH SELF-SUPERVISED AUXILIARY LEARNING)14.3020.85
" + }, + { + "category_id": 1, + "poly": [ + 297, + 713, + 1404, + 713, + 1404, + 899, + 297, + 899 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1205, + 1402, + 1205, + 1402, + 1329, + 299, + 1329 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 1973, + 1402, + 1973, + 1402, + 2033, + 298, + 2033 + ], + "score": 0.948 + }, + { + "category_id": 0, + "poly": [ + 299, + 1550, + 545, + 1550, + 545, + 1585, + 299, + 1585 + ], + "score": 0.914 + }, + { + "category_id": 2, + "poly": [ + 301, + 76, + 815, + 76, + 815, + 104, + 301, + 104 + ], + "score": 0.897 + }, + { + "category_id": 6, + "poly": [ + 345, + 170, + 1350, + 170, + 1350, + 204, + 345, + 204 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 301, + 1921, + 557, + 1921, + 557, + 1949, + 301, + 1949 + ], + "score": 0.877 + }, + { + "category_id": 2, + "poly": [ + 836, + 2089, + 862, + 2089, + 862, + 2112, + 836, + 2112 + ], + "score": 0.838 + }, + { + "category_id": 6, + "poly": [ + 297, + 455, + 1400, + 455, + 1400, + 488, + 297, + 488 + ], + "score": 0.822 + }, + { + "category_id": 6, + "poly": [ + 296, + 1348, + 1398, + 1348, + 1398, + 1381, + 296, + 1381 + ], + "score": 0.468 + }, + { + "category_id": 1, + "poly": [ + 296, + 1348, + 1398, + 1348, + 1398, + 1381, + 296, + 1381 + ], + "score": 0.462 + }, + { + "category_id": 13, + "poly": [ + 486, + 1272, + 506, + 1272, + 506, + 1295, + 486, + 1295 + ], + "score": 0.72, + "latex": "\\tau" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1543.0, + 550.0, + 1543.0, + 550.0, + 1596.0, + 290.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 818.0, + 73.0, + 818.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 166.0, + 1353.0, + 166.0, + 1353.0, + 208.0, + 343.0, + 208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1923.0, + 558.0, + 1923.0, + 558.0, + 1952.0, + 299.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 449.0, + 1404.0, + 449.0, + 1404.0, + 495.0, + 293.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1344.0, + 1403.0, + 1344.0, + 1403.0, + 1385.0, + 295.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 914.0, + 1404.0, + 914.0, + 1404.0, + 947.0, + 296.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 944.0, + 1404.0, + 944.0, + 1404.0, + 980.0, + 293.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 973.0, + 1405.0, + 973.0, + 1405.0, + 1012.0, + 292.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1003.0, + 1405.0, + 1003.0, + 1405.0, + 1042.0, + 293.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1034.0, + 1405.0, + 1034.0, + 1405.0, + 1074.0, + 293.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1063.0, + 1404.0, + 1063.0, + 1404.0, + 1107.0, + 291.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1093.0, + 1407.0, + 1093.0, + 1407.0, + 1137.0, + 292.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1126.0, + 1406.0, + 1126.0, + 1406.0, + 1163.0, + 292.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1158.0, + 1343.0, + 1158.0, + 1343.0, + 1194.0, + 296.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1613.0, + 1404.0, + 1613.0, + 1404.0, + 1646.0, + 297.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1642.0, + 1402.0, + 1642.0, + 1402.0, + 1674.0, + 296.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1672.0, + 1405.0, + 1672.0, + 1405.0, + 1708.0, + 294.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1700.0, + 1406.0, + 1700.0, + 1406.0, + 1741.0, + 293.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1734.0, + 1404.0, + 1734.0, + 1404.0, + 1767.0, + 296.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1765.0, + 1402.0, + 1765.0, + 1402.0, + 1798.0, + 296.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1794.0, + 1406.0, + 1794.0, + 1406.0, + 1828.0, + 292.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1821.0, + 1404.0, + 1821.0, + 1404.0, + 1862.0, + 293.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1857.0, + 696.0, + 1857.0, + 696.0, + 1889.0, + 296.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 712.0, + 1406.0, + 712.0, + 1406.0, + 747.0, + 295.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 743.0, + 1405.0, + 743.0, + 1405.0, + 778.0, + 295.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 773.0, + 1405.0, + 773.0, + 1405.0, + 810.0, + 294.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 805.0, + 1402.0, + 805.0, + 1402.0, + 839.0, + 294.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 835.0, + 1405.0, + 835.0, + 1405.0, + 871.0, + 294.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 865.0, + 1240.0, + 865.0, + 1240.0, + 903.0, + 294.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1206.0, + 1404.0, + 1206.0, + 1404.0, + 1239.0, + 297.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1233.0, + 1403.0, + 1233.0, + 1403.0, + 1272.0, + 293.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1267.0, + 485.0, + 1267.0, + 485.0, + 1303.0, + 294.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 1267.0, + 1404.0, + 1267.0, + 1404.0, + 1303.0, + 507.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1297.0, + 1324.0, + 1297.0, + 1324.0, + 1332.0, + 294.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1969.0, + 1404.0, + 1969.0, + 1404.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2002.0, + 1118.0, + 2002.0, + 1118.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1344.0, + 1403.0, + 1344.0, + 1403.0, + 1385.0, + 295.0, + 1385.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.831 + }, + { + "category_id": 1, + "poly": [ + 297, + 806, + 1402, + 806, + 1402, + 872, + 297, + 872 + ], + "score": 0.824 + }, + { + "category_id": 1, + "poly": [ + 292, + 723, + 1402, + 723, + 1402, + 788, + 292, + 788 + ], + "score": 0.805 + }, + { + "category_id": 1, + "poly": [ + 294, + 641, + 1402, + 641, + 1402, + 706, + 294, + 706 + ], + "score": 0.8 + }, + { + "category_id": 1, + "poly": [ + 297, + 557, + 1403, + 557, + 1403, + 622, + 297, + 622 + ], + "score": 0.796 + }, + { + "category_id": 1, + "poly": [ + 297, + 889, + 1380, + 889, + 1380, + 925, + 297, + 925 + ], + "score": 0.794 + }, + { + "category_id": 1, + "poly": [ + 298, + 1419, + 1396, + 1419, + 1396, + 1485, + 298, + 1485 + ], + "score": 0.794 + }, + { + "category_id": 1, + "poly": [ + 295, + 1503, + 1358, + 1503, + 1358, + 1538, + 295, + 1538 + ], + "score": 0.794 + }, + { + "category_id": 1, + "poly": [ + 297, + 1637, + 1399, + 1637, + 1399, + 1703, + 297, + 1703 + ], + "score": 0.788 + }, + { + "category_id": 1, + "poly": [ + 296, + 1025, + 1401, + 1025, + 1401, + 1091, + 296, + 1091 + ], + "score": 0.78 + }, + { + "category_id": 1, + "poly": [ + 294, + 1252, + 1402, + 1252, + 1402, + 1316, + 294, + 1316 + ], + "score": 0.779 + }, + { + "category_id": 1, + "poly": [ + 296, + 1721, + 1398, + 1721, + 1398, + 1786, + 296, + 1786 + ], + "score": 0.776 + }, + { + "category_id": 1, + "poly": [ + 298, + 1335, + 1398, + 1335, + 1398, + 1402, + 298, + 1402 + ], + "score": 0.771 + }, + { + "category_id": 1, + "poly": [ + 291, + 942, + 1400, + 942, + 1400, + 1009, + 291, + 1009 + ], + "score": 0.758 + }, + { + "category_id": 1, + "poly": [ + 297, + 1554, + 1395, + 1554, + 1395, + 1620, + 297, + 1620 + ], + "score": 0.756 + }, + { + "category_id": 1, + "poly": [ + 293, + 474, + 1401, + 474, + 1401, + 539, + 293, + 539 + ], + "score": 0.755 + }, + { + "category_id": 0, + "poly": [ + 299, + 228, + 489, + 228, + 489, + 261, + 299, + 261 + ], + "score": 0.743 + }, + { + "category_id": 1, + "poly": [ + 292, + 391, + 1401, + 391, + 1401, + 456, + 292, + 456 + ], + "score": 0.714 + }, + { + "category_id": 1, + "poly": [ + 297, + 1805, + 1399, + 1805, + 1399, + 1869, + 297, + 1869 + ], + "score": 0.685 + }, + { + "category_id": 1, + "poly": [ + 299, + 278, + 1402, + 278, + 1402, + 373, + 299, + 373 + ], + "score": 0.662 + }, + { + "category_id": 1, + "poly": [ + 298, + 1108, + 1404, + 1108, + 1404, + 1234, + 298, + 1234 + ], + "score": 0.655 + }, + { + "category_id": 1, + "poly": [ + 295, + 1971, + 1398, + 1971, + 1398, + 2034, + 295, + 2034 + ], + "score": 0.64 + }, + { + "category_id": 1, + "poly": [ + 297, + 1888, + 1398, + 1888, + 1398, + 1952, + 297, + 1952 + ], + "score": 0.614 + }, + { + "category_id": 1, + "poly": [ + 299, + 228, + 489, + 228, + 489, + 261, + 299, + 261 + ], + "score": 0.119 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 228.0, + 490.0, + 228.0, + 490.0, + 265.0, + 297.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 807.0, + 1404.0, + 807.0, + 1404.0, + 843.0, + 295.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 838.0, + 467.0, + 838.0, + 467.0, + 871.0, + 321.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 723.0, + 1404.0, + 723.0, + 1404.0, + 759.0, + 294.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 756.0, + 601.0, + 756.0, + 601.0, + 786.0, + 323.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 640.0, + 1404.0, + 640.0, + 1404.0, + 679.0, + 294.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 672.0, + 1205.0, + 672.0, + 1205.0, + 707.0, + 322.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 557.0, + 1404.0, + 557.0, + 1404.0, + 593.0, + 296.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 588.0, + 476.0, + 588.0, + 476.0, + 622.0, + 322.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 889.0, + 1376.0, + 889.0, + 1376.0, + 928.0, + 295.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1420.0, + 1402.0, + 1420.0, + 1402.0, + 1456.0, + 294.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1451.0, + 1136.0, + 1451.0, + 1136.0, + 1484.0, + 323.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1501.0, + 1360.0, + 1501.0, + 1360.0, + 1540.0, + 296.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1635.0, + 1404.0, + 1635.0, + 1404.0, + 1675.0, + 295.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1670.0, + 705.0, + 1670.0, + 705.0, + 1702.0, + 325.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1024.0, + 1405.0, + 1024.0, + 1405.0, + 1064.0, + 295.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1058.0, + 786.0, + 1058.0, + 786.0, + 1090.0, + 324.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1249.0, + 1404.0, + 1249.0, + 1404.0, + 1292.0, + 292.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1284.0, + 532.0, + 1284.0, + 532.0, + 1317.0, + 323.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1723.0, + 1402.0, + 1723.0, + 1402.0, + 1759.0, + 296.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1752.0, + 839.0, + 1752.0, + 839.0, + 1784.0, + 324.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1336.0, + 1404.0, + 1336.0, + 1404.0, + 1372.0, + 295.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1369.0, + 810.0, + 1369.0, + 810.0, + 1401.0, + 322.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 940.0, + 1403.0, + 940.0, + 1403.0, + 981.0, + 294.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 976.0, + 985.0, + 976.0, + 985.0, + 1008.0, + 324.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1552.0, + 1400.0, + 1552.0, + 1400.0, + 1592.0, + 294.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1586.0, + 1192.0, + 1586.0, + 1192.0, + 1621.0, + 324.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 473.0, + 1403.0, + 473.0, + 1403.0, + 513.0, + 293.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 507.0, + 973.0, + 507.0, + 973.0, + 539.0, + 321.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 389.0, + 1406.0, + 389.0, + 1406.0, + 430.0, + 292.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 424.0, + 609.0, + 424.0, + 609.0, + 453.0, + 321.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1805.0, + 1404.0, + 1805.0, + 1404.0, + 1841.0, + 295.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1838.0, + 711.0, + 1838.0, + 711.0, + 1868.0, + 323.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 276.0, + 1406.0, + 276.0, + 1406.0, + 317.0, + 293.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 311.0, + 1406.0, + 311.0, + 1406.0, + 345.0, + 324.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 340.0, + 1304.0, + 340.0, + 1304.0, + 378.0, + 323.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1109.0, + 1404.0, + 1109.0, + 1404.0, + 1145.0, + 293.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1140.0, + 1405.0, + 1140.0, + 1405.0, + 1176.0, + 321.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1171.0, + 1406.0, + 1171.0, + 1406.0, + 1207.0, + 322.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1202.0, + 473.0, + 1202.0, + 473.0, + 1236.0, + 322.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1968.0, + 1404.0, + 1968.0, + 1404.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 2003.0, + 714.0, + 2003.0, + 714.0, + 2034.0, + 323.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1885.0, + 1403.0, + 1885.0, + 1403.0, + 1924.0, + 293.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1919.0, + 636.0, + 1919.0, + 636.0, + 1951.0, + 323.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 228.0, + 490.0, + 228.0, + 490.0, + 265.0, + 297.0, + 265.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.844 + }, + { + "category_id": 1, + "poly": [ + 299, + 1006, + 1398, + 1006, + 1398, + 1070, + 299, + 1070 + ], + "score": 0.708 + }, + { + "category_id": 1, + "poly": [ + 295, + 924, + 1399, + 924, + 1399, + 988, + 295, + 988 + ], + "score": 0.698 + }, + { + "category_id": 1, + "poly": [ + 294, + 229, + 1402, + 229, + 1402, + 294, + 294, + 294 + ], + "score": 0.698 + }, + { + "category_id": 1, + "poly": [ + 295, + 1837, + 1403, + 1837, + 1403, + 1902, + 295, + 1902 + ], + "score": 0.698 + }, + { + "category_id": 1, + "poly": [ + 298, + 508, + 1400, + 508, + 1400, + 573, + 298, + 573 + ], + "score": 0.693 + }, + { + "category_id": 1, + "poly": [ + 297, + 1587, + 1398, + 1587, + 1398, + 1652, + 297, + 1652 + ], + "score": 0.688 + }, + { + "category_id": 1, + "poly": [ + 293, + 757, + 1402, + 757, + 1402, + 822, + 293, + 822 + ], + "score": 0.675 + }, + { + "category_id": 1, + "poly": [ + 298, + 1089, + 1397, + 1089, + 1397, + 1154, + 298, + 1154 + ], + "score": 0.672 + }, + { + "category_id": 1, + "poly": [ + 295, + 426, + 1402, + 426, + 1402, + 490, + 295, + 490 + ], + "score": 0.671 + }, + { + "category_id": 1, + "poly": [ + 297, + 1670, + 1399, + 1670, + 1399, + 1735, + 297, + 1735 + ], + "score": 0.67 + }, + { + "category_id": 1, + "poly": [ + 294, + 1753, + 1399, + 1753, + 1399, + 1818, + 294, + 1818 + ], + "score": 0.668 + }, + { + "category_id": 1, + "poly": [ + 297, + 1505, + 1398, + 1505, + 1398, + 1569, + 297, + 1569 + ], + "score": 0.666 + }, + { + "category_id": 1, + "poly": [ + 299, + 840, + 1398, + 840, + 1398, + 905, + 299, + 905 + ], + "score": 0.657 + }, + { + "category_id": 1, + "poly": [ + 303, + 311, + 1400, + 311, + 1400, + 406, + 303, + 406 + ], + "score": 0.651 + }, + { + "category_id": 1, + "poly": [ + 296, + 1421, + 1398, + 1421, + 1398, + 1486, + 296, + 1486 + ], + "score": 0.638 + }, + { + "category_id": 1, + "poly": [ + 297, + 592, + 1402, + 592, + 1402, + 656, + 297, + 656 + ], + "score": 0.625 + }, + { + "category_id": 1, + "poly": [ + 297, + 1172, + 1396, + 1172, + 1396, + 1238, + 297, + 1238 + ], + "score": 0.622 + }, + { + "category_id": 1, + "poly": [ + 297, + 674, + 1401, + 674, + 1401, + 739, + 297, + 739 + ], + "score": 0.604 + }, + { + "category_id": 1, + "poly": [ + 296, + 1338, + 1397, + 1338, + 1397, + 1404, + 296, + 1404 + ], + "score": 0.601 + }, + { + "category_id": 1, + "poly": [ + 296, + 1255, + 1398, + 1255, + 1398, + 1321, + 296, + 1321 + ], + "score": 0.594 + }, + { + "category_id": 13, + "poly": [ + 1257, + 1262, + 1291, + 1262, + 1291, + 1285, + 1257, + 1285 + ], + "score": 0.57, + "latex": "^ { + + }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1002.0, + 1403.0, + 1002.0, + 1403.0, + 1044.0, + 293.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1037.0, + 495.0, + 1037.0, + 495.0, + 1069.0, + 321.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 923.0, + 1403.0, + 923.0, + 1403.0, + 961.0, + 294.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 955.0, + 1063.0, + 955.0, + 1063.0, + 989.0, + 321.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 262.0, + 619.0, + 262.0, + 619.0, + 294.0, + 322.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1835.0, + 1406.0, + 1835.0, + 1406.0, + 1875.0, + 293.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1867.0, + 479.0, + 1867.0, + 479.0, + 1900.0, + 323.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 510.0, + 1401.0, + 510.0, + 1401.0, + 546.0, + 294.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 539.0, + 497.0, + 539.0, + 497.0, + 573.0, + 322.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1589.0, + 1403.0, + 1589.0, + 1403.0, + 1625.0, + 295.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1618.0, + 636.0, + 1618.0, + 636.0, + 1652.0, + 321.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 756.0, + 1404.0, + 756.0, + 1404.0, + 797.0, + 293.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 790.0, + 883.0, + 790.0, + 883.0, + 822.0, + 322.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1088.0, + 1403.0, + 1088.0, + 1403.0, + 1124.0, + 294.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1120.0, + 490.0, + 1120.0, + 490.0, + 1154.0, + 325.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 422.0, + 1406.0, + 422.0, + 1406.0, + 464.0, + 293.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 459.0, + 819.0, + 459.0, + 819.0, + 489.0, + 323.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1670.0, + 1403.0, + 1670.0, + 1403.0, + 1706.0, + 296.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1703.0, + 1041.0, + 1703.0, + 1041.0, + 1736.0, + 322.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1755.0, + 1404.0, + 1755.0, + 1404.0, + 1791.0, + 295.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1786.0, + 770.0, + 1786.0, + 770.0, + 1816.0, + 323.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1502.0, + 1402.0, + 1502.0, + 1402.0, + 1542.0, + 293.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1537.0, + 1035.0, + 1537.0, + 1035.0, + 1570.0, + 322.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 836.0, + 1406.0, + 836.0, + 1406.0, + 880.0, + 294.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 872.0, + 1196.0, + 872.0, + 1196.0, + 905.0, + 323.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 311.0, + 1403.0, + 311.0, + 1403.0, + 349.0, + 297.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 343.0, + 1404.0, + 343.0, + 1404.0, + 379.0, + 322.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 371.0, + 399.0, + 371.0, + 399.0, + 407.0, + 320.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1423.0, + 1403.0, + 1423.0, + 1403.0, + 1459.0, + 295.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1454.0, + 859.0, + 1454.0, + 859.0, + 1487.0, + 322.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 587.0, + 1403.0, + 587.0, + 1403.0, + 632.0, + 294.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 624.0, + 1096.0, + 624.0, + 1096.0, + 656.0, + 325.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1173.0, + 1402.0, + 1173.0, + 1402.0, + 1209.0, + 297.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1202.0, + 1249.0, + 1202.0, + 1249.0, + 1240.0, + 322.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 677.0, + 1403.0, + 677.0, + 1403.0, + 709.0, + 296.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 705.0, + 1166.0, + 705.0, + 1166.0, + 740.0, + 321.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1337.0, + 1402.0, + 1337.0, + 1402.0, + 1376.0, + 294.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1371.0, + 938.0, + 1371.0, + 938.0, + 1403.0, + 325.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1257.0, + 1256.0, + 1257.0, + 1256.0, + 1293.0, + 297.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1257.0, + 1403.0, + 1257.0, + 1403.0, + 1293.0, + 1292.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1285.0, + 1403.0, + 1285.0, + 1403.0, + 1323.0, + 320.0, + 1323.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 780, + 1405, + 780, + 1405, + 1002, + 296, + 1002 + ], + "score": 0.979 + }, + { + "category_id": 8, + "poly": [ + 501, + 343, + 1197, + 343, + 1197, + 769, + 501, + 769 + ], + "score": 0.946 + }, + { + "category_id": 0, + "poly": [ + 298, + 224, + 1015, + 224, + 1015, + 264, + 298, + 264 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 286, + 294, + 1371, + 294, + 1371, + 329, + 286, + 329 + ], + "score": 0.915 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 816, + 74, + 816, + 105, + 298, + 105 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.844 + }, + { + "category_id": 9, + "poly": [ + 1368, + 542, + 1402, + 542, + 1402, + 571, + 1368, + 571 + ], + "score": 0.749 + }, + { + "category_id": 9, + "poly": [ + 1368, + 543, + 1401, + 543, + 1401, + 571, + 1368, + 571 + ], + "score": 0.097 + }, + { + "category_id": 14, + "poly": [ + 497, + 344, + 1193, + 344, + 1193, + 772, + 497, + 772 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { i _ { t } = \\sigma ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + { \\bar { W } } _ { i h } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { i } ) } \\\\ & { g _ { t } = \\operatorname { t a n h } ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + W _ { h , p } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { p } ) } \\\\ & { f _ { t } = \\sigma ( W _ { s ^ { t } } + { \\bar { X } } _ { t } + W _ { h , p } + { \\bar { \\mathcal { H } } } _ { i - 1 } ^ { k } + b _ { f } ) } \\\\ & { i _ { t } ^ { * } = \\sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { X } } _ { t } ^ { k - 1 } + b _ { i } ) } \\\\ & { i _ { t } ^ { * } = \\operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { i } ^ { k - 1 } + b _ { i } ^ { * } ) } \\\\ & { f _ { t } ^ { * } = \\operatorname { t a n h } ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\\\ & { c _ { t } ^ { * } = \\sigma ( W _ { r s ^ { t } } ^ { - 1 } + { \\bar { X } } _ { t } + W _ { m + 1 } + { \\bar { \\mathcal { H } } } _ { t } ^ { k - 1 } + b _ { f } ^ { * } ) } \\\\ & { M _ { t } ^ { * } = i _ { t } \\odot g _ { t } + f _ { t } \\odot { \\bar { C } } _ { t - 1 } ^ { k } } \\\\ & { \\boldsymbol { M } _ { t } ^ { * } = i _ { t } ^ { * } \\odot g _ { t } ^ { * } + f _ { t } ^ { * } \\odot { \\bar { M } } _ { t } ^ { k - 1 } } \\\\ & o _ { s } = \\sigma ( W _ { r s ^ { t } } + { \\bar { X } } _ { t } + W _ { i s ^ { t } } + { \\bar { M } } _ { t - 1 } ^ { k } + W _ { i s ^ { t } } + \\bar \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1270, + 813, + 1347, + 813, + 1347, + 850, + 1270, + 850 + ], + "score": 0.92, + "latex": "\\mathbf { \\mathcal { M } } _ { t } ^ { k - 1 }" + }, + { + "category_id": 13, + "poly": [ + 686, + 845, + 750, + 845, + 750, + 882, + 686, + 882 + ], + "score": 0.92, + "latex": "\\mathcal { H } _ { t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 805, + 845, + 861, + 845, + 861, + 882, + 805, + 882 + ], + "score": 0.91, + "latex": "\\mathcal { C } _ { t - 1 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 297, + 936, + 416, + 936, + 416, + 966, + 297, + 966 + ], + "score": 0.91, + "latex": "\\mathbb { R } ^ { H \\times W \\times C }" + }, + { + "category_id": 13, + "poly": [ + 1196, + 878, + 1223, + 878, + 1223, + 911, + 1196, + 911 + ], + "score": 0.88, + "latex": "f _ { t } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1112, + 879, + 1139, + 879, + 1139, + 910, + 1112, + 910 + ], + "score": 0.87, + "latex": "f _ { t }" + }, + { + "category_id": 13, + "poly": [ + 862, + 881, + 885, + 881, + 885, + 909, + 862, + 909 + ], + "score": 0.86, + "latex": "i _ { t }" + }, + { + "category_id": 13, + "poly": [ + 565, + 817, + 598, + 817, + 598, + 848, + 565, + 848 + ], + "score": 0.86, + "latex": "\\mathcal { X } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 942, + 878, + 964, + 878, + 964, + 910, + 942, + 910 + ], + "score": 0.86, + "latex": "i _ { t } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1028, + 786, + 1053, + 786, + 1053, + 812, + 1028, + 812 + ], + "score": 0.84, + "latex": "\\odot" + }, + { + "category_id": 13, + "poly": [ + 1107, + 298, + 1125, + 298, + 1125, + 324, + 1107, + 324 + ], + "score": 0.83, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 428, + 913, + 454, + 913, + 454, + 939, + 428, + 939 + ], + "score": 0.82, + "latex": "o _ { t }" + }, + { + "category_id": 13, + "poly": [ + 371, + 789, + 391, + 789, + 391, + 810, + 371, + 810 + ], + "score": 0.78, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 981, + 300, + 994, + 300, + 994, + 324, + 981, + 324 + ], + "score": 0.77, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 657, + 788, + 676, + 788, + 676, + 810, + 657, + 810 + ], + "score": 0.77, + "latex": "^ *" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 224.0, + 1016.0, + 224.0, + 1016.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 818.0, + 72.0, + 818.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2122.0, + 832.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 784.0, + 370.0, + 784.0, + 370.0, + 816.0, + 298.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 784.0, + 656.0, + 784.0, + 656.0, + 816.0, + 392.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 784.0, + 1027.0, + 784.0, + 1027.0, + 816.0, + 677.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 784.0, + 1402.0, + 784.0, + 1402.0, + 816.0, + 1054.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 808.0, + 564.0, + 808.0, + 564.0, + 855.0, + 291.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 808.0, + 1269.0, + 808.0, + 1269.0, + 855.0, + 599.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 808.0, + 1408.0, + 808.0, + 1408.0, + 855.0, + 1348.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 837.0, + 685.0, + 837.0, + 685.0, + 892.0, + 288.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 837.0, + 804.0, + 837.0, + 804.0, + 892.0, + 751.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 837.0, + 1411.0, + 837.0, + 1411.0, + 892.0, + 862.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 876.0, + 861.0, + 876.0, + 861.0, + 915.0, + 294.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 876.0, + 941.0, + 876.0, + 941.0, + 915.0, + 886.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 876.0, + 1111.0, + 876.0, + 1111.0, + 915.0, + 965.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 876.0, + 1195.0, + 876.0, + 1195.0, + 915.0, + 1140.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 876.0, + 1406.0, + 876.0, + 1406.0, + 915.0, + 1224.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 907.0, + 427.0, + 907.0, + 427.0, + 943.0, + 294.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 907.0, + 1402.0, + 907.0, + 1402.0, + 943.0, + 455.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 927.0, + 296.0, + 927.0, + 296.0, + 982.0, + 289.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 927.0, + 1410.0, + 927.0, + 1410.0, + 982.0, + 417.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 970.0, + 985.0, + 970.0, + 985.0, + 1004.0, + 294.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 292.0, + 980.0, + 292.0, + 980.0, + 334.0, + 294.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 292.0, + 1106.0, + 292.0, + 1106.0, + 334.0, + 995.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 292.0, + 1377.0, + 292.0, + 1377.0, + 334.0, + 1126.0, + 334.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/B1lnbRNtwr/images/4e04d6ea7dd5bfb2ef01a6dfd6d452ab0dbf57732e286b392cbf2337a905571b.jpg b/parse/train/B1lnbRNtwr/images/4e04d6ea7dd5bfb2ef01a6dfd6d452ab0dbf57732e286b392cbf2337a905571b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2da87b7f93dec90b1d23dae77b24aa54bbcd3c4c --- /dev/null +++ b/parse/train/B1lnbRNtwr/images/4e04d6ea7dd5bfb2ef01a6dfd6d452ab0dbf57732e286b392cbf2337a905571b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b4e116b9d20df38224cd326e63c61905516c01a53143df1021e219979f221d8 +size 31574 diff --git a/parse/train/B1lnbRNtwr/images/4e7aea86a35c3a5154dd03b345cf8c91955b43f8821d291f304467e0d6db27fa.jpg b/parse/train/B1lnbRNtwr/images/4e7aea86a35c3a5154dd03b345cf8c91955b43f8821d291f304467e0d6db27fa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ed9c05010caf6f830e770336806c09de45eca01c --- /dev/null +++ b/parse/train/B1lnbRNtwr/images/4e7aea86a35c3a5154dd03b345cf8c91955b43f8821d291f304467e0d6db27fa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:209aeb772eb63818d3a72cf5a46daa3878dad39104f9532e71ae20dd75c713a2 +size 60351 diff --git a/parse/train/B1lnbRNtwr/images/891bb76b8b22cc3f22d8d35865dbfce9b0f0a82d9df045cca8830b19b5be6265.jpg b/parse/train/B1lnbRNtwr/images/891bb76b8b22cc3f22d8d35865dbfce9b0f0a82d9df045cca8830b19b5be6265.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6b95b00c43fd1b234d86d4d38ac8ab530b4952c4 --- /dev/null +++ b/parse/train/B1lnbRNtwr/images/891bb76b8b22cc3f22d8d35865dbfce9b0f0a82d9df045cca8830b19b5be6265.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:223ea1def0929baa780583482cbf6e0ee2be327e94d4fe8db95835f88ba18f7d +size 42832 diff --git a/parse/train/B1lnbRNtwr/images/96061ba572ebf13a72a349a63a2a4c26ec4970bd22d2b93d03adfbc9020773de.jpg b/parse/train/B1lnbRNtwr/images/96061ba572ebf13a72a349a63a2a4c26ec4970bd22d2b93d03adfbc9020773de.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9c169e9340a8d52cfa294628117b9c23a8c45846 --- /dev/null +++ b/parse/train/B1lnbRNtwr/images/96061ba572ebf13a72a349a63a2a4c26ec4970bd22d2b93d03adfbc9020773de.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99eec22f5a7e31bfd97b0ac6f42eb2b9e10fef707cb3369e2e1a06a3e8a7f9e8 +size 81831 diff --git a/parse/train/B1lnbRNtwr/images/a2588d44c5222e93ce00d4984202080de0bba24d5330bb29927852b42faca985.jpg b/parse/train/B1lnbRNtwr/images/a2588d44c5222e93ce00d4984202080de0bba24d5330bb29927852b42faca985.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4482a25127fd192b1cf8e9f8fee10fafb352b314 --- /dev/null +++ b/parse/train/B1lnbRNtwr/images/a2588d44c5222e93ce00d4984202080de0bba24d5330bb29927852b42faca985.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a78b40a8d4f553b109dd0baf6ba8d47e3c578e97a2076bcbfb2f7bba44ad6f0d +size 58462 diff --git a/parse/train/B1lnbRNtwr/images/a820488f09223f97e60319c333a9fc46ccb028312e07b0032199bdaa49aa909f.jpg b/parse/train/B1lnbRNtwr/images/a820488f09223f97e60319c333a9fc46ccb028312e07b0032199bdaa49aa909f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e6b8d861cb56f36d0b04da6d1f256c9dc1980781 --- /dev/null +++ b/parse/train/B1lnbRNtwr/images/a820488f09223f97e60319c333a9fc46ccb028312e07b0032199bdaa49aa909f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3e8332768fa84a5a7f7e625ed8cb6137c88d73e454813afc2da8dc88477cc8b +size 41931 diff --git a/parse/train/B1xGGTEtDH/images/005a9be986d36d130fea0002daed498fe4a8ef02624f4b22255a2db90f57ca15.jpg b/parse/train/B1xGGTEtDH/images/005a9be986d36d130fea0002daed498fe4a8ef02624f4b22255a2db90f57ca15.jpg new file mode 100644 index 0000000000000000000000000000000000000000..85c063c00cd545ec54c048711c8078309ff68aa9 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/005a9be986d36d130fea0002daed498fe4a8ef02624f4b22255a2db90f57ca15.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7e4b1bfd0e57753ed256f0df1f7ed7f4ead0f7d264da57ef657a938fdec249b +size 4307 diff --git a/parse/train/B1xGGTEtDH/images/010d230e8480987f35c395761e51d890f07056606c7bd01bf5f483044374d8cb.jpg b/parse/train/B1xGGTEtDH/images/010d230e8480987f35c395761e51d890f07056606c7bd01bf5f483044374d8cb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3e845f182efe586650ebe9b2a686ea532f6a7f27 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/010d230e8480987f35c395761e51d890f07056606c7bd01bf5f483044374d8cb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5226af5faff2b4dc2d579cb95d8a9271da9316cade0d3d3e6de1455b9225aaa +size 3159 diff --git a/parse/train/B1xGGTEtDH/images/01609c450a5b9a8279dc91aa24d8c88699819c4e9c8687e1f0906f85dfd6f520.jpg b/parse/train/B1xGGTEtDH/images/01609c450a5b9a8279dc91aa24d8c88699819c4e9c8687e1f0906f85dfd6f520.jpg new file mode 100644 index 0000000000000000000000000000000000000000..499d46cd6187670412f0ccfe46d0cbaed28a45f1 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/01609c450a5b9a8279dc91aa24d8c88699819c4e9c8687e1f0906f85dfd6f520.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b34eec9f94c7276f5874880b77f4940c9a3e827e9906e429c6778cee1a7e863 +size 3679 diff --git a/parse/train/B1xGGTEtDH/images/0f26ba81f1c75042a707b1c99df88e3e8eb00ee71f2be0b24436d8cb1b077e35.jpg b/parse/train/B1xGGTEtDH/images/0f26ba81f1c75042a707b1c99df88e3e8eb00ee71f2be0b24436d8cb1b077e35.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e92481dd5489eeed14f4c96c0e2fac82a1be00b7 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/0f26ba81f1c75042a707b1c99df88e3e8eb00ee71f2be0b24436d8cb1b077e35.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:191e5882a2f0b74605171472405189f15112e98c027fc95e5be9e144da5216b4 +size 4198 diff --git a/parse/train/B1xGGTEtDH/images/35ba781337ed71f0a131372d33d2dc660d3951779d0785429c0401dd7b6e7c7f.jpg b/parse/train/B1xGGTEtDH/images/35ba781337ed71f0a131372d33d2dc660d3951779d0785429c0401dd7b6e7c7f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..68e9e3ad0cba80ff54a3f82bdc8965e69364a6ca --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/35ba781337ed71f0a131372d33d2dc660d3951779d0785429c0401dd7b6e7c7f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed3d4ef12f885f6d505d25747a12927b089877adc85f7d53b4b5da06e4ef2b14 +size 4284 diff --git a/parse/train/B1xGGTEtDH/images/3d0e433b1f0701389abdd2a421eee41c4927a407bd0b1734451d94e541c5b095.jpg b/parse/train/B1xGGTEtDH/images/3d0e433b1f0701389abdd2a421eee41c4927a407bd0b1734451d94e541c5b095.jpg new file mode 100644 index 0000000000000000000000000000000000000000..583ba29243f845a2423724effe16e5fb66ffb850 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/3d0e433b1f0701389abdd2a421eee41c4927a407bd0b1734451d94e541c5b095.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:655877cfe8ca6fd81e9c19260de4e819bc1edf84b02e9088e1250575f992b94f +size 9051 diff --git a/parse/train/B1xGGTEtDH/images/4738d8e24386430ae848681079f2ebd138170adaf7e14aff4db064cc0fbd8001.jpg b/parse/train/B1xGGTEtDH/images/4738d8e24386430ae848681079f2ebd138170adaf7e14aff4db064cc0fbd8001.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1dcdc1158e0698c280991da7866ca43c933a3f63 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/4738d8e24386430ae848681079f2ebd138170adaf7e14aff4db064cc0fbd8001.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0574aa4d7cda5354b3fa0e4664c8c00df07ff7d2fe87ac323918137e5d27b095 +size 2271 diff --git a/parse/train/B1xGGTEtDH/images/4df0dc596ea71798f34aaccae20dd7daeeca10f0be2d667d27ba1b99b75ce4fd.jpg b/parse/train/B1xGGTEtDH/images/4df0dc596ea71798f34aaccae20dd7daeeca10f0be2d667d27ba1b99b75ce4fd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dab4c974083d1cdf5f24b6f63168a8614e6ed834 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/4df0dc596ea71798f34aaccae20dd7daeeca10f0be2d667d27ba1b99b75ce4fd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac555e71d91bf4e9ca0d00fad5aaa5733f18dc4a224d5b3c51687278143840ab +size 4108 diff --git a/parse/train/B1xGGTEtDH/images/569ab58274b72ece6d7ed56bf58530ba9d79a8e9360ea4a67f9f75a2e051c854.jpg b/parse/train/B1xGGTEtDH/images/569ab58274b72ece6d7ed56bf58530ba9d79a8e9360ea4a67f9f75a2e051c854.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d7719304fd7bbdf71846d5f067ccd3bb975e6c7c --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/569ab58274b72ece6d7ed56bf58530ba9d79a8e9360ea4a67f9f75a2e051c854.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ba96c53e54fb7e90dbf451b065fc5c49d589b550e21168a73b27523078e0fb5a +size 3427 diff --git a/parse/train/B1xGGTEtDH/images/60a4927ccdb05a77b16a792eab7efe5b0f543fd058fc512d5cc22ef3b7d6d9b3.jpg b/parse/train/B1xGGTEtDH/images/60a4927ccdb05a77b16a792eab7efe5b0f543fd058fc512d5cc22ef3b7d6d9b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d30b8d32aace94b5e3c40bd3da789f8114b1f865 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/60a4927ccdb05a77b16a792eab7efe5b0f543fd058fc512d5cc22ef3b7d6d9b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f82860f9045b4b2d76938bf2bd9ac5898758bbf74975a4f1fc21561dea048630 +size 3797 diff --git a/parse/train/B1xGGTEtDH/images/6377853cb048c309803332041350ad6d6988551eb6cd08a3f0d4d1596fbd63b2.jpg b/parse/train/B1xGGTEtDH/images/6377853cb048c309803332041350ad6d6988551eb6cd08a3f0d4d1596fbd63b2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf9712ac5b5cc31887424cd9fe39a968c88b245d --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/6377853cb048c309803332041350ad6d6988551eb6cd08a3f0d4d1596fbd63b2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0126e35b0b30f6b4ca22f50e452d1b7d291672bb6f2e764fdfbcaca902d8f7d +size 13708 diff --git a/parse/train/B1xGGTEtDH/images/76f362de9cfa71e4b13fef5602f6c11d29ca89f829c7fd6a514cd94326fc076a.jpg b/parse/train/B1xGGTEtDH/images/76f362de9cfa71e4b13fef5602f6c11d29ca89f829c7fd6a514cd94326fc076a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..22391370e969e36adb33f05a408317cba7814eed --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/76f362de9cfa71e4b13fef5602f6c11d29ca89f829c7fd6a514cd94326fc076a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e41b80f9068c739aa152de94c7018234c3c96eb166de7ddb396b5503f398362 +size 3405 diff --git a/parse/train/B1xGGTEtDH/images/77a1f1c40712be63d7bf1a1e9d1de9dcc7f38e3a02e4a29d54050d0305512ccb.jpg b/parse/train/B1xGGTEtDH/images/77a1f1c40712be63d7bf1a1e9d1de9dcc7f38e3a02e4a29d54050d0305512ccb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..007ff70893b110292b14372143a9605fa8402150 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/77a1f1c40712be63d7bf1a1e9d1de9dcc7f38e3a02e4a29d54050d0305512ccb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d33e6c6ee3382b2785adee5a5657109fb2bdb13b3b06f34d46a67b1d2d0ac9e +size 44030 diff --git a/parse/train/B1xGGTEtDH/images/83b867048d143c81fcfe359bb0e610677f3b20beca63cf7ad0e8e674bdff7604.jpg b/parse/train/B1xGGTEtDH/images/83b867048d143c81fcfe359bb0e610677f3b20beca63cf7ad0e8e674bdff7604.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7fe6c4e814637226ea9e3668865946f856282619 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/83b867048d143c81fcfe359bb0e610677f3b20beca63cf7ad0e8e674bdff7604.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2031ec135b8d83910c2e65d2c5d01fe8e09a26f8345b28c40ccf859ac467254 +size 15158 diff --git a/parse/train/B1xGGTEtDH/images/8bfe647369ce1c5f8ff0d03120eefb268cc14a53af69b7c42d3445d606ac9aa9.jpg b/parse/train/B1xGGTEtDH/images/8bfe647369ce1c5f8ff0d03120eefb268cc14a53af69b7c42d3445d606ac9aa9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b646fa3880e1d84dadad2640ce1978eca876c90d --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/8bfe647369ce1c5f8ff0d03120eefb268cc14a53af69b7c42d3445d606ac9aa9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:126d87b2080cd1d79e780a1ab283c4c23a4d3bafadf6e892f37198e5b0a4dcbb +size 6578 diff --git a/parse/train/B1xGGTEtDH/images/8defa91d163f54513042517266f86ed7402338804de2a06cb41feb06c3ef1572.jpg b/parse/train/B1xGGTEtDH/images/8defa91d163f54513042517266f86ed7402338804de2a06cb41feb06c3ef1572.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2e1982b73ed4507d0573c00a620d46837a9785bb --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/8defa91d163f54513042517266f86ed7402338804de2a06cb41feb06c3ef1572.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5bd76c76180d2692a7c80457a538df9e1c0adae216a2c89c7b50262bec1c554 +size 223013 diff --git a/parse/train/B1xGGTEtDH/images/af1a570e535f5908c5c85bb8e2800e0d3354b9df516226ad42fc3679c7f8313f.jpg b/parse/train/B1xGGTEtDH/images/af1a570e535f5908c5c85bb8e2800e0d3354b9df516226ad42fc3679c7f8313f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4e4d7a61a00728e3f384f9425819e113254dee66 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/af1a570e535f5908c5c85bb8e2800e0d3354b9df516226ad42fc3679c7f8313f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48e4b8dc42b93c0d852a99607aa8ce64a6ab37ea2e588207680a5d6997738b59 +size 2814 diff --git a/parse/train/B1xGGTEtDH/images/b6988f3422f9c6d56e7f4130e2f2fa230c957fb8f2ba8a0f238bc53d9b2bcb1f.jpg b/parse/train/B1xGGTEtDH/images/b6988f3422f9c6d56e7f4130e2f2fa230c957fb8f2ba8a0f238bc53d9b2bcb1f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..991dda8e62a933ca77ad2abcaed3d5c67d3803ac --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/b6988f3422f9c6d56e7f4130e2f2fa230c957fb8f2ba8a0f238bc53d9b2bcb1f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:301581006e7666f9620c60aeba93f9cb7dd70a706360407b39929a6905aeb1d1 +size 2660 diff --git a/parse/train/B1xGGTEtDH/images/b7304db5862598da35f9e657922523f8a65ef551dcc7897011d1de37a90ba8d5.jpg b/parse/train/B1xGGTEtDH/images/b7304db5862598da35f9e657922523f8a65ef551dcc7897011d1de37a90ba8d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8ec005055045a56e3275b9883cd7956fc5598ead --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/b7304db5862598da35f9e657922523f8a65ef551dcc7897011d1de37a90ba8d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:316c1238dda02276f2ede7a3c64ccc166cc0a2eda2f6c2f124975448bcee51b9 +size 2011 diff --git a/parse/train/B1xGGTEtDH/images/ce1621d47ac877e748d4018735dfaf56dc1c387ad4dd886abba76e685ab61889.jpg b/parse/train/B1xGGTEtDH/images/ce1621d47ac877e748d4018735dfaf56dc1c387ad4dd886abba76e685ab61889.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6175e22b6e0a9e86cc2082858e3a4c5a001b187a --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/ce1621d47ac877e748d4018735dfaf56dc1c387ad4dd886abba76e685ab61889.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c8aa546ba230d6b90e99c1508a7704d7f48fbd4d2e320f07d1742dca8c8d334 +size 6853 diff --git a/parse/train/B1xGGTEtDH/images/d8d861e9e77e44a85d11823870d8b8120a821d325d15dfdac7585047bbac1c94.jpg b/parse/train/B1xGGTEtDH/images/d8d861e9e77e44a85d11823870d8b8120a821d325d15dfdac7585047bbac1c94.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4539850e0d62b7c104b293aa26acafe90a6c6d1d --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/d8d861e9e77e44a85d11823870d8b8120a821d325d15dfdac7585047bbac1c94.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02dd824383efc74d5369bfe157e59ac16992e00850d2f91e9062cdbbaed5e46d +size 12024 diff --git a/parse/train/B1xGGTEtDH/images/f645608cc349ad456f0a6d179e919a4e41762d8e8d22904b07bcc9560f4d1099.jpg b/parse/train/B1xGGTEtDH/images/f645608cc349ad456f0a6d179e919a4e41762d8e8d22904b07bcc9560f4d1099.jpg new file mode 100644 index 0000000000000000000000000000000000000000..97c1c2685db680ae8589433a839371336b6d299b --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/f645608cc349ad456f0a6d179e919a4e41762d8e8d22904b07bcc9560f4d1099.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f82de12c45906b4538c6c340aa0cb3b83af517004566ac3e7a60539ce8322a44 +size 5006 diff --git a/parse/train/B1xGGTEtDH/images/f8aba92fb785ad43d50d7d69856a6c125c034251dd982e6ac39031fe541afb0f.jpg b/parse/train/B1xGGTEtDH/images/f8aba92fb785ad43d50d7d69856a6c125c034251dd982e6ac39031fe541afb0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d17d37774469343c6ff74a6dfd7f704d2e557ff1 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/f8aba92fb785ad43d50d7d69856a6c125c034251dd982e6ac39031fe541afb0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4cbac6162772737482d01af59a77c590a9b49144f60447aca99ecfbb2bacd2fe +size 9326 diff --git a/parse/train/B1xGGTEtDH/images/fd2fcf6b3f35a674ee7b85622c79d14ec69e8382b2118a9fe6c0463987f2cf41.jpg b/parse/train/B1xGGTEtDH/images/fd2fcf6b3f35a674ee7b85622c79d14ec69e8382b2118a9fe6c0463987f2cf41.jpg new file mode 100644 index 0000000000000000000000000000000000000000..34ff9421823ad4502cc682797fab995ff56e9156 --- /dev/null +++ b/parse/train/B1xGGTEtDH/images/fd2fcf6b3f35a674ee7b85622c79d14ec69e8382b2118a9fe6c0463987f2cf41.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7355e7feb4b319c18534109b84e4aef44820f32ea653f2c689210b549a94badf +size 10943 diff --git a/parse/train/BkMiWhR5K7/images/01e2937e20c234ab99fa4c31716fc29edbacc869cd093edc1602b268f8337c53.jpg b/parse/train/BkMiWhR5K7/images/01e2937e20c234ab99fa4c31716fc29edbacc869cd093edc1602b268f8337c53.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e37d61c8b3ad055ec75ce0bd08a300d1c2042330 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/01e2937e20c234ab99fa4c31716fc29edbacc869cd093edc1602b268f8337c53.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f363189ccf4bb482d40db1ab4748c571f044aba8a3d294a1c38f097fe97d9b9f +size 6568 diff --git a/parse/train/BkMiWhR5K7/images/033c09256bf40d9a938ead82d2a2e9329b1da0241b822822b9bfc8e9370f224f.jpg b/parse/train/BkMiWhR5K7/images/033c09256bf40d9a938ead82d2a2e9329b1da0241b822822b9bfc8e9370f224f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a65d17b164d7be04e03014ccfc23664d291d2573 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/033c09256bf40d9a938ead82d2a2e9329b1da0241b822822b9bfc8e9370f224f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d205a2d1e0270cbe1ba5103fea78f24135fa7dfad9d0f3dc9f6a92382b251ee +size 8887 diff --git a/parse/train/BkMiWhR5K7/images/03c138a46f1eea162c4e7c2ed7dde6132e79969cafcd063d94adc89e0ee3c82e.jpg b/parse/train/BkMiWhR5K7/images/03c138a46f1eea162c4e7c2ed7dde6132e79969cafcd063d94adc89e0ee3c82e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..356b3cea6515340441fbafac18c29b907bd8d8a8 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/03c138a46f1eea162c4e7c2ed7dde6132e79969cafcd063d94adc89e0ee3c82e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d52361c6887b2f221ef498b2095095b960f36c9aa9caede5aa8a55c5ea0e4e7 +size 49687 diff --git a/parse/train/BkMiWhR5K7/images/051ba5e9bfdba24001c5a53833c8e69d8b370640906ab1fd34adc4a674be15b5.jpg b/parse/train/BkMiWhR5K7/images/051ba5e9bfdba24001c5a53833c8e69d8b370640906ab1fd34adc4a674be15b5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2856a00a089797e7f0049e9789f28991dbf75d67 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/051ba5e9bfdba24001c5a53833c8e69d8b370640906ab1fd34adc4a674be15b5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82c0ca964a403f879d9fba650735399d72fc1c562dad84e1af17e5e34703321d +size 5865 diff --git a/parse/train/BkMiWhR5K7/images/0cfa962cfd687e288dd600ea48a34a0807d2c5e3052ee4b48836540dfe04695d.jpg b/parse/train/BkMiWhR5K7/images/0cfa962cfd687e288dd600ea48a34a0807d2c5e3052ee4b48836540dfe04695d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e931b8eddb2b00a3331d431498ee4865a761596b --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/0cfa962cfd687e288dd600ea48a34a0807d2c5e3052ee4b48836540dfe04695d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5236f5ed0f3c966b3b529cfa31d0e185313a40aab140d655d9fc861ba5bd8fd +size 11967 diff --git a/parse/train/BkMiWhR5K7/images/0e59b85447a205c777648a913b22566a62c463523315771922f179cefb152c47.jpg b/parse/train/BkMiWhR5K7/images/0e59b85447a205c777648a913b22566a62c463523315771922f179cefb152c47.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f37a793b8f5ab3beb8ed377a9da437069c105593 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/0e59b85447a205c777648a913b22566a62c463523315771922f179cefb152c47.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a05e78595e12a670c6b5698f11eaab51f9353ab9e55b19511b43d217f7aa7377 +size 9774 diff --git a/parse/train/BkMiWhR5K7/images/11b9f5e2242fccfcf84600d877a5977c77fc696edb2a54ec603871f397e6de76.jpg b/parse/train/BkMiWhR5K7/images/11b9f5e2242fccfcf84600d877a5977c77fc696edb2a54ec603871f397e6de76.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f5944839843e35c9fe63c83e2334f5e997e6efd1 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/11b9f5e2242fccfcf84600d877a5977c77fc696edb2a54ec603871f397e6de76.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a21a770eb50906f8971f137cf7c48811e440ecc85387418b043a555a9017bcc4 +size 16692 diff --git a/parse/train/BkMiWhR5K7/images/160a371af5bb1a84c89942353c84dcb61158efe51c2bd7c9953619b37f24e2e0.jpg b/parse/train/BkMiWhR5K7/images/160a371af5bb1a84c89942353c84dcb61158efe51c2bd7c9953619b37f24e2e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..25aea4f7bc8d17a8e73edd415c35d4149b9f7dc6 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/160a371af5bb1a84c89942353c84dcb61158efe51c2bd7c9953619b37f24e2e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9da14a4fd9bde0154b8f22892ae2a4b92dfdc960a65cbf016869df780cb8b7e5 +size 14387 diff --git a/parse/train/BkMiWhR5K7/images/169b372ceb32e163188692b5c7785e7736638c6bfff5cb3f49dc6f4875eada5b.jpg b/parse/train/BkMiWhR5K7/images/169b372ceb32e163188692b5c7785e7736638c6bfff5cb3f49dc6f4875eada5b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f63de9210afa36d14fcfdd91cbd48b7256059cd4 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/169b372ceb32e163188692b5c7785e7736638c6bfff5cb3f49dc6f4875eada5b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48be434013d4712b90f5829105384ded73451746bdb51c06c8754c2f907fc59c +size 5723 diff --git a/parse/train/BkMiWhR5K7/images/2601855269c883f52828dbcffd4305a80670e474720170f2e0bca04a3be75321.jpg b/parse/train/BkMiWhR5K7/images/2601855269c883f52828dbcffd4305a80670e474720170f2e0bca04a3be75321.jpg new file mode 100644 index 0000000000000000000000000000000000000000..16cbc7d411ab9a47f74ae9fbca8de1c7934d6795 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/2601855269c883f52828dbcffd4305a80670e474720170f2e0bca04a3be75321.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac4a6e40ef25080ea7f812f13612aa92f9c6f46b70a73dddccb35ddbd763516f +size 20058 diff --git a/parse/train/BkMiWhR5K7/images/26cff5c517828d67bf7018b934881e692a975988a698b2cc8713241a08eed4ac.jpg b/parse/train/BkMiWhR5K7/images/26cff5c517828d67bf7018b934881e692a975988a698b2cc8713241a08eed4ac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf6376dc36498ac2f3f97fa2ed2aaa56305ad2b6 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/26cff5c517828d67bf7018b934881e692a975988a698b2cc8713241a08eed4ac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4485b7736dda7854549236168df99c283c6b4a82402ad8d3e45b61a186c1b4c7 +size 9460 diff --git a/parse/train/BkMiWhR5K7/images/285ffe9df8012ae055451e0dd2e1429b99144085e1243388b001d66c030d5f85.jpg b/parse/train/BkMiWhR5K7/images/285ffe9df8012ae055451e0dd2e1429b99144085e1243388b001d66c030d5f85.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b7c7256ab29a04739b86e1b9a8ab5c176b9820e2 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/285ffe9df8012ae055451e0dd2e1429b99144085e1243388b001d66c030d5f85.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e7f15343b64594d326b0585dddd812ccbb7488fe4ae7f5c5b08dc458f4d475a +size 25724 diff --git a/parse/train/BkMiWhR5K7/images/2bc111464008ab092d2df08469e0badf39501a546cecb48dfe86e1c8b76b2022.jpg b/parse/train/BkMiWhR5K7/images/2bc111464008ab092d2df08469e0badf39501a546cecb48dfe86e1c8b76b2022.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c8bf1ba7c97690281b5bfe199311d09dec7d2e0b --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/2bc111464008ab092d2df08469e0badf39501a546cecb48dfe86e1c8b76b2022.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb541787256a7a19aa34a294b3b5d019fba79f2f93c6c0bf63946741de03e458 +size 4941 diff --git a/parse/train/BkMiWhR5K7/images/30c37af862088da7a5370f9becff751d224253a299673945a00af3400f57183b.jpg b/parse/train/BkMiWhR5K7/images/30c37af862088da7a5370f9becff751d224253a299673945a00af3400f57183b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f73327a14a8f898aedb28f7a7f2902c17562073b --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/30c37af862088da7a5370f9becff751d224253a299673945a00af3400f57183b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85f9b9910558528554e35633fad40a584f59693d1d7011ab4e8daf26a11c8585 +size 4590 diff --git a/parse/train/BkMiWhR5K7/images/41e9aa1e5aa47d713cb965fe4d15759c637d9a7ee3da6b6acd6787956fe185ea.jpg b/parse/train/BkMiWhR5K7/images/41e9aa1e5aa47d713cb965fe4d15759c637d9a7ee3da6b6acd6787956fe185ea.jpg new file mode 100644 index 0000000000000000000000000000000000000000..59157085d98178f19abbf261695ab5542553e051 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/41e9aa1e5aa47d713cb965fe4d15759c637d9a7ee3da6b6acd6787956fe185ea.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5312b0891922d1b86ccccf34cfcd259cfd268b952138d65f0ba7043837c2c842 +size 29817 diff --git a/parse/train/BkMiWhR5K7/images/4249b62006b3569a840ffe6db36bc969c0d2d7df9345590ea32a0f302a79dc91.jpg b/parse/train/BkMiWhR5K7/images/4249b62006b3569a840ffe6db36bc969c0d2d7df9345590ea32a0f302a79dc91.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a014433cf9d263cd1742f8956b1602e3f66e93fd --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/4249b62006b3569a840ffe6db36bc969c0d2d7df9345590ea32a0f302a79dc91.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:249407e0d8d973eb171a0e6e2b8f8bbef33d927c6649ad80f8ce8b6e7c609368 +size 8480 diff --git a/parse/train/BkMiWhR5K7/images/466866250b5a0632b0febe5db65e7b5279dc68c31a54f7d4bb514596bb8e4ead.jpg b/parse/train/BkMiWhR5K7/images/466866250b5a0632b0febe5db65e7b5279dc68c31a54f7d4bb514596bb8e4ead.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa680e67e164e8dc3bad9c180bedb8d8dc9ee473 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/466866250b5a0632b0febe5db65e7b5279dc68c31a54f7d4bb514596bb8e4ead.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd93a14f3520162ca49fa4d3f2a07797a0fdec2b9b5676ccfb0d03e3acd4e2e2 +size 22791 diff --git a/parse/train/BkMiWhR5K7/images/4bd6dcf16e6db72b782d6df27a10d8ad586cfe321469bc9db302b2b2394ea5bd.jpg b/parse/train/BkMiWhR5K7/images/4bd6dcf16e6db72b782d6df27a10d8ad586cfe321469bc9db302b2b2394ea5bd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dd9c00e450d026f42ebfc632cd0f047bab57e16a --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/4bd6dcf16e6db72b782d6df27a10d8ad586cfe321469bc9db302b2b2394ea5bd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:730e831ce9c6f905b358a708e92bfab8dca1334e85782b010a66ab2e6514a727 +size 7066 diff --git a/parse/train/BkMiWhR5K7/images/4c5f73d9e92dd775186cfb787fbabb29a977ee007b9c133f3ada98d5af1bdf3f.jpg b/parse/train/BkMiWhR5K7/images/4c5f73d9e92dd775186cfb787fbabb29a977ee007b9c133f3ada98d5af1bdf3f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c6a859a1aa2975a1c4e8193ee54e06e47d1ae0fd --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/4c5f73d9e92dd775186cfb787fbabb29a977ee007b9c133f3ada98d5af1bdf3f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7a1183d186ae64f4c62151fca8f532878a64e1648037fe41e3b412405434d97 +size 11843 diff --git a/parse/train/BkMiWhR5K7/images/4fb8aca011bbd05ea16c1ea922a0951fd9791d59961f5f506a92c854fd74f16e.jpg b/parse/train/BkMiWhR5K7/images/4fb8aca011bbd05ea16c1ea922a0951fd9791d59961f5f506a92c854fd74f16e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3638c44aa8b1efdb6eeb9998972eddcb1f7ce2d4 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/4fb8aca011bbd05ea16c1ea922a0951fd9791d59961f5f506a92c854fd74f16e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:524fd3cb30ad49b4f494229bfb34079c068e28ba01519edec6489ba115701c91 +size 6866 diff --git a/parse/train/BkMiWhR5K7/images/50e81a114935c44b9cafffa2e8474f97c1322c0d970b6c0fa8e4e1eefc7e61d5.jpg b/parse/train/BkMiWhR5K7/images/50e81a114935c44b9cafffa2e8474f97c1322c0d970b6c0fa8e4e1eefc7e61d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f45ac9eb952c3e33357ac62f5044c8af4c58f524 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/50e81a114935c44b9cafffa2e8474f97c1322c0d970b6c0fa8e4e1eefc7e61d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f2d0efd23b0f27d1628853c8c2125d3a81878d02dd167f51cb7f6e4c90d0d43 +size 13714 diff --git a/parse/train/BkMiWhR5K7/images/5642cef5f11ae036545391a317fd2136d085ad411de29db56bb71019a2482142.jpg b/parse/train/BkMiWhR5K7/images/5642cef5f11ae036545391a317fd2136d085ad411de29db56bb71019a2482142.jpg new file mode 100644 index 0000000000000000000000000000000000000000..445561df2cd965cf3f7fd306e600cb6ba5dddb56 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/5642cef5f11ae036545391a317fd2136d085ad411de29db56bb71019a2482142.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe787fa63d34cb153157873c9bb9acf181c7882ebef3dd92506fa78f3297f247 +size 10964 diff --git a/parse/train/BkMiWhR5K7/images/5902d28ee415e0b6f69c879916dcc4ea3e464af83f341cf9ff2051fdda47edff.jpg b/parse/train/BkMiWhR5K7/images/5902d28ee415e0b6f69c879916dcc4ea3e464af83f341cf9ff2051fdda47edff.jpg new file mode 100644 index 0000000000000000000000000000000000000000..31048845d5065ecc08d2d9d2d8218e3889501f34 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/5902d28ee415e0b6f69c879916dcc4ea3e464af83f341cf9ff2051fdda47edff.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df472c34461aaddc52a26276a54abf8de839556963ec4571da373419e0b21fc0 +size 1980 diff --git a/parse/train/BkMiWhR5K7/images/60fc3224ca2c77118a21ae31b54f1effbea1a4fc14a48736ba42bc23eab4ef01.jpg b/parse/train/BkMiWhR5K7/images/60fc3224ca2c77118a21ae31b54f1effbea1a4fc14a48736ba42bc23eab4ef01.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f5378ca6b340f6a7a1da7d7f0e1760788994578c --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/60fc3224ca2c77118a21ae31b54f1effbea1a4fc14a48736ba42bc23eab4ef01.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b8921ed752879de6b766a86571545ba009da490462e9e55af129cfff7432779 +size 4647 diff --git a/parse/train/BkMiWhR5K7/images/6554ac5f229e99c5dba15375d05aa2514b017ada58f9d05df9fa78f5e89f290c.jpg b/parse/train/BkMiWhR5K7/images/6554ac5f229e99c5dba15375d05aa2514b017ada58f9d05df9fa78f5e89f290c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cfd81adf6d47973e7ab49f79f5746f3baa9d81af --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/6554ac5f229e99c5dba15375d05aa2514b017ada58f9d05df9fa78f5e89f290c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3a3be7c375ad831da86eda369cd8ce83ce55b989f8bee923672962fab99a711 +size 9533 diff --git a/parse/train/BkMiWhR5K7/images/7177ddd785fc401efae7fb6e723179501f8154c819cc22e4952afb0e2c59f838.jpg b/parse/train/BkMiWhR5K7/images/7177ddd785fc401efae7fb6e723179501f8154c819cc22e4952afb0e2c59f838.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0f5602dbc58ce4f26543b2934756840e6e44a8f1 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/7177ddd785fc401efae7fb6e723179501f8154c819cc22e4952afb0e2c59f838.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:108dc560564ee426447957687da02496c3ed3c182c2458d0ac2cb63dcbdb3df4 +size 46000 diff --git a/parse/train/BkMiWhR5K7/images/7781c5c74950697a662802f9697348105e386b35d19d1a500bd91931f43e9553.jpg b/parse/train/BkMiWhR5K7/images/7781c5c74950697a662802f9697348105e386b35d19d1a500bd91931f43e9553.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dbf8ca6a33e74e236920ebf135085c8c1c2ef558 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/7781c5c74950697a662802f9697348105e386b35d19d1a500bd91931f43e9553.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27cb618108a9aa3e33ab27ec376d6bd89dcd94d502f4ce5c176df561febb5130 +size 31138 diff --git a/parse/train/BkMiWhR5K7/images/790dce92240bf4611549cbb710bcc157edb7d8e9795916a14056f779882e5a50.jpg b/parse/train/BkMiWhR5K7/images/790dce92240bf4611549cbb710bcc157edb7d8e9795916a14056f779882e5a50.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dfe0eb45210b62025786800cfe738871185e69bc --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/790dce92240bf4611549cbb710bcc157edb7d8e9795916a14056f779882e5a50.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef2fd9036499b380125b51da678d07544572fe82c0b75933d1df8d4f8c5d48be +size 21993 diff --git a/parse/train/BkMiWhR5K7/images/7c4d4e7b2f18ce110f16dc07e1f1f98ac3b662c73e810d31edba9cf1ccd25355.jpg b/parse/train/BkMiWhR5K7/images/7c4d4e7b2f18ce110f16dc07e1f1f98ac3b662c73e810d31edba9cf1ccd25355.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b233f22a2c50c88271cbb14fd6f19a7125578179 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/7c4d4e7b2f18ce110f16dc07e1f1f98ac3b662c73e810d31edba9cf1ccd25355.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9195469b226233605a80a38b70140784515fae1e2df264adf1f41d442ac16e56 +size 26898 diff --git a/parse/train/BkMiWhR5K7/images/7cb5b0643f637157d3213a58f98322773fb93af71454a2dd961f0d133e8b58ab.jpg b/parse/train/BkMiWhR5K7/images/7cb5b0643f637157d3213a58f98322773fb93af71454a2dd961f0d133e8b58ab.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f86d69e630c1c2f25a2f797d6571818c9b5f48d1 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/7cb5b0643f637157d3213a58f98322773fb93af71454a2dd961f0d133e8b58ab.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84e5bbca72d256730afa612e86a304670474a2e2865c80dac32c650b664852af +size 3678 diff --git a/parse/train/BkMiWhR5K7/images/7dbfcead8c881039f2c722f57194d5d3f72b4be86ecd1835b31be28670d02211.jpg b/parse/train/BkMiWhR5K7/images/7dbfcead8c881039f2c722f57194d5d3f72b4be86ecd1835b31be28670d02211.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c4c70b8bed5e4ba198554e437bce57fc08bb4607 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/7dbfcead8c881039f2c722f57194d5d3f72b4be86ecd1835b31be28670d02211.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b394c71fb35032b95961988dd424efccf1f0c0b57392d02cc229679ad8f0783e +size 5360 diff --git a/parse/train/BkMiWhR5K7/images/7e13ad9cfac045f656772116779cba22a75db51ff3c07d70559cbee4357d5480.jpg b/parse/train/BkMiWhR5K7/images/7e13ad9cfac045f656772116779cba22a75db51ff3c07d70559cbee4357d5480.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cedc89b67b9ab79d07e230db7ce335340c4fb32c --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/7e13ad9cfac045f656772116779cba22a75db51ff3c07d70559cbee4357d5480.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b935c569355594aa8f8de8abace518feb843fdccc79c33c936b715988e40cd68 +size 6084 diff --git a/parse/train/BkMiWhR5K7/images/85d158553e50bf61ca5a0c1d589eaf40bb29970c27c80a07869676c62a503653.jpg b/parse/train/BkMiWhR5K7/images/85d158553e50bf61ca5a0c1d589eaf40bb29970c27c80a07869676c62a503653.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6cf53ca8e61df0f76e5710b7ee364c223290564d --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/85d158553e50bf61ca5a0c1d589eaf40bb29970c27c80a07869676c62a503653.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9fa4efdc09270288d1662fe2e00744a593d489935982e2ad494c3bf10e67df37 +size 40292 diff --git a/parse/train/BkMiWhR5K7/images/8654310f4782271bb208e94239968669de6c04ac9eb4e17539c1e70e81516ec9.jpg b/parse/train/BkMiWhR5K7/images/8654310f4782271bb208e94239968669de6c04ac9eb4e17539c1e70e81516ec9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..61681c5b61648ce1ae6fdcdcda165decbe585235 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/8654310f4782271bb208e94239968669de6c04ac9eb4e17539c1e70e81516ec9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46aa501e25c30c343e2bcb0a03dd6b6a160d29945223e83c4830b7dadc86643a +size 3821 diff --git a/parse/train/BkMiWhR5K7/images/87bee4235646e942c9e3e5247800f8dcde60306a3b8849147fb444df10250347.jpg b/parse/train/BkMiWhR5K7/images/87bee4235646e942c9e3e5247800f8dcde60306a3b8849147fb444df10250347.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ebabda52b69fc2aa00c11ade975b0cd92bd917c7 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/87bee4235646e942c9e3e5247800f8dcde60306a3b8849147fb444df10250347.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:66daf25a0c13f32240734cdbfc5d9146a8d30020d5477db9df452178a6a5cb6f +size 53495 diff --git a/parse/train/BkMiWhR5K7/images/8a20ed2908fcc71aeeb5522db88b458d87105e915ee974bb285f16c4d1124ca2.jpg b/parse/train/BkMiWhR5K7/images/8a20ed2908fcc71aeeb5522db88b458d87105e915ee974bb285f16c4d1124ca2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3ac4619000131bd4ca5231fe80ed29034c81a828 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/8a20ed2908fcc71aeeb5522db88b458d87105e915ee974bb285f16c4d1124ca2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3867e7c631afb82884a218ac8341a1d724ea8dc990271f62688ab14ff4f0a50c +size 48321 diff --git a/parse/train/BkMiWhR5K7/images/91fc8b8711ab4007fba1bb807600ca8ceb3ac87726703de4bb6f7fa87a4dcc56.jpg b/parse/train/BkMiWhR5K7/images/91fc8b8711ab4007fba1bb807600ca8ceb3ac87726703de4bb6f7fa87a4dcc56.jpg new file mode 100644 index 0000000000000000000000000000000000000000..438cf45835ab36b329910fc1b2efadbbc0fcf4d8 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/91fc8b8711ab4007fba1bb807600ca8ceb3ac87726703de4bb6f7fa87a4dcc56.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:249af4e22c25c0e0517b3b36f59a125c7a4bec53346b8e9fdc5be6de9d94adbc +size 6949 diff --git a/parse/train/BkMiWhR5K7/images/960328f45ca90aded0bd7c68e0b32e9795b088bdaa7a4278e2065746de6cff85.jpg b/parse/train/BkMiWhR5K7/images/960328f45ca90aded0bd7c68e0b32e9795b088bdaa7a4278e2065746de6cff85.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bb8220e02539521feac61e62d1fb1a44a7a079b8 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/960328f45ca90aded0bd7c68e0b32e9795b088bdaa7a4278e2065746de6cff85.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4bf708a5f8aed9391a6657bb054359dfbfc0ecd378279d97ad559da10bd5a0a5 +size 5159 diff --git a/parse/train/BkMiWhR5K7/images/9bab0092265b6f5c1ce4ba90c040dddc5136b0b1d8015b27ed82b5f931ce74f0.jpg b/parse/train/BkMiWhR5K7/images/9bab0092265b6f5c1ce4ba90c040dddc5136b0b1d8015b27ed82b5f931ce74f0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9d197c6657f95d33448cff0229189901e4bacfa4 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/9bab0092265b6f5c1ce4ba90c040dddc5136b0b1d8015b27ed82b5f931ce74f0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a3e90384e1932c954b29642d4e51a74373d89574af43205bd8676b33eca49e7 +size 35288 diff --git a/parse/train/BkMiWhR5K7/images/9d2de50f9bd383a86c6cad3014bb2560e8d886ba604babf74d3ee18df0292b03.jpg b/parse/train/BkMiWhR5K7/images/9d2de50f9bd383a86c6cad3014bb2560e8d886ba604babf74d3ee18df0292b03.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c68018de9f3d2b78ff4e6c463d8c903ee61151b5 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/9d2de50f9bd383a86c6cad3014bb2560e8d886ba604babf74d3ee18df0292b03.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea1554b55194304b1e642720a87bed1ff40ad9d7cbfe3dd2a57d7eb311da202a +size 10835 diff --git a/parse/train/BkMiWhR5K7/images/9e62d1ac9ba9f8481e2055c68ce4bdb9586b8bd302d94a416e6c6948b89ec037.jpg b/parse/train/BkMiWhR5K7/images/9e62d1ac9ba9f8481e2055c68ce4bdb9586b8bd302d94a416e6c6948b89ec037.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a3117b5be09ee027afded63a5094699a4d4d1cb --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/9e62d1ac9ba9f8481e2055c68ce4bdb9586b8bd302d94a416e6c6948b89ec037.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e1e5f2c7d0b2af1b55ee3e0cd91be285c413f2ac3adcfa316f984446d632837 +size 14577 diff --git a/parse/train/BkMiWhR5K7/images/a92573aa67908bd10b0315bb2df3e6386d7b585c37159f2843ce088283b402d4.jpg b/parse/train/BkMiWhR5K7/images/a92573aa67908bd10b0315bb2df3e6386d7b585c37159f2843ce088283b402d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..58b78db219f562336a61361ab66adea8f0864b2f --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/a92573aa67908bd10b0315bb2df3e6386d7b585c37159f2843ce088283b402d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7f7bb5836b8730fe903d4936d8b2e981095584a6611bd58a0bc188bc8794dd5 +size 4887 diff --git a/parse/train/BkMiWhR5K7/images/ac940cbb89e720bb797295ad8e55502aac8cc4b0aa7ee68d41e81f5b205135f8.jpg b/parse/train/BkMiWhR5K7/images/ac940cbb89e720bb797295ad8e55502aac8cc4b0aa7ee68d41e81f5b205135f8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d3a72db14258da8ecc582bb1c6ebfea359e88303 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/ac940cbb89e720bb797295ad8e55502aac8cc4b0aa7ee68d41e81f5b205135f8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6fc552b31a6240b6856a512c96efb3949524544a7ef9915a1a8f2a57df301055 +size 6910 diff --git a/parse/train/BkMiWhR5K7/images/b2efd040ff2f25234b65bdeb5f780e4447f22237652ddd58656f934ec9bfd32f.jpg b/parse/train/BkMiWhR5K7/images/b2efd040ff2f25234b65bdeb5f780e4447f22237652ddd58656f934ec9bfd32f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93cb257eadbb5063f58462b3236cad9784594af1 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/b2efd040ff2f25234b65bdeb5f780e4447f22237652ddd58656f934ec9bfd32f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2651f19c24916a70e4c9e9c153539267b7658505cd0b606471c8894d5439565 +size 3148 diff --git a/parse/train/BkMiWhR5K7/images/b4b338415a61ea9d820b81697d6092f7866c6288450d33d4f8ef80bb230f3ea4.jpg b/parse/train/BkMiWhR5K7/images/b4b338415a61ea9d820b81697d6092f7866c6288450d33d4f8ef80bb230f3ea4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3459eb7cb5543721f32674d434fd504b639d8165 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/b4b338415a61ea9d820b81697d6092f7866c6288450d33d4f8ef80bb230f3ea4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7fd2c8478ba333ed3cdb0bb83640d834e3c35ea188d3327eccb7f19f61773e5a +size 37271 diff --git a/parse/train/BkMiWhR5K7/images/b93d131141ea03174d6d93e79bb2ea565d1ca4469781635975d5ebca91b11200.jpg b/parse/train/BkMiWhR5K7/images/b93d131141ea03174d6d93e79bb2ea565d1ca4469781635975d5ebca91b11200.jpg new file mode 100644 index 0000000000000000000000000000000000000000..79bd10a5aee9bdf8003b3558def93a8b8f997200 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/b93d131141ea03174d6d93e79bb2ea565d1ca4469781635975d5ebca91b11200.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a0d29f9f8813a63bcac9285c23854d4c446c922324f2c39a7067899c7258dcc8 +size 8182 diff --git a/parse/train/BkMiWhR5K7/images/bb42e1c8673d57e2b0724272b668ad171aa0357b7dbfe71d416a9971728da9e3.jpg b/parse/train/BkMiWhR5K7/images/bb42e1c8673d57e2b0724272b668ad171aa0357b7dbfe71d416a9971728da9e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..de70a643369fd7bb417dda85848e37d83a993a6c --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/bb42e1c8673d57e2b0724272b668ad171aa0357b7dbfe71d416a9971728da9e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a5c751bd8be8dae2df35dc1cdbeb24c1bfd63e6505f3723b2b0cedac15c997a +size 4172 diff --git a/parse/train/BkMiWhR5K7/images/bcae4637528921b9306cdf64000e85d669082709a21eff06dfea46e7826aaa67.jpg b/parse/train/BkMiWhR5K7/images/bcae4637528921b9306cdf64000e85d669082709a21eff06dfea46e7826aaa67.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fb57264a2182f099db8989561b4bd9ef8768cc52 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/bcae4637528921b9306cdf64000e85d669082709a21eff06dfea46e7826aaa67.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2de6079440d137180af16d93dd7369e5badb680dd79448cfec80940e98b761f8 +size 5095 diff --git a/parse/train/BkMiWhR5K7/images/bcc3313cbed6ba44eb213de14029a86ad0d44e0aab636611996e2776c7bc884e.jpg b/parse/train/BkMiWhR5K7/images/bcc3313cbed6ba44eb213de14029a86ad0d44e0aab636611996e2776c7bc884e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b71bd2c6e2d8e35f0d3d86a4d2a9a8295eb0ace9 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/bcc3313cbed6ba44eb213de14029a86ad0d44e0aab636611996e2776c7bc884e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8d5f4b05d4eac147603079639978d7948348fc905a15dd8bddb5d4e672215cc +size 4950 diff --git a/parse/train/BkMiWhR5K7/images/c271ea286ab448a1680b91a06ea20b068e500f581d3be8f1b0c5b7fb2d37ca88.jpg b/parse/train/BkMiWhR5K7/images/c271ea286ab448a1680b91a06ea20b068e500f581d3be8f1b0c5b7fb2d37ca88.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8bc410d2276fe67629f3a6e7f56f993d6dc7eeca --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/c271ea286ab448a1680b91a06ea20b068e500f581d3be8f1b0c5b7fb2d37ca88.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7abd0c5f35c2e36df0255553256ef4847bbd83b83aef5c2b274c78d64885d4cf +size 67086 diff --git a/parse/train/BkMiWhR5K7/images/c647d5a89b38c134b7db7303149427dc6e6fd1659ccf84f9ac1195c663c82b02.jpg b/parse/train/BkMiWhR5K7/images/c647d5a89b38c134b7db7303149427dc6e6fd1659ccf84f9ac1195c663c82b02.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93b2b04a5040af72ab83829d678052f0826364c5 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/c647d5a89b38c134b7db7303149427dc6e6fd1659ccf84f9ac1195c663c82b02.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b94e6fbb5a82fe6db0888b1f69d4529de6339eddf7c873af8a5664b0aaf954bf +size 4377 diff --git a/parse/train/BkMiWhR5K7/images/c71bee350e54ff77482d5e5c2123fa47d92bea255a30c497a2d8b6e83ad176f4.jpg b/parse/train/BkMiWhR5K7/images/c71bee350e54ff77482d5e5c2123fa47d92bea255a30c497a2d8b6e83ad176f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f8b6e64fd3eb5b74d7e1da16896e2c0b2253fc53 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/c71bee350e54ff77482d5e5c2123fa47d92bea255a30c497a2d8b6e83ad176f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d14388e90547d295a9a6b3f5ed3cadacedfe8b38b9d1bd73cf258c60313ae53c +size 34627 diff --git a/parse/train/BkMiWhR5K7/images/dba17539962d2519cbe43eac0e6a02d3af6f03077654827a4664d31e21a23ca7.jpg b/parse/train/BkMiWhR5K7/images/dba17539962d2519cbe43eac0e6a02d3af6f03077654827a4664d31e21a23ca7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3845cbda615bfb908f886b2d5788456fc4f749c8 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/dba17539962d2519cbe43eac0e6a02d3af6f03077654827a4664d31e21a23ca7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0ae99e7889a40562b4633e23ace4937c26808a35d25199ef50de8eda018a95f +size 54190 diff --git a/parse/train/BkMiWhR5K7/images/de945064141f9df299532cf285f40fcce245831e3616230428e17c6a41bdd44b.jpg b/parse/train/BkMiWhR5K7/images/de945064141f9df299532cf285f40fcce245831e3616230428e17c6a41bdd44b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..82c9214f9b62a90b437d53a54850f9a34858d5a6 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/de945064141f9df299532cf285f40fcce245831e3616230428e17c6a41bdd44b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd066cadfc94fecd02b067b6fbc7f7badfbc315306b0353c5d2d76cc251e9d58 +size 1967 diff --git a/parse/train/BkMiWhR5K7/images/dea14c3c9613083a7c89ddf3ccd0b6b7247dce5b5816ef5030e5daaa3d85d837.jpg b/parse/train/BkMiWhR5K7/images/dea14c3c9613083a7c89ddf3ccd0b6b7247dce5b5816ef5030e5daaa3d85d837.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e1cad85861a0591e00e668b506eef6349534d27 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/dea14c3c9613083a7c89ddf3ccd0b6b7247dce5b5816ef5030e5daaa3d85d837.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:769df228ad508cc2d13f3a3a02ad69d90c8a06da9f111357129b8524db31089e +size 6346 diff --git a/parse/train/BkMiWhR5K7/images/e142598a01f1fccc179c6ad8660b12f02eeb97a1f8064be62282bb445ff4ea6f.jpg b/parse/train/BkMiWhR5K7/images/e142598a01f1fccc179c6ad8660b12f02eeb97a1f8064be62282bb445ff4ea6f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ad03cc7f713cfa4acb5f4d8684b6740dba4af5ef --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/e142598a01f1fccc179c6ad8660b12f02eeb97a1f8064be62282bb445ff4ea6f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f92751430d89e4160a47b2acb8e86c77ca788e3719ad32f9116de89b04e0a61 +size 41550 diff --git a/parse/train/BkMiWhR5K7/images/e3d697228c28dd9ad40973e007fa6dc558e0e91770aae94b0fc88fec8a7a333f.jpg b/parse/train/BkMiWhR5K7/images/e3d697228c28dd9ad40973e007fa6dc558e0e91770aae94b0fc88fec8a7a333f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c3c51c1c280fbfff2fb3a3c8bc93bc4b05fb84aa --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/e3d697228c28dd9ad40973e007fa6dc558e0e91770aae94b0fc88fec8a7a333f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6a596ab8078308947f9d80d88a4bdd192cdb38370ae64c5947d27c34c81ce4e +size 32952 diff --git a/parse/train/BkMiWhR5K7/images/f19ec39a6dde6d08873b8c541bd55211e28a4c2c7aff11110bccb672d22f3707.jpg b/parse/train/BkMiWhR5K7/images/f19ec39a6dde6d08873b8c541bd55211e28a4c2c7aff11110bccb672d22f3707.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5707ef4141ec543288ad2407f320d139a494750b --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/f19ec39a6dde6d08873b8c541bd55211e28a4c2c7aff11110bccb672d22f3707.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63a57e984db6237b82dfafed209b8a1e79d59f8f8c81ac0487ab42a07c13833a +size 9091 diff --git a/parse/train/BkMiWhR5K7/images/f89f030e4bdfbb6f5f4ddce2c53646f3b757ba0f431e9fe2350b5807ac72576e.jpg b/parse/train/BkMiWhR5K7/images/f89f030e4bdfbb6f5f4ddce2c53646f3b757ba0f431e9fe2350b5807ac72576e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d01a1234ce0a88c3114724172d78cdde91aea2d9 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/f89f030e4bdfbb6f5f4ddce2c53646f3b757ba0f431e9fe2350b5807ac72576e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85ea239c93235fa952bcfeb942585c43560dd68d9ff39a13202b7f36b133d5fa +size 13604 diff --git a/parse/train/BkMiWhR5K7/images/fc5edb351081165fa9a9ae3df52f1659532205f4cf16adfb7b3c1fae69671bee.jpg b/parse/train/BkMiWhR5K7/images/fc5edb351081165fa9a9ae3df52f1659532205f4cf16adfb7b3c1fae69671bee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ff5c1f2f0b72a6169e3057db2541246dfbc53f0f --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/fc5edb351081165fa9a9ae3df52f1659532205f4cf16adfb7b3c1fae69671bee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6550eb7a24fb26e9150bfe87f553a419c9dc1d7a2501b7e6116f7fa0b1a3d8bd +size 37647 diff --git a/parse/train/BkMiWhR5K7/images/fc8c98def5043adde7c5d768e048c1ebe6812e927249ee9c887ae684b5faa9d1.jpg b/parse/train/BkMiWhR5K7/images/fc8c98def5043adde7c5d768e048c1ebe6812e927249ee9c887ae684b5faa9d1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e4dfbc2145f4579bd1ff97afadf766ea14ded443 --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/fc8c98def5043adde7c5d768e048c1ebe6812e927249ee9c887ae684b5faa9d1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8805d8d0d1bfae3ec1193811cf65bf7f769bd874cced829b620b9ddcab6f610f +size 5817 diff --git a/parse/train/BkMiWhR5K7/images/fce491d5ff18b89c2bd7c1290ca1eec4c2b2b0ba28919778be2736bddb1d7be4.jpg b/parse/train/BkMiWhR5K7/images/fce491d5ff18b89c2bd7c1290ca1eec4c2b2b0ba28919778be2736bddb1d7be4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9903412747a12f2ab1e1e30990fda9e5ccc8453f --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/fce491d5ff18b89c2bd7c1290ca1eec4c2b2b0ba28919778be2736bddb1d7be4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af4f32f621796690a91825d432517841cc82b54cae1d1c4cd8f81dd2bf3950e9 +size 7477 diff --git a/parse/train/BkMiWhR5K7/images/fdb7ee7f46bc2649a8f0d052db4e9dcb53bbce8c0946091699a167cd8450e3ec.jpg b/parse/train/BkMiWhR5K7/images/fdb7ee7f46bc2649a8f0d052db4e9dcb53bbce8c0946091699a167cd8450e3ec.jpg new file mode 100644 index 0000000000000000000000000000000000000000..450be5e95ae8d0775f81cdb5cd94d033c5aa142c --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/fdb7ee7f46bc2649a8f0d052db4e9dcb53bbce8c0946091699a167cd8450e3ec.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9d2d32d2b6cf600a50add0f7f81663055c5f91fc98379e52cc3e2e2decc02c99 +size 1478 diff --git a/parse/train/BkMiWhR5K7/images/ffbbe7c0edfcd928a20059262ccc40782473db20e14f213de8eecb39bd3a74c2.jpg b/parse/train/BkMiWhR5K7/images/ffbbe7c0edfcd928a20059262ccc40782473db20e14f213de8eecb39bd3a74c2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4009345ea329325a4dd83f1148c20a49a13f65dd --- /dev/null +++ b/parse/train/BkMiWhR5K7/images/ffbbe7c0edfcd928a20059262ccc40782473db20e14f213de8eecb39bd3a74c2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6f98e417a9a48b77d797df63f082f25a01156fe819a857e89a258c6f3801590 +size 5347 diff --git a/parse/train/EI2KOXKdnP/images/2c8c6654efa6c7166ff291d40e60d324d625c8579bcf986e46cbc1e3cf546ea9.jpg b/parse/train/EI2KOXKdnP/images/2c8c6654efa6c7166ff291d40e60d324d625c8579bcf986e46cbc1e3cf546ea9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6b4952764ff2d779d5c05d279d7eee563e39f0b1 --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/2c8c6654efa6c7166ff291d40e60d324d625c8579bcf986e46cbc1e3cf546ea9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb7e73e2a5ef202d9eb52f0bb7325dfd444c3e9a73f3b40d7f8cf1debc037c63 +size 12946 diff --git a/parse/train/EI2KOXKdnP/images/4d73edba72e7ee7ea5c72a0d41ac9476e577939a71c5c92b54605615df0e5d5c.jpg b/parse/train/EI2KOXKdnP/images/4d73edba72e7ee7ea5c72a0d41ac9476e577939a71c5c92b54605615df0e5d5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ae9fbf32777b027377219b010e4698e27a52a591 --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/4d73edba72e7ee7ea5c72a0d41ac9476e577939a71c5c92b54605615df0e5d5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8767370c39d7ea28c4f3b12035a73436ac8fc33827791a315bb605e0ff107ce3 +size 62053 diff --git a/parse/train/EI2KOXKdnP/images/520a8ab358be53d9597d292a724c5d33b52cd7a6c61bc1526554b9469dd172b6.jpg b/parse/train/EI2KOXKdnP/images/520a8ab358be53d9597d292a724c5d33b52cd7a6c61bc1526554b9469dd172b6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5f5c9ea6c41bb20d7241e3b9bd3d232536fb8f1b --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/520a8ab358be53d9597d292a724c5d33b52cd7a6c61bc1526554b9469dd172b6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:abf597c241350d84bce581260f33bbfaee1ee5d7bf36d59ac6a92038c385e93f +size 82318 diff --git a/parse/train/EI2KOXKdnP/images/6580eadc8dc26ea6d1683a42b8533e69114404d978ae2cc5853f9e28e8fb58f1.jpg b/parse/train/EI2KOXKdnP/images/6580eadc8dc26ea6d1683a42b8533e69114404d978ae2cc5853f9e28e8fb58f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8994985676003ef9583e073fdf05abd7e69b779a --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/6580eadc8dc26ea6d1683a42b8533e69114404d978ae2cc5853f9e28e8fb58f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c37d77083184c03fdc33af891127debf5c5360d09f27e1f4a48f096a26d4a570 +size 114062 diff --git a/parse/train/EI2KOXKdnP/images/759c8bb1b0f4ebf46f621f72824e0a25d409516ac63197eee33154d2126e49cb.jpg b/parse/train/EI2KOXKdnP/images/759c8bb1b0f4ebf46f621f72824e0a25d409516ac63197eee33154d2126e49cb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..12423444e5f205bd692f788cf065cf4bc5c27a02 --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/759c8bb1b0f4ebf46f621f72824e0a25d409516ac63197eee33154d2126e49cb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1504e19c5f2b68f4eb63a916a5528b5ca7f388eaa49b32f14243b0f9fc809d4 +size 61370 diff --git a/parse/train/EI2KOXKdnP/images/7e26b274eefa2449ead43edb17625410832b76d92162f3866cc372fdf2f626a0.jpg b/parse/train/EI2KOXKdnP/images/7e26b274eefa2449ead43edb17625410832b76d92162f3866cc372fdf2f626a0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7cf3818f7b5cfcc8951d98d603740e85eaefc29a --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/7e26b274eefa2449ead43edb17625410832b76d92162f3866cc372fdf2f626a0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5afebe6fcb5d7b70899906f6d7a1b26786f6774c1e08f1f4146fe3800ee5d186 +size 50266 diff --git a/parse/train/EI2KOXKdnP/images/7fb179a35692e5d7af2ac5139b1c7d513db42916c8578b35a36b1c4ad70aff29.jpg b/parse/train/EI2KOXKdnP/images/7fb179a35692e5d7af2ac5139b1c7d513db42916c8578b35a36b1c4ad70aff29.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d4a708a6914014752c970cb9e005b58e4e666374 --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/7fb179a35692e5d7af2ac5139b1c7d513db42916c8578b35a36b1c4ad70aff29.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26c6ac2898998130cf3bf54db4adcbe9de8fa1be6a6e6595bf4d4782e7e4a1c0 +size 167873 diff --git a/parse/train/EI2KOXKdnP/images/dcb60c2290be7785a89bed98d1d0cadb3b9e81795e4b5a5cc118b0fb971d4aa5.jpg b/parse/train/EI2KOXKdnP/images/dcb60c2290be7785a89bed98d1d0cadb3b9e81795e4b5a5cc118b0fb971d4aa5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..51c64c154d1af866b56edeafbef7eaec3cada352 --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/dcb60c2290be7785a89bed98d1d0cadb3b9e81795e4b5a5cc118b0fb971d4aa5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:669a0f8d8a7233533a792d505cb24cbe4ec371eb47d9998189a1825a92e90f7a +size 37119 diff --git a/parse/train/EI2KOXKdnP/images/e888ec18a5f1ffc5126a75c42bc7c88e5c7569217134a6f3ca6be02f3064c9b2.jpg b/parse/train/EI2KOXKdnP/images/e888ec18a5f1ffc5126a75c42bc7c88e5c7569217134a6f3ca6be02f3064c9b2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..02b69d01c3b80bdd89b8b7f225d4227b4208564e --- /dev/null +++ b/parse/train/EI2KOXKdnP/images/e888ec18a5f1ffc5126a75c42bc7c88e5c7569217134a6f3ca6be02f3064c9b2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:519644d3f2a763bcdda25d987ac9787ca9d77ae7c024c4642b06c19c7f396be3 +size 54903 diff --git a/parse/train/HJepXaVYDr/images/00bccf1850e4ef2ed56295f8f24bb4912746e1922754f812461f2edcdf33825d.jpg b/parse/train/HJepXaVYDr/images/00bccf1850e4ef2ed56295f8f24bb4912746e1922754f812461f2edcdf33825d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..070ee5aa0a26c2d058b0ff22e48307099689d077 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/00bccf1850e4ef2ed56295f8f24bb4912746e1922754f812461f2edcdf33825d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9423c7557e753f9ac71710c0dbce624888566d83efbd7ff3c8ded4c648aad689 +size 11856 diff --git a/parse/train/HJepXaVYDr/images/03ea3ee2f301c64dbc25b0b7f819f11fc74f9fa7b5cac9871976dd0d09464cde.jpg b/parse/train/HJepXaVYDr/images/03ea3ee2f301c64dbc25b0b7f819f11fc74f9fa7b5cac9871976dd0d09464cde.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fdc16d764ad61e42eac6049b65bb4b8644b3cafd --- /dev/null +++ b/parse/train/HJepXaVYDr/images/03ea3ee2f301c64dbc25b0b7f819f11fc74f9fa7b5cac9871976dd0d09464cde.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ff7976911b95ed47005811eefc09b057b0d7c9ccdae7529153dad4072d2fe74 +size 21064 diff --git a/parse/train/HJepXaVYDr/images/0b8a3d553f5b6daec578bb0ae1a48818152adf660bb8af685829362506af1745.jpg b/parse/train/HJepXaVYDr/images/0b8a3d553f5b6daec578bb0ae1a48818152adf660bb8af685829362506af1745.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a9dc3df14594a036aaffac6a8b146c17432c5588 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/0b8a3d553f5b6daec578bb0ae1a48818152adf660bb8af685829362506af1745.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:365234898ccd7cc4a503241b62dc3bfd5b5f90d012b9c9a6baec31779eb82ebf +size 21469 diff --git a/parse/train/HJepXaVYDr/images/0cd81253ac32a9325d12936b07a7275a2ebd3d3405a43ee34014f59f08f58717.jpg b/parse/train/HJepXaVYDr/images/0cd81253ac32a9325d12936b07a7275a2ebd3d3405a43ee34014f59f08f58717.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6e194621e5247ad01737d8d231f3709d9417c31a --- /dev/null +++ b/parse/train/HJepXaVYDr/images/0cd81253ac32a9325d12936b07a7275a2ebd3d3405a43ee34014f59f08f58717.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89f49f28d3e168c93c2e3e45406db3fb96c93ea12d225c671b53eb27335f2857 +size 10543 diff --git a/parse/train/HJepXaVYDr/images/113a6de40c0fbf466e53b4ef5fa1a7a58e138427ccb739d6d79bf203ef033110.jpg b/parse/train/HJepXaVYDr/images/113a6de40c0fbf466e53b4ef5fa1a7a58e138427ccb739d6d79bf203ef033110.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1e3dfefc6a09fd8d5b4d819a1e871a000e707f79 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/113a6de40c0fbf466e53b4ef5fa1a7a58e138427ccb739d6d79bf203ef033110.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26fc6460404f65031f99b6265940eefb926ba3cf6787d59e80143cba7ed9d01b +size 11545 diff --git a/parse/train/HJepXaVYDr/images/13efdc56da50ba2b2d875bec63c8a7d84e4d942cdae2ed1b10dd09eef75c8509.jpg b/parse/train/HJepXaVYDr/images/13efdc56da50ba2b2d875bec63c8a7d84e4d942cdae2ed1b10dd09eef75c8509.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c1fd91761fd7b3c32948f5f29e4e93899b54aac --- /dev/null +++ b/parse/train/HJepXaVYDr/images/13efdc56da50ba2b2d875bec63c8a7d84e4d942cdae2ed1b10dd09eef75c8509.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c502dfc2cae5e99b8a9e3dd412c55d3a90ecb14f77e9a57fe75ed6a08934bba +size 94170 diff --git a/parse/train/HJepXaVYDr/images/178050c2d7d8d10491cc2d21cd2b7c0756f5844bee7dff404b9d21171b8a8dc8.jpg b/parse/train/HJepXaVYDr/images/178050c2d7d8d10491cc2d21cd2b7c0756f5844bee7dff404b9d21171b8a8dc8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..585b23980f6a35e6fb700a55d382819b2d9b3db3 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/178050c2d7d8d10491cc2d21cd2b7c0756f5844bee7dff404b9d21171b8a8dc8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb6f26d3e0401ad283b02cb456ae3f6e2ecdb7ede98b47e0e8420ab95bb567d6 +size 8888 diff --git a/parse/train/HJepXaVYDr/images/1c3c6be671fb567a3bfefffea86be1b2451f8ead415b1ccc437f348cb2d02feb.jpg b/parse/train/HJepXaVYDr/images/1c3c6be671fb567a3bfefffea86be1b2451f8ead415b1ccc437f348cb2d02feb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..44bafd177fa987c6b7e4b7a42ffa8345221bec5e --- /dev/null +++ b/parse/train/HJepXaVYDr/images/1c3c6be671fb567a3bfefffea86be1b2451f8ead415b1ccc437f348cb2d02feb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6be47c989974b57a52bcd71914a8bfdc5f5b9565b595d06f7f83d9af2413efcc +size 17823 diff --git a/parse/train/HJepXaVYDr/images/1d7148f210a481ee3ba1df50baafbdb76e74f5d6e2829868713c478849c464ed.jpg b/parse/train/HJepXaVYDr/images/1d7148f210a481ee3ba1df50baafbdb76e74f5d6e2829868713c478849c464ed.jpg new file mode 100644 index 0000000000000000000000000000000000000000..87f1a55bd93ca42b0af9b8f90bb4564478ece0c9 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/1d7148f210a481ee3ba1df50baafbdb76e74f5d6e2829868713c478849c464ed.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5bfa3cd36446b34c7b9a34073a9fd4a12437b240f626d91e8e34af5cc3badf5 +size 14894 diff --git a/parse/train/HJepXaVYDr/images/1e6c53eaa96b226198da08639d51b42dab148e890fe4d6d16fd8a30229f86f84.jpg b/parse/train/HJepXaVYDr/images/1e6c53eaa96b226198da08639d51b42dab148e890fe4d6d16fd8a30229f86f84.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d70038ef9084a0c94aba46cdd0cc3e3c4b320a4e --- /dev/null +++ b/parse/train/HJepXaVYDr/images/1e6c53eaa96b226198da08639d51b42dab148e890fe4d6d16fd8a30229f86f84.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92936ef8bda7a1b1fa7b53e52b1919fa9146ba7ffd9d05ed463907c039b06bbb +size 7315 diff --git a/parse/train/HJepXaVYDr/images/2063edd58b26117cea5522f8b426f017544ef810739add48d597f2cbb3911b0f.jpg b/parse/train/HJepXaVYDr/images/2063edd58b26117cea5522f8b426f017544ef810739add48d597f2cbb3911b0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..24e4c545c24021b30d30bffa776a27f16deb6e3b --- /dev/null +++ b/parse/train/HJepXaVYDr/images/2063edd58b26117cea5522f8b426f017544ef810739add48d597f2cbb3911b0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b8b2adc4ee9f4ec6a83bdbefea3593609fc412c22b9db6b3b9a366be916ad7a +size 41624 diff --git a/parse/train/HJepXaVYDr/images/235c709771d53ba716d117959f0ee6b7ce3bfb75821a4c1c62953e24f4a6f3a9.jpg b/parse/train/HJepXaVYDr/images/235c709771d53ba716d117959f0ee6b7ce3bfb75821a4c1c62953e24f4a6f3a9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..50f28c81be06fcbea14507452410cc52c0f3353d --- /dev/null +++ b/parse/train/HJepXaVYDr/images/235c709771d53ba716d117959f0ee6b7ce3bfb75821a4c1c62953e24f4a6f3a9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7c1f5e8fb88867b08d0082837971971cf588232f506a15bb66c3f0e085904d2 +size 46842 diff --git a/parse/train/HJepXaVYDr/images/2493ad16e7cd2fe42772fd63d7fd59534500cf3875454c622e98b6427d845f14.jpg b/parse/train/HJepXaVYDr/images/2493ad16e7cd2fe42772fd63d7fd59534500cf3875454c622e98b6427d845f14.jpg new file mode 100644 index 0000000000000000000000000000000000000000..640b380d8b18254f9fdf92d61f09c92cfc8cec1b --- /dev/null +++ b/parse/train/HJepXaVYDr/images/2493ad16e7cd2fe42772fd63d7fd59534500cf3875454c622e98b6427d845f14.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e47ab5e5df591d1877ab3e759e03e5930e7702065b4a15bb14de26a67da7456b +size 77358 diff --git a/parse/train/HJepXaVYDr/images/264b1229523872bb695ee8f6693e7496f4af55236f73b56bc3ea87c5f2a084e9.jpg b/parse/train/HJepXaVYDr/images/264b1229523872bb695ee8f6693e7496f4af55236f73b56bc3ea87c5f2a084e9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a4a6f0fef9b01d99184e844258c162d34a578371 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/264b1229523872bb695ee8f6693e7496f4af55236f73b56bc3ea87c5f2a084e9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f143d49e84ebc2285b24633280460da3dfe6a4ceb4317161490b5990e1b59bbf +size 79618 diff --git a/parse/train/HJepXaVYDr/images/272827396ca0641bf853e7279a75a1a1e7129d9e4148c65028a5cba8ddc02f6c.jpg b/parse/train/HJepXaVYDr/images/272827396ca0641bf853e7279a75a1a1e7129d9e4148c65028a5cba8ddc02f6c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7046d1561dfb696084197f866369de43f9440b06 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/272827396ca0641bf853e7279a75a1a1e7129d9e4148c65028a5cba8ddc02f6c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe919babaa8a103c976f73a1a0330c977b737235311481adadb6d233ac727d24 +size 5465 diff --git a/parse/train/HJepXaVYDr/images/278fa1a4d2de7128851d850d49a22e43924037e55d386901933cfe9ee003206d.jpg b/parse/train/HJepXaVYDr/images/278fa1a4d2de7128851d850d49a22e43924037e55d386901933cfe9ee003206d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2af9520fea2009496166205e264702e23cb19c69 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/278fa1a4d2de7128851d850d49a22e43924037e55d386901933cfe9ee003206d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b60b8b8c091dd4d38654bd8762bc07e193800d6c670980f1a8c9616cffbc5826 +size 39884 diff --git a/parse/train/HJepXaVYDr/images/28e808fb21b534181aaf3011b1bed319afd8e533d63a07ca06c5f37e99ee8c60.jpg b/parse/train/HJepXaVYDr/images/28e808fb21b534181aaf3011b1bed319afd8e533d63a07ca06c5f37e99ee8c60.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e74c2ad7542f4ce180dd90a7ebb06f742b4f654e --- /dev/null +++ b/parse/train/HJepXaVYDr/images/28e808fb21b534181aaf3011b1bed319afd8e533d63a07ca06c5f37e99ee8c60.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:284675b61e341a3635a289ab515e5879d452a8d97ea2264b3a5c01ef5d105759 +size 11899 diff --git a/parse/train/HJepXaVYDr/images/2d008bdd66aea8d8d2c0caebc5118dc3ea02ae1aa4d4cb05f7455967d60b2304.jpg b/parse/train/HJepXaVYDr/images/2d008bdd66aea8d8d2c0caebc5118dc3ea02ae1aa4d4cb05f7455967d60b2304.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cc649702df59fa7ae08da8b3f240e866249d0528 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/2d008bdd66aea8d8d2c0caebc5118dc3ea02ae1aa4d4cb05f7455967d60b2304.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a45a547e0e113d26aa00e98478f285cad6eb2028d3cdc34cc1785dc3fbd5b351 +size 63623 diff --git a/parse/train/HJepXaVYDr/images/2fdeb38dd45d14a50f55b9e8d0995495a6db4f34b6894fec756009bb74dba557.jpg b/parse/train/HJepXaVYDr/images/2fdeb38dd45d14a50f55b9e8d0995495a6db4f34b6894fec756009bb74dba557.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d1aaa03a72be4b18e0190193071b77ead3db71d --- /dev/null +++ b/parse/train/HJepXaVYDr/images/2fdeb38dd45d14a50f55b9e8d0995495a6db4f34b6894fec756009bb74dba557.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5016997d326f8474d3ae20a297caf41c950151b2fda40e2f34e880a831269211 +size 19913 diff --git a/parse/train/HJepXaVYDr/images/313e9794d8774e7475aaaec15ab0ef61bba244715a3f3fbf09021a1c95e0d80a.jpg b/parse/train/HJepXaVYDr/images/313e9794d8774e7475aaaec15ab0ef61bba244715a3f3fbf09021a1c95e0d80a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..31b58c01ce2a819d0a4124233a0394441cafa28b --- /dev/null +++ b/parse/train/HJepXaVYDr/images/313e9794d8774e7475aaaec15ab0ef61bba244715a3f3fbf09021a1c95e0d80a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eea1082e3e026a38b4e8867c6ac8d6c824cb2e02be4d78e7fc33042687b602f8 +size 31038 diff --git a/parse/train/HJepXaVYDr/images/32126be47cbc8cfad2deae704e8281db60f29512363a515e5f427510d6771935.jpg b/parse/train/HJepXaVYDr/images/32126be47cbc8cfad2deae704e8281db60f29512363a515e5f427510d6771935.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cda099d540d8365f1b5fbc551255ae292d7fc899 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/32126be47cbc8cfad2deae704e8281db60f29512363a515e5f427510d6771935.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf381ce7380e17a27cc83653612bd30883aadedfe3a69ac2254d8ff065a75d45 +size 39066 diff --git a/parse/train/HJepXaVYDr/images/3a81c842c0109481ca98ef7bb6cfd893d63cc0a2a151fbc0afdb9bbd2148b777.jpg b/parse/train/HJepXaVYDr/images/3a81c842c0109481ca98ef7bb6cfd893d63cc0a2a151fbc0afdb9bbd2148b777.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d5a0c5dce4df2528f580b3ae12c48c3e04938428 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/3a81c842c0109481ca98ef7bb6cfd893d63cc0a2a151fbc0afdb9bbd2148b777.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc775ee4583bfe778415167ff453b3e64daa648458d4fb043eaeebe674ccd7a0 +size 12151 diff --git a/parse/train/HJepXaVYDr/images/3c334d07d93209d06135e4317587f33154337e7cc636662cd2eb9b30e27a8d74.jpg b/parse/train/HJepXaVYDr/images/3c334d07d93209d06135e4317587f33154337e7cc636662cd2eb9b30e27a8d74.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f8b1cd1f64c0d390aee9e9b7e7dcaf2e642b941e --- /dev/null +++ b/parse/train/HJepXaVYDr/images/3c334d07d93209d06135e4317587f33154337e7cc636662cd2eb9b30e27a8d74.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c8f7e4605a5dcf93ca7566f375fa67c3ae9a8662cf54ec0db4efb97eb02c1d8 +size 5086 diff --git a/parse/train/HJepXaVYDr/images/41501cb6f7c97396303d22ed8d509eb3726c6a97cd6b3faa51cffcef6befdf0f.jpg b/parse/train/HJepXaVYDr/images/41501cb6f7c97396303d22ed8d509eb3726c6a97cd6b3faa51cffcef6befdf0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bf1399b04f74971067b5708d4b79916db49fcb42 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/41501cb6f7c97396303d22ed8d509eb3726c6a97cd6b3faa51cffcef6befdf0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8104e2f7e937ce92c6d070796f214e20c5698cc275ac869c673b90c2c366db3 +size 12244 diff --git a/parse/train/HJepXaVYDr/images/4650d3d6467af532a3dc6c86c20ab6370c347511bc2e67bfd98c234a6943cc5c.jpg b/parse/train/HJepXaVYDr/images/4650d3d6467af532a3dc6c86c20ab6370c347511bc2e67bfd98c234a6943cc5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d1ddf00bd1a7ecb0b60cd01daf2a80178ba84636 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/4650d3d6467af532a3dc6c86c20ab6370c347511bc2e67bfd98c234a6943cc5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e6edeec0a4a79a00403ed6d37b1fe445e2b80eec551cf71e6ce7d45c998880b +size 50756 diff --git a/parse/train/HJepXaVYDr/images/46d7ebbd5b672c55d6c8bc77baa53755ea731e5d5bbaf71c06226bcba1d802b3.jpg b/parse/train/HJepXaVYDr/images/46d7ebbd5b672c55d6c8bc77baa53755ea731e5d5bbaf71c06226bcba1d802b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e88d8553b4eaedc8d0a9997433dad62289877cd7 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/46d7ebbd5b672c55d6c8bc77baa53755ea731e5d5bbaf71c06226bcba1d802b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06010e7338be5287535d5b9a349b5d3674ad217bee0d9068565140c7ed8db8f5 +size 55631 diff --git a/parse/train/HJepXaVYDr/images/484982d1f4e2b074bed83930b20a0b9a28ecb63cef88dc53040a9d93a05786d6.jpg b/parse/train/HJepXaVYDr/images/484982d1f4e2b074bed83930b20a0b9a28ecb63cef88dc53040a9d93a05786d6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d6a8956dde3e467d4c02c443310a14c15580f17b --- /dev/null +++ b/parse/train/HJepXaVYDr/images/484982d1f4e2b074bed83930b20a0b9a28ecb63cef88dc53040a9d93a05786d6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38d72052b43fb3fef4f2b7910ab122824a63e95c5177aaf9da668cf985656096 +size 11855 diff --git a/parse/train/HJepXaVYDr/images/4d5f91f21c1a8c9119cb2fd1f97b4af300d88010338786665d6a128bd27546a7.jpg b/parse/train/HJepXaVYDr/images/4d5f91f21c1a8c9119cb2fd1f97b4af300d88010338786665d6a128bd27546a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6f774e9f2f95e466c231fbf137c971ac31ee4996 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/4d5f91f21c1a8c9119cb2fd1f97b4af300d88010338786665d6a128bd27546a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84f131c9995a6655af59cd453dbc3408cf11d02fd360424d3dc6eaf41614644b +size 17724 diff --git a/parse/train/HJepXaVYDr/images/5063e94024f17fa373aceceb9598efd0d33a3ecd548feca65e5ec9133c686ccd.jpg b/parse/train/HJepXaVYDr/images/5063e94024f17fa373aceceb9598efd0d33a3ecd548feca65e5ec9133c686ccd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6f214c1b0d2fad343cd25c929fbfe07abe1a48ef --- /dev/null +++ b/parse/train/HJepXaVYDr/images/5063e94024f17fa373aceceb9598efd0d33a3ecd548feca65e5ec9133c686ccd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7a5541bb2587cb9761c6172354e1a332e71ea9bd5e2485f4e1a2149c8574bd6 +size 5018 diff --git a/parse/train/HJepXaVYDr/images/57e0d822ca438df5e062843ce525ee29be04898082dedc6918653673824da9ec.jpg b/parse/train/HJepXaVYDr/images/57e0d822ca438df5e062843ce525ee29be04898082dedc6918653673824da9ec.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4da5ff82e12ffd77b2f0448f1deab2f717cd570d --- /dev/null +++ b/parse/train/HJepXaVYDr/images/57e0d822ca438df5e062843ce525ee29be04898082dedc6918653673824da9ec.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7166bc7d3637f11d8b14311c00bf79e9b9880a006e5e6a141ac1edd6e2fe669 +size 10298 diff --git a/parse/train/HJepXaVYDr/images/59498540b2540717edf2795c5a54f48829a06f37dcf0de1b8a9507f71ff36c41.jpg b/parse/train/HJepXaVYDr/images/59498540b2540717edf2795c5a54f48829a06f37dcf0de1b8a9507f71ff36c41.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7df4bb87c047ac4b2754197d67b20703cea7dea0 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/59498540b2540717edf2795c5a54f48829a06f37dcf0de1b8a9507f71ff36c41.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39a4082474f249ad01642e08fd56a9d93f825e663134f9b85be22e9cc27cbd54 +size 8452 diff --git a/parse/train/HJepXaVYDr/images/5c13d362e5a13684060db0c9e87725a2e1e742aca545af87cf1bd6823c4cdd91.jpg b/parse/train/HJepXaVYDr/images/5c13d362e5a13684060db0c9e87725a2e1e742aca545af87cf1bd6823c4cdd91.jpg new file mode 100644 index 0000000000000000000000000000000000000000..57f0dbd547ef271eb4312305a1dfb32d37ad026f --- /dev/null +++ b/parse/train/HJepXaVYDr/images/5c13d362e5a13684060db0c9e87725a2e1e742aca545af87cf1bd6823c4cdd91.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4132a48f28c97760161b88c3d7a6b829400f7b037fd9f15822b88182a4e1dcd +size 32467 diff --git a/parse/train/HJepXaVYDr/images/62be499db5d850a5a7ef3cbb042057d3b348f97d9f14111f8ec257be76a99aa6.jpg b/parse/train/HJepXaVYDr/images/62be499db5d850a5a7ef3cbb042057d3b348f97d9f14111f8ec257be76a99aa6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c01eb593be55ae11a1d4adc1b57174957b975551 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/62be499db5d850a5a7ef3cbb042057d3b348f97d9f14111f8ec257be76a99aa6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:727edc4d17ae3f9c81cbf0640f757a3e7aa9cd4f11b6ee42054e66dd22ee3703 +size 7336 diff --git a/parse/train/HJepXaVYDr/images/62c4b3640cb4584cd7d0968ba66c311cbb56d025a2406c5dd7d54eee168555e2.jpg b/parse/train/HJepXaVYDr/images/62c4b3640cb4584cd7d0968ba66c311cbb56d025a2406c5dd7d54eee168555e2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..24e7bd920eda0efba7521178f497e57f4ca1a7cc --- /dev/null +++ b/parse/train/HJepXaVYDr/images/62c4b3640cb4584cd7d0968ba66c311cbb56d025a2406c5dd7d54eee168555e2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46eda48ac08240e75c3b4df5647b88fb70de1230730e577017b730c52b927898 +size 8673 diff --git a/parse/train/HJepXaVYDr/images/62df30c7234afada36b47e6993367c79f8c30fbc317db268255cdd550fcd36cf.jpg b/parse/train/HJepXaVYDr/images/62df30c7234afada36b47e6993367c79f8c30fbc317db268255cdd550fcd36cf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..94da4aa732529d27359efa212501b0bdcc47a954 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/62df30c7234afada36b47e6993367c79f8c30fbc317db268255cdd550fcd36cf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca723af5e91f668f7f1895d9881b358743d459ffc6f8ca1599f368404fcb10d2 +size 32407 diff --git a/parse/train/HJepXaVYDr/images/64cc085a65dfa98589295cdd2ba1593fb7ea8bc7531e5b83cc4689d74c80d95e.jpg b/parse/train/HJepXaVYDr/images/64cc085a65dfa98589295cdd2ba1593fb7ea8bc7531e5b83cc4689d74c80d95e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8e4638da384d4bb8bd9c5382b60a7798e68d21b4 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/64cc085a65dfa98589295cdd2ba1593fb7ea8bc7531e5b83cc4689d74c80d95e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ba3e955254760dca2acb6472fb580037302a53dd3b9e9571aef16ec994e10e8 +size 9070 diff --git a/parse/train/HJepXaVYDr/images/69a347657cfc2c8013c7ad88134f487c64e5f9b12f2194a14232741702ae17f8.jpg b/parse/train/HJepXaVYDr/images/69a347657cfc2c8013c7ad88134f487c64e5f9b12f2194a14232741702ae17f8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c631c6a88a59936d259e08304e27c5bca858f2f --- /dev/null +++ b/parse/train/HJepXaVYDr/images/69a347657cfc2c8013c7ad88134f487c64e5f9b12f2194a14232741702ae17f8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d183a329dbe24bfe1be3afc67f5a0a0e0fcb046d352d5140f8f4c158c98b201 +size 19528 diff --git a/parse/train/HJepXaVYDr/images/6e83d2439df843dc0a7c74cc3395f8169b953cd3fb5393690badca881e5ae336.jpg b/parse/train/HJepXaVYDr/images/6e83d2439df843dc0a7c74cc3395f8169b953cd3fb5393690badca881e5ae336.jpg new file mode 100644 index 0000000000000000000000000000000000000000..95e6cdcccd50733502888c3209024321a5a855b2 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/6e83d2439df843dc0a7c74cc3395f8169b953cd3fb5393690badca881e5ae336.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:687e87e1ea37db38818537580b3d3bfcc24e031b69d5e00ddb3deb0262b2086f +size 29713 diff --git a/parse/train/HJepXaVYDr/images/6f0403d2b1d27e6b947d8403f657e8727bd5ef121b4e5572ecd4dcafcb15832e.jpg b/parse/train/HJepXaVYDr/images/6f0403d2b1d27e6b947d8403f657e8727bd5ef121b4e5572ecd4dcafcb15832e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..25c8133e4e3ddc98f990e1aef85f65ea58d67fb6 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/6f0403d2b1d27e6b947d8403f657e8727bd5ef121b4e5572ecd4dcafcb15832e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7556e0cdd2338ae620c39384b62f10abc17c683210c4b1aaf7f73b324c9d2036 +size 5525 diff --git a/parse/train/HJepXaVYDr/images/715afae69228fc760e80024fbb27409bfe423dbd3c2ae93e9d8e6d873556cd0b.jpg b/parse/train/HJepXaVYDr/images/715afae69228fc760e80024fbb27409bfe423dbd3c2ae93e9d8e6d873556cd0b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a75ba41e9e73a69a01f4a0c184276e8cfe59c4e0 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/715afae69228fc760e80024fbb27409bfe423dbd3c2ae93e9d8e6d873556cd0b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59e5c7eee53fb46ab496b4e3b2512cfcd8f68e18850cb3c538c8a9221d02305f +size 5898 diff --git a/parse/train/HJepXaVYDr/images/75737c69c0147794b9dc29243ba63055af81a4f96a7dca1fe4289f2ab221192a.jpg b/parse/train/HJepXaVYDr/images/75737c69c0147794b9dc29243ba63055af81a4f96a7dca1fe4289f2ab221192a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ccf5f78ae091e6ec9f93f7858a0e908913ff5da6 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/75737c69c0147794b9dc29243ba63055af81a4f96a7dca1fe4289f2ab221192a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b21fe82cdbb065c6129efe0cdd466cb7e847c580f94e6f14e957a543189fa43d +size 14637 diff --git a/parse/train/HJepXaVYDr/images/77d1da590ee866663fd4ecab029d3353023d57172a0984f5ce5445d329ecfba3.jpg b/parse/train/HJepXaVYDr/images/77d1da590ee866663fd4ecab029d3353023d57172a0984f5ce5445d329ecfba3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..60efc6adc6000a2b49ea33d948f273d1a4b8e6f7 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/77d1da590ee866663fd4ecab029d3353023d57172a0984f5ce5445d329ecfba3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7665ce9917cff4612c6d0d168676dc83a63c1a5c6314df0ac32c21482a050df5 +size 7174 diff --git a/parse/train/HJepXaVYDr/images/7bdf9a190bdffdab956a7a7c4610f8f54d2d4986db39772db7461a70b926be6e.jpg b/parse/train/HJepXaVYDr/images/7bdf9a190bdffdab956a7a7c4610f8f54d2d4986db39772db7461a70b926be6e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dd09384a9dfbb308247b91dbd1adabf36e4052c8 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/7bdf9a190bdffdab956a7a7c4610f8f54d2d4986db39772db7461a70b926be6e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9f4c8c243b807682db7b76b08492354b788a84900c72fa0fac0f31e90f356e3 +size 9789 diff --git a/parse/train/HJepXaVYDr/images/7e7bfe5f2c289e87bd7f7a53299238ebc46ef68a4beaed5d9520db14b8624f9e.jpg b/parse/train/HJepXaVYDr/images/7e7bfe5f2c289e87bd7f7a53299238ebc46ef68a4beaed5d9520db14b8624f9e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b0366e9c4de4de85683213ac3400901697549717 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/7e7bfe5f2c289e87bd7f7a53299238ebc46ef68a4beaed5d9520db14b8624f9e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:376b69a38aa7e41ddda59702fef93fd326aae1dc2210bf5c6d45376dd29dc8ad +size 38168 diff --git a/parse/train/HJepXaVYDr/images/830be693bafa95212b748f603f4f3d28b79c1cd21466de4fcb70c6b34938fff3.jpg b/parse/train/HJepXaVYDr/images/830be693bafa95212b748f603f4f3d28b79c1cd21466de4fcb70c6b34938fff3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c0d839fc67c23ff53f33e0d22f13435574b4ada8 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/830be693bafa95212b748f603f4f3d28b79c1cd21466de4fcb70c6b34938fff3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a9f735e6aa689dbe11db4e3b792d5d0b8be89146c0776628e058ccce225ca440 +size 32690 diff --git a/parse/train/HJepXaVYDr/images/892d9dfbbfc4805cbfab5e273b9d612214a228bbbc4db8e78754ae64bed11a3a.jpg b/parse/train/HJepXaVYDr/images/892d9dfbbfc4805cbfab5e273b9d612214a228bbbc4db8e78754ae64bed11a3a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b9f1fab70f2c76820cc49a40298603e841f0def5 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/892d9dfbbfc4805cbfab5e273b9d612214a228bbbc4db8e78754ae64bed11a3a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6403d261fdca5f4f75165e96e7501b73bbc0bf3d6ca8152996143a1424db476 +size 20226 diff --git a/parse/train/HJepXaVYDr/images/8954a40ab0b1e9b288c5f747f397fd39bf37a1602d968e77857410aab8faafa4.jpg b/parse/train/HJepXaVYDr/images/8954a40ab0b1e9b288c5f747f397fd39bf37a1602d968e77857410aab8faafa4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d0c232d581aa9887ffb06da845febc0948ae86a --- /dev/null +++ b/parse/train/HJepXaVYDr/images/8954a40ab0b1e9b288c5f747f397fd39bf37a1602d968e77857410aab8faafa4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3056162d3d88411894ee4e5f29356d4e1685c2aa47d113936728e70cc5e5057 +size 11498 diff --git a/parse/train/HJepXaVYDr/images/8f24510c86a3460404dcfb65282664396903e2c8f76608a369d9168719bdeef3.jpg b/parse/train/HJepXaVYDr/images/8f24510c86a3460404dcfb65282664396903e2c8f76608a369d9168719bdeef3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..44c96126824a9c1bd9f7bcbe57648fc183ca6dac --- /dev/null +++ b/parse/train/HJepXaVYDr/images/8f24510c86a3460404dcfb65282664396903e2c8f76608a369d9168719bdeef3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e30dcb4e2fd9adda5e40c02624f871a5c9873e0321a31c6edea9a14916037a76 +size 7255 diff --git a/parse/train/HJepXaVYDr/images/927726ced42db88daa7a167a934e7d96dfb60b051c032174649c4f3e58da264f.jpg b/parse/train/HJepXaVYDr/images/927726ced42db88daa7a167a934e7d96dfb60b051c032174649c4f3e58da264f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9cb68815c50612bdd7601c9d7501a44d8c763fae --- /dev/null +++ b/parse/train/HJepXaVYDr/images/927726ced42db88daa7a167a934e7d96dfb60b051c032174649c4f3e58da264f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8cc9893d979a5ea2518875f407e1920e91a0a57a5cfc5191cccf3fb8793b91d5 +size 13329 diff --git a/parse/train/HJepXaVYDr/images/93ae4400c19cd34632cabc5a4de34c286cf43afc50d8abb18ef829ba9a770913.jpg b/parse/train/HJepXaVYDr/images/93ae4400c19cd34632cabc5a4de34c286cf43afc50d8abb18ef829ba9a770913.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5ccbab08e19b8fbfc25dd8a1641bba66473ee74 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/93ae4400c19cd34632cabc5a4de34c286cf43afc50d8abb18ef829ba9a770913.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2003d7b0af430c8581afa1f1cbadf7f4178e0da9c7989e373b41a975f1eb904 +size 10623 diff --git a/parse/train/HJepXaVYDr/images/960c03246ff5c849a0b3d6637af0953e93825b817676c7dbc47ff470b18be40f.jpg b/parse/train/HJepXaVYDr/images/960c03246ff5c849a0b3d6637af0953e93825b817676c7dbc47ff470b18be40f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d6396eac6bb89268c79b02c1d05deda45a87590f --- /dev/null +++ b/parse/train/HJepXaVYDr/images/960c03246ff5c849a0b3d6637af0953e93825b817676c7dbc47ff470b18be40f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:baf31a50f4cf6d8e28f07ac4ba8e523971af748f5041840154aaa66e3c30412e +size 7525 diff --git a/parse/train/HJepXaVYDr/images/982ac55ad19e5ef6e54878047181fec119e28273e1b5015ff001ce2c851e490c.jpg b/parse/train/HJepXaVYDr/images/982ac55ad19e5ef6e54878047181fec119e28273e1b5015ff001ce2c851e490c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e8b3774e25357a5d48406e691a9ba1071d528f81 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/982ac55ad19e5ef6e54878047181fec119e28273e1b5015ff001ce2c851e490c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2300d7888adf22f63ea7bcf718bac1f5bb38a04c6efcf82333dd644e13e4ef52 +size 12193 diff --git a/parse/train/HJepXaVYDr/images/985be6ac051023add5d3afe48877bc0693da6c9c16cc57827deff61567db0b03.jpg b/parse/train/HJepXaVYDr/images/985be6ac051023add5d3afe48877bc0693da6c9c16cc57827deff61567db0b03.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fa7de5d05b88f5934955a8645787ac93600fb7b4 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/985be6ac051023add5d3afe48877bc0693da6c9c16cc57827deff61567db0b03.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d26edaa4071f42a889b6a91eb95b517c54d76e357dbe900ad6f39de3b4afa231 +size 168723 diff --git a/parse/train/HJepXaVYDr/images/990f2ca627a3b6ed3022f0d70bce6bd108c27d4875487f3f0cb4529d9c662ba6.jpg b/parse/train/HJepXaVYDr/images/990f2ca627a3b6ed3022f0d70bce6bd108c27d4875487f3f0cb4529d9c662ba6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a7ac521f91a33af55f3d1e30dc2d1a218017501f --- /dev/null +++ b/parse/train/HJepXaVYDr/images/990f2ca627a3b6ed3022f0d70bce6bd108c27d4875487f3f0cb4529d9c662ba6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4365cf2ee66059337132d4b2dc0499cf2a014e205b3026c82d324b7fc6b295a +size 12921 diff --git a/parse/train/HJepXaVYDr/images/9ada70950cf746ae17c91554d1b585bbb9e026f32f09f1ec14a4fc7a06ca8288.jpg b/parse/train/HJepXaVYDr/images/9ada70950cf746ae17c91554d1b585bbb9e026f32f09f1ec14a4fc7a06ca8288.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e671cb713b57b3079dbea6d98b8bececf309852c --- /dev/null +++ b/parse/train/HJepXaVYDr/images/9ada70950cf746ae17c91554d1b585bbb9e026f32f09f1ec14a4fc7a06ca8288.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9c1712456878c5ffa2f9d16bbd82744659c9a19eebd8dec77d28cb495e68656 +size 193569 diff --git a/parse/train/HJepXaVYDr/images/a0b46ef2419da6129a24851193966a173c4ec62e97a085d8931a95ac5a6d1ba2.jpg b/parse/train/HJepXaVYDr/images/a0b46ef2419da6129a24851193966a173c4ec62e97a085d8931a95ac5a6d1ba2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..473d2c4b9556c130bf3dbfa44c472adf7be1b670 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/a0b46ef2419da6129a24851193966a173c4ec62e97a085d8931a95ac5a6d1ba2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15eed03b52de4d281f14424b7895c6b8b1f5c069bc4c4b4738f04c8129c5d703 +size 16514 diff --git a/parse/train/HJepXaVYDr/images/a35378749b6d1c306a375a9c817679c64d8ac1677dceccd92ece492d7d985c67.jpg b/parse/train/HJepXaVYDr/images/a35378749b6d1c306a375a9c817679c64d8ac1677dceccd92ece492d7d985c67.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2bf2454068887a82bf7423242e592021a178b139 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/a35378749b6d1c306a375a9c817679c64d8ac1677dceccd92ece492d7d985c67.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:999b2f7144ab1e1a8d33453c021eaa54e3fa99894f1aff94717f349558b0331f +size 8240 diff --git a/parse/train/HJepXaVYDr/images/a3e3a9542c0ec5dbf36ff54b8e2be97f9d2100780c4d98a64ef97b594f035967.jpg b/parse/train/HJepXaVYDr/images/a3e3a9542c0ec5dbf36ff54b8e2be97f9d2100780c4d98a64ef97b594f035967.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8d47c97735763f13487a282879dda38d249c942e --- /dev/null +++ b/parse/train/HJepXaVYDr/images/a3e3a9542c0ec5dbf36ff54b8e2be97f9d2100780c4d98a64ef97b594f035967.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:282307b2e1bc3dbac66977681b58bd2dfbadc14bf807b342fbeaaa70318a59cf +size 31452 diff --git a/parse/train/HJepXaVYDr/images/a3f54971f81204b692ece9f3a120d81925c2d0e57a36761f410f42747f8e91eb.jpg b/parse/train/HJepXaVYDr/images/a3f54971f81204b692ece9f3a120d81925c2d0e57a36761f410f42747f8e91eb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..07de498ffa84c39aff32028227c82a59ed21e636 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/a3f54971f81204b692ece9f3a120d81925c2d0e57a36761f410f42747f8e91eb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0a7a00c54ebb2c62a247ab2628434d2ebaff6ce101b45ee8e435c0892fd166e +size 14905 diff --git a/parse/train/HJepXaVYDr/images/a54594a22a9878549394fba42f0f13bafb645303a832e4d00ab1100746d8fd93.jpg b/parse/train/HJepXaVYDr/images/a54594a22a9878549394fba42f0f13bafb645303a832e4d00ab1100746d8fd93.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d99d25666279798c784e4879f75225acc0804871 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/a54594a22a9878549394fba42f0f13bafb645303a832e4d00ab1100746d8fd93.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:202cb3b516e51b956be09242ca66a4c20f8d0ba218b3d8320cf8c1d2f76ca990 +size 11494 diff --git a/parse/train/HJepXaVYDr/images/a68dde3d930cffdda7b3f1e241e7c0dd44bcd4928c5e99cec517748f79203dcb.jpg b/parse/train/HJepXaVYDr/images/a68dde3d930cffdda7b3f1e241e7c0dd44bcd4928c5e99cec517748f79203dcb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..906391d281af8988651fedf2997be2d3fa5a4674 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/a68dde3d930cffdda7b3f1e241e7c0dd44bcd4928c5e99cec517748f79203dcb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:604f4f6d8677ff7eb02af684d7c7085c6d4241b2f9af7ea27e38cbf5c4b8f330 +size 15607 diff --git a/parse/train/HJepXaVYDr/images/a74f871e4087035d43af3f37731fa36a140786e348b74c8e76348466c7ab6832.jpg b/parse/train/HJepXaVYDr/images/a74f871e4087035d43af3f37731fa36a140786e348b74c8e76348466c7ab6832.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e400e969b47748f30230b7416ec75e31aab3b29e --- /dev/null +++ b/parse/train/HJepXaVYDr/images/a74f871e4087035d43af3f37731fa36a140786e348b74c8e76348466c7ab6832.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1034bac45da3ec86b95c073ce7e08897eef4c1be85b17eb1149bb71c3091876b +size 15887 diff --git a/parse/train/HJepXaVYDr/images/abb707c1cb74766b0dc500f779ac92b175d09d3308a5e96c247bd748d1263f96.jpg b/parse/train/HJepXaVYDr/images/abb707c1cb74766b0dc500f779ac92b175d09d3308a5e96c247bd748d1263f96.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cbb1c764e0b263b8a46832cf7d464aae391c08df --- /dev/null +++ b/parse/train/HJepXaVYDr/images/abb707c1cb74766b0dc500f779ac92b175d09d3308a5e96c247bd748d1263f96.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6f86b304d303b0c0aa24b108ebd3d94a5344af2e74ccaf1249ae5039ea6348d +size 12230 diff --git a/parse/train/HJepXaVYDr/images/adc75b6d88daa12120dd336568700d36fb3675575d3eb2dbd567e08aaeb479d8.jpg b/parse/train/HJepXaVYDr/images/adc75b6d88daa12120dd336568700d36fb3675575d3eb2dbd567e08aaeb479d8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ce8928731a91c623cdb907ceb5a003fa4e4fe0db --- /dev/null +++ b/parse/train/HJepXaVYDr/images/adc75b6d88daa12120dd336568700d36fb3675575d3eb2dbd567e08aaeb479d8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18d13866886b9419e405fdd408e80dcbe692807cdb187de3565131b2b2185fbc +size 18163 diff --git a/parse/train/HJepXaVYDr/images/b19748070484cdd61dafedb762467c319d79de1e6a1d68dbc766b5f4ae85b90e.jpg b/parse/train/HJepXaVYDr/images/b19748070484cdd61dafedb762467c319d79de1e6a1d68dbc766b5f4ae85b90e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..26159f4d331ce7e92d3d63192feae614feeeeaa4 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/b19748070484cdd61dafedb762467c319d79de1e6a1d68dbc766b5f4ae85b90e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:965181e1187d89477440c70df9dd3cec1b2437baa28b55a6bb68d2f1509335e1 +size 6704 diff --git a/parse/train/HJepXaVYDr/images/b2bd7bc22d87ecc8c406046d99c48a69f7c938457bd8bd0cf4c8cb42e57742a3.jpg b/parse/train/HJepXaVYDr/images/b2bd7bc22d87ecc8c406046d99c48a69f7c938457bd8bd0cf4c8cb42e57742a3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c4a1061ab8b297961b43d74c23e352e72e32f878 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/b2bd7bc22d87ecc8c406046d99c48a69f7c938457bd8bd0cf4c8cb42e57742a3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a98f328e0b237b5e424e4e78571eb37ebcf6e27477a03280181ec7872f21b2db +size 9870 diff --git a/parse/train/HJepXaVYDr/images/b5e94966c767b8205facafa7b02c73b4c3519052cf96cf511d75fc6c0da30f7a.jpg b/parse/train/HJepXaVYDr/images/b5e94966c767b8205facafa7b02c73b4c3519052cf96cf511d75fc6c0da30f7a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6bfcbce6453cd5f447d40f20d0663f7555455cb2 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/b5e94966c767b8205facafa7b02c73b4c3519052cf96cf511d75fc6c0da30f7a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e1753f9508affa8f54aaff36029c74a34cda3e696a8e31db1ef2dd8cfc6248e +size 40957 diff --git a/parse/train/HJepXaVYDr/images/ba131224053c5c8087ea5971dc06528e6140341d7e80e0e5a9bfcb10c7f0f573.jpg b/parse/train/HJepXaVYDr/images/ba131224053c5c8087ea5971dc06528e6140341d7e80e0e5a9bfcb10c7f0f573.jpg new file mode 100644 index 0000000000000000000000000000000000000000..71876f9e9d2852249b484e44b26fb5eefe15c023 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/ba131224053c5c8087ea5971dc06528e6140341d7e80e0e5a9bfcb10c7f0f573.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0352c0570664f60221cbe577dfc9f24d26f924895acf5f7bb4d27cf086719101 +size 14532 diff --git a/parse/train/HJepXaVYDr/images/bb8667a6f2b34e03036dc6bb7180c1f850a67c6559a40cf900211991224fe49c.jpg b/parse/train/HJepXaVYDr/images/bb8667a6f2b34e03036dc6bb7180c1f850a67c6559a40cf900211991224fe49c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2ff4a12a51f4aca060df5691264006d97c67d134 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/bb8667a6f2b34e03036dc6bb7180c1f850a67c6559a40cf900211991224fe49c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2733afc772d30ccdb21c2f7a4c9bf543c3eca76d2005014a94255f99730c0d0 +size 7809 diff --git a/parse/train/HJepXaVYDr/images/ca705af3a3a934c7f20b6417c78fef87c4b620289c5fe2a09599c28f5a6a06a5.jpg b/parse/train/HJepXaVYDr/images/ca705af3a3a934c7f20b6417c78fef87c4b620289c5fe2a09599c28f5a6a06a5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f8b1cd1f64c0d390aee9e9b7e7dcaf2e642b941e --- /dev/null +++ b/parse/train/HJepXaVYDr/images/ca705af3a3a934c7f20b6417c78fef87c4b620289c5fe2a09599c28f5a6a06a5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c8f7e4605a5dcf93ca7566f375fa67c3ae9a8662cf54ec0db4efb97eb02c1d8 +size 5086 diff --git a/parse/train/HJepXaVYDr/images/cc93f709d056130ea521baf4c035ffd8105557fcc8227ab80bcc838d02d507f1.jpg b/parse/train/HJepXaVYDr/images/cc93f709d056130ea521baf4c035ffd8105557fcc8227ab80bcc838d02d507f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..28ec06b55f4056eb807eb705918381ecee86597d --- /dev/null +++ b/parse/train/HJepXaVYDr/images/cc93f709d056130ea521baf4c035ffd8105557fcc8227ab80bcc838d02d507f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3021fa274c2a19fbd00efd0c224b79d8297a0a004a2e2d6a7e05fb32394c476b +size 9416 diff --git a/parse/train/HJepXaVYDr/images/cef4f763c42d79a5a13f27faa32e74392d7ac1017faddc1ad90844df6c3a4e50.jpg b/parse/train/HJepXaVYDr/images/cef4f763c42d79a5a13f27faa32e74392d7ac1017faddc1ad90844df6c3a4e50.jpg new file mode 100644 index 0000000000000000000000000000000000000000..618d7e78eb41cbbc766bb4fb7a615791411e6268 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/cef4f763c42d79a5a13f27faa32e74392d7ac1017faddc1ad90844df6c3a4e50.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c04d0ac4383e12a59b7b1f0451a539a9f1f322df7ba0acf62f2622f8c56e3c87 +size 54483 diff --git a/parse/train/HJepXaVYDr/images/d0da19463097becb2b3c12cf6cf473bd18ef4917d0bc291d55c38b893b5d349a.jpg b/parse/train/HJepXaVYDr/images/d0da19463097becb2b3c12cf6cf473bd18ef4917d0bc291d55c38b893b5d349a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db07171aa5643ace8f26f6fe7f8e69cc1411aa9e --- /dev/null +++ b/parse/train/HJepXaVYDr/images/d0da19463097becb2b3c12cf6cf473bd18ef4917d0bc291d55c38b893b5d349a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3803039a784aaedd5561f78cc61a77ed3e11e745781ea0fc860cadd6d5fcd831 +size 28400 diff --git a/parse/train/HJepXaVYDr/images/d17a73638757336e98aa76c53e14fed7c8782157d810c7452998a7c5b4b74e4f.jpg b/parse/train/HJepXaVYDr/images/d17a73638757336e98aa76c53e14fed7c8782157d810c7452998a7c5b4b74e4f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a53d896d66fa2e14f96e8d73f8009c116ec3c9e4 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/d17a73638757336e98aa76c53e14fed7c8782157d810c7452998a7c5b4b74e4f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9d5c4a3dec1183990ad3d13712754bd002629ee5c9bfba1ea6268f8cf48fd04 +size 10195 diff --git a/parse/train/HJepXaVYDr/images/d1b00bd1ec18d887d03d1c7b9a649dfa5c6b8a7b84304ba6fe9c18ede6078115.jpg b/parse/train/HJepXaVYDr/images/d1b00bd1ec18d887d03d1c7b9a649dfa5c6b8a7b84304ba6fe9c18ede6078115.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dc257a68ff590f58f58ef0b7dfb62e25892cc978 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/d1b00bd1ec18d887d03d1c7b9a649dfa5c6b8a7b84304ba6fe9c18ede6078115.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d031d9a8bd57f4e9c525c3fc0f17665f330a642752e28c93aed241b5bc42ec0 +size 68708 diff --git a/parse/train/HJepXaVYDr/images/d52b2dbd0705f1370b49151ed9baa5cae82ec18d8abb6ed8666d23d8e6206ad9.jpg b/parse/train/HJepXaVYDr/images/d52b2dbd0705f1370b49151ed9baa5cae82ec18d8abb6ed8666d23d8e6206ad9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..17618eca11a58b35b0dbdce42577479d83beeb27 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/d52b2dbd0705f1370b49151ed9baa5cae82ec18d8abb6ed8666d23d8e6206ad9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aba93f4a7502ac77b5d590ec36309f7fa3cd7e8935f4acf87b6005cb4b0015a2 +size 29418 diff --git a/parse/train/HJepXaVYDr/images/d92e7c527ebfbf490be0ecd9191d86f8b11720f5f0591bdda9d9bb9b6c36fdd5.jpg b/parse/train/HJepXaVYDr/images/d92e7c527ebfbf490be0ecd9191d86f8b11720f5f0591bdda9d9bb9b6c36fdd5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7ea8c48ebd0671bd7dd66460cdf620762f6d1b69 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/d92e7c527ebfbf490be0ecd9191d86f8b11720f5f0591bdda9d9bb9b6c36fdd5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47c5a15a39bd93edb0fc5fd596fc06e9ae62f8de53e713d198d2052f86ad650e +size 118658 diff --git a/parse/train/HJepXaVYDr/images/dd9cdb18b3beb309c59a15aa79d269f5851e48d5ba4325ccdd2e3e42d9f56de8.jpg b/parse/train/HJepXaVYDr/images/dd9cdb18b3beb309c59a15aa79d269f5851e48d5ba4325ccdd2e3e42d9f56de8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a70b22ab0a57e5fdf0861b845e52b9ce26df8737 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/dd9cdb18b3beb309c59a15aa79d269f5851e48d5ba4325ccdd2e3e42d9f56de8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afb2ca20246f603090bb61758bab1547c58fe39df3e81bcf8616d8e2c71f967e +size 14202 diff --git a/parse/train/HJepXaVYDr/images/ddb820f13d9a345f6c22b2c6f079b0e76ae3fbb9fa4f3da40044af7a483867d6.jpg b/parse/train/HJepXaVYDr/images/ddb820f13d9a345f6c22b2c6f079b0e76ae3fbb9fa4f3da40044af7a483867d6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..94b0f1fc3ca519bca8c488ecf968ebfcae202508 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/ddb820f13d9a345f6c22b2c6f079b0e76ae3fbb9fa4f3da40044af7a483867d6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c02fca6bb57e2a0d2fc84c8dbdf1dbae8a5e2e63d42327d0fed5c466b1140ba2 +size 25314 diff --git a/parse/train/HJepXaVYDr/images/e244b562a506284250edb2d849854a5adc6eea0344d84d48e830b7b36d5480c2.jpg b/parse/train/HJepXaVYDr/images/e244b562a506284250edb2d849854a5adc6eea0344d84d48e830b7b36d5480c2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e0f201c6b77cb78b2323409e19060a5b1da2aa02 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/e244b562a506284250edb2d849854a5adc6eea0344d84d48e830b7b36d5480c2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf10c77a297e0853c985e0631cdb1948140ff7a67c80a07f24d6749ca5207e55 +size 11836 diff --git a/parse/train/HJepXaVYDr/images/e551cb1be8574e4d0ec5a6025d2999657dbcddfbf6470a50e66d9e2f5f0d4279.jpg b/parse/train/HJepXaVYDr/images/e551cb1be8574e4d0ec5a6025d2999657dbcddfbf6470a50e66d9e2f5f0d4279.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5670f8a71b197c9dc34b895c1eef4a436152d739 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/e551cb1be8574e4d0ec5a6025d2999657dbcddfbf6470a50e66d9e2f5f0d4279.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c42b2cb76bbe606059de3dc64a4c8acbb9f38ca16d7b25654fca6ccf8446d1f3 +size 11332 diff --git a/parse/train/HJepXaVYDr/images/e79c40714f1ee19d2aa5ec649e4c57b41f8614c5586a0ed4fe717432e067567b.jpg b/parse/train/HJepXaVYDr/images/e79c40714f1ee19d2aa5ec649e4c57b41f8614c5586a0ed4fe717432e067567b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ab0846610a8e9f5ac503670d81466030c0e1c38c --- /dev/null +++ b/parse/train/HJepXaVYDr/images/e79c40714f1ee19d2aa5ec649e4c57b41f8614c5586a0ed4fe717432e067567b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06f4b906aab64864dac587dde4311feaea49631f9f7686c3b61feb52283e8766 +size 5193 diff --git a/parse/train/HJepXaVYDr/images/ebae8269df04f1addf4a673e76fe88b849655d8e17f5fe4d934ea9cc15a43c78.jpg b/parse/train/HJepXaVYDr/images/ebae8269df04f1addf4a673e76fe88b849655d8e17f5fe4d934ea9cc15a43c78.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3bf1601b09159c6fa488832cf7c93980dce148c5 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/ebae8269df04f1addf4a673e76fe88b849655d8e17f5fe4d934ea9cc15a43c78.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b8d288c438db62c19f4a4fc49f2090442068c75d4d7a58af17c8dd6ad3d1bdf +size 22815 diff --git a/parse/train/HJepXaVYDr/images/ecd99e982a7e67217e06b89db56a613c7c0068a5b36ea0130298faf5088470fd.jpg b/parse/train/HJepXaVYDr/images/ecd99e982a7e67217e06b89db56a613c7c0068a5b36ea0130298faf5088470fd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b3b404517be92dddbc6cca48fa4a205448bffad --- /dev/null +++ b/parse/train/HJepXaVYDr/images/ecd99e982a7e67217e06b89db56a613c7c0068a5b36ea0130298faf5088470fd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b2a35bb6ad32df6268cbf76a6199ab2f5b68cf43f7becafca2cf03dfe185c2c +size 11962 diff --git a/parse/train/HJepXaVYDr/images/f4702576c267e50cc6c9894b234b658c393b81179e62c02c55ff1863884f0e05.jpg b/parse/train/HJepXaVYDr/images/f4702576c267e50cc6c9894b234b658c393b81179e62c02c55ff1863884f0e05.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9885945fd9ceba478c23bd65aea05449e2cbdfbc --- /dev/null +++ b/parse/train/HJepXaVYDr/images/f4702576c267e50cc6c9894b234b658c393b81179e62c02c55ff1863884f0e05.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7420a7d4e62b8e427597ad9d552ac591a0a3dc9a498071afb8eee1d84edc2a75 +size 15527 diff --git a/parse/train/HJepXaVYDr/images/f6649212459266d15ffae4d6db182deb5d73c8c5582fce8f28349a824993f0e9.jpg b/parse/train/HJepXaVYDr/images/f6649212459266d15ffae4d6db182deb5d73c8c5582fce8f28349a824993f0e9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..53fe7946db5002100048c67e2d3e8e5f3f3bee82 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/f6649212459266d15ffae4d6db182deb5d73c8c5582fce8f28349a824993f0e9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c690f8773eb949ea2ad1062a1fe7d5457945fe63df6d860b26706cc9ec9df15e +size 7819 diff --git a/parse/train/HJepXaVYDr/images/ffe5b64c57a23999f3fcac9d5a5ed9ea09aa3179b6107a121bbc9855ee4ad15e.jpg b/parse/train/HJepXaVYDr/images/ffe5b64c57a23999f3fcac9d5a5ed9ea09aa3179b6107a121bbc9855ee4ad15e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f5770e8f59dc23ef0e3634b89be8de93e466de9 --- /dev/null +++ b/parse/train/HJepXaVYDr/images/ffe5b64c57a23999f3fcac9d5a5ed9ea09aa3179b6107a121bbc9855ee4ad15e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d484c578f631d660387eb5e30612e4b2265fc1705c98980207550fdd933a25e +size 11623 diff --git a/parse/train/HJgXCV9xx/images/05ed67df80480876ac80fe82b973710f287b8b6e8af17249818d18d05e2d04c4.jpg b/parse/train/HJgXCV9xx/images/05ed67df80480876ac80fe82b973710f287b8b6e8af17249818d18d05e2d04c4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e4d2da38188116d17057916e8db1e3e0f43d3dda --- /dev/null +++ b/parse/train/HJgXCV9xx/images/05ed67df80480876ac80fe82b973710f287b8b6e8af17249818d18d05e2d04c4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d2a6dea322861e7a0e5750d2cf5403c3221425939e27b2d5d6e58454b0bab66 +size 139827 diff --git a/parse/train/HJgXCV9xx/images/074cfd0cfbfcd6081ffa9520c57a20b124eb707cf3c3bdb187d9bd68cb5ec126.jpg b/parse/train/HJgXCV9xx/images/074cfd0cfbfcd6081ffa9520c57a20b124eb707cf3c3bdb187d9bd68cb5ec126.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5032e6ab75216d3b1da26ba5da221f3c662e2096 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/074cfd0cfbfcd6081ffa9520c57a20b124eb707cf3c3bdb187d9bd68cb5ec126.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3ce523166a00a0ad361ef784b2e3a3d880cf5206bb9466e5b319faaf84b2d7bd +size 244422 diff --git a/parse/train/HJgXCV9xx/images/110eb7234d89a9b70f0bc1e75cae8365602646fd485a3ddf5c016021f0267bd7.jpg b/parse/train/HJgXCV9xx/images/110eb7234d89a9b70f0bc1e75cae8365602646fd485a3ddf5c016021f0267bd7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1aa102cecad203ae7b417fde48f9ec4438536a43 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/110eb7234d89a9b70f0bc1e75cae8365602646fd485a3ddf5c016021f0267bd7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:397e41d39a2efca37462981f11ac37dcdfa2fa050b718bda6641ee652e06a629 +size 6999 diff --git a/parse/train/HJgXCV9xx/images/14feac19494c09716c71796a567833aca13ef83b35e0c92e5f1ecc70edecdb33.jpg b/parse/train/HJgXCV9xx/images/14feac19494c09716c71796a567833aca13ef83b35e0c92e5f1ecc70edecdb33.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1101b5f84c5389a4ccb5eb45231d60f6f2a9ceb0 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/14feac19494c09716c71796a567833aca13ef83b35e0c92e5f1ecc70edecdb33.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a7d72967aa756ed36d8feadcbab5d13bbd2564b73d2f44fddbd1c26c24fcddf +size 3582 diff --git a/parse/train/HJgXCV9xx/images/281442d4e457affbd85fab31d4e43ffabee513e55e0c7745e9963c90e38b9dfd.jpg b/parse/train/HJgXCV9xx/images/281442d4e457affbd85fab31d4e43ffabee513e55e0c7745e9963c90e38b9dfd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..52a7063a16a5949798707c40a45dacad3b880b60 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/281442d4e457affbd85fab31d4e43ffabee513e55e0c7745e9963c90e38b9dfd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61d0fc9e37d8c134a78f2a5b50b947d03459dba8e08365a13fa320e91092a0d6 +size 23674 diff --git a/parse/train/HJgXCV9xx/images/32fa8ec819f7ad327e91b663e83c09e512124353ef0957e751a104651b7b6131.jpg b/parse/train/HJgXCV9xx/images/32fa8ec819f7ad327e91b663e83c09e512124353ef0957e751a104651b7b6131.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5153c64166e473c7c23f2e677a2a6861e373922 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/32fa8ec819f7ad327e91b663e83c09e512124353ef0957e751a104651b7b6131.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8adafe631a08e015e7613913300ff48c21f364fa0c50963405b457b9df9e31fe +size 13862 diff --git a/parse/train/HJgXCV9xx/images/3737f4baf3d6df13b680d0683c59004ec204522d0328caa3d43b3831fd5e616f.jpg b/parse/train/HJgXCV9xx/images/3737f4baf3d6df13b680d0683c59004ec204522d0328caa3d43b3831fd5e616f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b5ade8f0c851d6ffaadbf0acceeebe89e87180f2 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/3737f4baf3d6df13b680d0683c59004ec204522d0328caa3d43b3831fd5e616f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bba3235b04bc760978193f456780841da19587664d5fc2522fc8027175f30378 +size 5232 diff --git a/parse/train/HJgXCV9xx/images/4a41de306ba4fd6107b7556374b3ed095be6bbc65f77245d2966f27e797939e7.jpg b/parse/train/HJgXCV9xx/images/4a41de306ba4fd6107b7556374b3ed095be6bbc65f77245d2966f27e797939e7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..220c42d3ae2bd930dd7dba7f034902a34f4b4770 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/4a41de306ba4fd6107b7556374b3ed095be6bbc65f77245d2966f27e797939e7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1368175bc18bdcef14a614d3a272e720988921651fa6152c9ae48729d130cf1a +size 254668 diff --git a/parse/train/HJgXCV9xx/images/55d166d1fe656addb154b4a220ac8e33fc70cfcba49cd9aad48b0674d3808e01.jpg b/parse/train/HJgXCV9xx/images/55d166d1fe656addb154b4a220ac8e33fc70cfcba49cd9aad48b0674d3808e01.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6822be9b726302b6fa1c61a69370a3c50c8315e8 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/55d166d1fe656addb154b4a220ac8e33fc70cfcba49cd9aad48b0674d3808e01.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15d425c0aca641423dc02ff7db932ddd5bcdfe0390ffc7b8685216a8bd12c1ad +size 24536 diff --git a/parse/train/HJgXCV9xx/images/59a90a12286c1a9339195508cf12e2c6b9fbd0bac0171de968749ab26f5db2e0.jpg b/parse/train/HJgXCV9xx/images/59a90a12286c1a9339195508cf12e2c6b9fbd0bac0171de968749ab26f5db2e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7343615543dbf4d8ee3c2d5eebba1171b6b3442f --- /dev/null +++ b/parse/train/HJgXCV9xx/images/59a90a12286c1a9339195508cf12e2c6b9fbd0bac0171de968749ab26f5db2e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ff8bfb4f026a238cf2db406ad70dda1902750cd65910fc93bdd08a956d06ca5 +size 52752 diff --git a/parse/train/HJgXCV9xx/images/5f2293f3a1d7454aa518654814d44c0454b37a7bb6dc9dbc94c1a17afb356d2d.jpg b/parse/train/HJgXCV9xx/images/5f2293f3a1d7454aa518654814d44c0454b37a7bb6dc9dbc94c1a17afb356d2d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bfece1cb6a75ab0fd15c4f3178e1470b63ce6573 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/5f2293f3a1d7454aa518654814d44c0454b37a7bb6dc9dbc94c1a17afb356d2d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2e0170f44623b2b470a2de88afedff96229f68594252d373330c37665057098 +size 23766 diff --git a/parse/train/HJgXCV9xx/images/601367ddf264f8852e89e26d1be135aa6332d970b5b8bd479d17587c8f8c659f.jpg b/parse/train/HJgXCV9xx/images/601367ddf264f8852e89e26d1be135aa6332d970b5b8bd479d17587c8f8c659f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5c006fc370b6eb85fd50023c06df02d3979db230 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/601367ddf264f8852e89e26d1be135aa6332d970b5b8bd479d17587c8f8c659f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e03a88c9425e91fda45f4af703f12ba4ba4bd1c59c60bf7b5fc6430e37a1a538 +size 147864 diff --git a/parse/train/HJgXCV9xx/images/63d6867f9da236bb7f337bcd968570cf6e06f0d3d0cdfacc32f0d5857bd75dc7.jpg b/parse/train/HJgXCV9xx/images/63d6867f9da236bb7f337bcd968570cf6e06f0d3d0cdfacc32f0d5857bd75dc7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8937531eb191c16f214fe2dd6de009ac33f83afd --- /dev/null +++ b/parse/train/HJgXCV9xx/images/63d6867f9da236bb7f337bcd968570cf6e06f0d3d0cdfacc32f0d5857bd75dc7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:773b4c539dad03e95194d6f8c44861ee50b9d26d8d653ad31ecc2e220b56841d +size 143785 diff --git a/parse/train/HJgXCV9xx/images/6ba21ef128eca801fe84f39394dcadd73741ef6b87f153388d67db3208f9a34c.jpg b/parse/train/HJgXCV9xx/images/6ba21ef128eca801fe84f39394dcadd73741ef6b87f153388d67db3208f9a34c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ad0d8821e92224c5192110264274f329d0a1d8cb --- /dev/null +++ b/parse/train/HJgXCV9xx/images/6ba21ef128eca801fe84f39394dcadd73741ef6b87f153388d67db3208f9a34c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:687e7ab8a3ee1d0b71b7fdf2d3e3e30031d7731575fe47ee70dcd264fb3fc2b1 +size 241465 diff --git a/parse/train/HJgXCV9xx/images/73c5258cad5d6520b349f47df32b028d96e36dc0d65314a95967ea3d948acff6.jpg b/parse/train/HJgXCV9xx/images/73c5258cad5d6520b349f47df32b028d96e36dc0d65314a95967ea3d948acff6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3309704a542de5a5688e8fe6035d8d89213f8ceb --- /dev/null +++ b/parse/train/HJgXCV9xx/images/73c5258cad5d6520b349f47df32b028d96e36dc0d65314a95967ea3d948acff6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c7b6551d2dfa964be09034b180bd54478b88a20392a5bf62c082ac08e4d7446 +size 74317 diff --git a/parse/train/HJgXCV9xx/images/8230b363b455317273022351fdaedda93da694fd962df3e347f55bbe6e20e31e.jpg b/parse/train/HJgXCV9xx/images/8230b363b455317273022351fdaedda93da694fd962df3e347f55bbe6e20e31e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..26518b5a3052d6d49a428f092877f51308930653 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/8230b363b455317273022351fdaedda93da694fd962df3e347f55bbe6e20e31e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:742f8e45407f2fcbe1e8e891f85f018ef6ec76b68367e23aaa7d2adb16aaa919 +size 53387 diff --git a/parse/train/HJgXCV9xx/images/989ff1dc27990201aef46db85c2297a53b838e35913217e07e60fbaa019ebc78.jpg b/parse/train/HJgXCV9xx/images/989ff1dc27990201aef46db85c2297a53b838e35913217e07e60fbaa019ebc78.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c4805d4c100cb886a4ba2b99edd2ccdb5a60d733 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/989ff1dc27990201aef46db85c2297a53b838e35913217e07e60fbaa019ebc78.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9acd1f64d0eaabf6d57760972824a2c51d36fd3f315b4b68d7352eca09fb9b8e +size 5866 diff --git a/parse/train/HJgXCV9xx/images/b35fb660c29c1acc5103256fdee13fc7b7c467ebbfd241f5a08c27d32c49da14.jpg b/parse/train/HJgXCV9xx/images/b35fb660c29c1acc5103256fdee13fc7b7c467ebbfd241f5a08c27d32c49da14.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d96d2993f6cca958fcfca0b42e73add9ad14853 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/b35fb660c29c1acc5103256fdee13fc7b7c467ebbfd241f5a08c27d32c49da14.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be78a32244323af60f3b5d0a6d53c8db7f7cc0433d5a28ebef476f084d2eb864 +size 148502 diff --git a/parse/train/HJgXCV9xx/images/b8d403de4d781e88b1c2526b1628bbdbdc323a3bd1149c1df8046013d5c5cf20.jpg b/parse/train/HJgXCV9xx/images/b8d403de4d781e88b1c2526b1628bbdbdc323a3bd1149c1df8046013d5c5cf20.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6fcc0c70b69028d5e99475082d51665b43075c64 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/b8d403de4d781e88b1c2526b1628bbdbdc323a3bd1149c1df8046013d5c5cf20.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a68aa1360ef850bc2308ccfc2a9c7e19ca17ab45c18241523471bcc3ef62f3c +size 145950 diff --git a/parse/train/HJgXCV9xx/images/cbb2bce2922f54d67b77a545efdc03a0308050b7a8a16f1e0d970a7ce110f915.jpg b/parse/train/HJgXCV9xx/images/cbb2bce2922f54d67b77a545efdc03a0308050b7a8a16f1e0d970a7ce110f915.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8cbabbde38fbfa05b44144c7da9d3cfeb31fb8ac --- /dev/null +++ b/parse/train/HJgXCV9xx/images/cbb2bce2922f54d67b77a545efdc03a0308050b7a8a16f1e0d970a7ce110f915.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0744268b15947456a48b7f21c181214fd0a86050fa4d1496af08b9bcf7ded364 +size 252153 diff --git a/parse/train/HJgXCV9xx/images/e4ee0cf1375719c66f888c3dc093a1933c1440f0fedf3a088d9e194999485e44.jpg b/parse/train/HJgXCV9xx/images/e4ee0cf1375719c66f888c3dc093a1933c1440f0fedf3a088d9e194999485e44.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1c72d479f9b931ac0ccb2332de10dbecee16a163 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/e4ee0cf1375719c66f888c3dc093a1933c1440f0fedf3a088d9e194999485e44.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8b318e112478f75a23f3e99ec162c08641bcb04cc813f34695c2fd89d387c505 +size 228013 diff --git a/parse/train/HJgXCV9xx/images/e5575aa39cbf83409b60aab184a9ddf4ea01b684abfaadd96c3265e4886c3252.jpg b/parse/train/HJgXCV9xx/images/e5575aa39cbf83409b60aab184a9ddf4ea01b684abfaadd96c3265e4886c3252.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a6f851a887a4394db2ddad2b1e6f6f37657f35c4 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/e5575aa39cbf83409b60aab184a9ddf4ea01b684abfaadd96c3265e4886c3252.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d212810755f1fa1131f0e01e8ccbe0aec997ff6c540e8febce1f1692ccc7257 +size 85834 diff --git a/parse/train/HJgXCV9xx/images/e918250643fab52af2f22cc525822f337a08eaa6b3418c5a0a6ad8cafb26660c.jpg b/parse/train/HJgXCV9xx/images/e918250643fab52af2f22cc525822f337a08eaa6b3418c5a0a6ad8cafb26660c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b27ba9187f25d7eb46feae5af02a48d935768185 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/e918250643fab52af2f22cc525822f337a08eaa6b3418c5a0a6ad8cafb26660c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0bfc32fc232a5de1ba6bf0255746125f18636effa9aab576fe8f9ad391531374 +size 26178 diff --git a/parse/train/HJgXCV9xx/images/f7ee2c803c397cfc8ddd71b9a46a1f5da7456f408725e0c1f6a33eb775e84051.jpg b/parse/train/HJgXCV9xx/images/f7ee2c803c397cfc8ddd71b9a46a1f5da7456f408725e0c1f6a33eb775e84051.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c0b3cedc8955ef8ac2128436e892b8014b5c4d60 --- /dev/null +++ b/parse/train/HJgXCV9xx/images/f7ee2c803c397cfc8ddd71b9a46a1f5da7456f408725e0c1f6a33eb775e84051.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a894326dc106fbf84674ab83430c47df51a3035a38c70b9ff6367a053938651 +size 5322 diff --git a/parse/train/HJxwDiActX/images/12adc89fd98d0d6e7695b08f1fd72cd6beb3a85dd40b057fdaee4f3d9f3534d9.jpg b/parse/train/HJxwDiActX/images/12adc89fd98d0d6e7695b08f1fd72cd6beb3a85dd40b057fdaee4f3d9f3534d9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a623c540ec297f2780116c5c354c84081763d275 --- /dev/null +++ b/parse/train/HJxwDiActX/images/12adc89fd98d0d6e7695b08f1fd72cd6beb3a85dd40b057fdaee4f3d9f3534d9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d1293def57b92e67409e89c07965d8bd41439adf035c231fc87377a17e9ad07 +size 15592 diff --git a/parse/train/HJxwDiActX/images/19c961a8418d60a5c347aab8c59d6f84a9c439acdc514242eab85de0e83f4f4f.jpg b/parse/train/HJxwDiActX/images/19c961a8418d60a5c347aab8c59d6f84a9c439acdc514242eab85de0e83f4f4f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..daa9f4c2578abb47d804eb07bb329bd8d7a20e19 --- /dev/null +++ b/parse/train/HJxwDiActX/images/19c961a8418d60a5c347aab8c59d6f84a9c439acdc514242eab85de0e83f4f4f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:054206b66db0629d89acbd0a1737af5feadc8e7da8ebbd3c9db060e98c645bc7 +size 5590 diff --git a/parse/train/HJxwDiActX/images/2ff9c093b643afca8ef2c1e3f88855bcf31afbd13dd1528e6247eae8019aba54.jpg b/parse/train/HJxwDiActX/images/2ff9c093b643afca8ef2c1e3f88855bcf31afbd13dd1528e6247eae8019aba54.jpg new file mode 100644 index 0000000000000000000000000000000000000000..397ec5cfc3b924e56cb772f9406e88c41d0deb9a --- /dev/null +++ b/parse/train/HJxwDiActX/images/2ff9c093b643afca8ef2c1e3f88855bcf31afbd13dd1528e6247eae8019aba54.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a61a5560a3ff94412c1b45584fdf00fd3927c5f17caa87df32e00b61e141cf5 +size 20807 diff --git a/parse/train/HJxwDiActX/images/3393c1316e650209fdaeb85ab62a718088c1ada207a2377867aa89e7b8a81c48.jpg b/parse/train/HJxwDiActX/images/3393c1316e650209fdaeb85ab62a718088c1ada207a2377867aa89e7b8a81c48.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93221a2bb6a56f32ee6a5d30dc1d44eb3e8e9239 --- /dev/null +++ b/parse/train/HJxwDiActX/images/3393c1316e650209fdaeb85ab62a718088c1ada207a2377867aa89e7b8a81c48.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:878ffaccf2833f3d21afd0d6f0afa3c5b1c86b39c5e2db6b914d528d4283fcc6 +size 24454 diff --git a/parse/train/HJxwDiActX/images/40b22a0ad601377a1ce1e611194332e0a89f564a58c33a04e379d5c89fec841e.jpg b/parse/train/HJxwDiActX/images/40b22a0ad601377a1ce1e611194332e0a89f564a58c33a04e379d5c89fec841e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c760dee14e67435fd60167c385e86d98686ddeb6 --- /dev/null +++ b/parse/train/HJxwDiActX/images/40b22a0ad601377a1ce1e611194332e0a89f564a58c33a04e379d5c89fec841e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7f5d11767c9f53cbe90b0eea83610dc87c1b414e2f1aba0bd075c28233330d1 +size 4697 diff --git a/parse/train/HJxwDiActX/images/4328a2a5ada1d0da21560bfb3192aa5de292ddf81a01eb151890d443c82e0df6.jpg b/parse/train/HJxwDiActX/images/4328a2a5ada1d0da21560bfb3192aa5de292ddf81a01eb151890d443c82e0df6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ad8c484931827658c1e5a7d525bd09b5b43f5e81 --- /dev/null +++ b/parse/train/HJxwDiActX/images/4328a2a5ada1d0da21560bfb3192aa5de292ddf81a01eb151890d443c82e0df6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b0ec51f8a0a0d3c742a149b7b176cf1e70bd8fce0e4549a2fac842be9a5bed0 +size 3773 diff --git a/parse/train/HJxwDiActX/images/46746c2f2c4cd0c3cc97e7b171627657bfa7857c1c7b4aced2897811ebf2906a.jpg b/parse/train/HJxwDiActX/images/46746c2f2c4cd0c3cc97e7b171627657bfa7857c1c7b4aced2897811ebf2906a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..091ea7839ee9d017f074c76169483784ed5baecd --- /dev/null +++ b/parse/train/HJxwDiActX/images/46746c2f2c4cd0c3cc97e7b171627657bfa7857c1c7b4aced2897811ebf2906a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6fa1e151fae743cb3bc251ad2a8248160571042e6d0aa8ec1bf96055f1a543c +size 18376 diff --git a/parse/train/HJxwDiActX/images/4a80132369306393f978eca23cc50d9f6a07c414590dda72eeea87cdcb3a5008.jpg b/parse/train/HJxwDiActX/images/4a80132369306393f978eca23cc50d9f6a07c414590dda72eeea87cdcb3a5008.jpg new file mode 100644 index 0000000000000000000000000000000000000000..326c4dc60a44226751a43724fafc1c6c73e8809b --- /dev/null +++ b/parse/train/HJxwDiActX/images/4a80132369306393f978eca23cc50d9f6a07c414590dda72eeea87cdcb3a5008.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4692318e89d192bb90dd9d8f4ec84657b21277eec3eb0594e9bda5fa0a282884 +size 11395 diff --git a/parse/train/HJxwDiActX/images/51e22126b3933d20d8956c3997ec5ea365d386f3d94ee0e1646b2ef1103bf225.jpg b/parse/train/HJxwDiActX/images/51e22126b3933d20d8956c3997ec5ea365d386f3d94ee0e1646b2ef1103bf225.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c9780cc7d3ca3f15f4c075bca1a2c399aba7e2c1 --- /dev/null +++ b/parse/train/HJxwDiActX/images/51e22126b3933d20d8956c3997ec5ea365d386f3d94ee0e1646b2ef1103bf225.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e58b6fc152bfbf08efda58175fe762a511fc1ebd205a17c2a06887eac78033c +size 34266 diff --git a/parse/train/HJxwDiActX/images/55d0f33075fc7e1319ed4edb76d5ba47b60cc87af4c91fce599014a09ea5f6f7.jpg b/parse/train/HJxwDiActX/images/55d0f33075fc7e1319ed4edb76d5ba47b60cc87af4c91fce599014a09ea5f6f7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..43a01df5a1533894c039d3f3e18c581995364773 --- /dev/null +++ b/parse/train/HJxwDiActX/images/55d0f33075fc7e1319ed4edb76d5ba47b60cc87af4c91fce599014a09ea5f6f7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27f82ff901da3e4329cd76447c180281a0940e0ef0e135b3de31e04b7bc55706 +size 19173 diff --git a/parse/train/HJxwDiActX/images/5877850f6a696278032a065d087f4fa24a19d79aa23f47b2aa9c496d94e89f59.jpg b/parse/train/HJxwDiActX/images/5877850f6a696278032a065d087f4fa24a19d79aa23f47b2aa9c496d94e89f59.jpg new file mode 100644 index 0000000000000000000000000000000000000000..53287ee7ffd8eb9008f1af6714b12f7ce6f698f3 --- /dev/null +++ b/parse/train/HJxwDiActX/images/5877850f6a696278032a065d087f4fa24a19d79aa23f47b2aa9c496d94e89f59.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c37b158595d1567b4ec4c98c68368753f317a09c517425d5df66a5026a10a2e +size 22066 diff --git a/parse/train/HJxwDiActX/images/58ac904b17dfb6b9dfc15ba2f1714e4cba22987cb55f76e17797f5e7eebd4070.jpg b/parse/train/HJxwDiActX/images/58ac904b17dfb6b9dfc15ba2f1714e4cba22987cb55f76e17797f5e7eebd4070.jpg new file mode 100644 index 0000000000000000000000000000000000000000..15e1ff9bd01a053bf382bebd31067a10ea399221 --- /dev/null +++ b/parse/train/HJxwDiActX/images/58ac904b17dfb6b9dfc15ba2f1714e4cba22987cb55f76e17797f5e7eebd4070.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45f40fdd6d6f1cd1f74c65c115925a3902767fd54d94045dbc01f361641d06b3 +size 1983 diff --git a/parse/train/HJxwDiActX/images/593fc981486363436c5a296bd938a3f0cc0ade15f69f2a64e2e0f35d38ab5059.jpg b/parse/train/HJxwDiActX/images/593fc981486363436c5a296bd938a3f0cc0ade15f69f2a64e2e0f35d38ab5059.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5454a14ecf94077a83d6cde9a255594a236c9c40 --- /dev/null +++ b/parse/train/HJxwDiActX/images/593fc981486363436c5a296bd938a3f0cc0ade15f69f2a64e2e0f35d38ab5059.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf286d350839651f92cb10b836fab7e6aa3bc18dfae4190cc3374e283d518719 +size 17422 diff --git a/parse/train/HJxwDiActX/images/6583c490aa16fce28fe0294e0a0c39b064a9259ee1d6a36cec1c6b46e30bf286.jpg b/parse/train/HJxwDiActX/images/6583c490aa16fce28fe0294e0a0c39b064a9259ee1d6a36cec1c6b46e30bf286.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f15aef8517efdc7d6301ce489186316570d402d --- /dev/null +++ b/parse/train/HJxwDiActX/images/6583c490aa16fce28fe0294e0a0c39b064a9259ee1d6a36cec1c6b46e30bf286.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc9abe9d7a29439ee02fa2507c37d7d984afc4212254d124bf33d0351e52c11c +size 56596 diff --git a/parse/train/HJxwDiActX/images/6a2983fb32a44cb93b90d7dbcbefe6d6e236b400196c2e700f733b5896bd0c71.jpg b/parse/train/HJxwDiActX/images/6a2983fb32a44cb93b90d7dbcbefe6d6e236b400196c2e700f733b5896bd0c71.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7753f21a22e7279cd82806667a09aa4d7fad62dd --- /dev/null +++ b/parse/train/HJxwDiActX/images/6a2983fb32a44cb93b90d7dbcbefe6d6e236b400196c2e700f733b5896bd0c71.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0769148ad16688b94404d64b7434af09e651e65ae82468244bd06793fcbfd417 +size 22023 diff --git a/parse/train/HJxwDiActX/images/728dba15c496270334ec9bb3633eadfed395b33b19efa8da841769dd5ee42c7c.jpg b/parse/train/HJxwDiActX/images/728dba15c496270334ec9bb3633eadfed395b33b19efa8da841769dd5ee42c7c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f828cfba7d4d94ad6419178a9fe8019707f6d8a8 --- /dev/null +++ b/parse/train/HJxwDiActX/images/728dba15c496270334ec9bb3633eadfed395b33b19efa8da841769dd5ee42c7c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf15d7b673d8bd600957226fb3362b4c44ea9a9ed70c063cd6619dce757883fd +size 5688 diff --git a/parse/train/HJxwDiActX/images/78f6dcb25f87433f876a05ebda8e11d0a83e41e7db7b03a40163cee0b2baad0d.jpg b/parse/train/HJxwDiActX/images/78f6dcb25f87433f876a05ebda8e11d0a83e41e7db7b03a40163cee0b2baad0d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf1f4240f53f75fd83e683b2b53fce938dcfbc95 --- /dev/null +++ b/parse/train/HJxwDiActX/images/78f6dcb25f87433f876a05ebda8e11d0a83e41e7db7b03a40163cee0b2baad0d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f6eb1041bec98687048e18ce0cdb89afc69c22b34782027950feaa98e439ed0 +size 3523 diff --git a/parse/train/HJxwDiActX/images/8cc9340e9f6cf001385745ac8cc4b37b39a59349aa733cfc788bdd925935ff93.jpg b/parse/train/HJxwDiActX/images/8cc9340e9f6cf001385745ac8cc4b37b39a59349aa733cfc788bdd925935ff93.jpg new file mode 100644 index 0000000000000000000000000000000000000000..581609105c04e86b15878ff957752c2a165036c5 --- /dev/null +++ b/parse/train/HJxwDiActX/images/8cc9340e9f6cf001385745ac8cc4b37b39a59349aa733cfc788bdd925935ff93.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:29d3ea4e5f7b5d94c0fa99ca18c01b1b0281fdf3c345e7401641cf183bd1f265 +size 4299 diff --git a/parse/train/HJxwDiActX/images/918c6afe99629ee0b28104d8a9108c9698ad147b66336844239de69be60b6761.jpg b/parse/train/HJxwDiActX/images/918c6afe99629ee0b28104d8a9108c9698ad147b66336844239de69be60b6761.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8984214b514f8746cf449e827e56e5a19fc766d0 --- /dev/null +++ b/parse/train/HJxwDiActX/images/918c6afe99629ee0b28104d8a9108c9698ad147b66336844239de69be60b6761.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e371d2ecf210541f74466a113c0430223715a91f591c4c738e9952bf81e5631d +size 43837 diff --git a/parse/train/HJxwDiActX/images/b558e893981f87fa96e3a405f141ba3ff4246cbec97db62af7c5c9f55d69fe38.jpg b/parse/train/HJxwDiActX/images/b558e893981f87fa96e3a405f141ba3ff4246cbec97db62af7c5c9f55d69fe38.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6797969130a41abb2a8caac414075b712e16644b --- /dev/null +++ b/parse/train/HJxwDiActX/images/b558e893981f87fa96e3a405f141ba3ff4246cbec97db62af7c5c9f55d69fe38.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23401bd07932b6ed60b0ed62c6137b0c82a9ebf46b3daf2cc8f677f160479e41 +size 23142 diff --git a/parse/train/HJxwDiActX/images/b90cbb0741b34dcf3c71a8aa8ca4a1061c3be49ed7bff08682c3c328047045b9.jpg b/parse/train/HJxwDiActX/images/b90cbb0741b34dcf3c71a8aa8ca4a1061c3be49ed7bff08682c3c328047045b9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..af6043978a1cc04bc4237fc95772170dd61d6f58 --- /dev/null +++ b/parse/train/HJxwDiActX/images/b90cbb0741b34dcf3c71a8aa8ca4a1061c3be49ed7bff08682c3c328047045b9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bee99e18f1eebade8489f32075c67f49c09746894f00afc0dc1f3f146e1ed70a +size 4352 diff --git a/parse/train/HJxwDiActX/images/ba005ed854b84d3c2a3e3ee6f1ad177c538f73860517a0f464ae80d613871d54.jpg b/parse/train/HJxwDiActX/images/ba005ed854b84d3c2a3e3ee6f1ad177c538f73860517a0f464ae80d613871d54.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eddbd1cfefdf4f4bd246f24ee30c17d56985702b --- /dev/null +++ b/parse/train/HJxwDiActX/images/ba005ed854b84d3c2a3e3ee6f1ad177c538f73860517a0f464ae80d613871d54.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b442f1b390b7d42cbc92d9c34bf48e40333310b65a8f90bac895130a57659b0d +size 9433 diff --git a/parse/train/HJxwDiActX/images/bd7ac006442084af2ca3a9be2fe4440769f067883a0e1ff4e66a21e9e75a978d.jpg b/parse/train/HJxwDiActX/images/bd7ac006442084af2ca3a9be2fe4440769f067883a0e1ff4e66a21e9e75a978d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..48e372a5cf198d1a17b9e309d754510ee1b4e4ea --- /dev/null +++ b/parse/train/HJxwDiActX/images/bd7ac006442084af2ca3a9be2fe4440769f067883a0e1ff4e66a21e9e75a978d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e89759d082b1e57e9ade4d12502cf1740d828a42b19c8454f9ebcf7e6cb0705a +size 7030 diff --git a/parse/train/HJxwDiActX/images/c225ea23a10e81fccc5bb5ac12beaf88d3dd4a4380938ca733214e6a03ea550a.jpg b/parse/train/HJxwDiActX/images/c225ea23a10e81fccc5bb5ac12beaf88d3dd4a4380938ca733214e6a03ea550a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5f78835d4bc81cc8896edbd5df1eb88fb1ddb76a --- /dev/null +++ b/parse/train/HJxwDiActX/images/c225ea23a10e81fccc5bb5ac12beaf88d3dd4a4380938ca733214e6a03ea550a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9731dc2f690abaf2d229bd73235792244c38687492634f1f633bb65411d94687 +size 10037 diff --git a/parse/train/HJxwDiActX/images/cd0fb8b0b8397f69a4446cdcc90821d1797d6aee654882edd8dc4328fc66c0f9.jpg b/parse/train/HJxwDiActX/images/cd0fb8b0b8397f69a4446cdcc90821d1797d6aee654882edd8dc4328fc66c0f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b433cd4e6f2559e6267dc4da8dd0f0c598242807 --- /dev/null +++ b/parse/train/HJxwDiActX/images/cd0fb8b0b8397f69a4446cdcc90821d1797d6aee654882edd8dc4328fc66c0f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ecd548d2bc7817a0d66882770d4d9045c407567c53bd999a62f9e4d887317bed +size 29009 diff --git a/parse/train/HJxwDiActX/images/e0b0f3f62c87099c3be63048b1cf65e4a3dcc30599750b6cdfd1200ac64da07e.jpg b/parse/train/HJxwDiActX/images/e0b0f3f62c87099c3be63048b1cf65e4a3dcc30599750b6cdfd1200ac64da07e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f1ecd7fbab9283a50ade187bad3ca2dc3ec607c7 --- /dev/null +++ b/parse/train/HJxwDiActX/images/e0b0f3f62c87099c3be63048b1cf65e4a3dcc30599750b6cdfd1200ac64da07e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7ca9234b49e56fdf70b335cedfbc194cd5b793ec94e1b8a94a533cd946e104de +size 14539 diff --git a/parse/train/HJxwDiActX/images/e51f6aa6863d062961ab788f55e2115111bed60f9704e81160eb899861150adf.jpg b/parse/train/HJxwDiActX/images/e51f6aa6863d062961ab788f55e2115111bed60f9704e81160eb899861150adf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ac2ae2de088635b1bd0d9bec746af6b379580b7e --- /dev/null +++ b/parse/train/HJxwDiActX/images/e51f6aa6863d062961ab788f55e2115111bed60f9704e81160eb899861150adf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18b2424e941dfff884cc8d6f4078d8377bc0ffd50c039973c6f62ad292c2873a +size 16792 diff --git a/parse/train/HJxwDiActX/images/ea7e6900d0585b611e13695dbd8d6dd019e443c57b0b9d67df8d74b05e118cf3.jpg b/parse/train/HJxwDiActX/images/ea7e6900d0585b611e13695dbd8d6dd019e443c57b0b9d67df8d74b05e118cf3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..780323fc84a985cbb531c629021ab8b8c9100662 --- /dev/null +++ b/parse/train/HJxwDiActX/images/ea7e6900d0585b611e13695dbd8d6dd019e443c57b0b9d67df8d74b05e118cf3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8ab240ff25268a5b35294c422d67125382006f2238af205ea2635b4b0657542 +size 7953 diff --git a/parse/train/HJxwDiActX/images/efa474886deb1f2b0d27a17f8d9eb62f73e171b17f3ef63af211a6a6ebb42bf2.jpg b/parse/train/HJxwDiActX/images/efa474886deb1f2b0d27a17f8d9eb62f73e171b17f3ef63af211a6a6ebb42bf2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..729b08953a0572d1c25a749c2319fd31d98a7ae6 --- /dev/null +++ b/parse/train/HJxwDiActX/images/efa474886deb1f2b0d27a17f8d9eb62f73e171b17f3ef63af211a6a6ebb42bf2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33d28e5615f6cdaaa9d981be6c9768fb64297cab8506779460ce56b0bc900294 +size 13229 diff --git a/parse/train/HJxwDiActX/images/fa80a5f32051b56fde49f5102a99ebeb6bcaea836173663b563a5f20ce540db0.jpg b/parse/train/HJxwDiActX/images/fa80a5f32051b56fde49f5102a99ebeb6bcaea836173663b563a5f20ce540db0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4f0d9215eecb357a63cf44d3d4398d5ac1d67cdf --- /dev/null +++ b/parse/train/HJxwDiActX/images/fa80a5f32051b56fde49f5102a99ebeb6bcaea836173663b563a5f20ce540db0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42d22d4bf0b327da30838c808a8ce1543c2d13aa5b7aa41f8ca982b2c3214b66 +size 3801 diff --git a/parse/train/HJxwDiActX/images/fbf03a0ebb808615c2b8f5fdff2ad1c44f0dce152416ea929956926aaaba3b35.jpg b/parse/train/HJxwDiActX/images/fbf03a0ebb808615c2b8f5fdff2ad1c44f0dce152416ea929956926aaaba3b35.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9bf6b1cc16d6dd4b5fdbcf274ad7225b2122256a --- /dev/null +++ b/parse/train/HJxwDiActX/images/fbf03a0ebb808615c2b8f5fdff2ad1c44f0dce152416ea929956926aaaba3b35.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59fe9e85eae87b814720a54c219411d01d1a658124599fb19ad7750ef2c7a931 +size 11495 diff --git a/parse/train/HJxwDiActX/images/fda5720d46cc52464b46743c9dc970b8d18a8f078a4ae084fafe70b19c4123c0.jpg b/parse/train/HJxwDiActX/images/fda5720d46cc52464b46743c9dc970b8d18a8f078a4ae084fafe70b19c4123c0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..40e70cce3619c7b19551a37107a3ba95ff552af2 --- /dev/null +++ b/parse/train/HJxwDiActX/images/fda5720d46cc52464b46743c9dc970b8d18a8f078a4ae084fafe70b19c4123c0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59bdf4af2f720f39f45c95f9a1c2f14d136a90b3c3b3c010f06ba7ee26680e7c +size 13670 diff --git a/parse/train/Hke-WTVtwr/images/01faabd0975590a2a37566da11434014bfbc29f94dde966000f6cc151a5e403b.jpg b/parse/train/Hke-WTVtwr/images/01faabd0975590a2a37566da11434014bfbc29f94dde966000f6cc151a5e403b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..46066b9cea7c199be40d5d3fa3b30ca57671f07d --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/01faabd0975590a2a37566da11434014bfbc29f94dde966000f6cc151a5e403b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e831f6728592a2a12d4992692a8e95d8229597a622e9863f2954878262624800 +size 6815 diff --git a/parse/train/Hke-WTVtwr/images/02fe029bb38f319e348753ab7e72854b6e995d5a6b9ab63dac2933107e0b98c4.jpg b/parse/train/Hke-WTVtwr/images/02fe029bb38f319e348753ab7e72854b6e995d5a6b9ab63dac2933107e0b98c4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ab250132c32c26e7b4557a8ffca63b90fdde5f38 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/02fe029bb38f319e348753ab7e72854b6e995d5a6b9ab63dac2933107e0b98c4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5591f790d24b8b3b7893981501579adfdb253275f0ae84bbe59fd9b3c471fa7e +size 12699 diff --git a/parse/train/Hke-WTVtwr/images/0eb4f7bc17fac9dafe8f29328509ea3a49988b0cc118ca248f9f536791ad15c7.jpg b/parse/train/Hke-WTVtwr/images/0eb4f7bc17fac9dafe8f29328509ea3a49988b0cc118ca248f9f536791ad15c7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f30ccbf23f39c2002f9086f0f814908dc3b03a08 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/0eb4f7bc17fac9dafe8f29328509ea3a49988b0cc118ca248f9f536791ad15c7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4a80998142e4f773b2c9e42e3931e74940134aafdf4dcc4a6feea12afdc8fce +size 31162 diff --git a/parse/train/Hke-WTVtwr/images/11c83652acf7a3b0ac5f911a9040fc2e1ef56561033821f7ca3bc19bbd96067f.jpg b/parse/train/Hke-WTVtwr/images/11c83652acf7a3b0ac5f911a9040fc2e1ef56561033821f7ca3bc19bbd96067f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e3701ee7e86dc6824e606570d50ab60b3da3564 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/11c83652acf7a3b0ac5f911a9040fc2e1ef56561033821f7ca3bc19bbd96067f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70c585a6630d1a0a9f9f6339d24efce296a012032ae8b684f818e3dc4fe1fccc +size 5435 diff --git a/parse/train/Hke-WTVtwr/images/18234793b3ffc63153ac63f3b07e3f300c4f122f0e2692fe53c4e74e604a276b.jpg b/parse/train/Hke-WTVtwr/images/18234793b3ffc63153ac63f3b07e3f300c4f122f0e2692fe53c4e74e604a276b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c1b02f9caeb7c6b228d0a4394c2e1a90f58f9d75 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/18234793b3ffc63153ac63f3b07e3f300c4f122f0e2692fe53c4e74e604a276b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b5bd692baccb041560999e27b183e1a955c9f7eda99610f2ef19522befb393b +size 10270 diff --git a/parse/train/Hke-WTVtwr/images/2d3112ebe37ae3445f9917a47031f8a67ac78b06bafa42d5c4d1684d7c0f1d0d.jpg b/parse/train/Hke-WTVtwr/images/2d3112ebe37ae3445f9917a47031f8a67ac78b06bafa42d5c4d1684d7c0f1d0d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e537b3935952ab5af9331be2891a237faef938d3 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/2d3112ebe37ae3445f9917a47031f8a67ac78b06bafa42d5c4d1684d7c0f1d0d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c1f4d4f73f54d3562ac314d6af9754712acce98223991ce400d1c64a0a3a60d +size 102846 diff --git a/parse/train/Hke-WTVtwr/images/3e281f61b64e81bd85bc32ab1d241a5dd6a5878639505a4ffd1356dda400c6a3.jpg b/parse/train/Hke-WTVtwr/images/3e281f61b64e81bd85bc32ab1d241a5dd6a5878639505a4ffd1356dda400c6a3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e99f422a14abe499de812849753f4e418a7effad --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/3e281f61b64e81bd85bc32ab1d241a5dd6a5878639505a4ffd1356dda400c6a3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65695d80352faaf18341f3f7192cf683e656e9098fb06b13bf78083b88908fa9 +size 29894 diff --git a/parse/train/Hke-WTVtwr/images/4200795c7256b106cf6b11d3f6f812f89fa1ca3dcf9833290c41d544ffbced76.jpg b/parse/train/Hke-WTVtwr/images/4200795c7256b106cf6b11d3f6f812f89fa1ca3dcf9833290c41d544ffbced76.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a47a22cb9b00c2bc6be124fe2d889bb589a9078d --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/4200795c7256b106cf6b11d3f6f812f89fa1ca3dcf9833290c41d544ffbced76.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d1b095eaa970bd3339d46c8ade34e7616ae49732b582c3e28bbd5589d3968ce +size 6880 diff --git a/parse/train/Hke-WTVtwr/images/4313ba26dee6f520c296fda6cc12724c2fafd66f5247b95630e90d1ad1825a8f.jpg b/parse/train/Hke-WTVtwr/images/4313ba26dee6f520c296fda6cc12724c2fafd66f5247b95630e90d1ad1825a8f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9635f08766b10bf2d43b0e296f228a3585f112ae --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/4313ba26dee6f520c296fda6cc12724c2fafd66f5247b95630e90d1ad1825a8f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f33938c069ea1142c991d6201dc2b552d578cde90dac811359e65d3ed52da3ee +size 4373 diff --git a/parse/train/Hke-WTVtwr/images/446f31a81a4ea34d6d45ae9ae88f3f4b1bfa136ba94f9eedde5d432bc81f94df.jpg b/parse/train/Hke-WTVtwr/images/446f31a81a4ea34d6d45ae9ae88f3f4b1bfa136ba94f9eedde5d432bc81f94df.jpg new file mode 100644 index 0000000000000000000000000000000000000000..35cf9845a978d50e2b2566fffb6af05069abced0 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/446f31a81a4ea34d6d45ae9ae88f3f4b1bfa136ba94f9eedde5d432bc81f94df.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d040cbdb1256f44d51a595ee61e7fb47fe52c182d1e4cba37ec811637ad68d61 +size 8570 diff --git a/parse/train/Hke-WTVtwr/images/46a7e39cd995f7ff8468017cfdf5c3ec1990a3339a362390a8b9beb91692c702.jpg b/parse/train/Hke-WTVtwr/images/46a7e39cd995f7ff8468017cfdf5c3ec1990a3339a362390a8b9beb91692c702.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cecc4a9c3edb86ff0f1c75210297d5096d66ab89 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/46a7e39cd995f7ff8468017cfdf5c3ec1990a3339a362390a8b9beb91692c702.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b86e630468ee4d42034b67919611c1a44053602f28262e7740f9a9727e97d4ee +size 4002 diff --git a/parse/train/Hke-WTVtwr/images/53de9ce9bf8de49eeca107551940bd7611309416116e1ee648f96a3557cb0a0e.jpg b/parse/train/Hke-WTVtwr/images/53de9ce9bf8de49eeca107551940bd7611309416116e1ee648f96a3557cb0a0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6337220e6714f52b123d39c479ce47016fe3e823 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/53de9ce9bf8de49eeca107551940bd7611309416116e1ee648f96a3557cb0a0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe0aa9c7edce33e1fa96031eaa7ca1f98c1449204f741d2d1f38812d75b06fd1 +size 4822 diff --git a/parse/train/Hke-WTVtwr/images/56b40debdb8bcfe755f8890fa267df347982b6635b593811567d82720a1931b1.jpg b/parse/train/Hke-WTVtwr/images/56b40debdb8bcfe755f8890fa267df347982b6635b593811567d82720a1931b1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e41431d8cb2f9ffbca3f9dc2af5da97e26f3d0d2 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/56b40debdb8bcfe755f8890fa267df347982b6635b593811567d82720a1931b1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:acb46ecb7744658816555f061e1a44cd8a07202bfdf31892575a228d691be2f2 +size 3882 diff --git a/parse/train/Hke-WTVtwr/images/6d0795a5607bb360d12a9ff9b0f12a4fbbe5da63bed7601073fa004070938f20.jpg b/parse/train/Hke-WTVtwr/images/6d0795a5607bb360d12a9ff9b0f12a4fbbe5da63bed7601073fa004070938f20.jpg new file mode 100644 index 0000000000000000000000000000000000000000..58659fb9dfc2523b56a2adf4cb87e3a5e0b7dc59 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/6d0795a5607bb360d12a9ff9b0f12a4fbbe5da63bed7601073fa004070938f20.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d7f43b2bdfdf85eda217a00518b00de3edc64febeb9a8796b1cd0024e9552c5 +size 8493 diff --git a/parse/train/Hke-WTVtwr/images/71d6fddbcbb19c028e644b6683e1214f76462e01010b5cc7a01f43777ff7d1e8.jpg b/parse/train/Hke-WTVtwr/images/71d6fddbcbb19c028e644b6683e1214f76462e01010b5cc7a01f43777ff7d1e8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1ca684657c2246fd1d53a6aab49e52b7180c054f --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/71d6fddbcbb19c028e644b6683e1214f76462e01010b5cc7a01f43777ff7d1e8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3fa6575b3edf983889273cff5c471f3c50b8da03fa73766674d129d7c8f05da +size 42540 diff --git a/parse/train/Hke-WTVtwr/images/73c83963e37e356ccac9f1beece911eb94ee770c88669cc2d4f3ecc5d7df34db.jpg b/parse/train/Hke-WTVtwr/images/73c83963e37e356ccac9f1beece911eb94ee770c88669cc2d4f3ecc5d7df34db.jpg new file mode 100644 index 0000000000000000000000000000000000000000..640ba1ee4df590082008b750fde87f264117d322 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/73c83963e37e356ccac9f1beece911eb94ee770c88669cc2d4f3ecc5d7df34db.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53d3eeec98a3deb28aed131db4c74888bd8bc4dd87b1ab62e216a4062f5d86f4 +size 5715 diff --git a/parse/train/Hke-WTVtwr/images/777c66237d8f2c17c53e01b5aeff65834a3b57f5fcc7a6323d0ff3d017987acc.jpg b/parse/train/Hke-WTVtwr/images/777c66237d8f2c17c53e01b5aeff65834a3b57f5fcc7a6323d0ff3d017987acc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7538946f46b66a98486ae72b36fac1f36773f987 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/777c66237d8f2c17c53e01b5aeff65834a3b57f5fcc7a6323d0ff3d017987acc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd2d73c46a5b5ca16d16ed3a0cca8819ed161bb5903b3fd46aa4b5d04f259749 +size 150615 diff --git a/parse/train/Hke-WTVtwr/images/ac76f83825c53b1916373c5c9767b64608d583d08b9aeaa21b04acff6e5b50b1.jpg b/parse/train/Hke-WTVtwr/images/ac76f83825c53b1916373c5c9767b64608d583d08b9aeaa21b04acff6e5b50b1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1f42c42471dfa125186a05bbdf02c1733837d457 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/ac76f83825c53b1916373c5c9767b64608d583d08b9aeaa21b04acff6e5b50b1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23d375467d17eea5f05cf12041ed1ccb0b0b0e9257bcc809392f37b195aebc1f +size 3952 diff --git a/parse/train/Hke-WTVtwr/images/b161fc29c882b041c0e94712ee052872c4b3dde58d2170a08f1847105500aa0e.jpg b/parse/train/Hke-WTVtwr/images/b161fc29c882b041c0e94712ee052872c4b3dde58d2170a08f1847105500aa0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..718d34200a5977dca25fb34f8c5d3c5e21a420c2 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/b161fc29c882b041c0e94712ee052872c4b3dde58d2170a08f1847105500aa0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e4b568773a7f4ca75094953e183b8e143164603021c489d9c63d66eb1df8c4d +size 112860 diff --git a/parse/train/Hke-WTVtwr/images/ce1e7d7d5a8e8bd0be6504b3fdce0e2b05007d59a4113e47e09cdc41334ddb9b.jpg b/parse/train/Hke-WTVtwr/images/ce1e7d7d5a8e8bd0be6504b3fdce0e2b05007d59a4113e47e09cdc41334ddb9b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c909bad5684f05a5da4d0290e73504ffea830ce6 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/ce1e7d7d5a8e8bd0be6504b3fdce0e2b05007d59a4113e47e09cdc41334ddb9b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2b0722f0c118786f8229b59b118492be627836be5ed06b0e0a6fb041c255c91 +size 8499 diff --git a/parse/train/Hke-WTVtwr/images/ddd7d4e72917a127a89ebd1ccac157624d13881d396572841bf2b40b7f9bc09e.jpg b/parse/train/Hke-WTVtwr/images/ddd7d4e72917a127a89ebd1ccac157624d13881d396572841bf2b40b7f9bc09e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38ac198679887096af67691a67a9d49a286f661e --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/ddd7d4e72917a127a89ebd1ccac157624d13881d396572841bf2b40b7f9bc09e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24b1cf06f98baa51340337cdc15aaea6b47614da192d0ae2650bb02dc7191598 +size 7577 diff --git a/parse/train/Hke-WTVtwr/images/de7f52f5ee6760b22f8a39852d68d9c1d6d44ac42d72ca7c92dbece30230b7bb.jpg b/parse/train/Hke-WTVtwr/images/de7f52f5ee6760b22f8a39852d68d9c1d6d44ac42d72ca7c92dbece30230b7bb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ca4081e04ed59aaa675648364728e41f49a20698 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/de7f52f5ee6760b22f8a39852d68d9c1d6d44ac42d72ca7c92dbece30230b7bb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db5bfb8c12d8259a4cc3d671a6bd92bef24c94bd3fa4ba058e7bef1d9142e520 +size 50763 diff --git a/parse/train/Hke-WTVtwr/images/e18d67190d27ad89d72343f0f7e580d4499ce23a054a7022eb0448e518ff1704.jpg b/parse/train/Hke-WTVtwr/images/e18d67190d27ad89d72343f0f7e580d4499ce23a054a7022eb0448e518ff1704.jpg new file mode 100644 index 0000000000000000000000000000000000000000..69f8a1090e89c45dee77132124421f73153c5273 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/e18d67190d27ad89d72343f0f7e580d4499ce23a054a7022eb0448e518ff1704.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6026cb8d58737ec45f1d12bb041fa9a51e9f7646e935b9604ccc09f94fae859f +size 21375 diff --git a/parse/train/Hke-WTVtwr/images/e9af78adb2dadc1a549d01b8486ae40b138cd2f0c29508ab1ba281b078d55c6d.jpg b/parse/train/Hke-WTVtwr/images/e9af78adb2dadc1a549d01b8486ae40b138cd2f0c29508ab1ba281b078d55c6d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ec8f2e9817f7bffdee5f8d5aeeafca1627cd0355 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/e9af78adb2dadc1a549d01b8486ae40b138cd2f0c29508ab1ba281b078d55c6d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c7dcae8265b6f33006311e4100024e75a592544f0c35753771cfd10d4ec5cd6 +size 3698 diff --git a/parse/train/Hke-WTVtwr/images/ee1df64c33d8fcbca82c0506d4ba4959600580db8fbc24d9ca169bde95813e90.jpg b/parse/train/Hke-WTVtwr/images/ee1df64c33d8fcbca82c0506d4ba4959600580db8fbc24d9ca169bde95813e90.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4f82e6805291b8057c03609849b17199be86cb0b --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/ee1df64c33d8fcbca82c0506d4ba4959600580db8fbc24d9ca169bde95813e90.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff7b585a7432affeac57cffc9c8f16e25d2e49131b4a5aaa3a1e8b9d8fdb1590 +size 25198 diff --git a/parse/train/Hke-WTVtwr/images/f123726c10f6426226fd420db516f924007181387d1d51653b5c3c6bedc2ef2a.jpg b/parse/train/Hke-WTVtwr/images/f123726c10f6426226fd420db516f924007181387d1d51653b5c3c6bedc2ef2a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..61520ca4d081fbfb91e59294c27cad6d1e8153c7 --- /dev/null +++ b/parse/train/Hke-WTVtwr/images/f123726c10f6426226fd420db516f924007181387d1d51653b5c3c6bedc2ef2a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:293ce9a7cd121115796894922d4f87cfde119b245524af0d7d9d549d6cdc92e7 +size 33306 diff --git a/parse/train/Hkg9HgBYwH/images/00947a914d6f11b4f28e5ab82a6459e872ffdb95d5e4faee942d111deebf480b.jpg b/parse/train/Hkg9HgBYwH/images/00947a914d6f11b4f28e5ab82a6459e872ffdb95d5e4faee942d111deebf480b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4570f0a2c13f351b047344d94042a38eeda1744d --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/00947a914d6f11b4f28e5ab82a6459e872ffdb95d5e4faee942d111deebf480b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f6584b833a2da8f426f044d96f512b71df126b3db2f1a0697daf6131fcef050 +size 16088 diff --git a/parse/train/Hkg9HgBYwH/images/169edf5d45f7dc40da8e89f154db60d512947d0313ecc2015ef0fe44b377935f.jpg b/parse/train/Hkg9HgBYwH/images/169edf5d45f7dc40da8e89f154db60d512947d0313ecc2015ef0fe44b377935f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5604b17925b1a6a78fa6f77a3192d3731dc1eac1 --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/169edf5d45f7dc40da8e89f154db60d512947d0313ecc2015ef0fe44b377935f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b6d9c17339b1228d41bb758e575c777064bb19595d34035b22192fafc30b4b7 +size 65275 diff --git a/parse/train/Hkg9HgBYwH/images/3e19fe817ce4a99ff396bce5f0974f6c494fb6de5aa1471310accb6358a90ff0.jpg b/parse/train/Hkg9HgBYwH/images/3e19fe817ce4a99ff396bce5f0974f6c494fb6de5aa1471310accb6358a90ff0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ccc55412b1348331026ce447470d22fadde98b76 --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/3e19fe817ce4a99ff396bce5f0974f6c494fb6de5aa1471310accb6358a90ff0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cef160d70bb3099eb73d90888b9483be159ca204aff7eb504a0c4a8f37c41732 +size 53199 diff --git a/parse/train/Hkg9HgBYwH/images/67408dbb7c9ae86dbe2128a85229b51a8827797707b5eb78032c4218695d5694.jpg b/parse/train/Hkg9HgBYwH/images/67408dbb7c9ae86dbe2128a85229b51a8827797707b5eb78032c4218695d5694.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5775c813740e1491de3296e19ac097b0479eed9d --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/67408dbb7c9ae86dbe2128a85229b51a8827797707b5eb78032c4218695d5694.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0433c5a098a08032af29e5544da09b6c29df2871d03e72f6133a596546718482 +size 30976 diff --git a/parse/train/Hkg9HgBYwH/images/709133722cd8eb5b31e91b6a93b84f89295eb2c658fb84eb35a1a695cc7617d7.jpg b/parse/train/Hkg9HgBYwH/images/709133722cd8eb5b31e91b6a93b84f89295eb2c658fb84eb35a1a695cc7617d7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9dee78dec5ade8ba6bff21260a87a780b860baf8 --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/709133722cd8eb5b31e91b6a93b84f89295eb2c658fb84eb35a1a695cc7617d7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f43389b0bc6bc469a82471c48753d4ed8812618b4029f95c5e74d9df56e36386 +size 29795 diff --git a/parse/train/Hkg9HgBYwH/images/743e474368fbf561237b0c29fe3d5975e4bbc36327b4ae2809973edb592256c7.jpg b/parse/train/Hkg9HgBYwH/images/743e474368fbf561237b0c29fe3d5975e4bbc36327b4ae2809973edb592256c7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8ab83c03d33fa66785d7787050db0bae866ff014 --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/743e474368fbf561237b0c29fe3d5975e4bbc36327b4ae2809973edb592256c7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:842c731c702f9529d2e2f33c825bec99fe89f198cac8933ea58322bcce09d9e6 +size 37533 diff --git a/parse/train/Hkg9HgBYwH/images/75efeab00569c13b28563387655884feec0002aca2c13fab5b665278b73c3031.jpg b/parse/train/Hkg9HgBYwH/images/75efeab00569c13b28563387655884feec0002aca2c13fab5b665278b73c3031.jpg new file mode 100644 index 0000000000000000000000000000000000000000..87db80369e782d016d5255c9ac01ac71e46fed15 --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/75efeab00569c13b28563387655884feec0002aca2c13fab5b665278b73c3031.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32ff74e4c7644d88cef2b6fc5cdd0f463382645822e22ea0600d5ed08e46c08a +size 37938 diff --git a/parse/train/Hkg9HgBYwH/images/7c9073195f56bbc57d2a14c7a45342f17da059b1fc837236892d6730e164904f.jpg b/parse/train/Hkg9HgBYwH/images/7c9073195f56bbc57d2a14c7a45342f17da059b1fc837236892d6730e164904f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..63c9b1c147eb8aa393072e1f01452eebcf5a768e --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/7c9073195f56bbc57d2a14c7a45342f17da059b1fc837236892d6730e164904f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1cee6df0fbdec504cd20fef2b4540a6c8c9533269ba2e2c114659dbaaaf05d34 +size 15620 diff --git a/parse/train/Hkg9HgBYwH/images/9143cf4263afb01001cd56408b4e5d97dd06e0da98ce3b6b9819bac7c4a15181.jpg b/parse/train/Hkg9HgBYwH/images/9143cf4263afb01001cd56408b4e5d97dd06e0da98ce3b6b9819bac7c4a15181.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f77b2573f4a92e4dddb847acc3a59ae33599121d --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/9143cf4263afb01001cd56408b4e5d97dd06e0da98ce3b6b9819bac7c4a15181.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5496bc5b7e22cc382592766326272b206d65617c72c5b676f6052fc086b48c5d +size 19421 diff --git a/parse/train/Hkg9HgBYwH/images/a2d9f2ffb6690827d30510f9470f7c6cac7f0f4e5a80a2d787ee6cbc8c0d2ef4.jpg b/parse/train/Hkg9HgBYwH/images/a2d9f2ffb6690827d30510f9470f7c6cac7f0f4e5a80a2d787ee6cbc8c0d2ef4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..adf8e868736f5d51c6af9ecf73fc16573c58b0f4 --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/a2d9f2ffb6690827d30510f9470f7c6cac7f0f4e5a80a2d787ee6cbc8c0d2ef4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d68aabee0b2a4ce429c512b97617b962ab35032a55774bd4d3eafc7400a4da4 +size 97663 diff --git a/parse/train/Hkg9HgBYwH/images/b66a13d74361917ee7ccccacc6121b5c49899e962dcd86bd78a1a467af26109b.jpg b/parse/train/Hkg9HgBYwH/images/b66a13d74361917ee7ccccacc6121b5c49899e962dcd86bd78a1a467af26109b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1c8e15a15b5e12f6f1564e10a18e1d9d546e632c --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/b66a13d74361917ee7ccccacc6121b5c49899e962dcd86bd78a1a467af26109b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b20e6c8967f73b3a4d25a7ec91046aaf1796e74c9874d1abd45809c43fd39a60 +size 162866 diff --git a/parse/train/Hkg9HgBYwH/images/ba9cf42e5f640a7644af8396e335d8e5ca1c273744f87052f503d4dd8e3d7b41.jpg b/parse/train/Hkg9HgBYwH/images/ba9cf42e5f640a7644af8396e335d8e5ca1c273744f87052f503d4dd8e3d7b41.jpg new file mode 100644 index 0000000000000000000000000000000000000000..efe2a77eb6f28d939165672353955ad1bfcaaf25 --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/ba9cf42e5f640a7644af8396e335d8e5ca1c273744f87052f503d4dd8e3d7b41.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:875921d7438f0d49e1e6ef91c0e4bab6f09a3c9e762a891903e962c68bf37f6f +size 180784 diff --git a/parse/train/Hkg9HgBYwH/images/bc82bd157bca56953080c77933040d21ce1a37ea33c5d67ed123aa59090ddf61.jpg b/parse/train/Hkg9HgBYwH/images/bc82bd157bca56953080c77933040d21ce1a37ea33c5d67ed123aa59090ddf61.jpg new file mode 100644 index 0000000000000000000000000000000000000000..43af61e2c6566bbef89b6cd3dd5d9c6c1a25674c --- /dev/null +++ b/parse/train/Hkg9HgBYwH/images/bc82bd157bca56953080c77933040d21ce1a37ea33c5d67ed123aa59090ddf61.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2626913d730ff6e7e66feafa76f0bef758665cfa2a58b14d2fbdac10cdeeaebb +size 27292 diff --git a/parse/train/Hkx6hANtwH/images/0704bc0c7e8c893792ec823fb48f17a51aad7e1fc559005791ffd19d40365088.jpg b/parse/train/Hkx6hANtwH/images/0704bc0c7e8c893792ec823fb48f17a51aad7e1fc559005791ffd19d40365088.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa8f357ef18faf8be8b15336e81980f5d48f5916 --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/0704bc0c7e8c893792ec823fb48f17a51aad7e1fc559005791ffd19d40365088.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2f982d03072089bfae795ea5211d82f88c0f235790c3258ddcedd65ee485b52 +size 8686 diff --git a/parse/train/Hkx6hANtwH/images/31c68df64eb2110e4109736d455d3ea9cdaa31a6f60c0b757ace4acc03bab5d8.jpg b/parse/train/Hkx6hANtwH/images/31c68df64eb2110e4109736d455d3ea9cdaa31a6f60c0b757ace4acc03bab5d8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..65390319cc26eaa9378f0ca113d18201fe4d827a --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/31c68df64eb2110e4109736d455d3ea9cdaa31a6f60c0b757ace4acc03bab5d8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e08ee5c28def65cc583c0d60267721d042c84b730bfc1f16bce013c58d5e6570 +size 17318 diff --git a/parse/train/Hkx6hANtwH/images/76a4914e3ed97a06eaeb1df44f451c9eadd1a690c6cdf2fa9da57ca02985eee1.jpg b/parse/train/Hkx6hANtwH/images/76a4914e3ed97a06eaeb1df44f451c9eadd1a690c6cdf2fa9da57ca02985eee1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9eaaace55d6a8431beccf7b309026143dcc48d56 --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/76a4914e3ed97a06eaeb1df44f451c9eadd1a690c6cdf2fa9da57ca02985eee1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:916e5d41199f92b6b40a4e650c5a3169fdbff1bf65f1e35ee7cc2a70312db8e8 +size 30806 diff --git a/parse/train/Hkx6hANtwH/images/867f9b5b9beb63cf9bb4e93c830c897a5e7d7f696a1ca14d863c872dea0aa122.jpg b/parse/train/Hkx6hANtwH/images/867f9b5b9beb63cf9bb4e93c830c897a5e7d7f696a1ca14d863c872dea0aa122.jpg new file mode 100644 index 0000000000000000000000000000000000000000..750d21b40d4b67c35e6479c10705807f25831ada --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/867f9b5b9beb63cf9bb4e93c830c897a5e7d7f696a1ca14d863c872dea0aa122.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed4b497bb0e7c9f0e5a5f992bd60d13192a2b7cb895a8ee064d7fe688a3181d4 +size 30886 diff --git a/parse/train/Hkx6hANtwH/images/8b366af6a97aff66356802892132639cec7f2785168c58d8368fa5d044531487.jpg b/parse/train/Hkx6hANtwH/images/8b366af6a97aff66356802892132639cec7f2785168c58d8368fa5d044531487.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5c93c765e55d3ee82204c3c2c1875ddac9162436 --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/8b366af6a97aff66356802892132639cec7f2785168c58d8368fa5d044531487.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb1f8e89b4d2d42430fcf26fba585087c3a5215d5edf2fdf77616e0899d78044 +size 7353 diff --git a/parse/train/Hkx6hANtwH/images/9c0591eac7853e28eaa2e94a9188412913e417de04594be75f5393ad2cb70643.jpg b/parse/train/Hkx6hANtwH/images/9c0591eac7853e28eaa2e94a9188412913e417de04594be75f5393ad2cb70643.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0f727e289092b49ca4331e3b250df0d345c0e507 --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/9c0591eac7853e28eaa2e94a9188412913e417de04594be75f5393ad2cb70643.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab1f12c27dd59efab215c0e5b4a0d591a05b1a9185fd50b75c4b2682ec2369f8 +size 31499 diff --git a/parse/train/Hkx6hANtwH/images/adcd764ac59db7fadd7cdac465302b9dd68de82aa6212801e6625211229c7a48.jpg b/parse/train/Hkx6hANtwH/images/adcd764ac59db7fadd7cdac465302b9dd68de82aa6212801e6625211229c7a48.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a334bb7fdbd1888fa21345896e440370087252fe --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/adcd764ac59db7fadd7cdac465302b9dd68de82aa6212801e6625211229c7a48.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67f5dabdf5a6f99c8f1e30f439177d40bd7b27628499ec10fbe3a32e42050a10 +size 60251 diff --git a/parse/train/Hkx6hANtwH/images/b74803a52c6327a0394255bf6d75a77a568af1f66436189ff151a0ca70cb8ef8.jpg b/parse/train/Hkx6hANtwH/images/b74803a52c6327a0394255bf6d75a77a568af1f66436189ff151a0ca70cb8ef8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93f82908e34975f261ccc5751dba0f1f18e754a3 --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/b74803a52c6327a0394255bf6d75a77a568af1f66436189ff151a0ca70cb8ef8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22b0e3b2e307aeb12ac8b876a3b63b8f1e2c355b3c54e574bcfd2235688c630b +size 6617 diff --git a/parse/train/Hkx6hANtwH/images/c050fe5203c0a63cdaa3b27dd12048cc5c4310fdfaf2384df0fc2667797885f5.jpg b/parse/train/Hkx6hANtwH/images/c050fe5203c0a63cdaa3b27dd12048cc5c4310fdfaf2384df0fc2667797885f5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..566b43808f8bfdbc2ae121e91a5fc76f0f8c450e --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/c050fe5203c0a63cdaa3b27dd12048cc5c4310fdfaf2384df0fc2667797885f5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a87571d329edfeae579e6e963797d0ff6c08b7e2f4493ac32d601959544edcec +size 17687 diff --git a/parse/train/Hkx6hANtwH/images/c379e43ee1c078f294bc4ea605359a81dff643be05acb373a176d554360d5b82.jpg b/parse/train/Hkx6hANtwH/images/c379e43ee1c078f294bc4ea605359a81dff643be05acb373a176d554360d5b82.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b6528bcb634151d48b2652a1219b28353468aa6 --- /dev/null +++ b/parse/train/Hkx6hANtwH/images/c379e43ee1c078f294bc4ea605359a81dff643be05acb373a176d554360d5b82.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:160ca86fe91c3dc4d6049b7ba375487d391669852e291392e3fdd90e75b2bef9 +size 31806 diff --git a/parse/train/HyeJf1HKvS/images/0468e18e5419dad1156cfd4e259d1e59bcc17a39d4075fadb8b26c97db19ab11.jpg b/parse/train/HyeJf1HKvS/images/0468e18e5419dad1156cfd4e259d1e59bcc17a39d4075fadb8b26c97db19ab11.jpg new file mode 100644 index 0000000000000000000000000000000000000000..776c8d6692e477ffa703aed87b0a5fd74c7e35e5 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/0468e18e5419dad1156cfd4e259d1e59bcc17a39d4075fadb8b26c97db19ab11.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:376e67171494562e10051444b7708e7eb1eaf293fee762f90295c5ac2e7617fc +size 10229 diff --git a/parse/train/HyeJf1HKvS/images/13beab06402240b0d310b5939f080e92c6fc7d9d3b637e6b71c450d14c9f5a49.jpg b/parse/train/HyeJf1HKvS/images/13beab06402240b0d310b5939f080e92c6fc7d9d3b637e6b71c450d14c9f5a49.jpg new file mode 100644 index 0000000000000000000000000000000000000000..111d916ca9bf807f53dd2877c1adacdb532b5d2b --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/13beab06402240b0d310b5939f080e92c6fc7d9d3b637e6b71c450d14c9f5a49.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8efe63ab45c1177e81ecbb9f78efc852a00492af122dd2ca242d5367d7c25975 +size 128010 diff --git a/parse/train/HyeJf1HKvS/images/2a2883e3c5be6048cc015519ae7237de3ac6fdac5cd609a473183d7bad4155ab.jpg b/parse/train/HyeJf1HKvS/images/2a2883e3c5be6048cc015519ae7237de3ac6fdac5cd609a473183d7bad4155ab.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b4ecbe6ba6660c2d13f8809265efa210a8b41586 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/2a2883e3c5be6048cc015519ae7237de3ac6fdac5cd609a473183d7bad4155ab.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5cdfc571a576ed84be8955d985510b0858dee25ea04bec8a41b009ae1a2a01c7 +size 11744 diff --git a/parse/train/HyeJf1HKvS/images/404e980e78689b2604e8aad2c376cd5ac70b527f72c0c55bd42e6868053dc192.jpg b/parse/train/HyeJf1HKvS/images/404e980e78689b2604e8aad2c376cd5ac70b527f72c0c55bd42e6868053dc192.jpg new file mode 100644 index 0000000000000000000000000000000000000000..62ca22b3b45a2296be9bffae2f8745e691dbf2c1 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/404e980e78689b2604e8aad2c376cd5ac70b527f72c0c55bd42e6868053dc192.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb4d3707ffff22a3e095cda618c9fdaa1b7721f3bb901bdb7e7b6dfda74e2c8b +size 46028 diff --git a/parse/train/HyeJf1HKvS/images/53bf7d6437b8391703593375d96d8ee82820e4dd785307adabe24baba6016609.jpg b/parse/train/HyeJf1HKvS/images/53bf7d6437b8391703593375d96d8ee82820e4dd785307adabe24baba6016609.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c7db889f0797b9b0ac102c07d78a989943d48002 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/53bf7d6437b8391703593375d96d8ee82820e4dd785307adabe24baba6016609.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:407c962c13e07fb64af47dca2405c40355c7d6351bdf6a94871a99d704e8017b +size 16099 diff --git a/parse/train/HyeJf1HKvS/images/548116f0d0fa49f2b6d39da185a39878fe603b78af1fd7c1ddd96364371e2b62.jpg b/parse/train/HyeJf1HKvS/images/548116f0d0fa49f2b6d39da185a39878fe603b78af1fd7c1ddd96364371e2b62.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ec5a7f8a4a7e0c480879a481263e5371963b9d28 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/548116f0d0fa49f2b6d39da185a39878fe603b78af1fd7c1ddd96364371e2b62.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:615234ec9bae66fcc5beef92ba38eb952f3dcedda5b2e148c4fc0021d9bab94b +size 56516 diff --git a/parse/train/HyeJf1HKvS/images/599b206c594e0edc36bae597479ecfb82304b75388da400706918f91b9d8fa4e.jpg b/parse/train/HyeJf1HKvS/images/599b206c594e0edc36bae597479ecfb82304b75388da400706918f91b9d8fa4e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..12ce5deff97064b869888c082c45454bf86d2625 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/599b206c594e0edc36bae597479ecfb82304b75388da400706918f91b9d8fa4e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:541abbac7aaea2ac2e57874a947b25b725fd45417869c7a9fdab961daf1b4a74 +size 64781 diff --git a/parse/train/HyeJf1HKvS/images/63be76c73ecdecaff00e5a4647069501f21ac96b7f640087aff376ebb858edee.jpg b/parse/train/HyeJf1HKvS/images/63be76c73ecdecaff00e5a4647069501f21ac96b7f640087aff376ebb858edee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..565c0ff7194a457b2e7c510d2b3aac916f744d67 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/63be76c73ecdecaff00e5a4647069501f21ac96b7f640087aff376ebb858edee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d18bd23d018f7540abdf7eff68bb4e6dc7b74e74840ed9cd0fadd170988bb591 +size 69530 diff --git a/parse/train/HyeJf1HKvS/images/6c99c40493cbfc26f8dd26256651bce7aaade6eb5b23eb335ee9533603cd845a.jpg b/parse/train/HyeJf1HKvS/images/6c99c40493cbfc26f8dd26256651bce7aaade6eb5b23eb335ee9533603cd845a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db8a4fd42f80984493a5147562249d8a3f06e8d9 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/6c99c40493cbfc26f8dd26256651bce7aaade6eb5b23eb335ee9533603cd845a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c78e9006deb6709f057f7280a355bdc53d83dc0f752f7d65fbf9457c4c92231a +size 30506 diff --git a/parse/train/HyeJf1HKvS/images/70795f70059c7b4ca66cce1daf2efbe0e56e26571cd0f87f0406f5d8b474f9f9.jpg b/parse/train/HyeJf1HKvS/images/70795f70059c7b4ca66cce1daf2efbe0e56e26571cd0f87f0406f5d8b474f9f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f030f9f682f98b9e891ab3196d02c738cdcc2a73 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/70795f70059c7b4ca66cce1daf2efbe0e56e26571cd0f87f0406f5d8b474f9f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee188e60867c22f3360cf84e5e6377b97ea2c6218e5e3bb60aad676ba361ea9d +size 143789 diff --git a/parse/train/HyeJf1HKvS/images/7e35ec444d039d8b099d1c3820861f62074f18c9e84cc08297f8d7644b0cc990.jpg b/parse/train/HyeJf1HKvS/images/7e35ec444d039d8b099d1c3820861f62074f18c9e84cc08297f8d7644b0cc990.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f64c15f3949165c982bf0ae36a82d7639d1b1d55 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/7e35ec444d039d8b099d1c3820861f62074f18c9e84cc08297f8d7644b0cc990.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e296346cbf2a863259567c2ed168da5b4df76785e971cd6fda2af1187b77f923 +size 7538 diff --git a/parse/train/HyeJf1HKvS/images/817f96e5f54841c46f9e0ed0cc739ecd674774c0b7bcae61a45e9c3302a424e9.jpg b/parse/train/HyeJf1HKvS/images/817f96e5f54841c46f9e0ed0cc739ecd674774c0b7bcae61a45e9c3302a424e9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..da4e10b5ff506169455a1adb49992755175a3f9a --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/817f96e5f54841c46f9e0ed0cc739ecd674774c0b7bcae61a45e9c3302a424e9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a586e79a287aa9941d75b63e4ab6911ffa621faa9658cb9bd05a3afd2489e759 +size 8928 diff --git a/parse/train/HyeJf1HKvS/images/8b1c526f84e485700e340f0c180d80ee1e6cbaa6d2bf759219ed6ca70979bd76.jpg b/parse/train/HyeJf1HKvS/images/8b1c526f84e485700e340f0c180d80ee1e6cbaa6d2bf759219ed6ca70979bd76.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e44f13a0dd7b0f690998b644dba91236503f63d8 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/8b1c526f84e485700e340f0c180d80ee1e6cbaa6d2bf759219ed6ca70979bd76.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05b26ce666d216f0101f304c98e7d673a39519a43461d7d2018fb3f1dfc94f34 +size 8688 diff --git a/parse/train/HyeJf1HKvS/images/a0e5898d9d7df30b927e787f931541a787811e8a67f5286c0fd7853a8763a7f3.jpg b/parse/train/HyeJf1HKvS/images/a0e5898d9d7df30b927e787f931541a787811e8a67f5286c0fd7853a8763a7f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..266d384a14ecdb08e3e79c13fd5c8ed41652ee51 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/a0e5898d9d7df30b927e787f931541a787811e8a67f5286c0fd7853a8763a7f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b5c4d5caefb20054a0b98d77f3cc059d8b28f42db5bec71fbb9a20276b9c72e +size 8599 diff --git a/parse/train/HyeJf1HKvS/images/a3b333cc38aad26ac1f4f423f5901789e07d24010b45307963331409e36c52fc.jpg b/parse/train/HyeJf1HKvS/images/a3b333cc38aad26ac1f4f423f5901789e07d24010b45307963331409e36c52fc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f296eb42c53791b8515e38fd809b38449c3c4714 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/a3b333cc38aad26ac1f4f423f5901789e07d24010b45307963331409e36c52fc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4ad625ff5e1293f295bd58ee4111b8d738ac59efb68b568dbb1caf5b77e5f2e +size 13306 diff --git a/parse/train/HyeJf1HKvS/images/adf82f7a4358c9739d020e2cd27cbbaffd776b4919115ca5a87be154185c194b.jpg b/parse/train/HyeJf1HKvS/images/adf82f7a4358c9739d020e2cd27cbbaffd776b4919115ca5a87be154185c194b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f4e1d05e413deeb0db44ed7cee35aab579ac2479 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/adf82f7a4358c9739d020e2cd27cbbaffd776b4919115ca5a87be154185c194b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ac6c925d560bd7ca21414d37dca7757e0314cb7bcb62b04d116a6b09009536b +size 7565 diff --git a/parse/train/HyeJf1HKvS/images/b0523d1f37be373c21da1e1dcc3b14dc934b4e47d589e5b75f9e63fdd3b4f6b7.jpg b/parse/train/HyeJf1HKvS/images/b0523d1f37be373c21da1e1dcc3b14dc934b4e47d589e5b75f9e63fdd3b4f6b7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..73ed0cf82d12a390c6061df30c03c009ef7f2651 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/b0523d1f37be373c21da1e1dcc3b14dc934b4e47d589e5b75f9e63fdd3b4f6b7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:233c40181b5c1f1d45c89e09dd73bfe03df43b17d3c32b99e7913ecaf6ce556a +size 211016 diff --git a/parse/train/HyeJf1HKvS/images/b83508e97025fb611e42b21a88425c0416bec13e223d0ca5d946ab75ffa8c1eb.jpg b/parse/train/HyeJf1HKvS/images/b83508e97025fb611e42b21a88425c0416bec13e223d0ca5d946ab75ffa8c1eb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f2f2d8cd62819e107bc7bc53b802a8481d490545 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/b83508e97025fb611e42b21a88425c0416bec13e223d0ca5d946ab75ffa8c1eb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c152a8124f9c88b263803ab388a21d01ffba8f2dc2359cab62a8494fef8a109e +size 117385 diff --git a/parse/train/HyeJf1HKvS/images/bd9ecad57e28ef5a936826b755ce300f667b7ee489e02ddec66320650f3241ae.jpg b/parse/train/HyeJf1HKvS/images/bd9ecad57e28ef5a936826b755ce300f667b7ee489e02ddec66320650f3241ae.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0d4e39e8d7b6e9f1ffbc9825ada398804cd6e523 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/bd9ecad57e28ef5a936826b755ce300f667b7ee489e02ddec66320650f3241ae.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3795fb7e6ae962b3118bc7c84c6e0734d9a897a3043ba17709b50687d8a14e34 +size 26100 diff --git a/parse/train/HyeJf1HKvS/images/c8d318e4b487c1ada80eedb4e906d33d89f16f7e534a37283c4488cddd3ec3aa.jpg b/parse/train/HyeJf1HKvS/images/c8d318e4b487c1ada80eedb4e906d33d89f16f7e534a37283c4488cddd3ec3aa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e06c499533fa5b277659280909ecfdd139477e1 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/c8d318e4b487c1ada80eedb4e906d33d89f16f7e534a37283c4488cddd3ec3aa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2f8a2aedcc69cf07676dba74918c39cc5f4cdd6693129b22a0a77ad9d81466f +size 7720 diff --git a/parse/train/HyeJf1HKvS/images/d03d720ba20fcef7834445d3c64bb6d990e8d735002be1b7bbc9473bde46614c.jpg b/parse/train/HyeJf1HKvS/images/d03d720ba20fcef7834445d3c64bb6d990e8d735002be1b7bbc9473bde46614c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b1541248d3ea5ce16abe2a97a774051226f75345 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/d03d720ba20fcef7834445d3c64bb6d990e8d735002be1b7bbc9473bde46614c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f2fc39cfed587a72d31a99de142851da082f376f422470b0bca558a8180896e +size 35652 diff --git a/parse/train/HyeJf1HKvS/images/d07deecbef8492f81efbdfef42ace93d5cdb33f45724c7dd74963da1e02ea73e.jpg b/parse/train/HyeJf1HKvS/images/d07deecbef8492f81efbdfef42ace93d5cdb33f45724c7dd74963da1e02ea73e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1e856a6fbf5eb33afb967a14a47cbec335eb98dc --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/d07deecbef8492f81efbdfef42ace93d5cdb33f45724c7dd74963da1e02ea73e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5b861f63083c2d9dd8abcd449717acd84bffa6442b904c155642fcf3acc1213 +size 3728 diff --git a/parse/train/HyeJf1HKvS/images/df4f73bc25ecad10c24fdf6e9b7bff221bd60c1329e2201b138b668efb3ee62a.jpg b/parse/train/HyeJf1HKvS/images/df4f73bc25ecad10c24fdf6e9b7bff221bd60c1329e2201b138b668efb3ee62a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6049f2e26c0e76fca486e04e0e17b41c94cb85d9 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/df4f73bc25ecad10c24fdf6e9b7bff221bd60c1329e2201b138b668efb3ee62a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3eff7bd0a158ccf64f3456811ef7239d7b5b1cf576291693ad28587122c57d1 +size 9875 diff --git a/parse/train/HyeJf1HKvS/images/e6f6d7dccd8d7bff51cfdc9d102060ee1d487e315e018197dbce828cdc711e21.jpg b/parse/train/HyeJf1HKvS/images/e6f6d7dccd8d7bff51cfdc9d102060ee1d487e315e018197dbce828cdc711e21.jpg new file mode 100644 index 0000000000000000000000000000000000000000..180fc8be215b20127188ce6623645c45c0c539a5 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/e6f6d7dccd8d7bff51cfdc9d102060ee1d487e315e018197dbce828cdc711e21.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9861ea60e50fff2d9231567dee5e31b0ac08f565fa7feceda92b3b7d3bf20a80 +size 6719 diff --git a/parse/train/HyeJf1HKvS/images/eb2f2ffa796d7cc12ebf2a24963018acf36573b26d76c7ffbe8b2559ff83fa44.jpg b/parse/train/HyeJf1HKvS/images/eb2f2ffa796d7cc12ebf2a24963018acf36573b26d76c7ffbe8b2559ff83fa44.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7dd0b06bc3cb7184f12945027bef10aa34dcaca7 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/eb2f2ffa796d7cc12ebf2a24963018acf36573b26d76c7ffbe8b2559ff83fa44.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:91ad436ab339e63fedb4f749354de8ac6d3a7c8ceb3e7df3db3f2371451ccf4c +size 8139 diff --git a/parse/train/HyeJf1HKvS/images/f675450bd58e319430ad0b8c3924d5c9d9bae89bb8d96cde3758fa2b3b400702.jpg b/parse/train/HyeJf1HKvS/images/f675450bd58e319430ad0b8c3924d5c9d9bae89bb8d96cde3758fa2b3b400702.jpg new file mode 100644 index 0000000000000000000000000000000000000000..572174feea4cf1c3076f73455f692ee2ea3684dc --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/f675450bd58e319430ad0b8c3924d5c9d9bae89bb8d96cde3758fa2b3b400702.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:571c7509a934afd6c0ff4c28ad74596a722c9fdee1dfe7072c80646b25f45cf6 +size 80441 diff --git a/parse/train/HyeJf1HKvS/images/f702ffe552b281f6d4a39846e226c817afe94ecc987d0c040551bc4bd98692c3.jpg b/parse/train/HyeJf1HKvS/images/f702ffe552b281f6d4a39846e226c817afe94ecc987d0c040551bc4bd98692c3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..04df8ad820b6538c8de35b11a30124bcc058bb25 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/f702ffe552b281f6d4a39846e226c817afe94ecc987d0c040551bc4bd98692c3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02a8804384a467a11074748aa2f76690108b25fa3c28592f4de98b4b9b11f4ad +size 11030 diff --git a/parse/train/HyeJf1HKvS/images/fd162f8bddabe23eae7c0d9bec5748fe877f766df6775a2b9f8542f86c139e6e.jpg b/parse/train/HyeJf1HKvS/images/fd162f8bddabe23eae7c0d9bec5748fe877f766df6775a2b9f8542f86c139e6e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ec9a31d137ce3d90953d67d76c23eb870924caf3 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/fd162f8bddabe23eae7c0d9bec5748fe877f766df6775a2b9f8542f86c139e6e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1e3019dd2d4d8d4faa8ecfa5ef7679e4eacdd02cad3f3b977c651c5a8b78e3d +size 71619 diff --git a/parse/train/HyeJf1HKvS/images/fe71d89acc1d0741d5674b99d71eb12c797281c8ad04e95b7f72ba9078314c58.jpg b/parse/train/HyeJf1HKvS/images/fe71d89acc1d0741d5674b99d71eb12c797281c8ad04e95b7f72ba9078314c58.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c1ab05519de37aa8232f4eb158e8c5c0105aba6 --- /dev/null +++ b/parse/train/HyeJf1HKvS/images/fe71d89acc1d0741d5674b99d71eb12c797281c8ad04e95b7f72ba9078314c58.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b586d5ee5bd3e5600bf00e178ab75514ba8c5b160649828b421a7f1ce52aed6 +size 55197 diff --git a/parse/train/LJjC6DmSkgT/images/02bf370265e6243b99d3e16eb7c0b1722e28c196e9ddfc5d32fc62f7c3c014e1.jpg b/parse/train/LJjC6DmSkgT/images/02bf370265e6243b99d3e16eb7c0b1722e28c196e9ddfc5d32fc62f7c3c014e1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..427aaee7485ee3c42632d5894a5037b07467da7c --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/02bf370265e6243b99d3e16eb7c0b1722e28c196e9ddfc5d32fc62f7c3c014e1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9dd291badc97efd317521fe12a11486085ec58bc8d5f1abe2ad54ace362e456 +size 42820 diff --git a/parse/train/LJjC6DmSkgT/images/0ac0839a77314eef3f415108a944bbd40d71d498bce0e862bb483ebc1931b9d5.jpg b/parse/train/LJjC6DmSkgT/images/0ac0839a77314eef3f415108a944bbd40d71d498bce0e862bb483ebc1931b9d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..122270eb46636040bf11cf0362c843b74937398b --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/0ac0839a77314eef3f415108a944bbd40d71d498bce0e862bb483ebc1931b9d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09a913488ab66033d60c28d5dd4b047899ebd4c4edb4ed07e349f138effd2906 +size 78160 diff --git a/parse/train/LJjC6DmSkgT/images/0f0345f61ac65711d6d305b1481ef67eeec525b956f9ba83de24ca2b119153a6.jpg b/parse/train/LJjC6DmSkgT/images/0f0345f61ac65711d6d305b1481ef67eeec525b956f9ba83de24ca2b119153a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4afa786e135ff2e8671b68c4178fe28244259cc4 --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/0f0345f61ac65711d6d305b1481ef67eeec525b956f9ba83de24ca2b119153a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df5b0118db69bb7e8a7409b04263e4067e7044f0c500fa1527724aeefc8521cb +size 154137 diff --git a/parse/train/LJjC6DmSkgT/images/13b03e1aec38cbde4911d46835b11fd8b31a69fd07dffdd977f563adc068d68c.jpg b/parse/train/LJjC6DmSkgT/images/13b03e1aec38cbde4911d46835b11fd8b31a69fd07dffdd977f563adc068d68c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e2287977a7580ab46689bca557206daf5f2f69ea --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/13b03e1aec38cbde4911d46835b11fd8b31a69fd07dffdd977f563adc068d68c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e62d1125a486c5908f8e80ba08d2786c9de90c95563070b937ca20d5ac543da +size 20459 diff --git a/parse/train/LJjC6DmSkgT/images/15e7c403758902bb5ea5868ea8692568be4f804673d45808fabf30d3b81f2ef4.jpg b/parse/train/LJjC6DmSkgT/images/15e7c403758902bb5ea5868ea8692568be4f804673d45808fabf30d3b81f2ef4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f7f82d19cc3239327956f8aa8caf30433c81e0ad --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/15e7c403758902bb5ea5868ea8692568be4f804673d45808fabf30d3b81f2ef4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7492f169fd44fa147586c524f54e6920f4bf2d997dcaa6bd0866748e11ef770e +size 5467 diff --git a/parse/train/LJjC6DmSkgT/images/17e4a056bfa4a79f052b0fa9a2bda2d98a0218cf22b906687e416b47a2b6930a.jpg b/parse/train/LJjC6DmSkgT/images/17e4a056bfa4a79f052b0fa9a2bda2d98a0218cf22b906687e416b47a2b6930a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b9f61b2fae9d53116c11b38f020b89d04b583f8a --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/17e4a056bfa4a79f052b0fa9a2bda2d98a0218cf22b906687e416b47a2b6930a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:55c543de98f076ac1799168d6b1670ff217b0de5d5f6b01bc5f04023d551fc73 +size 6567 diff --git a/parse/train/LJjC6DmSkgT/images/33a49ca74a1f9f86d0a220ac3ac14cbacf032abb693e719e68854815f2c8c6a9.jpg b/parse/train/LJjC6DmSkgT/images/33a49ca74a1f9f86d0a220ac3ac14cbacf032abb693e719e68854815f2c8c6a9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..09d32e9c37156b3a141636922297e878966f7d42 --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/33a49ca74a1f9f86d0a220ac3ac14cbacf032abb693e719e68854815f2c8c6a9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f85a2efeda8f47b792d1a5074b45d35dd50f72dc436d87acd1182544369ab71b +size 8087 diff --git a/parse/train/LJjC6DmSkgT/images/367b2877a665e3be00acd4276a4e3836222c4f9175a235dad79236f82ab43ec4.jpg b/parse/train/LJjC6DmSkgT/images/367b2877a665e3be00acd4276a4e3836222c4f9175a235dad79236f82ab43ec4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3263ba1035a57207d4f1191481eb05990ce131ad --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/367b2877a665e3be00acd4276a4e3836222c4f9175a235dad79236f82ab43ec4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ed7056f81640a7d74cc7b026a9f03aaf4bdc521a181dc22edbceaaeb6dbd426 +size 2578 diff --git a/parse/train/LJjC6DmSkgT/images/637696b9b75108d43c64f5d8d33137e3bcbb1fc4d129dee043a64ee8e9ade9f4.jpg b/parse/train/LJjC6DmSkgT/images/637696b9b75108d43c64f5d8d33137e3bcbb1fc4d129dee043a64ee8e9ade9f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f0c3ee6d619b0a3b11884615af54613631b6432 --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/637696b9b75108d43c64f5d8d33137e3bcbb1fc4d129dee043a64ee8e9ade9f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:34b06e8b6d26b0dd69e825d763ebeeaef500b0b58c71b1ffc3ff01cf3c7c25bf +size 38980 diff --git a/parse/train/LJjC6DmSkgT/images/79e5610f1b14f8153e55f838afeb9202b2b770be5adfa97cac2fc32f449df0f7.jpg b/parse/train/LJjC6DmSkgT/images/79e5610f1b14f8153e55f838afeb9202b2b770be5adfa97cac2fc32f449df0f7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..908245bf1c6d06dadc69a8ad7ac98a37dd101440 --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/79e5610f1b14f8153e55f838afeb9202b2b770be5adfa97cac2fc32f449df0f7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c3a0802c901127a022fe345ab0331a330f1355db1269d77ae5e34445464e9f50 +size 46785 diff --git a/parse/train/LJjC6DmSkgT/images/8867454a7883ef58e59dde60ff69c591f6b66f78c2f2ac5b4e2b6f4612a905dc.jpg b/parse/train/LJjC6DmSkgT/images/8867454a7883ef58e59dde60ff69c591f6b66f78c2f2ac5b4e2b6f4612a905dc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a19ab90b78eddd65b2c474a2ad9e8e72d2ce6e70 --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/8867454a7883ef58e59dde60ff69c591f6b66f78c2f2ac5b4e2b6f4612a905dc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d278a740d26e23a93d871adb4b33c24a748a5ac8476fdc5c8c308a61af0a46a +size 8629 diff --git a/parse/train/LJjC6DmSkgT/images/c54db01831735e8c0e806f91a31128c9c5a56951978e8b26f9c6c8a513e8e144.jpg b/parse/train/LJjC6DmSkgT/images/c54db01831735e8c0e806f91a31128c9c5a56951978e8b26f9c6c8a513e8e144.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b57d0143eb4768695bf03cd87256356455591b3a --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/c54db01831735e8c0e806f91a31128c9c5a56951978e8b26f9c6c8a513e8e144.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3335d8bea946ed9fecaf0aee839f246d81dff44778f585cc2f9b8420c5fb68cb +size 8600 diff --git a/parse/train/LJjC6DmSkgT/images/cee7ef3efed08b2b5b6c43a2e93b94d054c7453ba4a529fb6d49724686ed0844.jpg b/parse/train/LJjC6DmSkgT/images/cee7ef3efed08b2b5b6c43a2e93b94d054c7453ba4a529fb6d49724686ed0844.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ae76d58f7d0c8815f6ef295002811744d0f26b39 --- /dev/null +++ b/parse/train/LJjC6DmSkgT/images/cee7ef3efed08b2b5b6c43a2e93b94d054c7453ba4a529fb6d49724686ed0844.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7948174f7d6284ae09d95f1d997703de1dd186b2d9e06b22ca242a88ad8d6bd +size 14615 diff --git a/parse/train/S1ejj64YvS/S1ejj64YvS_content_list.json b/parse/train/S1ejj64YvS/S1ejj64YvS_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..6032a8186d7c1489eee18404eef392eb3f1c8d0d --- /dev/null +++ b/parse/train/S1ejj64YvS/S1ejj64YvS_content_list.json @@ -0,0 +1,2325 @@ +[ + { + "type": "text", + "text": "GOOD SEMI-SUPERVISED VAE REQUIRES TIGHTEREVIDENCE LOWER BOUND", + "text_level": 1, + "bbox": [ + 176, + 98, + 821, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Anonymous authors Paper under double-blind review ", + "bbox": [ + 183, + 171, + 398, + 198 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 234, + 544, + 251 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Semi-supervised learning approaches based on generative models have now encountered 3 challenges: (1) The two-stage training strategy is not robust. (2) Good semi-supervised learning results and good generative performance can not be obtained at the same time. (3) Even at the expense of sacrificing generative performance, the semi-supervised classification results are still not satisfactory. To address these problems, we propose One-stage Semi-suPervised Optimal Transport VAE (OSPOT-VAE), a one-stage deep generative model that theoretically unifies the generation and classification loss in one ELBO framework and achieves a tighter ELBO by applying the optimal transport scheme to the distribution of latent variables. We show that with tighter ELBO, our OSPOT-VAE surpasses the best semi-supervised generative models by a large margin across many benchmark datasets. For example, we reduce the error rate from $1 4 . 4 1 \\%$ to $6 . 1 1 \\%$ on Cifar-10 with 4k labels and achieve state-of-the-art performance with $2 5 . 3 0 \\%$ on Cifar-100 with 10k labels. We also demonstrate that good generative models and semi-supervised results can be achieved simultaneously by OSPOT-VAE. ", + "bbox": [ + 233, + 267, + 764, + 474 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 502, + 336, + 518 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The rise of deep neural networks has led to breakthroughs in computer vision, natural language processing, and many other domains. Most of these models are trained on large labeled datasets via supervised learning. However, in many scenarios, although it is easy to acquire a large amount of the original data, obtaining corresponding labels is often very costly or even infeasible. Semisupervised learning (Thomas, 2009) is proposed to address this problem by training classifiers with sufficient unlabeled data and a small fraction of labeled data. ", + "bbox": [ + 174, + 534, + 825, + 617 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Recent works on semi-supervised learning can be grouped into three categories: (1) disagreement based learning via data perturbation (Miyato et al., 2019) and consistency enforcing (Verma et al., 2019), (2) metric learning (Wu et al., 2018), (3) generative approaches via generative adversarial network (GAN) (Springenberg, 2016) and variational autoencoder (VAE) (Kingma et al., 2014). Compared with the first two categories, generative approaches have great advantages in interpretability. Based on the latent variable assumption (Doersch, 2016), the generative model has an explicit variational inference form, so it can learn the marginal probability distribution of the raw data as well as the conditional distribution of the latent variables given the input data, which makes predictions more reasonable. Besides, generative approaches not only learn the required classification representations, but also capture the semantics-disentangled factors that generate the data, making it easier to generalize to different tasks (Narayanaswamy et al., 2017). ", + "bbox": [ + 174, + 625, + 825, + 777 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "However, in practice, semi-supervised generative approaches often encounter three major challenges: (1) The two-stage training process is not robust. Semi-supervised VAE (Kingma et al., 2014) needs to be trained carefully with a two-stage hierarchical strategy, while the training process of GAN is a two-stage adversarial game (Chrysos et al., 2019). (2) Good semi-supervised learning results and good generative performance can not be obtained at the same time. In GAN, good semi-supervised learning performance will lead to a mismatch between the generated results and the real data distribution (Dai et al., 2017). While in VAE, the evidence lower bound (ELBO) objective is irrelevant to the classification loss, making it difficult to learn from the labels directly (Narayanaswamy et al., 2017). (3) Even at the expense of sacrificing generative performance, the semi-supervised classification results are still not satisfactory. In practice, disagreement-based methods (Xie et al., 2019; Berthelot et al., 2019) have dramatically improved the state-of-the-art results on several standard datasets, surpassing generative approaches by a large margin. These challenges naturally raise a question: What limits the performance of generative approaches in semi-supervised learning? ", + "bbox": [ + 174, + 785, + 825, + 924 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/6a7096b8d04a9c60b237cc335b2a0bad726f4db3bc04839eec2139f9c99e967e.jpg", + "image_caption": [ + "Figure 1: The schematic of OSPOT-VAE " + ], + "image_footnote": [], + "bbox": [ + 173, + 65, + 825, + 207 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 242, + 825, + 299 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this work, we propose One-stage Semi-suPervised Optimal Transport VAE (OSPOT-VAE) to address these challenges, which consists of two improvements: (1) a one-stage semi-supervised VAE model that unifies the generation and classification loss in one ELBO framework. (2) an estimation of the margin between true log-likelihood and the ELBO that exports a tighter evidence lower bound by applying optimal transport (Ambrosio & Gigli, 2013) scheme to the distribution of latent variables. ", + "bbox": [ + 174, + 305, + 825, + 388 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Our model has the following contributions: ", + "bbox": [ + 176, + 396, + 457, + 411 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• We show that OSPOT-VAE can be well trained with a direct one-stage strategy. • We show that OSPOT-VAE can achieve both good generative performance and semisupervised learning results simultaneously on a series of benchmark datasets. • We point out that it is the large margin between the ELBO and the log-likelihood of the input data that limits the performance of semi-supervised VAE. Besides, we evaluate this assumption across many standard datasets and show that with the proposed tighter ELBO, OSPOT-VAE surpasses the best semi-supervised generative models by a large margin and achieves state-of-the-art performance on Cifar-100 with 10k labels. ", + "bbox": [ + 215, + 424, + 825, + 546 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 SEMI-SUPERVISED LEARNING METHODS ", + "text_level": 1, + "bbox": [ + 174, + 571, + 547, + 588 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In supervised learning (SL), we are facing with training data that appears as input-target pairs $( \\mathbf { X } , \\bar { \\mathbf { y } } ) \\ \\in \\ \\mathbb { D } _ { L }$ sampled from an unknown distribution $p ( \\mathbf { X } , \\mathbf { y } )$ . Our goal is to learn a function $f ( \\mathbf { X } ; \\phi )$ parameterized by $\\phi$ that makes the correct inference $\\mathbf { y }$ for unseen samples from $p ( \\mathbf { X } )$ . While in semi-supervised learning (SSL), we can obtain an extra collection of unlabeled data $\\mathbf { X } \\in \\mathbb { D } _ { U }$ sampled from the same distribution $p ( \\mathbf { X } )$ . We hope to leverage the data from both $\\mathbb { D } _ { L }$ and $\\mathbb { D } _ { U }$ to achieve a more accurate model than what would have been obtained by only using $\\mathbb { D } _ { L }$ . ", + "bbox": [ + 174, + 604, + 825, + 688 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this section, we review some existing methods for SSL. We mainly focus on those who have reached state-of-the-art results, as well as generative approaches which are strongly connected with our model; the more comprehensive overview is beyond the scope of this paper, we refer readers to (Oliver et al., 2018). ", + "bbox": [ + 174, + 695, + 825, + 751 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.1 DISAGREEMENT BASED LEARNING ", + "text_level": 1, + "bbox": [ + 176, + 767, + 460, + 782 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Disagreement-based learning refers to the general approaches of imposing disagreement among multiple learners on the same task or multiple predictions from a single learner. By eliminating the disagreement, we can enforce the generalization of the model on unseen data. A common technique for creating disagreement is data augmentation, which applies transformations or perturbations on the input data and leaves class semantics unchanged. For $\\mathbf { X } \\in \\mathbb { D } _ { U }$ , loss term can be derived as ", + "bbox": [ + 174, + 794, + 825, + 864 + ], + "page_idx": 1 + }, + { + "type": "equation", + "img_path": "images/35a0f9cac4ba95ade39d9324cca98481124ca123c67fcfaa3c6e63eb03b2a5ef.jpg", + "text": "$$\n\\| f ( \\operatorname { A u g m e n t } ( \\mathbf { X } ) ; \\phi ) - f ( \\mathbf { X } ; \\phi ) \\| _ { 2 } ^ { 2 }\n$$", + "text_format": "latex", + "bbox": [ + 382, + 871, + 614, + 888 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "where the Augment $( \\mathbf { X } )$ is a stochastic function which can be obtained by image transformation (Xie et al., 2019), virtual adversarial training (Miyato et al., 2019), or mixup method (Verma et al. 2018; ", + "bbox": [ + 174, + 895, + 825, + 924 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Berthelot et al. 2019). Another disagreement construction technique is to train multiple learners on the same dataset and utilize the loss ", + "bbox": [ + 171, + 103, + 823, + 132 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/0c004f168acc94a1992074a86420002524f703ff025de068d36c9ba7bc30ae81.jpg", + "text": "$$\n\\| f ( \\mathbf { X } ; \\boldsymbol { \\phi } _ { 1 } ) - f ( \\mathbf { X } ; \\boldsymbol { \\phi } _ { 2 } ) \\| _ { 2 } ^ { 2 }\n$$", + "text_format": "latex", + "bbox": [ + 411, + 133, + 586, + 152 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "to enforce the predictive consistency of different models, for example, “Mean Teacher” (Tarvainen & Valpola, 2017) and “Teacher Graph” (Luo et al., 2018). The generalization of the models gets enhanced. ", + "bbox": [ + 174, + 154, + 826, + 196 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.2 GENERATIVE APPROACHES", + "text_level": 1, + "bbox": [ + 176, + 212, + 403, + 227 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In generative approaches, input $\\mathbf { X }$ is supposed to have corresponding continuous and discrete latent variables, which we denote by $\\mathbf { z }$ and c respectively. ", + "bbox": [ + 173, + 238, + 823, + 267 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Feature matching (FM) GANs (Salimans et al., 2016; Dai et al., 2017) apply GANs to semisupervised learning on K-classification tasks by specifying a $( \\mathsf { K } { + } 1 )$ -class objective for the discriminator. Instead of binary classification, true samples are classified into the first K classes respectively and fake samples are classified into the $( \\mathsf { K } { + } 1 )$ -th class. This target function achieves strong empirical results by matching the generator distribution with true data distribution and improves semi-supervised classification performance. ", + "bbox": [ + 173, + 272, + 825, + 358 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Semi-supervised VAEs (Kingma et al. 2014; Narayanaswamy et al. 2017) construct a probabilistic model parameterized by $\\pmb \\theta$ and $\\phi$ that respectively describe the generation and inference process between $\\mathbf { X }$ and latent variables $\\mathbf { z }$ , c. The generation process of $\\mathbf { X }$ by $\\mathbf { z }$ and $\\mathbf { c }$ is : ", + "bbox": [ + 173, + 364, + 823, + 406 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/6a0e722f34fadd3f21f6961319e121414b4e995300bd18aa4894b8b3fe107b3d.jpg", + "text": "$$\np ( \\mathbf { z } ) = { \\mathcal { N } } ( z ; \\mathbf { 0 } , I ) ; \\qquad p ( \\mathbf { c } ) = \\mathbf { M } \\mathbf { u } \\mathbf { l } \\mathbf { t } ( \\mathbf { c } ; K , \\pi ) ; \\qquad p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) = f ( \\mathbf { X } ; \\mathbf { z } , \\mathbf { c } , \\theta )\n$$", + "text_format": "latex", + "bbox": [ + 235, + 409, + 764, + 426 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\operatorname { M u l t } ( K , \\pi )$ is the multinomial distribution with class $K$ and parameter $\\pi$ . $f ( \\mathbf { X } ; \\mathbf { z } , \\mathbf { c } , \\theta )$ is a suitable likelihood function, e.g. a Bernoulli or Gaussian distribution, parameterized by a non-linear transformation of the latent variables $\\mathbf { z }$ and $\\mathbf { c }$ . The class label $\\mathbf { y }$ is treated as c if given. For the inference process, with the following hypothesis ", + "bbox": [ + 174, + 429, + 825, + 484 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/7eeab28a239b62ea0da33b0354545d840db957c2a80c354b5f78cb9df6883931.jpg", + "text": "$$\n\\begin{array} { r } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) = q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) ; \\qquad p ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) = p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) ; \\qquad p ( \\mathbf { z } , \\mathbf { c } ) = p ( \\mathbf { z } ) p ( \\mathbf { c } ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 191, + 487, + 789, + 505 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "evidence lower bound (ELBO) is used as objective to predict the posterior distribution of latent variables as follows (see Appendix A.1 for proof): ", + "bbox": [ + 173, + 507, + 823, + 535 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/75ddcd8bbd9da9ee1832cdb364751c3d55dd3b48cb65b32a7cc354352c9cb2f7.jpg", + "text": "$$\n\\log p ( \\mathbf { X } ) \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } ) ) = \\mathrm { E L B O } ( \\phi ) ,\n$$", + "text_format": "latex", + "bbox": [ + 181, + 537, + 823, + 556 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "For the likelihood $\\log p ( \\mathbf { X } )$ is infeasible, VAE maximizes its evidence lower bound instead, which derives the negative ELBO loss function $\\mathcal { L } ( \\mathbf { X } ; \\boldsymbol { \\phi } , \\pmb { \\theta } ) = - \\mathrm { E L B O }$ . The predictions for the classification label $\\mathbf { y }$ can be obtained from the inferred posterior distribution $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ . When class label $\\mathbf { y }$ is not given, missing label sampling technique is used to sample from $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ as ", + "bbox": [ + 174, + 566, + 825, + 623 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/4f1c9d93744e0c9b9802fb463e544d7fb080013f2aff04c0adf6d01fa83e9490.jpg", + "text": "$$\n\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } f ( \\mathbf { X } ; \\mathbf { c } , \\theta ) = \\sum _ { k = 1 } ^ { K } q _ { \\phi } ( \\mathbf { y } _ { k } | \\mathbf { X } ) f ( \\mathbf { X } ; \\mathbf { y } _ { k } , \\theta )\n$$", + "text_format": "latex", + "bbox": [ + 336, + 626, + 661, + 670 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Here $\\mathbf { y } _ { k }$ represents a one-hot vector with 1 in $\\mathbf { k }$ -th dimension. Utilizing this sampling method, the algorithmic complexity of VAE is proportional to $\\mathrm { K }$ so it is computationally inefficient. Note that in objective (4), the label predictive distribution $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ only contributes to the generative performance. To remedy this, existing models simply add a cross-entropy loss to the negative ELBO loss such that the distribution $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ can also learn classification rules from the labeled data. The extended objective loss is ", + "bbox": [ + 173, + 671, + 825, + 755 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/923d4e4c4856ff8134d5593b9e08d387ca6c768d3fb16dd65fd0b636372c03b2.jpg", + "text": "$$\n\\operatorname* { m i n } _ { \\phi , \\theta } \\mathbb { E } _ { \\mathbf { X } \\sim \\mathbb { D } _ { U } } \\mathcal { L } \\big ( \\mathbf { X } ; \\phi , \\pmb { \\theta } ) + \\mathbb { E } _ { ( \\mathbf { X } , \\mathbf { y } ) \\sim \\mathbb { D } _ { L } } \\big [ \\mathcal { L } \\big ( \\mathbf { X } , \\mathbf { c } = \\mathbf { y } ; \\phi , \\pmb { \\theta } \\big ) - \\log q _ { \\phi } ( \\mathbf { y } | \\mathbf { X } ) \\big ]\n$$", + "text_format": "latex", + "bbox": [ + 258, + 757, + 741, + 782 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Two-stage Training Strategy: In practice, (Kingma et al., 2014) finds that directly training the one-stage objective (7) will lead to a bad semi-supervised learning result, so a two-stage training strategy is proposed to improve the model. The two-stage training strategy consists of two parts, M1 and M2. M1 means to learn a new continuous latent representation $\\mathbf { z } _ { 1 }$ first, and M2 means to train a semi-supervised model (7) with the embedding $\\mathbf { z } _ { 1 }$ from M1 instead of the raw data $\\mathbf { X }$ . This $\\mathbf { M } 1 { + } \\mathbf { M } 2$ strategy builds a deep VAE with two layers of random variables: $\\begin{array} { r l r } { \\mathrm { \\nabla } p _ { \\pmb \\theta } ( \\mathbf { X } , \\mathbf { z } _ { 1 } , \\mathbf { z } _ { 2 } , \\mathbf { c } ) } & { { } = } & { } \\end{array}$ $p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { z } _ { 1 } ) p _ { \\pmb { \\theta } } ( \\mathbf { z } _ { 1 } | \\mathbf { z } _ { 2 } , \\bar { \\mathbf { c } } ) p ( \\mathbf { z } _ { 2 } ) p ( \\mathbf { c } )$ , which can dramatically improve the performance of the inference $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ but is not robust in training. Moreover, GAN’s training process can also be considered as two-stage with generator and discriminator competing with each other, and the two-stage adversarial game adds the difficulty in training. ", + "bbox": [ + 173, + 784, + 825, + 924 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 ONE-STAGE SEMI-SUPERVISED OPTIMAL TRANSPORT VAE", + "text_level": 1, + "bbox": [ + 173, + 102, + 702, + 118 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this section, we introduce our semi-supervised VAE framework, OSPOT-VAE. Firstly, we derive a one-stage loss function that unifies the generation and classification loss under one ELBO without introducing any additional auxiliary loss items like (7). Then, we analyze a phenomenon that good ELBO values do not guarantee good semi-supervised performance and propose the optimal transport estimation to deal with it. At last, combining the two parts, we give the detailed algorithm of OSPOTVAE and discuss some problems in model optimization. ", + "bbox": [ + 173, + 131, + 825, + 217 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.1 ONE-STAGE SEMI-SUPERVISED VAE", + "text_level": 1, + "bbox": [ + 178, + 232, + 468, + 247 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Following the notations and assumptions $( 3 , 4 )$ in Section 2.2, we derive our one-stage semisupervised VAE. With the empirical distribution $p _ { e m p } ( \\mathbf { X } ; \\mathbb { D } ) = { \\frac { 1 } { | \\mathbb { D } | } } \\sum _ { \\mathbf { X } ^ { \\prime } \\in \\mathbb { D } } \\mathbf { 1 } _ { \\mathbf { X } = \\mathbf { X } ^ { \\prime } }$ , we utilize the decomposition in (Zhao et al., 2017) and rewrite the second part of (5) into (proof in Appendix A.2) ", + "bbox": [ + 173, + 257, + 825, + 321 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/e45412571be7347757b52de94b85bc837cd0ab42234dead9a5dca5f5e646be14.jpg", + "text": "$$\n\\begin{array} { r } { \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) \\big ) = \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } ) \\| p ( \\mathbf { z } ) \\big ) \\geq \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 238, + 323, + 758, + 342 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $q _ { \\phi } ( \\mathbf { z } ) = \\frac { 1 } { | \\mathbb { D } | } \\sum _ { \\mathbf { X } \\in \\mathbb { D } } q _ { \\phi } ( \\mathbf { z } | x )$ and $\\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } )$ is the mutual information between $\\mathbf { X }$ and $\\mathbf { z }$ . The left part of (8) equals to 0 when $\\mathbf { X }$ and $\\mathbf { z }$ are independent. This is undesirable, so $\\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } )$ can be regarded as the lower bound of controlled mutual information. The continuous variables in (8) can be easily extend to discrete variables $\\mathbf { c }$ . We can use $\\mathbf { I _ { z } }$ and $\\mathbf { I _ { c } }$ to denote the controlled information capacity and derive the objective for the unlabeled dataset $\\mathbb { D } _ { U }$ ", + "bbox": [ + 173, + 343, + 825, + 435 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/e47bd738aba74b86eb37673f2a87cc738f4e6ff9a0f4d33604e11a454b465a51.jpg", + "text": "$$\n\\begin{array} { r l } & { { \\mathcal { L } } _ { \\mathbb { D } _ { U } } ( { \\mathbf { X } } ; \\pmb { \\theta } , \\phi ) = { \\mathbb { E } } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] + \\beta _ { \\mathbf { z } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } ) - \\mathbf { I } _ { \\mathbf { z } } | } \\\\ & { \\quad \\quad \\quad \\quad + \\beta _ { \\mathbf { c } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { c } ) ) - \\mathbf { I } _ { \\mathbf { c } } | } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 238, + 435, + 759, + 474 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $\\beta , \\mathbf { I _ { z } } , \\mathbf { I _ { c } }$ are all hyper-parameters forcing the KL divergence term to match the mutual information capacities of $\\mathbf { z }$ and $\\mathbf { c }$ . ", + "bbox": [ + 173, + 474, + 825, + 502 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "For the labeled subset $\\mathbb { D } _ { L }$ , instead of directly employing class label y as sampled $\\mathbf { c }$ , we view it as the parameter of the true posterior distribution, i.e. $p ( \\mathbf { c } | \\mathbf { X } ) = \\mathbf { M u l t } ( \\mathbf { c } ; K , \\mathbf { y } )$ and derive the following one-stage ELBO form: ", + "bbox": [ + 174, + 508, + 825, + 551 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/f80090e70218c819d33a52d82e700b959d22225248b220c0e7f82186a6bafc88.jpg", + "text": "$$\n\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\log \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } [ \\log p ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) + \\log \\frac { p ( \\mathbf { z } ) p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } ] = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\log p ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) } \\\\ & { - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( p ( \\mathbf { c } | \\mathbf { X } ) \\| q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) ) + \\mathbb { E } _ { p ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 191, + 553, + 779, + 659 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Notice that $D _ { \\mathrm { K L } } ( p ( \\mathbf { c } | \\mathbf { X } ) \\| q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) )$ is equal to the common cross-entropy loss for y is a one-hot vector. In this respect, the margin between $p ( \\mathbf { c } | \\mathbf { X } )$ and $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ can be significantly small when the suitable optimization method is chosen. This allows us to utilize the approximation $p ( \\mathbf { c } | \\mathbf { X } ) \\approx$ $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ to modify $\\mathbb { E } _ { p ( \\mathbf { c } | \\mathbf { X } ) } \\log { \\frac { p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } }$ in (10), resulting in a consist ELBO with $\\mathcal { L } _ { \\mathbb { D } _ { U } } ( \\mathbf { X } ; \\pmb { \\theta } , \\phi )$ : ", + "bbox": [ + 173, + 659, + 825, + 734 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/2351c089caa35b04cd2b3ae813eaf36c669d3069f9c420f6e0484be262265240.jpg", + "text": "$$\n\\mathbb { E } _ { p ( { \\mathbf { c } } | { \\mathbf { X } } ) } \\log { \\frac { p ( { \\mathbf { c } } ) } { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } } \\approx { \\mathrm { ( } } { \\mathrm { w h e n ~ } } p ( { \\mathbf { c } } | { \\mathbf { X } } ) \\approx q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) { \\mathrm { ) } } \\mathbb { E } _ { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } \\log { \\frac { p ( { \\mathbf { c } } ) } { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } } = D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { c } } ) )\n$$", + "text_format": "latex", + "bbox": [ + 184, + 734, + 787, + 768 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Combining the ELBO form (10) of $\\mathbb { D } _ { L }$ with the mutual information decomposition (8) and the approximation (11), the new objective for semi-supervised VAE is: ", + "bbox": [ + 174, + 771, + 823, + 797 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/e66145bbe8ea7f846663612151d4d0a5731458eddb81a69f9f408a7e32e6c97a.jpg", + "text": "$$\n\\begin{array} { r l } & { { \\mathcal { L } } _ { \\mathbb { D } _ { L } } ( { \\mathbf { X } } , { \\mathbf { y } } ; \\pmb { \\theta } , \\phi ) = { \\mathbb { E } } _ { q _ { \\phi } ( { \\mathbf { z } } | { \\mathbf { X } } ) , p ( { \\mathbf { c } } | { \\mathbf { X } } ) } [ - \\log p _ { \\theta } ( { \\mathbf { X } } | { \\mathbf { z } } , { \\mathbf { c } } ) ] + \\beta _ { \\mathbf { z } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { z } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { z } } ) ) - { \\mathbf { I } } _ { \\mathbf { z } } | } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad + \\beta _ { \\mathbf { c } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { c } } ) ) - { \\mathbf { I } } _ { \\mathbf { c } } | + D _ { \\mathrm { K L } } ( p ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 197, + 799, + 774, + 839 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "With (9) and (12), the objective for the entire dataset is now ", + "bbox": [ + 173, + 838, + 571, + 853 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/bb2c07a93d0d6b41efa21c4dd7c9218009935c23ad85692a25d6cc9527ef7399.jpg", + "text": "$$\n\\operatorname* { m i n } _ { \\phi , \\theta } \\mathbb { E } _ { \\mathbf { X } \\sim p _ { e m p } ( \\mathbf { X } ; \\mathbb { D } _ { U } ) } \\mathcal { L } _ { \\mathbb { D } _ { U } } ( \\mathbf { X } ; \\theta , \\phi ) + \\mathbb { E } _ { ( \\mathbf { X } , \\mathbf { y } ) \\sim p _ { e m p } ( ( \\mathbf { X } , \\mathbf { y } ) ; \\mathbb { D } _ { L } ) } \\mathcal { L } _ { \\mathbb { D } _ { L } } ( \\mathbf { X } , \\mathbf { y } ; \\theta , \\phi )\n$$", + "text_format": "latex", + "bbox": [ + 246, + 854, + 754, + 880 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "This one-stage objective with a simple approximate transformation (9) unifies the generation loss as well as the target of SSL and results in improved performance of semi-supervised learning, which we demonstrate in Section 4.1. ", + "bbox": [ + 174, + 881, + 823, + 924 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Algorithm 1 Optimal transport estimation ingests a batch of observation $\\mathbf { X }$ as well as the representation $q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } )$ , $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ inferred from the original VAE and returns the estimation of the margin $D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) )$ and $D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { \\bar { c } } | \\mathbf { X } ) )$ . ", + "bbox": [ + 174, + 102, + 823, + 147 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Input: ", + "text_level": 1, + "bbox": [ + 178, + 152, + 223, + 165 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Batch of observation $\\mathbf { X }$ sampled from $p _ { e m p } ( \\mathbf { X } )$ ; Inferred parameter $( \\mu , \\mathrm { d i a g } ( \\sigma ^ { 2 } ) )$ of $q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) = \\mathcal { N } ( \\mathbf { z } ; \\pmb { \\mu } , \\mathrm { d i a g } ( \\pmb { \\sigma } ^ { 2 } ) ) ;$ ; Inferred parameter $\\pi$ of $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) = \\mathbf { M } \\mathbf { u } \\mathrm { l t } ( \\mathbf { c } ; K , \\pi )$ ; Hyperparameter $\\alpha$ for mixup vicinal distribution $p _ { m i x u p } ( \\mathbf { X } )$ ", + "bbox": [ + 200, + 165, + 648, + 223 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Output: ", + "text_level": 1, + "bbox": [ + 179, + 223, + 236, + 237 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "$\\tilde { \\mathbf { X } }$ sampled from $p _ { m i x u p } ( \\mathbf { X } )$ ; Estimation $L _ { M \\mathbf { z } }$ of the margin $D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) | | p ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) )$ Estimation $L _ { M _ { \\mathbf { c } } }$ of the margin $D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) | | p ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) )$ 1: $: \\mathrm { \\bf ~ X } ^ { \\prime } , \\pi ^ { \\prime } , \\mu ^ { \\prime } , \\sigma ^ { \\prime 2 } = \\mathrm { \\bf F }$ andomPermutation $( \\mathbf { X } , \\pi , \\mu , \\sigma ^ { 2 } )$ 2: $\\tilde { \\mathbf { X } } = \\lambda * \\mathbf { X } + \\left( 1 - \\lambda \\right) * \\mathbf { X } ^ { \\prime }$ , $\\lambda \\in \\beta ( \\alpha , \\alpha )$ 3: π˜ = OptimalTransportC $\\prime ( \\pi , \\pi ^ { \\prime } , \\lambda )$ 4: $( \\tilde { \\mu } , \\tilde { \\sigma } ^ { 2 } ) =$ OptimalTransportZ((µ, σ2), (µ0, σ02), λ) 5: $q _ { \\phi } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) , q _ { \\phi } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) = \\mathrm { V A E } ( \\tilde { \\mathbf { X } } )$ 6: $\\tilde { p } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) = \\mathcal { N } ( z ; \\tilde { \\mu } , \\mathrm { d i a g } ( \\tilde { \\pmb { \\sigma } } ^ { 2 } ) )$ 7: $\\tilde { p } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) = \\mathbf { M u l t } ( \\mathbf { c } ; K , \\tilde { \\pi } )$ 8: ${ \\cal L } _ { M _ { \\mathbf { z } } } = { \\cal D } _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { z } } | \\tilde { \\mathbf { X } } ) | | \\tilde { p } ( { \\mathbf { z } } | \\tilde { \\mathbf { X } } ) )$ 9: ${ \\cal L } _ { M _ { \\mathbf { c } } } = { \\cal D } _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | \\tilde { \\mathbf { X } } ) | | \\tilde { p } ( { \\mathbf { c } } | \\tilde { \\mathbf { X } } ) )$ 10: return $\\tilde { \\mathbf { X } } , L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } }$ ", + "bbox": [ + 181, + 237, + 565, + 443 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.2 OPTIMAL TRANSPORT ESTIMATION ", + "text_level": 1, + "bbox": [ + 176, + 472, + 460, + 486 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "To summarize the above, VAE aims to learn the useful representation $q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } )$ and $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ by reducing the KL divergence between the empirical distribution $p _ { e m p } ( \\mathbf { X } )$ and the model marginal $\\begin{array} { r } { p ( \\mathbf { X } ) \\ { \\stackrel { } { = } } \\ \\int _ { \\mathbf { z } } \\int _ { \\mathbf { c } } p ( \\mathbf { X } ) p ( \\mathbf { \\tilde { z } } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) d \\mathbf { z } d \\mathbf { c } } \\end{array}$ . Instead of minimizing $\\hat { D _ { \\mathrm { K L } } } ( p _ { e m p } ( \\mathbf { X } ) | | p ( \\mathbf { X } ) )$ directly, VAE models use the expected ELBO mentioned in (5) as target via the following inequality ", + "bbox": [ + 173, + 497, + 825, + 556 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/6d5acdd29c8437cdc290bb9221d06570314efa5a1c575c569f47582b6d75ef48.jpg", + "text": "$$\nD _ { \\mathrm { K L } } ( p _ { e m p } ( \\mathbf { X } ) \\| p ( \\mathbf { X } ) ) \\leq H ( p _ { e m p } ( \\mathbf { X } ) ) - \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\mathrm { E L B O }\n$$", + "text_format": "latex", + "bbox": [ + 305, + 563, + 692, + 582 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "However, one phenomenon is that good ELBO values do not imply accurate inference. A typical example has been discussed in (Zhao et al., 2017). Here we mainly focus on the cause of this phenomenon and propose optimal transport estimation to alleviate this problem in semi-supervised learning. Following the work in (Rezende et al., 2014), we write down the closed form of the expected margin between true log-likelihood and ELBO as (proof in Appendix A.3): ", + "bbox": [ + 173, + 585, + 825, + 656 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/1094a5ca0aeba83fb1bfdd67768ecac528b1fc67dc9103847c2d257a00e6ad3c.jpg", + "text": "$$\n\\mathbb { E } _ { p _ { \\epsilon m p } ( \\mathbf { X } ) } [ \\log p ( \\mathbf { X } ) - \\mathrm { E L B O } ] = \\mathbb { E } _ { p _ { \\epsilon m p } ( \\mathbf { X } ) } [ D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { c } | \\mathbf { X } ) ) ]\n$$", + "text_format": "latex", + "bbox": [ + 186, + 661, + 810, + 681 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Combined with the decomposition (8), training the expected ELBO target can only reduce the difference between marginal distributions $q _ { \\phi } ( \\mathbf { c } ) , \\bar { q } _ { \\phi } ( \\mathbf { z } )$ and $p ( \\mathbf { c } ) \\mathbf { , } p ( \\mathbf { z } )$ . It means that even with a good ELBO, the margin $\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\dot { \\mathbf { X } } ) \\big | \\big | p ( \\dot { \\mathbf { z } } | \\mathbf { X } ) \\big )$ and $\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) \\big )$ in (15) can still be large. In this scenario, the consistent optimization of ELBO will contribute no more to the semi-supervised classification performance. However, optimizing the margin in (15) directly is impossible, for $p ( \\mathbf { c } | \\mathbf { X } )$ and $p ( \\mathbf { z } | \\mathbf { X } )$ are unknown. To remedy this, we extend the empirically effective approximation in (Zhang et al., 2018) to our VAE framework with the form ", + "bbox": [ + 173, + 690, + 825, + 790 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/cc4ef86cc3bbd69ae4053aa0af711278880e6e00b9f74d53d807949a3f913750.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) \\approx \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) ( \\alpha \\to 0 ) } \\\\ & { \\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) \\approx \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) ( \\alpha \\to 0 ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 230, + 794, + 766, + 835 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $p _ { m i x u p } ( \\mathbf { X } )$ is the mixup vicinal distribution (Zhang et al., 2018) and $\\alpha$ is the related parameter. Then we propose optimal transport estimation to construct the estimations of $\\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\big | \\big | p \\big ( \\mathbf { z } | \\bar { \\mathbf { X } } \\big ) \\big )$ as wellariables s $\\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\big | \\big | p ( \\mathbf { z } | \\mathbf { X } ) \\big )$ by applying op-ptimal transport $\\mathbf { z }$ $\\mathbf { c }$ \nestimation are provided in Algorithm 1, and we present the details of the optimal transport scheme in the rest of this section. ", + "bbox": [ + 173, + 839, + 825, + 924 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Algorithm 2 OSPOT-VAE training process with epoch $t$ ", + "text_level": 1, + "bbox": [ + 174, + 103, + 547, + 118 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "text_level": 1, + "bbox": [ + 178, + 125, + 223, + 137 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Batch of labeled pairs $( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ) \\in \\mathbb { D } _ { L }$ ,Batch of unlabeled examples $\\mathbf { X } _ { U } \\in \\mathbb { D } _ { U }$ ; \nELBO hyperparameters: $\\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } = \\mathrm { E L B O S c h e d u l e r } ( t )$ ; \nOptimal transport estimation weights: $w _ { M _ { \\mathbf { z } } }$ $M _ { \\mathbf { z } } \\mathbf { , } w _ { } M _ { \\mathbf { c } } =$ WeightScheduler $\\cdot ( t )$ ; \nModel parameters: $\\pmb \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) }$ ; \nModel optimizer: SGD ", + "bbox": [ + 200, + 136, + 725, + 208 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Output: ", + "text_level": 1, + "bbox": [ + 178, + 208, + 236, + 220 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "1: Updated parameters: $\\begin{array} { r l } & { \\mathrm { U p d a t e d ~ p a r a m e t e r s : ~ } \\theta ^ { \\mathrm { { t x } ^ { \\prime } } } , \\phi ^ { \\mathrm { { t x } ^ { \\prime } } } , \\phi ^ { \\mathrm { { t x } ^ { \\prime } } } } \\\\ & { L _ { L } = \\mathcal { L } _ { \\mathbb { D } _ { L } } \\big ( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ; \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big ) } \\\\ & { L _ { U } = \\mathcal { L } _ { \\mathbb { D } _ { U } } \\big ( \\mathbf { X } _ { U } ; \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big ) } \\\\ & { L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } } = \\mathrm { O p t i m a l T r a n s p o r t E s t i m a t i o n } \\big ( \\mathbf { X } _ { U } , q _ { \\phi } \\big ( \\mathbf { z } \\big | \\mathbf { X } _ { U } \\big ) , q _ { \\phi } \\big ( \\mathbf { c } | \\mathbf { X } _ { U } \\big ) \\big ) } \\\\ & { L = L _ { L } + L _ { U } + w _ { M _ { \\mathbf { z } } } L _ { M _ { \\mathbf { z } } } + w _ { M _ { \\mathbf { c } } } L _ { M _ { \\mathbf { c } } } } \\\\ & { \\theta ^ { ( t ) } , \\phi ^ { ( t ) } = \\mathrm { S G D } \\big ( \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } , \\frac { \\partial L } { \\partial \\theta } , \\frac { \\partial L } { \\partial \\phi } \\big ) } \\end{array}$ $\\pmb { \\theta } ^ { ( t ) } , \\phi ^ { ( t ) }$ \n2: \n3: \n4: \n5: \n6: return $\\theta ^ { ( t ) } , \\phi ^ { ( t ) }$ ", + "bbox": [ + 179, + 219, + 671, + 330 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Optimal Transport Scheme: The mixup vicinal distribution can be understood as applying linear transport between the points $\\mathbf { X } , \\mathbf { X ^ { \\prime } } \\in \\mathbb { D }$ , extending the original dataset with new points falling on one straight line $\\tilde { \\mathbf { X } } = \\lambda * \\mathbf { X } + ( 1 - \\lambda ) * \\mathbf { X ^ { \\prime } } , \\lambda \\in [ 0 , 1 ] .$ . For $\\mathbf { \\tilde { X } }$ , it is a natural thought that this linear transformation could associate with the shortest-path transport in the latent space. Based on this, we calculate the distributions $\\tilde { p } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) , \\tilde { p } ( \\mathbf { c } | \\tilde { \\mathbf { X } } )$ of $\\mathbf { z } , \\mathbf { c }$ and consider them as the estimation of the true posterior distributions. Following the work of (Ambrosio & Gigli, 2013), the norm-2 based optimal transport scheme $\\gamma ( \\mathbf { x } , \\mathbf { y } )$ between two distributions $p ( \\mathbf { x } )$ and $p ( \\mathsf { y } )$ satisfy: ", + "bbox": [ + 173, + 354, + 825, + 458 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/2116513f91ae226591325da222f39f13d12ceacb6040877ced75229faef559bb.jpg", + "text": "$$\n\\begin{array} { c } { { \\displaystyle \\operatorname* { i n f } _ { \\gamma ( \\mathbf { x } , \\mathbf { y } ) } \\int _ { \\mathbf { x } } \\int _ { \\mathbf { y } } \\| \\mathbf { x } - \\mathbf { y } \\| _ { 2 } ^ { 2 } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { x } d \\mathbf { y } } } \\\\ { { \\mathrm { s . t . } ~ \\displaystyle \\int _ { \\mathbf { y } } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { y } = p ( \\mathbf { x } ) ; \\displaystyle \\int _ { \\mathbf { x } } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { x } = p ( \\mathbf { y } ) } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 349, + 460, + 650, + 530 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "For the continuous variable $\\mathbf z \\sim \\mathcal N ( \\pmb \\mu , \\mathrm { d i a g } ( \\pmb \\sigma ^ { 2 } ) )$ and discrete variable $\\mathbf { c } \\sim \\mathbf { M } \\mathbf { u } \\mathrm { l t } ( K , \\pi )$ , the following 2 propositions are proposed to calculate the shortest-path based on optimal transport scheme (see Appendix A.4 for proof). ", + "bbox": [ + 174, + 534, + 823, + 577 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Proposition 3.1. The shortest-path derived from optimal transport scheme (17) between $\\mathbf { z } _ { 1 } ~ \\sim$ $\\mathcal { N } ( \\bar { \\mu } _ { 1 } , d i a g ( \\sigma _ { 1 } ^ { 2 } ) )$ and $\\mathbf { z } _ { 2 } \\sim \\mathcal { N } ( \\bar { \\pmb { \\mu } } _ { 2 } , d i a g ( \\pmb { \\sigma } _ { 2 } ^ { 2 } ) )$ with $\\lambda \\in [ 0 , 1 ]$ is ", + "bbox": [ + 171, + 579, + 821, + 608 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/c10fbe1c072f505ca83c0c1cb24197e36de3bdeb0bb5ba0043666b4eef2fcefd.jpg", + "text": "$$\n\\begin{array} { r } { \\tilde { \\pmb { \\mu } } = \\lambda \\pmb { \\mu } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\mu } _ { 2 } } \\\\ { \\tilde { \\pmb { \\sigma } } = \\lambda \\pmb { \\sigma } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\sigma } _ { 2 } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 419, + 609, + 576, + 647 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Proposition 3.2. The shortest-path derived from $K L$ divergence based optimal transport scheme between $\\mathbf { c } _ { 1 } \\sim M u l t ( K , \\pmb { \\pi } _ { 1 } )$ and $\\mathbf { c } _ { 2 } \\sim M u l t ( K , \\pmb { \\pi } _ { 2 } )$ with $\\lambda \\in [ 0 , 1 ]$ is ", + "bbox": [ + 171, + 647, + 826, + 678 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/aa7a6f003d5e3a5a5f6228d01b90d7176fa15f0bf5e398fef355eecf2ef68797.jpg", + "text": "$$\n\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 }\n$$", + "text_format": "latex", + "bbox": [ + 421, + 679, + 575, + 696 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Algorithm 1 yields the optimal transport estimation of the margin in (15), which leads to a tighter ELBO. In Section 4.2, we demonstrate that with this tighter ELBO, the inference performance of semi-supervised VAE is significantly improved on many benchmark datasets. ", + "bbox": [ + 174, + 707, + 825, + 750 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "3.3 OPTIMIZATION OF OSPOT-VAE ", + "text_level": 1, + "bbox": [ + 176, + 765, + 437, + 780 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Combining one-stage semi-supervised VAE and optimal transport estimation, we can get the complete OSPOT-VAE model. The full OSPOT-VAE algorithm is provided in Algorithm 2, and a schematic is shown in Figure 1. Note that the conditions for the approximations used in Algorithm 1,2 satisfy (1) $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\approx p ( \\mathbf { c } | \\mathbf { X } )$ and (2) the VAE model has already achieved a good ELBO. Therefore, the warm-up schedule (Higgins et al., 2017) is used to set parameters $\\mathbf { I } _ { z } , \\mathbf { I } _ { c } , \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } }$ and $w _ { M _ { \\mathbf { z } } } , w _ { M _ { \\mathbf { c } } }$ . We list the details of “ELBOScheduler $( t )$ ” and “WeightScheduler $( t ) ^ { , }$ in Appendix A.5. ", + "bbox": [ + 173, + 790, + 825, + 876 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In Algorithm 2, we apply stochastic gradient descent (SGD) as optimizer, which needs to calculate the gradient $\\nabla _ { \\pmb { \\theta } , \\pmb { \\phi } } L$ . The target loss $L$ consists of KL divergence and the expected loglikelihood . The derivation of KL divergence part has a closed form, while calculating the gradient ", + "bbox": [ + 176, + 881, + 823, + 924 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/0fbc7c4661ac493feb70faa4b6cb7b05b60e960895adf465d9253ef3c448e160.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
MNIST(100 labels)SVHN(1k labels)
BackBoneMethod
Same with M1+M2Disentangled VAE9.71(±0.91)38.91(±1.06)
(Narayanaswamy et al., 2017)11.97(±1.71)54.33(±0.11)
M1(Kingma et al., 2014)3.33(±0.14)36.02(±0.10)
M1+M2(Kingma et al., 2014)
One-stage VAE3.14(±0.19)27.38(±0.78)
", + "bbox": [ + 173, + 47, + 828, + 154 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/dad9f1da926bf83694743684e6abbbe6639a58b1d56565be7fe305c5166c18ee.jpg", + "table_caption": [ + "Table 1: One-stage VAE error rate in MNIST and SVHN. " + ], + "table_footnote": [], + "table_body": "
BackBoneModel categoryModelCifar10(4k labels)
WRN-28-2DisagreementTemporal Ensembling(TE) (Laine & Aila,2017)16.37
Mean Teacher(Tarvainen & Valpola,2017)15.87
13.13
VAT+EntMin(Miyato et al., 2019) MixMatch(Berthelot et al., 2019)6.37
GS-BadGANt*(Li et al., 2019)17.11
WRN-28-10 GenerativeGenerativeOSPOT-VAE8.51(±0.32)
AutoAugment(Cubuk et al., 2019)14.1
DisagreementTemporal Ensembing(Laine & Aila, 2017)12.16
MixMatch*(Berthelot et al., 2019)4.95
GS-BadGANt*(Li et al., 2019) GAN combine TE‡*(Wei et al., 2018)14.41
", + "bbox": [ + 173, + 195, + 849, + 401 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Table 2: Error rate in Cifar10. $\\dagger$ denotes the best semi-supervised generative approach result. $\\ddagger$ denotes the model ensemble two categories. $^ *$ denotes the corresponding backbone is not exactly WideResNet (Zagoruyko & Komodakis, 2016), but belongs to one kind of its variations with a comparable amount of parameters. ", + "bbox": [ + 173, + 410, + 826, + 467 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "of $\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } )$ is difficult. To this end, we follow the work of (Rezende et al., 2014) and (Jang et al., 2017), using the reparameterization trick as ", + "bbox": [ + 173, + 474, + 825, + 503 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/2f16d44efa6c310194e924f6fce7a5608d4e7b7cdb4543c02f19e655fc237739.jpg", + "text": "$$\n\\begin{array} { r l } & { \\nabla _ { \\theta , \\phi } \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } ) = \\mathbb { E } _ { \\mathcal { N } ( \\epsilon ; \\mathbf { 0 } , \\mathbf { I } ) } \\nabla _ { \\theta , \\phi } \\log p _ { \\theta } ( \\mathbf { X } | \\mu + \\sigma \\cdot \\epsilon ) } \\\\ & { \\mathbb { E } _ { \\mathrm { G u m b e l } ( \\epsilon ; \\mathbf { 0 } , \\mathbf { 1 } ) } ) \\nabla _ { \\theta , \\phi } \\log p _ { \\theta } ( \\mathbf { X } | \\mathrm { S o f t m a x } ( \\frac { \\log \\pi + \\epsilon } { \\tau } ) ) \\to \\nabla _ { \\theta , \\phi } \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } ) ( \\tau \\to 0 ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 183, + 508, + 784, + 560 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Note that with (20), the algorithmic complexity of one-stage semi-supervised VAE is independent with the class number $K$ , making it easier to extend to large-scale classification tasks. ", + "bbox": [ + 173, + 564, + 825, + 593 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4 EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 176, + 613, + 326, + 628 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In this section, we demonstrate the 3 contributions of our OSPOT-VAE model with sufficient experiments on 4 standard SSL benchmark datasets, that is, MNIST, SVHN, Cifar10, and Cifar100. In Section 4.1, we show the validity of our one-stage semi-supervised VAE objective (13) by comparing with other one-stage and two-stage VAE models. Then, we evaluate the performance of OSPOT-VAE under “WideResNet”(Zagoruyko & Komodakis, 2016) backbone and compare with other state-of-the-art SSL models mentioned in Section 2. Besides, We provide an ablation study to verify the contribution of the optimal transport estimation. As an additional application, we show that good generative models and semi-supervised results can be obtained at the same time by OSPOT-VAE (Section 4.3). The source code is available at https: //github.com/PaperCodeSubmission/OSPOT-VAE; more details are available in Appendix A.6. ", + "bbox": [ + 173, + 643, + 825, + 797 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.1 ONE-STAGE SEMI-SUPERVISED VAE", + "text_level": 1, + "bbox": [ + 178, + 814, + 468, + 828 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We evaluate the effectiveness of the one-stage semi-supervised VAE objective on 2 standard benchmarks, MNIST and SVHN. As for baseline models, we consider two VAE-based SSL models, which are one-stage disentangled VAE (Narayanaswamy et al., 2017) and two-stage VAE $[ \\mathbf { M } 1 + \\mathbf { M } 2 ]$ ) (Kingma et al., 2014). For fairness, except the target loss functions, all models use the same structure as is used in $\\mathbf { M } 1 { + } \\mathbf { M } 2$ (Kingma et al., 2014). The results are presented in Table 1, and our model achieves the best performance. ", + "bbox": [ + 174, + 839, + 825, + 924 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/df43f7645c30e3e246c8d138916fed7f8b94841ff7bba869abee8e5251a4c8b1.jpg", + "table_caption": [ + "Table 3: Error in Cifar100. $\\dagger$ and $^ *$ have the same meaning as described in Table 2. " + ], + "table_footnote": [], + "table_body": "
BackBoneModelCifar100(4k labels)Cifar100(10k labels)
WRN-28-2II - Model(Laine & Aila, 2017)39.19
GS-BadGANt*(Li et al., 2019)45.1137.16
LP*(Iscen et al., 2019)43.7335.92
OSPOT-VAE40.58(±0.48)31.41(±0.21)
WRN-28-10MixMatch* (Berthelot et al., 2019)25.88
OSPOT-VAE33.76(±0.53)25.30(±0.31)
", + "bbox": [ + 173, + 87, + 839, + 195 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/4a6868c1c37140527432e50a5891d56788d6ea26e5419e19e71f238f827f019e.jpg", + "table_caption": [ + "Table 4: Ablation study with SVHN, Cifar10, and Cifar100. " + ], + "table_footnote": [], + "table_body": "
MethodsSVHN(1k labels)Cifar10(4k labels)Cifar100(10k labels)
One-stage VAE10.53(±0.17)18.26(±0.51)38.62(±0.67)
Optimal transport estimation (with encoder only)6.54(±0.62)10.71(±0.44)36.21(±0.29)
OSPOT-VAE5.79(±0.15)8.51(±0.32)31.41(±0.21)
", + "bbox": [ + 174, + 231, + 826, + 306 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.2 OSPOT-VAE ", + "text_level": 1, + "bbox": [ + 174, + 316, + 308, + 330 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We compare the results of OSPOT-VAE with two categories of state-of-the-art models mentioned in Section2. In all experiments, we use the “WideResNet-28” model or other deep models with a comparable amount of parameters as the backbone. The results in Table 2,3 demonstrate that our model outperforms most of the existing methods and surpasses state-of-the-art semi-supervised generative models (Dai et al., 2017) by a large margin. Notice that recently, data-augmentation based method, MixMatch, (Berthelot et al., 2019) achieves the absolute state-of-the-art results in all benchmarks. It uses pre-designed sophisticated data augmentation strategies for different datasets and outperforms our model. We list its results fairly as a comparison, while OSPOT-VAE surpasses it in Cifar100 dataset. ", + "bbox": [ + 173, + 342, + 825, + 468 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Ablation Study: The OSPOT-VAE model consists of two parts: (1) a one-stage VAE objective and (2) an optimal transport estimation. In ablation study, we analyze the effect of each component in our model with the backbone “WideResNet-28-2”. To study the independent effects of transport estimation, we combine it with the encoder part of OSPOT-VAE to build a classifier with loss function $L _ { M _ { \\mathbf { c } } }$ . The improved classification error rates in Table 4 show that, with optimal transport estimation, the posterior inference $q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )$ gets closer to the true distribution $p ( \\mathbf { c } | \\mathbf { X } )$ . It indicates that our optimal transport estimation does reduce the gap between ELBO and the log-likelihood of the input data and yield a tighter ELBO, which leads to a better semi-supervised performance. ", + "bbox": [ + 173, + 473, + 825, + 587 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/9ce60c9faf8f59fa59253a8299b0174ce2fbbee584b15e9749f147ca0247b678.jpg", + "table_caption": [ + "Table 5: Generative performance measured by ELBO with EL $\\mathbf { B O } \\leq \\log p ( \\mathbf { X } )$ " + ], + "table_footnote": [], + "table_body": "
ModelCifar10Cifar100
Pure VAE-226.25(±14.25)−1292.91(±1.10)
OSPOT-VAE-237.62(±6.27)-1271.82(±24.15)
", + "bbox": [ + 295, + 616, + 696, + 664 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.3 GENERATIVE PERFORMANCE ", + "text_level": 1, + "bbox": [ + 174, + 679, + 413, + 693 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "$\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\mathrm { E L B O }$ measures the margin between the true data distribution and the distribution learned by generation models (Doersch, 2016). By comparing the $\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) }$ value of pure unsupervised VAE and our semi-supervised VAE model under the same “WideResNet-28-2” backbone, we demonstrate that good generative models and semi-supervised results can be obtained at the same time in OSPOTVAE. The results in Table 5 show that the data generative distribution learned by our OSPOT-VAE model is as good as the pure VAE model. Further generated results are available in Appendix A.7. ", + "bbox": [ + 173, + 704, + 825, + 789 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5 CONCLUSION ", + "text_level": 1, + "bbox": [ + 174, + 809, + 318, + 825 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In this work, we pointed out that it was the large margin between ELBO and the true log-likelihood of the raw data that limits the performance of semi-supervised VAE. To this end, we introduced OSPOT-VAE, a one-stage generative model that unified the classification and generation objective and achieved a tighter ELBO by optimal transport estimation. We demonstrated our assertion through extensive experiments, and our semi-supervised results significantly outperform former state-of-the-art generative SSL methods by a large margin on Cifar10 and Cifar100. ", + "bbox": [ + 174, + 839, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 174, + 102, + 287, + 118 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Luigi Ambrosio and Nicola Gigli. A users guide to optimal transport. 2013. ", + "bbox": [ + 173, + 126, + 671, + 142 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "David Berthelot, Nicholas Carlini, Ian J. Goodfellow, Nicolas Papernot, Avital Oliver, and Colin Raffel. Mixmatch: A holistic approach to semi-supervised learning. CoRR, abs/1905.02249, 2019. URL http://arxiv.org/abs/1905.02249. ", + "bbox": [ + 178, + 152, + 823, + 195 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Grigorios G. Chrysos, Jean Kossaifi, and Stefanos Zafeiriou. Robust conditional generative adversarial networks. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019, 2019. URL https://openreview.net/forum?id= Byg0DsCqYQ. ", + "bbox": [ + 174, + 205, + 825, + 262 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Ekin D. Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasudevan, and Quoc V. Le. Autoaugment: Learning augmentation strategies from data. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2019. ", + "bbox": [ + 173, + 273, + 826, + 315 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Zihang Dai, Zhilin Yang, Fan Yang, William W. Cohen, and Ruslan Salakhutdinov. Good semi-supervised learning that requires a bad GAN. In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, 4-9 December 2017, Long Beach, CA, USA, pp. 6510–6520, 2017. URL http://papers.nips.cc/paper/ 7229-good-semi-supervised-learning-that-requires-a-bad-gan. ", + "bbox": [ + 173, + 327, + 825, + 411 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Carl Doersch. Tutorial on variational autoencoders. CoRR, abs/1606.05908, 2016. URL http: //arxiv.org/abs/1606.05908. ", + "bbox": [ + 174, + 422, + 820, + 452 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Irina Higgins, Lo¨ıc Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick, Shakir Mohamed, and Alexander Lerchner. beta-vae: Learning basic visual concepts with a constrained variational framework. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017. URL https://openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ Sy2fzU9gl. ", + "bbox": [ + 173, + 462, + 825, + 534 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, and Ondrej Chum. Label propagation for deep semisupervised learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5070–5079, 2019. ", + "bbox": [ + 174, + 544, + 825, + 587 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24- 26, 2017, Conference Track Proceedings, 2017. URL https://openreview.net/forum? id $=$ rkE3y85ee. ", + "bbox": [ + 173, + 597, + 825, + 654 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Diederik P. Kingma, Shakir Mohamed, Danilo Jimenez Rezende, and Max Welling. Semisupervised learning with deep generative models. In Advances in Neural Information Processing Systems 27: Annual Conference on Neural Information Processing Systems 2014, December 8-13 2014, Montreal, Quebec, Canada, pp. 3581–3589, 2014. URL http://papers.nips.cc/ paper/5352-semi-supervised-learning-with-deep-generative-models. ", + "bbox": [ + 173, + 665, + 825, + 736 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Max Kuang and Esteban G. Tabak. Preconditioning of optimal transport. SIAM J. Scientific Computing, 39(4), 2017. doi: 10.1137/16M1074953. URL https://doi.org/10.1137/ 16M1074953. ", + "bbox": [ + 173, + 746, + 825, + 789 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017. URL https://openreview.net/forum? id $=$ BJ6oOfqge. ", + "bbox": [ + 173, + 800, + 825, + 856 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Wenyuan Li, Zichen Wang, Jiayun Li, Jennifer Polson, William Speier, and Corey W. Arnold. Semisupervised learning based on generative adversarial network: a comparison between good GAN and bad GAN approach. CoRR, abs/1905.06484, 2019. URL http://arxiv.org/abs/ 1905.06484. ", + "bbox": [ + 174, + 867, + 825, + 922 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Yucen Luo, Jun Zhu, Mengxi Li, Yong Ren, and Bo Zhang. Smooth neighbors on teacher graphs for semi-supervised learning. In 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018, pp. 8896–8905, 2018. doi: 10.1109/ CVPR.2018.00927. URL http://openaccess.thecvf.com/content_cvpr_2018/ html/Luo_Smooth_Neighbors_on_CVPR_2018_paper.html. ", + "bbox": [ + 176, + 103, + 825, + 174 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Takeru Miyato, Shin-ichi Maeda, Masanori Koyama, and Shin Ishii. Virtual adversarial training: A regularization method for supervised and semi-supervised learning. IEEE Trans. Pattern Anal. Mach. Intell., 41(8):1979–1993, 2019. doi: 10.1109/TPAMI.2018.2858821. URL https:// doi.org/10.1109/TPAMI.2018.2858821. ", + "bbox": [ + 174, + 183, + 825, + 239 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Siddharth Narayanaswamy, Brooks Paige, Jan-Willem van de Meent, Alban Desmaison, Noah D. Goodman, Pushmeet Kohli, Frank D. Wood, and Philip H. S. Torr. Learning disentangled representations with semi-supervised deep generative models. In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, 4-9 December 2017, Long Beach, CA, USA, pp. 5925–5935, 2017. ", + "bbox": [ + 173, + 248, + 825, + 319 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Avital Oliver, Augustus Odena, Colin Raffel, Ekin D. Cubuk, and Ian J. Goodfellow. Realistic evaluation of semi-supervised learning algorithms. In 6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Workshop Track Proceedings, 2018. URL https://openreview.net/forum?id $=$ ByCZsFyPf. ", + "bbox": [ + 173, + 328, + 825, + 385 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and approximate inference in deep generative models. In Proceedings of the 31th International Conference on Machine Learning, ICML 2014, Beijing, China, 21-26 June 2014, pp. 1278–1286, 2014. URL http://proceedings.mlr.press/v32/rezende14.html. ", + "bbox": [ + 173, + 393, + 825, + 450 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, Xi Chen, and Xi Chen. Improved techniques for training gans. In D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon, and R. Garnett (eds.), Advances in Neural Information Processing Systems 29, pp. 2234–2242. Curran Associates, Inc., 2016. URL http://papers.nips.cc/paper/ 6125-improved-techniques-for-training-gans.pdf. ", + "bbox": [ + 174, + 459, + 825, + 530 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Jost Tobias Springenberg. Unsupervised and semi-supervised learning with categorical generative adversarial networks. In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings, 2016. URL http: //arxiv.org/abs/1511.06390. ", + "bbox": [ + 174, + 539, + 825, + 594 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Workshop Track Proceedings, 2017. URL https://openreview.net/forum?id $=$ ry8u21rtl. ", + "bbox": [ + 174, + 604, + 825, + 661 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Philippe Thomas. Semi-supervised learning by olivier chapelle, bernhard scholkopf, and alexander ¨ zien (review). IEEE Trans. Neural Networks, 20(3):542, 2009. doi: 10.1109/TNN.2009.2015974. URL https://doi.org/10.1109/TNN.2009.2015974. ", + "bbox": [ + 174, + 670, + 825, + 713 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Vikas Verma, Alex Lamb, Christopher Beckham, Aaron C. Courville, Ioannis Mitliagkas, and Yoshua Bengio. Manifold mixup: Encouraging meaningful on-manifold interpolation as a regularizer. CoRR, abs/1806.05236, 2018. URL http://arxiv.org/abs/1806.05236. ", + "bbox": [ + 174, + 722, + 825, + 765 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Vikas Verma, Alex Lamb, Juho Kannala, Yoshua Bengio, and David Lopez-Paz. Interpolation consistency training for semi-supervised learning. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI 2019, Macao, China, August 10-16, 2019, pp. 3635–3641, 2019. doi: 10.24963/ijcai.2019/504. URL https://doi.org/10. 24963/ijcai.2019/504. ", + "bbox": [ + 173, + 773, + 825, + 844 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Xiang Wei, Boqing Gong, Zixia Liu, Wei Lu, and Liqiang Wang. Improving the improved training of wasserstein gans: A consistency term and its dual effect. In 6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings, 2018. URL https://openreview.net/forum? id=SJx9GQb0-. ", + "bbox": [ + 174, + 854, + 825, + 922 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Zhirong Wu, Yuanjun Xiong, Stella X. Yu, and Dahua Lin. Unsupervised feature learning via non-parametric instance discrimination. In 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018, pp. 3733–3742, 2018. doi: 10.1109/CVPR.2018. 00393. URL http://openaccess.thecvf.com/content_cvpr_2018/html/Wu_ Unsupervised_Feature_Learning_CVPR_2018_paper.html. ", + "bbox": [ + 173, + 103, + 825, + 188 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Qizhe Xie, Zihang Dai, Eduard H. Hovy, Minh-Thang Luong, and Quoc V. Le. Unsupervised data augmentation. CoRR, abs/1904.12848, 2019. URL http://arxiv.org/abs/1904. 12848. ", + "bbox": [ + 174, + 195, + 821, + 237 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. CoRR, abs/1605.07146, 2016. URL http://arxiv.org/abs/1605.07146. ", + "bbox": [ + 174, + 244, + 820, + 273 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Hongyi Zhang, Moustapha Cisse, Yann N. Dauphin, and David Lopez-Paz. mixup: Beyond empiri- ´ cal risk minimization. In 6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings, 2018. URL https://openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ r1Ddp1-Rb. ", + "bbox": [ + 173, + 280, + 825, + 338 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Shengjia Zhao, Jiaming Song, and Stefano Ermon. Infovae: Information maximizing variational autoencoders. CoRR, abs/1706.02262, 2017. URL http://arxiv.org/abs/1706.02262. ", + "bbox": [ + 174, + 345, + 820, + 373 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "A APPENDIX ", + "text_level": 1, + "bbox": [ + 176, + 398, + 299, + 414 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "A.1 BASIC INEQUALITY OF ELBO ", + "bbox": [ + 176, + 428, + 428, + 444 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Proposition A.1. The Basic inequality of ELBO is ", + "bbox": [ + 174, + 454, + 506, + 469 + ], + "page_idx": 10 + }, + { + "type": "equation", + "img_path": "images/612d808a01d0d89ef90412fd6c19d21ab0340f377dd2441e51d376bbb90036ee.jpg", + "text": "$$\n\\log p ( \\mathbf { X } ) \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } ) )\n$$", + "text_format": "latex", + "bbox": [ + 210, + 473, + 787, + 491 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "proof ", + "bbox": [ + 173, + 501, + 215, + 515 + ], + "page_idx": 10 + }, + { + "type": "equation", + "img_path": "images/7b06614e163105f312f4bdf8734e569068cbd46e884e30ae0efe770b28018490.jpg", + "text": "$$\n\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\log \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } = \\log \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\log \\frac { p ( \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } + \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) } \\\\ & { = - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) | p ( \\mathbf { z } , \\mathbf { c } ) ) + \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] } \\\\ & { = \\mathrm { w i t h ~ a s s u m p t i o n } \\left( 3 . 4 \\right) \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | p ( \\mathbf { c } ) ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 536, + 794, + 686 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "A.2 DECOMPOSITION OF ELBO IN INFO-VAE", + "text_level": 1, + "bbox": [ + 174, + 696, + 509, + 712 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Proposition A.2. The expected ELBO with empirical distribution satisfies ", + "bbox": [ + 176, + 722, + 663, + 738 + ], + "page_idx": 10 + }, + { + "type": "equation", + "img_path": "images/b1363b4acbb9e8505530319cfe147d33224841d2d1effe447c6b47d4d9ee17b1.jpg", + "text": "$$\n\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) \\big ) = \\mathbf { I } _ { q _ { \\phi } } \\big ( \\mathbf { X } ; \\mathbf { z } \\big ) + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } ) \\| p ( \\mathbf { z } ) \\big )\n$$", + "text_format": "latex", + "bbox": [ + 282, + 742, + 714, + 760 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "proof ", + "bbox": [ + 173, + 768, + 215, + 784 + ], + "page_idx": 10 + }, + { + "type": "equation", + "img_path": "images/f78348420b8b221d1f87f4a036fd463c20b56ccf333cfc0d107bdef15cdc595f.jpg", + "text": "$$\n\\begin{array} { r l } & { \\displaystyle \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } ) ) = \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } { p ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } } \\\\ & { \\displaystyle = \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } { q _ { \\phi } ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } + \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } ) } { p ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } } \\\\ & { \\displaystyle = \\int _ { \\mathbf { X } } \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { X } ) \\log \\frac { q _ { \\phi } ( \\mathbf { z } , \\mathbf { X } ) } { q _ { \\phi } ( \\mathbf { z } ) p _ { e m p } ( \\mathbf { X } ) } d \\mathbf { z } d \\mathbf { X } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } ) | | p ( \\mathbf { z } ) ) } \\\\ & { \\displaystyle = \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } ) | | p ( \\mathbf { z } ) ) } \\end{array} \\overset { \\mathrm { U L } } { \\mathop : }\n$$", + "text_format": "latex", + "bbox": [ + 207, + 806, + 790, + 931 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "A.3 THE EQUATION FORM OF ELBO ", + "text_level": 1, + "bbox": [ + 174, + 103, + 442, + 118 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Proposition A.3. The expected margin between the true log-likelihood $\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\log p ( \\mathbf { X } )$ and $\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } E L B O$ is ", + "bbox": [ + 169, + 128, + 825, + 159 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/cb6a8d2fae314051384231eb0da47cce2013ff7800ca46f50afbfd817c45a02d.jpg", + "text": "$$\n\\mathbb { E } _ { p _ { c m p } ( \\mathbf { X } ) } [ \\log p ( \\mathbf { X } ) - E L B O ] = \\mathbb { E } _ { p _ { c m p } ( \\mathbf { X } ) } [ D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) ]\n$$", + "text_format": "latex", + "bbox": [ + 186, + 165, + 808, + 184 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "proof ", + "text_level": 1, + "bbox": [ + 173, + 194, + 215, + 208 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We just need to prove the following equation ", + "bbox": [ + 174, + 215, + 468, + 231 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/dab21f179631ac39d23d8f914db2bb86f9def4cf8a35c232b1b5f0dc8b4bd497.jpg", + "text": "$$\n\\log p ( \\mathbf { X } ) - \\mathrm { E L B O } = D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) )\n$$", + "text_format": "latex", + "bbox": [ + 256, + 234, + 740, + 252 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "and the proof under assumption $( 3 , 4 )$ is ", + "bbox": [ + 174, + 257, + 437, + 272 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/80740ddc57f57e984002db29dfb772d961712b7adeb1e2bbb08fa8b680584e08.jpg", + "text": "$$\n\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\displaystyle \\int _ { z , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log p ( \\mathbf { X } ) d \\mathbf { z } d \\mathbf { c } = \\displaystyle \\int _ { z , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } d \\mathbf { z } d \\mathbf { c } } \\\\ & { = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } + \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } { p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\mathrm { E L B O } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 277, + 834, + 369 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "A.4 OPTIMAL TRANSPORT SCHEME ", + "text_level": 1, + "bbox": [ + 176, + 381, + 437, + 396 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Proposition A.4. ", + "text_level": 1, + "bbox": [ + 173, + 407, + 289, + 422 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "1. The shortest-path derived from optimal transport scheme (17) between $\\mathbf z _ { 1 } \\sim \\mathcal N ( \\pmb { \\mu } _ { 1 } , \\mathrm { d i a g } ( \\pmb { \\sigma } _ { 1 } ^ { 2 } ) )$ and $\\mathbf z _ { 2 } \\sim \\mathcal N ( \\pmb { \\mu } _ { 2 } , \\mathrm { d i a g } ( \\pmb { \\sigma } _ { 2 } ^ { 2 } ) )$ with $\\lambda \\in [ 0 , 1 ]$ is ", + "bbox": [ + 173, + 433, + 821, + 463 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/1a5a90aae83b2c000b08123e680b76b3be8bde46513ca2ad4be17032da3680d2.jpg", + "text": "$$\n\\begin{array} { r } { \\tilde { \\pmb { \\mu } } = \\lambda \\pmb { \\mu } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\mu } _ { 2 } } \\\\ { \\tilde { \\pmb { \\sigma } } = \\lambda \\pmb { \\sigma } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\sigma } _ { 2 } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 419, + 467, + 576, + 503 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "proof ", + "text_level": 1, + "bbox": [ + 173, + 507, + 215, + 522 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Utilizing the conclusions in Kuang & Tabak (2017), the closed form of optimal transport from one multi-normal distribution $\\mathcal { N } ( \\mathbf { z } _ { 1 } ; \\mu _ { 1 } , \\pmb { \\Sigma } _ { 1 } )$ to another normal distribution $\\bar { \\mathcal { N } } ( \\mathbf { z } _ { 2 } ; \\mu _ { 2 } , \\Sigma _ { 2 } )$ is ", + "bbox": [ + 173, + 529, + 823, + 558 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/06967874ef6391478d9c20495ad2985d13efc1f8f4a3f5904488e6273f878eb5.jpg", + "text": "$$\n\\begin{array} { r } { { \\mathbf z } \\to { \\mathcal T } ( { \\mathbf z } ) = \\mu _ { 2 } + { \\mathbf T } ( { \\mathbf z } - { \\boldsymbol \\mu _ { 1 } } ) ; \\qquad { \\mathbf T } = { \\mathbf { \\Sigma } } _ { 1 } ^ { - \\frac { 1 } { 2 } } \\bigr ( { \\mathbf { \\Sigma } } _ { 1 } ^ { \\frac { 1 } { 2 } } { \\mathbf { \\Sigma } } _ { 2 } { \\mathbf { \\Sigma } } _ { 1 } ^ { \\frac { 1 } { 2 } } \\bigr ) ^ { \\frac { 1 } { 2 } } { \\mathbf { \\Sigma } } _ { 1 } ^ { - \\frac { 1 } { 2 } } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 277, + 561, + 717, + 585 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Utilize the diag matrix assumption, the optimal transport scheme with $\\lambda$ is ", + "bbox": [ + 176, + 589, + 661, + 604 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/8ed6d6dfe07b316d50a012889236cdfb64cad97d1f007d67edca2f2219b7905b.jpg", + "text": "$$\n\\begin{array} { r } { \\mathbf { z } _ { \\lambda } = ( 1 - \\lambda ) \\mathbf { z } _ { 1 } + \\lambda T ( \\mathbf { z } _ { 1 } ) } \\\\ { \\mathbf { T } = d i a g ( \\pmb { \\sigma } _ { 1 } / \\pmb { \\sigma } _ { 2 } ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 408, + 608, + 589, + 645 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Utilize (A.2), we can get (A.1) as ", + "bbox": [ + 173, + 648, + 400, + 665 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/fe26ce5355e26a16d568164dd51d1262af9e361cbae73fb249d507ba3db905f5.jpg", + "text": "$$\n{ \\bf z } _ { \\lambda } \\sim \\mathcal { N } ( \\tilde { \\pmb { \\mu } } , d i a g ( \\tilde { \\pmb { \\sigma } } ^ { 2 } ) ) ; \\qquad \\tilde { \\pmb { \\mu } } = \\lambda { \\pmb { \\mu } } _ { 1 } + ( 1 - \\lambda ) { \\pmb { \\mu } } _ { 2 } ; \\qquad \\tilde { \\pmb { \\sigma } } = \\lambda { \\pmb { \\sigma } } _ { 1 } + ( 1 - \\lambda ) { \\pmb { \\sigma } } _ { 2 } \\quad [ \\lambda \\in \\mathcal { N } ^ { \\pmb { \\mu } } ] .\n$$", + "text_format": "latex", + "bbox": [ + 217, + 670, + 769, + 688 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "2. The shortest-path derived from KL divergence based optimal transport scheme between $\\mathbf { c } _ { 1 } \\sim$ $\\mathrm { M u l t } ( K , \\pi _ { 1 } )$ and ${ \\bf c } _ { 2 } \\sim \\mathrm { M u l t } ( K , \\pi _ { 2 } )$ with $\\lambda \\in [ 0 , 1 ]$ is ", + "bbox": [ + 171, + 691, + 823, + 722 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/0f5495ae1d02266c38058a36e99eea685a61a09835dd32f0f0a2b01da853e839.jpg", + "text": "$$\n\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 }\n$$", + "text_format": "latex", + "bbox": [ + 421, + 726, + 576, + 743 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "proof ", + "text_level": 1, + "bbox": [ + 173, + 748, + 215, + 762 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "As the definition (17) has no closed-form solution for multinomial distribution, we use $\\mathrm { K L }$ divergence instead. The KL divergence based optimal transport target $\\mathbf { c } _ { \\lambda } \\sim \\mathbf { M u l t } ( K , \\tilde { \\pi } )$ between two multinomial distribution ${ \\bf c } _ { 1 } \\sim \\mathrm { M u l t } ( K , \\pi _ { 1 } )$ and ${ \\bf c } _ { 2 } \\sim \\bf { M u l t } ( { \\cal K } , \\pi _ { 2 } )$ with $\\lambda \\in [ 0 , 1 ]$ satisfy ", + "bbox": [ + 174, + 768, + 825, + 813 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/91abaf2591281c43c4b97376ffcbedc9d914c9e77d16d2eab169e9f3adc31089.jpg", + "text": "$$\n\\operatorname* { m i n } _ { \\tilde { \\pi } } \\lambda D _ { \\mathrm { K L } } ( \\pi _ { 1 } \\| \\tilde { \\pi } ) + ( 1 - \\lambda ) D _ { \\mathrm { K L } } ( \\pi _ { 2 } \\| \\tilde { \\pi } ) \\qquad s . t . \\sum _ { i = 1 } ^ { K } \\tilde { \\pi } _ { i } = 1\n$$", + "text_format": "latex", + "bbox": [ + 290, + 816, + 705, + 861 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "The Lagrange multiplier form of (A.4) is ", + "bbox": [ + 174, + 864, + 446, + 880 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/1937c5d3482a04a0b865cadf84482e4f7fc8637a676bcb628f7b4bd463552f6b.jpg", + "text": "$$\n\\mathcal { L } ( \\tilde { \\pi } , t ) = \\lambda D _ { \\mathrm { K L } } ( \\pi _ { 1 } \\| \\tilde { \\pi } ) + ( 1 - \\lambda ) D _ { \\mathrm { K L } } ( \\pi _ { 2 } \\| \\tilde { \\pi } ) + t * ( \\sum _ { i = 1 } ^ { K } \\tilde { \\pi } _ { i } - 1 )\n$$", + "text_format": "latex", + "bbox": [ + 276, + 885, + 722, + 928 + ], + "page_idx": 11 + }, + { + "type": "table", + "img_path": "images/391bdb23dc0c933375929b262309ab1307a56f3b9a9bb2b19986783d65f23084.jpg", + "table_caption": [ + "Table 6: Schedule Parameters " + ], + "table_footnote": [], + "table_body": "
MNISTSVHN(one-stage)SVHNCifar10Cifar100
hmaxtmaxhmaxtmaxhmaxtmaxhmaxtmaxhmaxtmax
305011751e-31501e-31501e-1150
B3050117511501e-31501e-3150
117.5505017512801502001501280150
Ic1750501752.31502.31504.6150
WMz///一1e-31501e-31501e-1150
WMc/1140012801280
", + "bbox": [ + 186, + 125, + 812, + 241 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "and the related KKT conditions are ", + "bbox": [ + 173, + 270, + 406, + 285 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/a47c8579fdd46832ed1ce705a4d8e646e3b49efb82a32523535c1f7b79b1e4c4.jpg", + "text": "$$\n\\begin{array} { r } { \\frac { \\partial \\mathcal { L } ( \\tilde { \\boldsymbol { \\pi } } , t ) } { \\partial \\tilde { \\boldsymbol { \\pi } } } = t - \\frac { \\lambda \\boldsymbol { \\pi } _ { 1 } + ( 1 - \\lambda ) \\boldsymbol { \\pi } _ { 2 } } { \\tilde { \\boldsymbol { \\pi } } } = 0 } \\\\ { t \\ast ( \\displaystyle \\sum _ { i = 1 } ^ { K } \\tilde { \\boldsymbol { \\pi } } _ { i } - 1 ) = 0 } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 366, + 292, + 632, + 369 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Solve the equation (A.5), we can get the closed form of $\\tilde { \\pi }$ as ", + "bbox": [ + 173, + 380, + 575, + 396 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/70cd9fe91d1124c77a6c6ac100b6a430b601579ae84870e3978ef0332e5c4e3f.jpg", + "text": "$$\n\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 } \\quad \\bigsqcup\n$$", + "text_format": "latex", + "bbox": [ + 406, + 406, + 591, + 424 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "A.5 SCHEDULE ANALYSIS ", + "text_level": 1, + "bbox": [ + 176, + 445, + 370, + 459 + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/712ae7b375b430e62be3a30c858dcdfec56c3e221e02585de9e0516814126163.jpg", + "image_caption": [ + "Figure 2: The Exponential Function-Based Scheduler " + ], + "image_footnote": [], + "bbox": [ + 248, + 511, + 710, + 752 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "The warm-up scheduler aims to slowly increase the parameters until they reach their maximum. For a certain hyperparameter $h$ , there are 2 parameters control its warm-up process, the target value $h _ { m a x }$ and the total epoch $t _ { m a x }$ to reach the target value. We use the exponential function to get the middle value $h _ { t }$ as ", + "bbox": [ + 173, + 806, + 825, + 863 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/9ae375ce93609156d7fd38df51dd5de71145d94170619f98c18b778c8534f02b.jpg", + "text": "$$\nh _ { t } = h _ { m a x } \\times \\exp { [ - 5 * ( 1 - \\operatorname* { m i n } ( 1 , t / t _ { m a x } ) ) ^ { 2 } ] }\n$$", + "text_format": "latex", + "bbox": [ + 338, + 867, + 658, + 886 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "The curve of (A.6) is shown in Figure 2, and we list the scheduler parameters of 4 benchmark datasets, i.e. MNIST, SVHN, Cifar10, Cifar100, in Table 6. ", + "bbox": [ + 173, + 895, + 823, + 924 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/6570e9dfa5eec3f1555477b23cafd74049c581d87a93e5e2dcffad761cee71c9.jpg", + "table_caption": [ + "Table 7: Details of Training Process " + ], + "table_footnote": [], + "table_body": "
MNISTSVHN(one-stage)SVHNCifar10Cifar100
Latent Dim(z/c)10/1032/10128/10128/10128/100
Mutual Info(z/c)17.5/17.050/501280/2.3200/2.31280/4.6
loss of -log pe(X|c,z)BCEBCEMSEMSEBCE
a of pmixup(X)一一222
optimizerAdamSGD
learning rate5e-41e-30.1
lr scheduler(decay ratio)一一every 50 epoch after 200-th(0.5)[500,600,650](0.2)
weight decay005e-4
", + "bbox": [ + 173, + 125, + 857, + 270 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "A.6 DETAILS OF TRAINING PROCESS ", + "text_level": 1, + "bbox": [ + 176, + 306, + 444, + 321 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Here we list some important items need to set in the training process for different process. We classify these items into 2 categories: (1) items related to the loss function and (2) items related to the optimization strategy. The details are as follows and we list the exact value in Table 7: ", + "bbox": [ + 174, + 338, + 825, + 381 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "1. Items related to the loss function ", + "bbox": [ + 214, + 398, + 447, + 412 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "• Latent dim The latent dim of discrete variable c is the same as the number of classifications, that is, $K$ . For continuous variable $\\mathbf { z }$ , the latent dim is determined by experiments. \n• Mutual information We find the value of continuous mutual information will affect the generative performance, but have little impact on semi-supervised learning results, so we choose a suitable value to get the best generative performance. For discrete information, we choose the value in the ideal scene, that is, $\\mathbf { I } _ { \\mathbf { c } } = \\log ( K )$ . \n• Calculation of $\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )$ $p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )$ has two forms: (1) normal distribution with $\\mathcal { N } ( f _ { \\boldsymbol { \\theta } } ( \\mathbf { c } , \\mathbf { z } ) , \\mathbf { I } )$ and (2) multinomial distribution with $\\mathbf { M u l t } ( d i m ( \\mathbf { X } ) , f _ { \\theta } ( \\mathbf { c } , \\mathbf { z } ) )$ . For the two forms, the loss function of $- \\log p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )$ is mean square error(MSE) and binary cross entropy(BCE) respectively. We use reparameterization trick in (20) to approximate expectation, and the sampling frequency is 1. \n• $\\alpha$ of $p _ { m i x u p } ( \\mathbf { X } )$ The $\\beta ( \\alpha , \\alpha )$ for mixup vicinal distribution will strongly affect SSL performance. We set it to 2 in all experiments. ", + "bbox": [ + 246, + 429, + 826, + 693 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "2. Items related to the optimization strategy ", + "bbox": [ + 214, + 709, + 501, + 723 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "• Optimizer In one-stage VAE, we use Adam. In OSPOT-VAE, we use SGD with momentum 0.9. \n• Learning rate We set the initial learning rate to 0.1 in OSPOT-VAE, which obtains best SSL performance. The scheduler of adjusting learning rate We decay the learning ratio in some milestones with the specified decay rate. \n• Weight decay Weight decay controls the strength of $L _ { 2 }$ regularization. ", + "bbox": [ + 248, + 739, + 825, + 891 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "We also list some standard training curves on benchmark datasets in Figure 3 and 4. ", + "bbox": [ + 173, + 909, + 722, + 924 + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/cb00a1218b54c04419c02f3df09a2f7602585690b55a8620fed2b95acb2b231b.jpg", + "image_caption": [ + "Figure 3: The training curves of Cifar10. Left: classification performance. Right: $\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )$ " + ], + "image_footnote": [], + "bbox": [ + 176, + 103, + 818, + 263 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/c99910939dd2e58faf73c25c385af2b0e5d3daf4d34a0192b639b515b048f86f.jpg", + "image_caption": [ + "Figure 4: The training curves of Cifar100. Left: classification performance. Right: $\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )$ " + ], + "image_footnote": [], + "bbox": [ + 174, + 363, + 816, + 522 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "A.7 GENERATIVE PERFORMANCE ", + "text_level": 1, + "bbox": [ + 176, + 601, + 421, + 614 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Following figures5-7 show the generation performance of our OSPOT-VAE. ", + "bbox": [ + 174, + 631, + 671, + 647 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/25e8427bc09e3e5a473bfd67e0a4ebd896f3b330cc776a7003253b54f424ad84.jpg", + "image_caption": [ + "Figure 5: The generative performance of OSPOT-VAE in MNIST and SVHN " + ], + "image_footnote": [], + "bbox": [ + 205, + 671, + 790, + 893 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/ee3780017d783d4bf7dd38dea93a2d2af64e16334a08785719316d9fd5e1961d.jpg", + "image_caption": [ + "Figure 6: Compare the generative performance of pure VAE and OSPOT-VAE in Cifar10 " + ], + "image_footnote": [], + "bbox": [ + 205, + 101, + 790, + 321 + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/5871b177cdb29994688d651de1f8c9d3eec78a3fb02e32f4b5033317a4d4ac6b.jpg", + "image_caption": [ + "Figure 7: Compare the generative performance of pure VAE and OSPOT-VAE in Cifar100 " + ], + "image_footnote": [], + "bbox": [ + 207, + 377, + 790, + 599 + ], + "page_idx": 15 + } +] \ No newline at end of file diff --git a/parse/train/S1ejj64YvS/S1ejj64YvS_middle.json b/parse/train/S1ejj64YvS/S1ejj64YvS_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..7fafec8653d2c68d30edc307de085447ee944b33 --- /dev/null +++ b/parse/train/S1ejj64YvS/S1ejj64YvS_middle.json @@ -0,0 +1,43291 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 503, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 77, + 505, + 99 + ], + "spans": [ + { + "bbox": [ + 105, + 77, + 505, + 99 + ], + "score": 1.0, + "content": "GOOD SEMI-SUPERVISED VAE REQUIRES TIGHTER", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 306, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 306, + 118 + ], + "score": 1.0, + "content": "EVIDENCE LOWER BOUND", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 136, + 244, + 157 + ], + "lines": [ + { + "bbox": [ + 113, + 136, + 201, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 136, + 201, + 147 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 112, + 146, + 245, + 159 + ], + "spans": [ + { + "bbox": [ + 112, + 146, + 245, + 159 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 278, + 186, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 277, + 186, + 335, + 200 + ], + "spans": [ + { + "bbox": [ + 277, + 186, + 335, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 212, + 468, + 376 + ], + "lines": [ + { + "bbox": [ + 141, + 212, + 470, + 224 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 470, + 224 + ], + "score": 1.0, + "content": "Semi-supervised learning approaches based on generative models have now en-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 223, + 470, + 235 + ], + "spans": [ + { + "bbox": [ + 141, + 223, + 470, + 235 + ], + "score": 1.0, + "content": "countered 3 challenges: (1) The two-stage training strategy is not robust. (2) Good", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 234, + 470, + 247 + ], + "spans": [ + { + "bbox": [ + 141, + 234, + 470, + 247 + ], + "score": 1.0, + "content": "semi-supervised learning results and good generative performance can not be ob-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 245, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 141, + 245, + 469, + 257 + ], + "score": 1.0, + "content": "tained at the same time. (3) Even at the expense of sacrificing generative perfor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 256, + 470, + 267 + ], + "spans": [ + { + "bbox": [ + 141, + 256, + 470, + 267 + ], + "score": 1.0, + "content": "mance, the semi-supervised classification results are still not satisfactory. To ad-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 267, + 470, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 470, + 280 + ], + "score": 1.0, + "content": "dress these problems, we propose One-stage Semi-suPervised Optimal Transport", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 277, + 470, + 290 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 470, + 290 + ], + "score": 1.0, + "content": "VAE (OSPOT-VAE), a one-stage deep generative model that theoretically uni-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 288, + 470, + 301 + ], + "spans": [ + { + "bbox": [ + 141, + 288, + 470, + 301 + ], + "score": 1.0, + "content": "fies the generation and classification loss in one ELBO framework and achieves", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "score": 1.0, + "content": "a tighter ELBO by applying the optimal transport scheme to the distribution of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 310, + 469, + 322 + ], + "spans": [ + { + "bbox": [ + 141, + 310, + 469, + 322 + ], + "score": 1.0, + "content": "latent variables. We show that with tighter ELBO, our OSPOT-VAE surpasses", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 321, + 470, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 470, + 334 + ], + "score": 1.0, + "content": "the best semi-supervised generative models by a large margin across many bench-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 333, + 469, + 344 + ], + "spans": [ + { + "bbox": [ + 142, + 333, + 383, + 344 + ], + "score": 1.0, + "content": "mark datasets. For example, we reduce the error rate from", + "type": "text" + }, + { + "bbox": [ + 384, + 333, + 415, + 343 + ], + "score": 0.87, + "content": "1 4 . 4 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 333, + 428, + 344 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 428, + 333, + 455, + 343 + ], + "score": 0.88, + "content": "6 . 1 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 333, + 469, + 344 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 343, + 470, + 356 + ], + "spans": [ + { + "bbox": [ + 142, + 343, + 423, + 356 + ], + "score": 1.0, + "content": "Cifar-10 with 4k labels and achieve state-of-the-art performance with", + "type": "text" + }, + { + "bbox": [ + 423, + 344, + 455, + 354 + ], + "score": 0.88, + "content": "2 5 . 3 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 343, + 470, + 356 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 354, + 469, + 366 + ], + "spans": [ + { + "bbox": [ + 142, + 354, + 469, + 366 + ], + "score": 1.0, + "content": "Cifar-100 with 10k labels. We also demonstrate that good generative models and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 366, + 436, + 377 + ], + "spans": [ + { + "bbox": [ + 141, + 366, + 436, + 377 + ], + "score": 1.0, + "content": "semi-supervised results can be achieved simultaneously by OSPOT-VAE.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 108, + 398, + 206, + 411 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 208, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 208, + 414 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 105, + 422, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 506, + 438 + ], + "score": 1.0, + "content": "The rise of deep neural networks has led to breakthroughs in computer vision, natural language", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "processing, and many other domains. Most of these models are trained on large labeled datasets", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "via supervised learning. However, in many scenarios, although it is easy to acquire a large amount", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "of the original data, obtaining corresponding labels is often very costly or even infeasible. Semi-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 468, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 479 + ], + "score": 1.0, + "content": "supervised learning (Thomas, 2009) is proposed to address this problem by training classifiers with", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 478, + 350, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 350, + 490 + ], + "score": 1.0, + "content": "sufficient unlabeled data and a small fraction of labeled data.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 505, + 616 + ], + "lines": [ + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "score": 1.0, + "content": "Recent works on semi-supervised learning can be grouped into three categories: (1) disagreement", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "based learning via data perturbation (Miyato et al., 2019) and consistency enforcing (Verma et al.,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 516, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 531 + ], + "score": 1.0, + "content": "2019), (2) metric learning (Wu et al., 2018), (3) generative approaches via generative adversar-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 527, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 541 + ], + "score": 1.0, + "content": "ial network (GAN) (Springenberg, 2016) and variational autoencoder (VAE) (Kingma et al., 2014).", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "score": 1.0, + "content": "Compared with the first two categories, generative approaches have great advantages in interpretabil-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "score": 1.0, + "content": "ity. Based on the latent variable assumption (Doersch, 2016), the generative model has an explicit", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "score": 1.0, + "content": "variational inference form, so it can learn the marginal probability distribution of the raw data as", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "well as the conditional distribution of the latent variables given the input data, which makes pre-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "dictions more reasonable. Besides, generative approaches not only learn the required classification", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 595, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 607 + ], + "score": 1.0, + "content": "representations, but also capture the semantics-disentangled factors that generate the data, making it", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 379, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 379, + 618 + ], + "score": 1.0, + "content": "easier to generalize to different tasks (Narayanaswamy et al., 2017).", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 622, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "However, in practice, semi-supervised generative approaches often encounter three major chal-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "score": 1.0, + "content": "lenges: (1) The two-stage training process is not robust. Semi-supervised VAE (Kingma et al.,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 642, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 658 + ], + "score": 1.0, + "content": "2014) needs to be trained carefully with a two-stage hierarchical strategy, while the training pro-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "cess of GAN is a two-stage adversarial game (Chrysos et al., 2019). (2) Good semi-supervised", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "learning results and good generative performance can not be obtained at the same time. In GAN,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "good semi-supervised learning performance will lead to a mismatch between the generated results", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "and the real data distribution (Dai et al., 2017). While in VAE, the evidence lower bound (ELBO)", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "objective is irrelevant to the classification loss, making it difficult to learn from the labels directly", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 711, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 722 + ], + "score": 1.0, + "content": "(Narayanaswamy et al., 2017). (3) Even at the expense of sacrificing generative performance, the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "semi-supervised classification results are still not satisfactory. In practice, disagreement-based meth-", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 42.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 503, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 77, + 505, + 99 + ], + "spans": [ + { + "bbox": [ + 105, + 77, + 505, + 99 + ], + "score": 1.0, + "content": "GOOD SEMI-SUPERVISED VAE REQUIRES TIGHTER", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 306, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 306, + 118 + ], + "score": 1.0, + "content": "EVIDENCE LOWER BOUND", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 136, + 244, + 157 + ], + "lines": [ + { + "bbox": [ + 113, + 136, + 201, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 136, + 201, + 147 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 112, + 146, + 245, + 159 + ], + "spans": [ + { + "bbox": [ + 112, + 146, + 245, + 159 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 112, + 136, + 245, + 159 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 186, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 277, + 186, + 335, + 200 + ], + "spans": [ + { + "bbox": [ + 277, + 186, + 335, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 212, + 468, + 376 + ], + "lines": [ + { + "bbox": [ + 141, + 212, + 470, + 224 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 470, + 224 + ], + "score": 1.0, + "content": "Semi-supervised learning approaches based on generative models have now en-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 223, + 470, + 235 + ], + "spans": [ + { + "bbox": [ + 141, + 223, + 470, + 235 + ], + "score": 1.0, + "content": "countered 3 challenges: (1) The two-stage training strategy is not robust. (2) Good", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 234, + 470, + 247 + ], + "spans": [ + { + "bbox": [ + 141, + 234, + 470, + 247 + ], + "score": 1.0, + "content": "semi-supervised learning results and good generative performance can not be ob-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 245, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 141, + 245, + 469, + 257 + ], + "score": 1.0, + "content": "tained at the same time. (3) Even at the expense of sacrificing generative perfor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 256, + 470, + 267 + ], + "spans": [ + { + "bbox": [ + 141, + 256, + 470, + 267 + ], + "score": 1.0, + "content": "mance, the semi-supervised classification results are still not satisfactory. To ad-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 267, + 470, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 470, + 280 + ], + "score": 1.0, + "content": "dress these problems, we propose One-stage Semi-suPervised Optimal Transport", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 277, + 470, + 290 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 470, + 290 + ], + "score": 1.0, + "content": "VAE (OSPOT-VAE), a one-stage deep generative model that theoretically uni-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 288, + 470, + 301 + ], + "spans": [ + { + "bbox": [ + 141, + 288, + 470, + 301 + ], + "score": 1.0, + "content": "fies the generation and classification loss in one ELBO framework and achieves", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "score": 1.0, + "content": "a tighter ELBO by applying the optimal transport scheme to the distribution of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 310, + 469, + 322 + ], + "spans": [ + { + "bbox": [ + 141, + 310, + 469, + 322 + ], + "score": 1.0, + "content": "latent variables. We show that with tighter ELBO, our OSPOT-VAE surpasses", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 321, + 470, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 470, + 334 + ], + "score": 1.0, + "content": "the best semi-supervised generative models by a large margin across many bench-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 333, + 469, + 344 + ], + "spans": [ + { + "bbox": [ + 142, + 333, + 383, + 344 + ], + "score": 1.0, + "content": "mark datasets. For example, we reduce the error rate from", + "type": "text" + }, + { + "bbox": [ + 384, + 333, + 415, + 343 + ], + "score": 0.87, + "content": "1 4 . 4 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 333, + 428, + 344 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 428, + 333, + 455, + 343 + ], + "score": 0.88, + "content": "6 . 1 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 333, + 469, + 344 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 343, + 470, + 356 + ], + "spans": [ + { + "bbox": [ + 142, + 343, + 423, + 356 + ], + "score": 1.0, + "content": "Cifar-10 with 4k labels and achieve state-of-the-art performance with", + "type": "text" + }, + { + "bbox": [ + 423, + 344, + 455, + 354 + ], + "score": 0.88, + "content": "2 5 . 3 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 343, + 470, + 356 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 354, + 469, + 366 + ], + "spans": [ + { + "bbox": [ + 142, + 354, + 469, + 366 + ], + "score": 1.0, + "content": "Cifar-100 with 10k labels. We also demonstrate that good generative models and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 366, + 436, + 377 + ], + "spans": [ + { + "bbox": [ + 141, + 366, + 436, + 377 + ], + "score": 1.0, + "content": "semi-supervised results can be achieved simultaneously by OSPOT-VAE.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 12, + "bbox_fs": [ + 141, + 212, + 470, + 377 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 398, + 206, + 411 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 208, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 208, + 414 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 105, + 422, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 506, + 438 + ], + "score": 1.0, + "content": "The rise of deep neural networks has led to breakthroughs in computer vision, natural language", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "processing, and many other domains. Most of these models are trained on large labeled datasets", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "via supervised learning. However, in many scenarios, although it is easy to acquire a large amount", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "of the original data, obtaining corresponding labels is often very costly or even infeasible. Semi-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 468, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 479 + ], + "score": 1.0, + "content": "supervised learning (Thomas, 2009) is proposed to address this problem by training classifiers with", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 478, + 350, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 350, + 490 + ], + "score": 1.0, + "content": "sufficient unlabeled data and a small fraction of labeled data.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 422, + 506, + 490 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 505, + 616 + ], + "lines": [ + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "score": 1.0, + "content": "Recent works on semi-supervised learning can be grouped into three categories: (1) disagreement", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "based learning via data perturbation (Miyato et al., 2019) and consistency enforcing (Verma et al.,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 516, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 531 + ], + "score": 1.0, + "content": "2019), (2) metric learning (Wu et al., 2018), (3) generative approaches via generative adversar-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 527, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 541 + ], + "score": 1.0, + "content": "ial network (GAN) (Springenberg, 2016) and variational autoencoder (VAE) (Kingma et al., 2014).", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "score": 1.0, + "content": "Compared with the first two categories, generative approaches have great advantages in interpretabil-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "score": 1.0, + "content": "ity. Based on the latent variable assumption (Doersch, 2016), the generative model has an explicit", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "score": 1.0, + "content": "variational inference form, so it can learn the marginal probability distribution of the raw data as", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "well as the conditional distribution of the latent variables given the input data, which makes pre-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "dictions more reasonable. Besides, generative approaches not only learn the required classification", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 595, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 607 + ], + "score": 1.0, + "content": "representations, but also capture the semantics-disentangled factors that generate the data, making it", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 379, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 379, + 618 + ], + "score": 1.0, + "content": "easier to generalize to different tasks (Narayanaswamy et al., 2017).", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 496, + 505, + 618 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 622, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "However, in practice, semi-supervised generative approaches often encounter three major chal-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "score": 1.0, + "content": "lenges: (1) The two-stage training process is not robust. Semi-supervised VAE (Kingma et al.,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 642, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 658 + ], + "score": 1.0, + "content": "2014) needs to be trained carefully with a two-stage hierarchical strategy, while the training pro-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "cess of GAN is a two-stage adversarial game (Chrysos et al., 2019). (2) Good semi-supervised", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "learning results and good generative performance can not be obtained at the same time. In GAN,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "good semi-supervised learning performance will lead to a mismatch between the generated results", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "and the real data distribution (Dai et al., 2017). While in VAE, the evidence lower bound (ELBO)", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "objective is irrelevant to the classification loss, making it difficult to learn from the labels directly", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 711, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 722 + ], + "score": 1.0, + "content": "(Narayanaswamy et al., 2017). (3) Even at the expense of sacrificing generative performance, the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "semi-supervised classification results are still not satisfactory. In practice, disagreement-based meth-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 192, + 505, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 505, + 204 + ], + "score": 1.0, + "content": "ods (Xie et al., 2019; Berthelot et al., 2019) have dramatically improved the state-of-the-art results", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 204, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 216 + ], + "score": 1.0, + "content": "on several standard datasets, surpassing generative approaches by a large margin. These challenges", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 214, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 505, + 227 + ], + "score": 1.0, + "content": "naturally raise a question: What limits the performance of generative approaches in semi-supervised", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 225, + 149, + 239 + ], + "spans": [ + { + "bbox": [ + 104, + 225, + 149, + 239 + ], + "score": 1.0, + "content": "learning?", + "type": "text", + "cross_page": true + } + ], + "index": 7 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 621, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 52, + 505, + 164 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 52, + 505, + 164 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 52, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 52, + 505, + 164 + ], + "score": 0.97, + "type": "image", + "image_path": "6a7096b8d04a9c60b237cc335b2a0bad726f4db3bc04839eec2139f9c99e967e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 52, + 505, + 89.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 89.33333333333334, + 505, + 126.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 126.66666666666669, + 505, + 164.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 223, + 173, + 389, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 222, + 172, + 389, + 186 + ], + "spans": [ + { + "bbox": [ + 222, + 172, + 389, + 186 + ], + "score": 1.0, + "content": "Figure 1: The schematic of OSPOT-VAE", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 192, + 505, + 237 + ], + "lines": [ + { + "bbox": [ + 106, + 192, + 505, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 505, + 204 + ], + "score": 1.0, + "content": "ods (Xie et al., 2019; Berthelot et al., 2019) have dramatically improved the state-of-the-art results", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 204, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 216 + ], + "score": 1.0, + "content": "on several standard datasets, surpassing generative approaches by a large margin. These challenges", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 214, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 505, + 227 + ], + "score": 1.0, + "content": "naturally raise a question: What limits the performance of generative approaches in semi-supervised", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 225, + 149, + 239 + ], + "spans": [ + { + "bbox": [ + 104, + 225, + 149, + 239 + ], + "score": 1.0, + "content": "learning?", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 107, + 242, + 505, + 308 + ], + "lines": [ + { + "bbox": [ + 105, + 242, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 506, + 255 + ], + "score": 1.0, + "content": "In this work, we propose One-stage Semi-suPervised Optimal Transport VAE (OSPOT-VAE) to", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "score": 1.0, + "content": "address these challenges, which consists of two improvements: (1) a one-stage semi-supervised", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 263, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 506, + 278 + ], + "score": 1.0, + "content": "VAE model that unifies the generation and classification loss in one ELBO framework. (2) an", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "estimation of the margin between true log-likelihood and the ELBO that exports a tighter evidence", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 286, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 506, + 300 + ], + "score": 1.0, + "content": "lower bound by applying optimal transport (Ambrosio & Gigli, 2013) scheme to the distribution of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 296, + 172, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 296, + 172, + 309 + ], + "score": 1.0, + "content": "latent variables.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 108, + 314, + 280, + 326 + ], + "lines": [ + { + "bbox": [ + 106, + 313, + 281, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 281, + 327 + ], + "score": 1.0, + "content": "Our model has the following contributions:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 132, + 336, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 131, + 335, + 460, + 350 + ], + "spans": [ + { + "bbox": [ + 131, + 335, + 460, + 350 + ], + "score": 1.0, + "content": "• We show that OSPOT-VAE can be well trained with a direct one-stage strategy.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 132, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 132, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "• We show that OSPOT-VAE can achieve both good generative performance and semi-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 363, + 453, + 375 + ], + "spans": [ + { + "bbox": [ + 141, + 363, + 453, + 375 + ], + "score": 1.0, + "content": "supervised learning results simultaneously on a series of benchmark datasets.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 133, + 378, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 133, + 378, + 505, + 389 + ], + "score": 1.0, + "content": "• We point out that it is the large margin between the ELBO and the log-likelihood of the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 141, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "input data that limits the performance of semi-supervised VAE. Besides, we evaluate this", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 400, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 142, + 400, + 505, + 411 + ], + "score": 1.0, + "content": "assumption across many standard datasets and show that with the proposed tighter ELBO,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 142, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "OSPOT-VAE surpasses the best semi-supervised generative models by a large margin and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 421, + 412, + 433 + ], + "spans": [ + { + "bbox": [ + 141, + 421, + 412, + 433 + ], + "score": 1.0, + "content": "achieves state-of-the-art performance on Cifar-100 with 10k labels.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18.5 + }, + { + "type": "title", + "bbox": [ + 107, + 453, + 335, + 466 + ], + "lines": [ + { + "bbox": [ + 104, + 452, + 336, + 469 + ], + "spans": [ + { + "bbox": [ + 104, + 452, + 336, + 469 + ], + "score": 1.0, + "content": "2 SEMI-SUPERVISED LEARNING METHODS", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 479, + 505, + 545 + ], + "lines": [ + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "score": 1.0, + "content": "In supervised learning (SL), we are facing with training data that appears as input-target pairs", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 490, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 107, + 490, + 165, + 502 + ], + "score": 0.89, + "content": "( \\mathbf { X } , \\bar { \\mathbf { y } } ) \\ \\in \\ \\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 490, + 332, + 503 + ], + "score": 1.0, + "content": "sampled from an unknown distribution", + "type": "text" + }, + { + "bbox": [ + 332, + 490, + 365, + 502 + ], + "score": 0.93, + "content": "p ( \\mathbf { X } , \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 490, + 505, + 503 + ], + "score": 1.0, + "content": ". Our goal is to learn a function", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 501, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 107, + 502, + 141, + 513 + ], + "score": 0.91, + "content": "f ( \\mathbf { X } ; \\phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 501, + 217, + 514 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 218, + 502, + 226, + 513 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 501, + 363, + 514 + ], + "score": 1.0, + "content": "that makes the correct inference", + "type": "text" + }, + { + "bbox": [ + 363, + 503, + 371, + 513 + ], + "score": 0.37, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 501, + 478, + 514 + ], + "score": 1.0, + "content": "for unseen samples from", + "type": "text" + }, + { + "bbox": [ + 479, + 501, + 501, + 513 + ], + "score": 0.91, + "content": "p ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 501, + 505, + 514 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 511, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 506, + 524 + ], + "score": 1.0, + "content": "While in semi-supervised learning (SSL), we can obtain an extra collection of unlabeled data", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 522, + 504, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 145, + 534 + ], + "score": 0.9, + "content": "\\mathbf { X } \\in \\mathbb { D } _ { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 522, + 294, + 536 + ], + "score": 1.0, + "content": "sampled from the same distribution", + "type": "text" + }, + { + "bbox": [ + 295, + 523, + 317, + 535 + ], + "score": 0.91, + "content": "p ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 522, + 489, + 536 + ], + "score": 1.0, + "content": ". We hope to leverage the data from both", + "type": "text" + }, + { + "bbox": [ + 489, + 523, + 504, + 534 + ], + "score": 0.87, + "content": "\\mathbb { D } _ { L }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 532, + 500, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 123, + 548 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 534, + 138, + 545 + ], + "score": 0.87, + "content": "\\mathbb { D } _ { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 532, + 479, + 548 + ], + "score": 1.0, + "content": "to achieve a more accurate model than what would have been obtained by only using", + "type": "text" + }, + { + "bbox": [ + 480, + 534, + 494, + 545 + ], + "score": 0.87, + "content": "\\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 532, + 500, + 548 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 551, + 505, + 595 + ], + "lines": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "In this section, we review some existing methods for SSL. We mainly focus on those who have", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 562, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 574 + ], + "score": 1.0, + "content": "reached state-of-the-art results, as well as generative approaches which are strongly connected with", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "score": 1.0, + "content": "our model; the more comprehensive overview is beyond the scope of this paper, we refer readers to", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 584, + 190, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 190, + 596 + ], + "score": 1.0, + "content": "(Oliver et al., 2018).", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5 + }, + { + "type": "title", + "bbox": [ + 108, + 608, + 282, + 620 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 282, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 282, + 621 + ], + "score": 1.0, + "content": "2.1 DISAGREEMENT BASED LEARNING", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 629, + 505, + 685 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 506, + 644 + ], + "score": 1.0, + "content": "Disagreement-based learning refers to the general approaches of imposing disagreement among", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 640, + 505, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 654 + ], + "score": 1.0, + "content": "multiple learners on the same task or multiple predictions from a single learner. By eliminating the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 651, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 505, + 665 + ], + "score": 1.0, + "content": "disagreement, we can enforce the generalization of the model on unseen data. A common technique", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 662, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 676 + ], + "score": 1.0, + "content": "for creating disagreement is data augmentation, which applies transformations or perturbations on", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 673, + 485, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 336, + 686 + ], + "score": 1.0, + "content": "the input data and leaves class semantics unchanged. For", + "type": "text" + }, + { + "bbox": [ + 336, + 673, + 371, + 685 + ], + "score": 0.91, + "content": "\\mathbf { X } \\in \\mathbb { D } _ { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 673, + 485, + 686 + ], + "score": 1.0, + "content": ", loss term can be derived as", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37 + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 690, + 376, + 704 + ], + "lines": [ + { + "bbox": [ + 234, + 690, + 376, + 704 + ], + "spans": [ + { + "bbox": [ + 234, + 690, + 376, + 704 + ], + "score": 0.9, + "content": "\\| f ( \\operatorname { A u g m e n t } ( \\mathbf { X } ) ; \\phi ) - f ( \\mathbf { X } ; \\phi ) \\| _ { 2 } ^ { 2 }", + "type": "interline_equation", + "image_path": "35a0f9cac4ba95ade39d9324cca98481124ca123c67fcfaa3c6e63eb03b2a5ef.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 234, + 690, + 376, + 704 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 185, + 722 + ], + "score": 1.0, + "content": "where the Augment", + "type": "text" + }, + { + "bbox": [ + 185, + 710, + 202, + 721 + ], + "score": 0.66, + "content": "( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "is a stochastic function which can be obtained by image transformation (Xie", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "et al., 2019), virtual adversarial training (Miyato et al., 2019), or mixup method (Verma et al. 2018;", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 52, + 505, + 164 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 52, + 505, + 164 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 52, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 52, + 505, + 164 + ], + "score": 0.97, + "type": "image", + "image_path": "6a7096b8d04a9c60b237cc335b2a0bad726f4db3bc04839eec2139f9c99e967e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 52, + 505, + 89.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 89.33333333333334, + 505, + 126.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 126.66666666666669, + 505, + 164.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 223, + 173, + 389, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 222, + 172, + 389, + 186 + ], + "spans": [ + { + "bbox": [ + 222, + 172, + 389, + 186 + ], + "score": 1.0, + "content": "Figure 1: The schematic of OSPOT-VAE", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 192, + 505, + 237 + ], + "lines": [], + "index": 5.5, + "bbox_fs": [ + 104, + 192, + 505, + 239 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 242, + 505, + 308 + ], + "lines": [ + { + "bbox": [ + 105, + 242, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 506, + 255 + ], + "score": 1.0, + "content": "In this work, we propose One-stage Semi-suPervised Optimal Transport VAE (OSPOT-VAE) to", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "score": 1.0, + "content": "address these challenges, which consists of two improvements: (1) a one-stage semi-supervised", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 263, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 506, + 278 + ], + "score": 1.0, + "content": "VAE model that unifies the generation and classification loss in one ELBO framework. (2) an", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "estimation of the margin between true log-likelihood and the ELBO that exports a tighter evidence", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 286, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 506, + 300 + ], + "score": 1.0, + "content": "lower bound by applying optimal transport (Ambrosio & Gigli, 2013) scheme to the distribution of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 296, + 172, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 296, + 172, + 309 + ], + "score": 1.0, + "content": "latent variables.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 242, + 506, + 309 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 314, + 280, + 326 + ], + "lines": [ + { + "bbox": [ + 106, + 313, + 281, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 281, + 327 + ], + "score": 1.0, + "content": "Our model has the following contributions:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 313, + 281, + 327 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 336, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 131, + 335, + 460, + 350 + ], + "spans": [ + { + "bbox": [ + 131, + 335, + 460, + 350 + ], + "score": 1.0, + "content": "• We show that OSPOT-VAE can be well trained with a direct one-stage strategy.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 132, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 132, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "• We show that OSPOT-VAE can achieve both good generative performance and semi-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 363, + 453, + 375 + ], + "spans": [ + { + "bbox": [ + 141, + 363, + 453, + 375 + ], + "score": 1.0, + "content": "supervised learning results simultaneously on a series of benchmark datasets.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 133, + 378, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 133, + 378, + 505, + 389 + ], + "score": 1.0, + "content": "• We point out that it is the large margin between the ELBO and the log-likelihood of the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 141, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "input data that limits the performance of semi-supervised VAE. Besides, we evaluate this", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 400, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 142, + 400, + 505, + 411 + ], + "score": 1.0, + "content": "assumption across many standard datasets and show that with the proposed tighter ELBO,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 142, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "OSPOT-VAE surpasses the best semi-supervised generative models by a large margin and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 421, + 412, + 433 + ], + "spans": [ + { + "bbox": [ + 141, + 421, + 412, + 433 + ], + "score": 1.0, + "content": "achieves state-of-the-art performance on Cifar-100 with 10k labels.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18.5, + "bbox_fs": [ + 131, + 335, + 506, + 433 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 453, + 335, + 466 + ], + "lines": [ + { + "bbox": [ + 104, + 452, + 336, + 469 + ], + "spans": [ + { + "bbox": [ + 104, + 452, + 336, + 469 + ], + "score": 1.0, + "content": "2 SEMI-SUPERVISED LEARNING METHODS", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 479, + 505, + 545 + ], + "lines": [ + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "score": 1.0, + "content": "In supervised learning (SL), we are facing with training data that appears as input-target pairs", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 490, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 107, + 490, + 165, + 502 + ], + "score": 0.89, + "content": "( \\mathbf { X } , \\bar { \\mathbf { y } } ) \\ \\in \\ \\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 490, + 332, + 503 + ], + "score": 1.0, + "content": "sampled from an unknown distribution", + "type": "text" + }, + { + "bbox": [ + 332, + 490, + 365, + 502 + ], + "score": 0.93, + "content": "p ( \\mathbf { X } , \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 490, + 505, + 503 + ], + "score": 1.0, + "content": ". Our goal is to learn a function", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 501, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 107, + 502, + 141, + 513 + ], + "score": 0.91, + "content": "f ( \\mathbf { X } ; \\phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 501, + 217, + 514 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 218, + 502, + 226, + 513 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 501, + 363, + 514 + ], + "score": 1.0, + "content": "that makes the correct inference", + "type": "text" + }, + { + "bbox": [ + 363, + 503, + 371, + 513 + ], + "score": 0.37, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 501, + 478, + 514 + ], + "score": 1.0, + "content": "for unseen samples from", + "type": "text" + }, + { + "bbox": [ + 479, + 501, + 501, + 513 + ], + "score": 0.91, + "content": "p ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 501, + 505, + 514 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 511, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 506, + 524 + ], + "score": 1.0, + "content": "While in semi-supervised learning (SSL), we can obtain an extra collection of unlabeled data", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 522, + 504, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 145, + 534 + ], + "score": 0.9, + "content": "\\mathbf { X } \\in \\mathbb { D } _ { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 522, + 294, + 536 + ], + "score": 1.0, + "content": "sampled from the same distribution", + "type": "text" + }, + { + "bbox": [ + 295, + 523, + 317, + 535 + ], + "score": 0.91, + "content": "p ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 522, + 489, + 536 + ], + "score": 1.0, + "content": ". We hope to leverage the data from both", + "type": "text" + }, + { + "bbox": [ + 489, + 523, + 504, + 534 + ], + "score": 0.87, + "content": "\\mathbb { D } _ { L }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 532, + 500, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 123, + 548 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 534, + 138, + 545 + ], + "score": 0.87, + "content": "\\mathbb { D } _ { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 532, + 479, + 548 + ], + "score": 1.0, + "content": "to achieve a more accurate model than what would have been obtained by only using", + "type": "text" + }, + { + "bbox": [ + 480, + 534, + 494, + 545 + ], + "score": 0.87, + "content": "\\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 532, + 500, + 548 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 478, + 506, + 548 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 551, + 505, + 595 + ], + "lines": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "In this section, we review some existing methods for SSL. We mainly focus on those who have", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 562, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 574 + ], + "score": 1.0, + "content": "reached state-of-the-art results, as well as generative approaches which are strongly connected with", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "score": 1.0, + "content": "our model; the more comprehensive overview is beyond the scope of this paper, we refer readers to", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 584, + 190, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 190, + 596 + ], + "score": 1.0, + "content": "(Oliver et al., 2018).", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 550, + 506, + 596 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 608, + 282, + 620 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 282, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 282, + 621 + ], + "score": 1.0, + "content": "2.1 DISAGREEMENT BASED LEARNING", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 629, + 505, + 685 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 506, + 644 + ], + "score": 1.0, + "content": "Disagreement-based learning refers to the general approaches of imposing disagreement among", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 640, + 505, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 654 + ], + "score": 1.0, + "content": "multiple learners on the same task or multiple predictions from a single learner. By eliminating the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 651, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 505, + 665 + ], + "score": 1.0, + "content": "disagreement, we can enforce the generalization of the model on unseen data. A common technique", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 662, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 676 + ], + "score": 1.0, + "content": "for creating disagreement is data augmentation, which applies transformations or perturbations on", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 673, + 485, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 336, + 686 + ], + "score": 1.0, + "content": "the input data and leaves class semantics unchanged. For", + "type": "text" + }, + { + "bbox": [ + 336, + 673, + 371, + 685 + ], + "score": 0.91, + "content": "\\mathbf { X } \\in \\mathbb { D } _ { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 673, + 485, + 686 + ], + "score": 1.0, + "content": ", loss term can be derived as", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 628, + 506, + 686 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 690, + 376, + 704 + ], + "lines": [ + { + "bbox": [ + 234, + 690, + 376, + 704 + ], + "spans": [ + { + "bbox": [ + 234, + 690, + 376, + 704 + ], + "score": 0.9, + "content": "\\| f ( \\operatorname { A u g m e n t } ( \\mathbf { X } ) ; \\phi ) - f ( \\mathbf { X } ; \\phi ) \\| _ { 2 } ^ { 2 }", + "type": "interline_equation", + "image_path": "35a0f9cac4ba95ade39d9324cca98481124ca123c67fcfaa3c6e63eb03b2a5ef.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 234, + 690, + 376, + 704 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 185, + 722 + ], + "score": 1.0, + "content": "where the Augment", + "type": "text" + }, + { + "bbox": [ + 185, + 710, + 202, + 721 + ], + "score": 0.66, + "content": "( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "is a stochastic function which can be obtained by image transformation (Xie", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "et al., 2019), virtual adversarial training (Miyato et al., 2019), or mixup method (Verma et al. 2018;", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Berthelot et al. 2019). Another disagreement construction technique is to train multiple learners on", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 251, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 251, + 105 + ], + "score": 1.0, + "content": "the same dataset and utilize the loss", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 252, + 106, + 359, + 121 + ], + "lines": [ + { + "bbox": [ + 252, + 106, + 359, + 121 + ], + "spans": [ + { + "bbox": [ + 252, + 106, + 359, + 121 + ], + "score": 0.91, + "content": "\\| f ( \\mathbf { X } ; \\boldsymbol { \\phi } _ { 1 } ) - f ( \\mathbf { X } ; \\boldsymbol { \\phi } _ { 2 } ) \\| _ { 2 } ^ { 2 }", + "type": "interline_equation", + "image_path": "0c004f168acc94a1992074a86420002524f703ff025de068d36c9ba7bc30ae81.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 252, + 106, + 359, + 121 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 122, + 506, + 156 + ], + "lines": [ + { + "bbox": [ + 106, + 122, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 505, + 135 + ], + "score": 1.0, + "content": "to enforce the predictive consistency of different models, for example, “Mean Teacher” (Tarvainen", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 133, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 505, + 145 + ], + "score": 1.0, + "content": "& Valpola, 2017) and “Teacher Graph” (Luo et al., 2018). The generalization of the models gets", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 145, + 149, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 149, + 156 + ], + "score": 1.0, + "content": "enhanced.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 108, + 168, + 247, + 180 + ], + "lines": [ + { + "bbox": [ + 105, + 167, + 249, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 249, + 182 + ], + "score": 1.0, + "content": "2.2 GENERATIVE APPROACHES", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 189, + 504, + 212 + ], + "lines": [ + { + "bbox": [ + 105, + 189, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 233, + 202 + ], + "score": 1.0, + "content": "In generative approaches, input", + "type": "text" + }, + { + "bbox": [ + 234, + 190, + 244, + 200 + ], + "score": 0.55, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 189, + 506, + 202 + ], + "score": 1.0, + "content": "is supposed to have corresponding continuous and discrete latent", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 199, + 313, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 229, + 213 + ], + "score": 1.0, + "content": "variables, which we denote by", + "type": "text" + }, + { + "bbox": [ + 230, + 202, + 236, + 210 + ], + "score": 0.49, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 199, + 313, + 213 + ], + "score": 1.0, + "content": "and c respectively.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 216, + 505, + 284 + ], + "lines": [ + { + "bbox": [ + 106, + 217, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 505, + 229 + ], + "score": 1.0, + "content": "Feature matching (FM) GANs (Salimans et al., 2016; Dai et al., 2017) apply GANs to semi-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 228, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 352, + 240 + ], + "score": 1.0, + "content": "supervised learning on K-classification tasks by specifying a", + "type": "text" + }, + { + "bbox": [ + 352, + 229, + 377, + 240 + ], + "score": 0.83, + "content": "( \\mathsf { K } { + } 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 228, + 505, + 240 + ], + "score": 1.0, + "content": "-class objective for the discrim-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 238, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 505, + 252 + ], + "score": 1.0, + "content": "inator. Instead of binary classification, true samples are classified into the first K classes respec-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 249, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 293, + 263 + ], + "score": 1.0, + "content": "tively and fake samples are classified into the", + "type": "text" + }, + { + "bbox": [ + 294, + 250, + 318, + 262 + ], + "score": 0.8, + "content": "( \\mathsf { K } { + } 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 249, + 505, + 263 + ], + "score": 1.0, + "content": "-th class. This target function achieves strong", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "empirical results by matching the generator distribution with true data distribution and improves", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 272, + 284, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 284, + 285 + ], + "score": 1.0, + "content": "semi-supervised classification performance.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 289, + 504, + 322 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "Semi-supervised VAEs (Kingma et al. 2014; Narayanaswamy et al. 2017) construct a probabilistic", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 208, + 312 + ], + "score": 1.0, + "content": "model parameterized by", + "type": "text" + }, + { + "bbox": [ + 208, + 300, + 216, + 310 + ], + "score": 0.76, + "content": "\\pmb \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 300, + 235, + 312 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 236, + 300, + 245, + 311 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 300, + 505, + 312 + ], + "score": 1.0, + "content": "that respectively describe the generation and inference process", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 310, + 430, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 142, + 324 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 142, + 311, + 152, + 321 + ], + "score": 0.58, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 310, + 232, + 324 + ], + "score": 1.0, + "content": "and latent variables", + "type": "text" + }, + { + "bbox": [ + 233, + 313, + 239, + 321 + ], + "score": 0.46, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 310, + 360, + 324 + ], + "score": 1.0, + "content": ", c. The generation process of", + "type": "text" + }, + { + "bbox": [ + 360, + 312, + 370, + 321 + ], + "score": 0.6, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 310, + 383, + 324 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 383, + 313, + 390, + 321 + ], + "score": 0.68, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 310, + 407, + 324 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 313, + 414, + 321 + ], + "score": 0.54, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 310, + 430, + 324 + ], + "score": 1.0, + "content": "is :", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 324, + 468, + 338 + ], + "lines": [ + { + "bbox": [ + 144, + 324, + 468, + 338 + ], + "spans": [ + { + "bbox": [ + 144, + 324, + 468, + 338 + ], + "score": 0.88, + "content": "p ( \\mathbf { z } ) = { \\mathcal { N } } ( z ; \\mathbf { 0 } , I ) ; \\qquad p ( \\mathbf { c } ) = \\mathbf { M } \\mathbf { u } \\mathbf { l } \\mathbf { t } ( \\mathbf { c } ; K , \\pi ) ; \\qquad p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) = f ( \\mathbf { X } ; \\mathbf { z } , \\mathbf { c } , \\theta )", + "type": "interline_equation", + "image_path": "6a0e722f34fadd3f21f6961319e121414b4e995300bd18aa4894b8b3fe107b3d.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 144, + 324, + 468, + 338 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 133, + 352 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 340, + 182, + 352 + ], + "score": 0.92, + "content": "\\operatorname { M u l t } ( K , \\pi )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 339, + 349, + 352 + ], + "score": 1.0, + "content": "is the multinomial distribution with class", + "type": "text" + }, + { + "bbox": [ + 350, + 340, + 360, + 350 + ], + "score": 0.84, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 339, + 421, + 352 + ], + "score": 1.0, + "content": "and parameter", + "type": "text" + }, + { + "bbox": [ + 421, + 342, + 429, + 350 + ], + "score": 0.65, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 339, + 434, + 352 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 435, + 340, + 487, + 352 + ], + "score": 0.91, + "content": "f ( \\mathbf { X } ; \\mathbf { z } , \\mathbf { c } , \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "is a", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 351, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 505, + 363 + ], + "score": 1.0, + "content": "suitable likelihood function, e.g. a Bernoulli or Gaussian distribution, parameterized by a non-linear", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 259, + 375 + ], + "score": 1.0, + "content": "transformation of the latent variables", + "type": "text" + }, + { + "bbox": [ + 260, + 364, + 267, + 372 + ], + "score": 0.65, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 362, + 286, + 375 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 286, + 364, + 293, + 372 + ], + "score": 0.4, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 362, + 363, + 375 + ], + "score": 1.0, + "content": ". The class label", + "type": "text" + }, + { + "bbox": [ + 363, + 363, + 372, + 373 + ], + "score": 0.68, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "is treated as c if given. For the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 374, + 302, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 302, + 385 + ], + "score": 1.0, + "content": "inference process, with the following hypothesis", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5 + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 386, + 483, + 400 + ], + "lines": [ + { + "bbox": [ + 117, + 386, + 483, + 400 + ], + "spans": [ + { + "bbox": [ + 117, + 386, + 483, + 400 + ], + "score": 0.86, + "content": "\\begin{array} { r } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) = q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) ; \\qquad p ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) = p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) ; \\qquad p ( \\mathbf { z } , \\mathbf { c } ) = p ( \\mathbf { z } ) p ( \\mathbf { c } ) } \\end{array}", + "type": "interline_equation", + "image_path": "7eeab28a239b62ea0da33b0354545d840db957c2a80c354b5f78cb9df6883931.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 117, + 386, + 483, + 400 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 402, + 504, + 424 + ], + "lines": [ + { + "bbox": [ + 106, + 402, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 506, + 414 + ], + "score": 1.0, + "content": "evidence lower bound (ELBO) is used as objective to predict the posterior distribution of latent", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 413, + 309, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 309, + 426 + ], + "score": 1.0, + "content": "variables as follows (see Appendix A.1 for proof):", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 426, + 504, + 441 + ], + "lines": [ + { + "bbox": [ + 111, + 426, + 504, + 441 + ], + "spans": [ + { + "bbox": [ + 111, + 426, + 504, + 441 + ], + "score": 0.9, + "content": "\\log p ( \\mathbf { X } ) \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } ) ) = \\mathrm { E L B O } ( \\phi ) ,", + "type": "interline_equation", + "image_path": "75ddcd8bbd9da9ee1832cdb364751c3d55dd3b48cb65b32a7cc354352c9cb2f7.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 111, + 426, + 504, + 441 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 449, + 505, + 494 + ], + "lines": [ + { + "bbox": [ + 105, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 181, + 461 + ], + "score": 1.0, + "content": "For the likelihood", + "type": "text" + }, + { + "bbox": [ + 181, + 449, + 218, + 461 + ], + "score": 0.92, + "content": "\\log p ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "is infeasible, VAE maximizes its evidence lower bound instead, which", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 271, + 473 + ], + "score": 1.0, + "content": "derives the negative ELBO loss function", + "type": "text" + }, + { + "bbox": [ + 271, + 460, + 364, + 473 + ], + "score": 0.9, + "content": "\\mathcal { L } ( \\mathbf { X } ; \\boldsymbol { \\phi } , \\pmb { \\theta } ) = - \\mathrm { E L B O }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 460, + 505, + 473 + ], + "score": 1.0, + "content": ". The predictions for the classifica-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 146, + 484 + ], + "score": 1.0, + "content": "tion label", + "type": "text" + }, + { + "bbox": [ + 146, + 473, + 154, + 483 + ], + "score": 0.25, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 471, + 376, + 484 + ], + "score": 1.0, + "content": "can be obtained from the inferred posterior distribution", + "type": "text" + }, + { + "bbox": [ + 377, + 471, + 412, + 483 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 471, + 487, + 484 + ], + "score": 1.0, + "content": ". When class label", + "type": "text" + }, + { + "bbox": [ + 487, + 473, + 495, + 483 + ], + "score": 0.63, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 482, + 425, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 377, + 495 + ], + "score": 1.0, + "content": "not given, missing label sampling technique is used to sample from", + "type": "text" + }, + { + "bbox": [ + 377, + 483, + 412, + 495 + ], + "score": 0.92, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 482, + 425, + 495 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28.5 + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 496, + 405, + 531 + ], + "lines": [ + { + "bbox": [ + 206, + 496, + 405, + 531 + ], + "spans": [ + { + "bbox": [ + 206, + 496, + 405, + 531 + ], + "score": 0.93, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } f ( \\mathbf { X } ; \\mathbf { c } , \\theta ) = \\sum _ { k = 1 } ^ { K } q _ { \\phi } ( \\mathbf { y } _ { k } | \\mathbf { X } ) f ( \\mathbf { X } ; \\mathbf { y } _ { k } , \\theta )", + "type": "interline_equation", + "image_path": "4f1c9d93744e0c9b9802fb463e544d7fb080013f2aff04c0adf6d01fa83e9490.jpg" + } + ] + } + ], + "index": 31.5, + "virtual_lines": [ + { + "bbox": [ + 206, + 496, + 405, + 513.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 206, + 513.5, + 405, + 531.0 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 532, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 129, + 545 + ], + "score": 1.0, + "content": "Here", + "type": "text" + }, + { + "bbox": [ + 129, + 534, + 142, + 544 + ], + "score": 0.85, + "content": "\\mathbf { y } _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 532, + 300, + 545 + ], + "score": 1.0, + "content": "represents a one-hot vector with 1 in", + "type": "text" + }, + { + "bbox": [ + 301, + 533, + 308, + 542 + ], + "score": 0.71, + "content": "\\mathbf { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "-th dimension. Utilizing this sampling method,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 326, + 556 + ], + "score": 1.0, + "content": "the algorithmic complexity of VAE is proportional to", + "type": "text" + }, + { + "bbox": [ + 326, + 544, + 335, + 553 + ], + "score": 0.35, + "content": "\\mathrm { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "so it is computationally inefficient. Note", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 553, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 326, + 567 + ], + "score": 1.0, + "content": "that in objective (4), the label predictive distribution", + "type": "text" + }, + { + "bbox": [ + 327, + 554, + 362, + 567 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 553, + 506, + 567 + ], + "score": 1.0, + "content": "only contributes to the generative", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "score": 1.0, + "content": "performance. To remedy this, existing models simply add a cross-entropy loss to the negative ELBO", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 576, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 226, + 588 + ], + "score": 1.0, + "content": "loss such that the distribution", + "type": "text" + }, + { + "bbox": [ + 226, + 576, + 262, + 588 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 576, + 506, + 588 + ], + "score": 1.0, + "content": "can also learn classification rules from the labeled data. The", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 587, + 212, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 212, + 600 + ], + "score": 1.0, + "content": "extended objective loss is", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5 + }, + { + "type": "interline_equation", + "bbox": [ + 158, + 600, + 454, + 620 + ], + "lines": [ + { + "bbox": [ + 158, + 600, + 454, + 620 + ], + "spans": [ + { + "bbox": [ + 158, + 600, + 454, + 620 + ], + "score": 0.89, + "content": "\\operatorname* { m i n } _ { \\phi , \\theta } \\mathbb { E } _ { \\mathbf { X } \\sim \\mathbb { D } _ { U } } \\mathcal { L } \\big ( \\mathbf { X } ; \\phi , \\pmb { \\theta } ) + \\mathbb { E } _ { ( \\mathbf { X } , \\mathbf { y } ) \\sim \\mathbb { D } _ { L } } \\big [ \\mathcal { L } \\big ( \\mathbf { X } , \\mathbf { c } = \\mathbf { y } ; \\phi , \\pmb { \\theta } \\big ) - \\log q _ { \\phi } ( \\mathbf { y } | \\mathbf { X } ) \\big ]", + "type": "interline_equation", + "image_path": "923d4e4c4856ff8134d5593b9e08d387ca6c768d3fb16dd65fd0b636372c03b2.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 158, + 600, + 454, + 620 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 621, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "Two-stage Training Strategy: In practice, (Kingma et al., 2014) finds that directly training the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 631, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 506, + 647 + ], + "score": 1.0, + "content": "one-stage objective (7) will lead to a bad semi-supervised learning result, so a two-stage training", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "strategy is proposed to improve the model. The two-stage training strategy consists of two parts,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 406, + 667 + ], + "score": 1.0, + "content": "M1 and M2. M1 means to learn a new continuous latent representation", + "type": "text" + }, + { + "bbox": [ + 406, + 657, + 417, + 666 + ], + "score": 0.86, + "content": "\\mathbf { z } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "first, and M2 means", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 665, + 504, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 342, + 678 + ], + "score": 1.0, + "content": "to train a semi-supervised model (7) with the embedding", + "type": "text" + }, + { + "bbox": [ + 342, + 667, + 353, + 677 + ], + "score": 0.85, + "content": "\\mathbf { z } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 665, + 491, + 678 + ], + "score": 1.0, + "content": "from M1 instead of the raw data", + "type": "text" + }, + { + "bbox": [ + 492, + 666, + 501, + 676 + ], + "score": 0.54, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 665, + 504, + 678 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 126, + 690 + ], + "score": 1.0, + "content": "This", + "type": "text" + }, + { + "bbox": [ + 127, + 677, + 162, + 687 + ], + "score": 0.83, + "content": "\\mathbf { M } 1 { + } \\mathbf { M } 2", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 677, + 427, + 690 + ], + "score": 1.0, + "content": "strategy builds a deep VAE with two layers of random variables:", + "type": "text" + }, + { + "bbox": [ + 427, + 677, + 505, + 689 + ], + "score": 0.9, + "content": "\\begin{array} { r l r } { \\mathrm { \\nabla } p _ { \\pmb \\theta } ( \\mathbf { X } , \\mathbf { z } _ { 1 } , \\mathbf { z } _ { 2 } , \\mathbf { c } ) } & { { } = } & { } \\end{array}", + "type": "inline_equation" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 687, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 236, + 700 + ], + "score": 0.92, + "content": "p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { z } _ { 1 } ) p _ { \\pmb { \\theta } } ( \\mathbf { z } _ { 1 } | \\mathbf { z } _ { 2 } , \\bar { \\mathbf { c } } ) p ( \\mathbf { z } _ { 2 } ) p ( \\mathbf { c } )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 687, + 506, + 701 + ], + "score": 1.0, + "content": ", which can dramatically improve the performance of the inference", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 107, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 107, + 699, + 142, + 711 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "but is not robust in training. Moreover, GAN’s training process can also be considered as", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "two-stage with generator and discriminator competing with each other, and the two-stage adversarial", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 251, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 251, + 734 + ], + "score": 1.0, + "content": "game adds the difficulty in training.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 44.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Berthelot et al. 2019). Another disagreement construction technique is to train multiple learners on", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 251, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 251, + 105 + ], + "score": 1.0, + "content": "the same dataset and utilize the loss", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 505, + 105 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 252, + 106, + 359, + 121 + ], + "lines": [ + { + "bbox": [ + 252, + 106, + 359, + 121 + ], + "spans": [ + { + "bbox": [ + 252, + 106, + 359, + 121 + ], + "score": 0.91, + "content": "\\| f ( \\mathbf { X } ; \\boldsymbol { \\phi } _ { 1 } ) - f ( \\mathbf { X } ; \\boldsymbol { \\phi } _ { 2 } ) \\| _ { 2 } ^ { 2 }", + "type": "interline_equation", + "image_path": "0c004f168acc94a1992074a86420002524f703ff025de068d36c9ba7bc30ae81.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 252, + 106, + 359, + 121 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 122, + 506, + 156 + ], + "lines": [ + { + "bbox": [ + 106, + 122, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 505, + 135 + ], + "score": 1.0, + "content": "to enforce the predictive consistency of different models, for example, “Mean Teacher” (Tarvainen", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 133, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 505, + 145 + ], + "score": 1.0, + "content": "& Valpola, 2017) and “Teacher Graph” (Luo et al., 2018). The generalization of the models gets", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 145, + 149, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 149, + 156 + ], + "score": 1.0, + "content": "enhanced.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 122, + 505, + 156 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 168, + 247, + 180 + ], + "lines": [ + { + "bbox": [ + 105, + 167, + 249, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 249, + 182 + ], + "score": 1.0, + "content": "2.2 GENERATIVE APPROACHES", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 189, + 504, + 212 + ], + "lines": [ + { + "bbox": [ + 105, + 189, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 233, + 202 + ], + "score": 1.0, + "content": "In generative approaches, input", + "type": "text" + }, + { + "bbox": [ + 234, + 190, + 244, + 200 + ], + "score": 0.55, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 189, + 506, + 202 + ], + "score": 1.0, + "content": "is supposed to have corresponding continuous and discrete latent", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 199, + 313, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 229, + 213 + ], + "score": 1.0, + "content": "variables, which we denote by", + "type": "text" + }, + { + "bbox": [ + 230, + 202, + 236, + 210 + ], + "score": 0.49, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 199, + 313, + 213 + ], + "score": 1.0, + "content": "and c respectively.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 189, + 506, + 213 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 216, + 505, + 284 + ], + "lines": [ + { + "bbox": [ + 106, + 217, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 505, + 229 + ], + "score": 1.0, + "content": "Feature matching (FM) GANs (Salimans et al., 2016; Dai et al., 2017) apply GANs to semi-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 228, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 352, + 240 + ], + "score": 1.0, + "content": "supervised learning on K-classification tasks by specifying a", + "type": "text" + }, + { + "bbox": [ + 352, + 229, + 377, + 240 + ], + "score": 0.83, + "content": "( \\mathsf { K } { + } 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 228, + 505, + 240 + ], + "score": 1.0, + "content": "-class objective for the discrim-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 238, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 505, + 252 + ], + "score": 1.0, + "content": "inator. Instead of binary classification, true samples are classified into the first K classes respec-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 249, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 293, + 263 + ], + "score": 1.0, + "content": "tively and fake samples are classified into the", + "type": "text" + }, + { + "bbox": [ + 294, + 250, + 318, + 262 + ], + "score": 0.8, + "content": "( \\mathsf { K } { + } 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 249, + 505, + 263 + ], + "score": 1.0, + "content": "-th class. This target function achieves strong", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "empirical results by matching the generator distribution with true data distribution and improves", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 272, + 284, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 284, + 285 + ], + "score": 1.0, + "content": "semi-supervised classification performance.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 217, + 505, + 285 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 289, + 504, + 322 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "Semi-supervised VAEs (Kingma et al. 2014; Narayanaswamy et al. 2017) construct a probabilistic", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 208, + 312 + ], + "score": 1.0, + "content": "model parameterized by", + "type": "text" + }, + { + "bbox": [ + 208, + 300, + 216, + 310 + ], + "score": 0.76, + "content": "\\pmb \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 300, + 235, + 312 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 236, + 300, + 245, + 311 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 300, + 505, + 312 + ], + "score": 1.0, + "content": "that respectively describe the generation and inference process", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 310, + 430, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 142, + 324 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 142, + 311, + 152, + 321 + ], + "score": 0.58, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 310, + 232, + 324 + ], + "score": 1.0, + "content": "and latent variables", + "type": "text" + }, + { + "bbox": [ + 233, + 313, + 239, + 321 + ], + "score": 0.46, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 310, + 360, + 324 + ], + "score": 1.0, + "content": ", c. The generation process of", + "type": "text" + }, + { + "bbox": [ + 360, + 312, + 370, + 321 + ], + "score": 0.6, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 310, + 383, + 324 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 383, + 313, + 390, + 321 + ], + "score": 0.68, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 310, + 407, + 324 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 313, + 414, + 321 + ], + "score": 0.54, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 310, + 430, + 324 + ], + "score": 1.0, + "content": "is :", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 289, + 505, + 324 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 324, + 468, + 338 + ], + "lines": [ + { + "bbox": [ + 144, + 324, + 468, + 338 + ], + "spans": [ + { + "bbox": [ + 144, + 324, + 468, + 338 + ], + "score": 0.88, + "content": "p ( \\mathbf { z } ) = { \\mathcal { N } } ( z ; \\mathbf { 0 } , I ) ; \\qquad p ( \\mathbf { c } ) = \\mathbf { M } \\mathbf { u } \\mathbf { l } \\mathbf { t } ( \\mathbf { c } ; K , \\pi ) ; \\qquad p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) = f ( \\mathbf { X } ; \\mathbf { z } , \\mathbf { c } , \\theta )", + "type": "interline_equation", + "image_path": "6a0e722f34fadd3f21f6961319e121414b4e995300bd18aa4894b8b3fe107b3d.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 144, + 324, + 468, + 338 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 133, + 352 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 340, + 182, + 352 + ], + "score": 0.92, + "content": "\\operatorname { M u l t } ( K , \\pi )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 339, + 349, + 352 + ], + "score": 1.0, + "content": "is the multinomial distribution with class", + "type": "text" + }, + { + "bbox": [ + 350, + 340, + 360, + 350 + ], + "score": 0.84, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 339, + 421, + 352 + ], + "score": 1.0, + "content": "and parameter", + "type": "text" + }, + { + "bbox": [ + 421, + 342, + 429, + 350 + ], + "score": 0.65, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 339, + 434, + 352 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 435, + 340, + 487, + 352 + ], + "score": 0.91, + "content": "f ( \\mathbf { X } ; \\mathbf { z } , \\mathbf { c } , \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "is a", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 351, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 505, + 363 + ], + "score": 1.0, + "content": "suitable likelihood function, e.g. a Bernoulli or Gaussian distribution, parameterized by a non-linear", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 259, + 375 + ], + "score": 1.0, + "content": "transformation of the latent variables", + "type": "text" + }, + { + "bbox": [ + 260, + 364, + 267, + 372 + ], + "score": 0.65, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 362, + 286, + 375 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 286, + 364, + 293, + 372 + ], + "score": 0.4, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 362, + 363, + 375 + ], + "score": 1.0, + "content": ". The class label", + "type": "text" + }, + { + "bbox": [ + 363, + 363, + 372, + 373 + ], + "score": 0.68, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "is treated as c if given. For the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 374, + 302, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 302, + 385 + ], + "score": 1.0, + "content": "inference process, with the following hypothesis", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 339, + 506, + 385 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 386, + 483, + 400 + ], + "lines": [ + { + "bbox": [ + 117, + 386, + 483, + 400 + ], + "spans": [ + { + "bbox": [ + 117, + 386, + 483, + 400 + ], + "score": 0.86, + "content": "\\begin{array} { r } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) = q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) ; \\qquad p ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) = p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) ; \\qquad p ( \\mathbf { z } , \\mathbf { c } ) = p ( \\mathbf { z } ) p ( \\mathbf { c } ) } \\end{array}", + "type": "interline_equation", + "image_path": "7eeab28a239b62ea0da33b0354545d840db957c2a80c354b5f78cb9df6883931.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 117, + 386, + 483, + 400 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 402, + 504, + 424 + ], + "lines": [ + { + "bbox": [ + 106, + 402, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 506, + 414 + ], + "score": 1.0, + "content": "evidence lower bound (ELBO) is used as objective to predict the posterior distribution of latent", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 413, + 309, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 309, + 426 + ], + "score": 1.0, + "content": "variables as follows (see Appendix A.1 for proof):", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 402, + 506, + 426 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 426, + 504, + 441 + ], + "lines": [ + { + "bbox": [ + 111, + 426, + 504, + 441 + ], + "spans": [ + { + "bbox": [ + 111, + 426, + 504, + 441 + ], + "score": 0.9, + "content": "\\log p ( \\mathbf { X } ) \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } ) ) = \\mathrm { E L B O } ( \\phi ) ,", + "type": "interline_equation", + "image_path": "75ddcd8bbd9da9ee1832cdb364751c3d55dd3b48cb65b32a7cc354352c9cb2f7.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 111, + 426, + 504, + 441 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 449, + 505, + 494 + ], + "lines": [ + { + "bbox": [ + 105, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 181, + 461 + ], + "score": 1.0, + "content": "For the likelihood", + "type": "text" + }, + { + "bbox": [ + 181, + 449, + 218, + 461 + ], + "score": 0.92, + "content": "\\log p ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "is infeasible, VAE maximizes its evidence lower bound instead, which", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 271, + 473 + ], + "score": 1.0, + "content": "derives the negative ELBO loss function", + "type": "text" + }, + { + "bbox": [ + 271, + 460, + 364, + 473 + ], + "score": 0.9, + "content": "\\mathcal { L } ( \\mathbf { X } ; \\boldsymbol { \\phi } , \\pmb { \\theta } ) = - \\mathrm { E L B O }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 460, + 505, + 473 + ], + "score": 1.0, + "content": ". The predictions for the classifica-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 146, + 484 + ], + "score": 1.0, + "content": "tion label", + "type": "text" + }, + { + "bbox": [ + 146, + 473, + 154, + 483 + ], + "score": 0.25, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 471, + 376, + 484 + ], + "score": 1.0, + "content": "can be obtained from the inferred posterior distribution", + "type": "text" + }, + { + "bbox": [ + 377, + 471, + 412, + 483 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 471, + 487, + 484 + ], + "score": 1.0, + "content": ". When class label", + "type": "text" + }, + { + "bbox": [ + 487, + 473, + 495, + 483 + ], + "score": 0.63, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 482, + 425, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 377, + 495 + ], + "score": 1.0, + "content": "not given, missing label sampling technique is used to sample from", + "type": "text" + }, + { + "bbox": [ + 377, + 483, + 412, + 495 + ], + "score": 0.92, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 482, + 425, + 495 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 449, + 506, + 495 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 496, + 405, + 531 + ], + "lines": [ + { + "bbox": [ + 206, + 496, + 405, + 531 + ], + "spans": [ + { + "bbox": [ + 206, + 496, + 405, + 531 + ], + "score": 0.93, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } f ( \\mathbf { X } ; \\mathbf { c } , \\theta ) = \\sum _ { k = 1 } ^ { K } q _ { \\phi } ( \\mathbf { y } _ { k } | \\mathbf { X } ) f ( \\mathbf { X } ; \\mathbf { y } _ { k } , \\theta )", + "type": "interline_equation", + "image_path": "4f1c9d93744e0c9b9802fb463e544d7fb080013f2aff04c0adf6d01fa83e9490.jpg" + } + ] + } + ], + "index": 31.5, + "virtual_lines": [ + { + "bbox": [ + 206, + 496, + 405, + 513.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 206, + 513.5, + 405, + 531.0 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 532, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 129, + 545 + ], + "score": 1.0, + "content": "Here", + "type": "text" + }, + { + "bbox": [ + 129, + 534, + 142, + 544 + ], + "score": 0.85, + "content": "\\mathbf { y } _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 532, + 300, + 545 + ], + "score": 1.0, + "content": "represents a one-hot vector with 1 in", + "type": "text" + }, + { + "bbox": [ + 301, + 533, + 308, + 542 + ], + "score": 0.71, + "content": "\\mathbf { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "-th dimension. Utilizing this sampling method,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 326, + 556 + ], + "score": 1.0, + "content": "the algorithmic complexity of VAE is proportional to", + "type": "text" + }, + { + "bbox": [ + 326, + 544, + 335, + 553 + ], + "score": 0.35, + "content": "\\mathrm { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "so it is computationally inefficient. Note", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 553, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 326, + 567 + ], + "score": 1.0, + "content": "that in objective (4), the label predictive distribution", + "type": "text" + }, + { + "bbox": [ + 327, + 554, + 362, + 567 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 553, + 506, + 567 + ], + "score": 1.0, + "content": "only contributes to the generative", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "score": 1.0, + "content": "performance. To remedy this, existing models simply add a cross-entropy loss to the negative ELBO", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 576, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 226, + 588 + ], + "score": 1.0, + "content": "loss such that the distribution", + "type": "text" + }, + { + "bbox": [ + 226, + 576, + 262, + 588 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 576, + 506, + 588 + ], + "score": 1.0, + "content": "can also learn classification rules from the labeled data. The", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 587, + 212, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 212, + 600 + ], + "score": 1.0, + "content": "extended objective loss is", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 532, + 506, + 600 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 158, + 600, + 454, + 620 + ], + "lines": [ + { + "bbox": [ + 158, + 600, + 454, + 620 + ], + "spans": [ + { + "bbox": [ + 158, + 600, + 454, + 620 + ], + "score": 0.89, + "content": "\\operatorname* { m i n } _ { \\phi , \\theta } \\mathbb { E } _ { \\mathbf { X } \\sim \\mathbb { D } _ { U } } \\mathcal { L } \\big ( \\mathbf { X } ; \\phi , \\pmb { \\theta } ) + \\mathbb { E } _ { ( \\mathbf { X } , \\mathbf { y } ) \\sim \\mathbb { D } _ { L } } \\big [ \\mathcal { L } \\big ( \\mathbf { X } , \\mathbf { c } = \\mathbf { y } ; \\phi , \\pmb { \\theta } \\big ) - \\log q _ { \\phi } ( \\mathbf { y } | \\mathbf { X } ) \\big ]", + "type": "interline_equation", + "image_path": "923d4e4c4856ff8134d5593b9e08d387ca6c768d3fb16dd65fd0b636372c03b2.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 158, + 600, + 454, + 620 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 621, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "Two-stage Training Strategy: In practice, (Kingma et al., 2014) finds that directly training the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 631, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 506, + 647 + ], + "score": 1.0, + "content": "one-stage objective (7) will lead to a bad semi-supervised learning result, so a two-stage training", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "strategy is proposed to improve the model. The two-stage training strategy consists of two parts,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 406, + 667 + ], + "score": 1.0, + "content": "M1 and M2. M1 means to learn a new continuous latent representation", + "type": "text" + }, + { + "bbox": [ + 406, + 657, + 417, + 666 + ], + "score": 0.86, + "content": "\\mathbf { z } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "first, and M2 means", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 665, + 504, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 342, + 678 + ], + "score": 1.0, + "content": "to train a semi-supervised model (7) with the embedding", + "type": "text" + }, + { + "bbox": [ + 342, + 667, + 353, + 677 + ], + "score": 0.85, + "content": "\\mathbf { z } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 665, + 491, + 678 + ], + "score": 1.0, + "content": "from M1 instead of the raw data", + "type": "text" + }, + { + "bbox": [ + 492, + 666, + 501, + 676 + ], + "score": 0.54, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 665, + 504, + 678 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 126, + 690 + ], + "score": 1.0, + "content": "This", + "type": "text" + }, + { + "bbox": [ + 127, + 677, + 162, + 687 + ], + "score": 0.83, + "content": "\\mathbf { M } 1 { + } \\mathbf { M } 2", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 677, + 427, + 690 + ], + "score": 1.0, + "content": "strategy builds a deep VAE with two layers of random variables:", + "type": "text" + }, + { + "bbox": [ + 427, + 677, + 505, + 689 + ], + "score": 0.9, + "content": "\\begin{array} { r l r } { \\mathrm { \\nabla } p _ { \\pmb \\theta } ( \\mathbf { X } , \\mathbf { z } _ { 1 } , \\mathbf { z } _ { 2 } , \\mathbf { c } ) } & { { } = } & { } \\end{array}", + "type": "inline_equation" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 687, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 236, + 700 + ], + "score": 0.92, + "content": "p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { z } _ { 1 } ) p _ { \\pmb { \\theta } } ( \\mathbf { z } _ { 1 } | \\mathbf { z } _ { 2 } , \\bar { \\mathbf { c } } ) p ( \\mathbf { z } _ { 2 } ) p ( \\mathbf { c } )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 687, + 506, + 701 + ], + "score": 1.0, + "content": ", which can dramatically improve the performance of the inference", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 107, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 107, + 699, + 142, + 711 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "but is not robust in training. Moreover, GAN’s training process can also be considered as", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "two-stage with generator and discriminator competing with each other, and the two-stage adversarial", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 251, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 251, + 734 + ], + "score": 1.0, + "content": "game adds the difficulty in training.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 621, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 81, + 430, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 432, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 432, + 96 + ], + "score": 1.0, + "content": "3 ONE-STAGE SEMI-SUPERVISED OPTIMAL TRANSPORT VAE", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 104, + 505, + 172 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 117 + ], + "score": 1.0, + "content": "In this section, we introduce our semi-supervised VAE framework, OSPOT-VAE. Firstly, we derive", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 117, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 104, + 117, + 506, + 128 + ], + "score": 1.0, + "content": "a one-stage loss function that unifies the generation and classification loss under one ELBO without", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 128, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 505, + 140 + ], + "score": 1.0, + "content": "introducing any additional auxiliary loss items like (7). Then, we analyze a phenomenon that good", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "score": 1.0, + "content": "ELBO values do not guarantee good semi-supervised performance and propose the optimal transport", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 148, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 162 + ], + "score": 1.0, + "content": "estimation to deal with it. At last, combining the two parts, we give the detailed algorithm of OSPOT-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 159, + 333, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 333, + 173 + ], + "score": 1.0, + "content": "VAE and discuss some problems in model optimization.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 109, + 184, + 287, + 196 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 288, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 288, + 196 + ], + "score": 1.0, + "content": "3.1 ONE-STAGE SEMI-SUPERVISED VAE", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 204, + 505, + 255 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 282, + 219 + ], + "score": 1.0, + "content": "Following the notations and assumptions", + "type": "text" + }, + { + "bbox": [ + 282, + 205, + 305, + 217 + ], + "score": 0.88, + "content": "( 3 , 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 203, + 506, + 219 + ], + "score": 1.0, + "content": "in Section 2.2, we derive our one-stage semi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 216, + 507, + 244 + ], + "spans": [ + { + "bbox": [ + 104, + 217, + 307, + 240 + ], + "score": 1.0, + "content": "supervised VAE. With the empirical distribution", + "type": "text" + }, + { + "bbox": [ + 307, + 216, + 441, + 244 + ], + "score": 0.93, + "content": "p _ { e m p } ( \\mathbf { X } ; \\mathbb { D } ) = { \\frac { 1 } { | \\mathbb { D } | } } \\sum _ { \\mathbf { X } ^ { \\prime } \\in \\mathbb { D } } \\mathbf { 1 } _ { \\mathbf { X } = \\mathbf { X } ^ { \\prime } }", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 217, + 507, + 240 + ], + "score": 1.0, + "content": ", we utilize the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 241, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 257 + ], + "score": 1.0, + "content": "decomposition in (Zhao et al., 2017) and rewrite the second part of (5) into (proof in Appendix A.2)", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 256, + 464, + 271 + ], + "lines": [ + { + "bbox": [ + 146, + 256, + 464, + 271 + ], + "spans": [ + { + "bbox": [ + 146, + 256, + 464, + 271 + ], + "score": 0.76, + "content": "\\begin{array} { r } { \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) \\big ) = \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } ) \\| p ( \\mathbf { z } ) \\big ) \\geq \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) } \\end{array}", + "type": "interline_equation", + "image_path": "e45412571be7347757b52de94b85bc837cd0ab42234dead9a5dca5f5e646be14.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 146, + 256, + 464, + 271 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 272, + 505, + 345 + ], + "lines": [ + { + "bbox": [ + 105, + 272, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 133, + 295 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 272, + 241, + 300 + ], + "score": 0.9, + "content": "q _ { \\phi } ( \\mathbf { z } ) = \\frac { 1 } { | \\mathbb { D } | } \\sum _ { \\mathbf { X } \\in \\mathbb { D } } q _ { \\phi } ( \\mathbf { z } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 275, + 260, + 293 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 261, + 277, + 302, + 291 + ], + "score": 0.9, + "content": "\\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 275, + 445, + 293 + ], + "score": 1.0, + "content": "is the mutual information between", + "type": "text" + }, + { + "bbox": [ + 446, + 278, + 455, + 288 + ], + "score": 0.75, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 275, + 474, + 293 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 475, + 280, + 481, + 288 + ], + "score": 0.68, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 275, + 506, + 293 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 297, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 237, + 313 + ], + "score": 1.0, + "content": "left part of (8) equals to 0 when", + "type": "text" + }, + { + "bbox": [ + 237, + 300, + 247, + 310 + ], + "score": 0.57, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 297, + 265, + 313 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 265, + 302, + 272, + 309 + ], + "score": 0.68, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 297, + 434, + 313 + ], + "score": 1.0, + "content": "are independent. This is undesirable, so", + "type": "text" + }, + { + "bbox": [ + 434, + 299, + 475, + 312 + ], + "score": 0.92, + "content": "\\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 297, + 506, + 313 + ], + "score": 1.0, + "content": "can be", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "score": 1.0, + "content": "regarded as the lower bound of controlled mutual information. The continuous variables in (8) can", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 322, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 256, + 335 + ], + "score": 1.0, + "content": "be easily extend to discrete variables", + "type": "text" + }, + { + "bbox": [ + 257, + 325, + 263, + 333 + ], + "score": 0.34, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 322, + 316, + 335 + ], + "score": 1.0, + "content": ". We can use", + "type": "text" + }, + { + "bbox": [ + 317, + 323, + 327, + 334 + ], + "score": 0.88, + "content": "\\mathbf { I _ { z } }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 322, + 345, + 335 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 345, + 323, + 356, + 334 + ], + "score": 0.88, + "content": "\\mathbf { I _ { c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 322, + 505, + 335 + ], + "score": 1.0, + "content": "to denote the controlled information", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 333, + 354, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 339, + 346 + ], + "score": 1.0, + "content": "capacity and derive the objective for the unlabeled dataset", + "type": "text" + }, + { + "bbox": [ + 339, + 334, + 354, + 345 + ], + "score": 0.87, + "content": "\\mathbb { D } _ { U }", + "type": "inline_equation" + } + ], + "index": 16 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 345, + 465, + 376 + ], + "lines": [ + { + "bbox": [ + 146, + 345, + 465, + 376 + ], + "spans": [ + { + "bbox": [ + 146, + 345, + 465, + 376 + ], + "score": 0.87, + "content": "\\begin{array} { r l } & { { \\mathcal { L } } _ { \\mathbb { D } _ { U } } ( { \\mathbf { X } } ; \\pmb { \\theta } , \\phi ) = { \\mathbb { E } } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] + \\beta _ { \\mathbf { z } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } ) - \\mathbf { I } _ { \\mathbf { z } } | } \\\\ & { \\quad \\quad \\quad \\quad + \\beta _ { \\mathbf { c } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { c } ) ) - \\mathbf { I } _ { \\mathbf { c } } | } \\end{array}", + "type": "interline_equation", + "image_path": "e47bd738aba74b86eb37673f2a87cc738f4e6ff9a0f4d33604e11a454b465a51.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 146, + 345, + 465, + 355.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 146, + 355.3333333333333, + 465, + 365.66666666666663 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 146, + 365.66666666666663, + 465, + 375.99999999999994 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 376, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 375, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 133, + 389 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 376, + 168, + 388 + ], + "score": 0.91, + "content": "\\beta , \\mathbf { I _ { z } } , \\mathbf { I _ { c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 375, + 506, + 389 + ], + "score": 1.0, + "content": "are all hyper-parameters forcing the KL divergence term to match the mutual infor-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 386, + 224, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 189, + 399 + ], + "score": 1.0, + "content": "mation capacities of", + "type": "text" + }, + { + "bbox": [ + 189, + 389, + 196, + 397 + ], + "score": 0.77, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 386, + 213, + 399 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 214, + 389, + 220, + 397 + ], + "score": 0.4, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 386, + 224, + 399 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 107, + 403, + 505, + 437 + ], + "lines": [ + { + "bbox": [ + 106, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 194, + 416 + ], + "score": 1.0, + "content": "For the labeled subset", + "type": "text" + }, + { + "bbox": [ + 195, + 405, + 208, + 415 + ], + "score": 0.89, + "content": "\\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 404, + 427, + 416 + ], + "score": 1.0, + "content": ", instead of directly employing class label y as sampled", + "type": "text" + }, + { + "bbox": [ + 428, + 406, + 434, + 414 + ], + "score": 0.42, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 404, + 505, + 416 + ], + "score": 1.0, + "content": ", we view it as the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 414, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 300, + 429 + ], + "score": 1.0, + "content": "parameter of the true posterior distribution, i.e.", + "type": "text" + }, + { + "bbox": [ + 301, + 415, + 402, + 427 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } | \\mathbf { X } ) = \\mathbf { M u l t } ( \\mathbf { c } ; K , \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 414, + 506, + 429 + ], + "score": 1.0, + "content": "and derive the following", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 425, + 201, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 201, + 438 + ], + "score": 1.0, + "content": "one-stage ELBO form:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 438, + 477, + 522 + ], + "lines": [ + { + "bbox": [ + 117, + 438, + 477, + 522 + ], + "spans": [ + { + "bbox": [ + 117, + 438, + 477, + 522 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\log \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } [ \\log p ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) + \\log \\frac { p ( \\mathbf { z } ) p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } ] = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\log p ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) } \\\\ & { - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( p ( \\mathbf { c } | \\mathbf { X } ) \\| q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) ) + \\mathbb { E } _ { p ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } } \\end{array}", + "type": "interline_equation", + "image_path": "f80090e70218c819d33a52d82e700b959d22225248b220c0e7f82186a6bafc88.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 117, + 438, + 477, + 466.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 117, + 466.0, + 477, + 494.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 117, + 494.0, + 477, + 522.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 522, + 505, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 155, + 535 + ], + "score": 1.0, + "content": "Notice that", + "type": "text" + }, + { + "bbox": [ + 155, + 522, + 252, + 535 + ], + "score": 0.9, + "content": "D _ { \\mathrm { K L } } ( p ( \\mathbf { c } | \\mathbf { X } ) \\| q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 522, + 506, + 535 + ], + "score": 1.0, + "content": "is equal to the common cross-entropy loss for y is a one-hot", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 285, + 546 + ], + "score": 1.0, + "content": "vector. In this respect, the margin between", + "type": "text" + }, + { + "bbox": [ + 285, + 533, + 316, + 545 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 534, + 335, + 546 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 336, + 533, + 371, + 546 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "can be significantly small when", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 462, + 556 + ], + "score": 1.0, + "content": "the suitable optimization method is chosen. This allows us to utilize the approximation", + "type": "text" + }, + { + "bbox": [ + 462, + 545, + 505, + 556 + ], + "score": 0.91, + "content": "p ( \\mathbf { c } | \\mathbf { X } ) \\approx", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 554, + 501, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 142, + 575 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 554, + 185, + 582 + ], + "score": 1.0, + "content": "to modify", + "type": "text" + }, + { + "bbox": [ + 185, + 555, + 270, + 582 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { p ( \\mathbf { c } | \\mathbf { X } ) } \\log { \\frac { p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 554, + 438, + 582 + ], + "score": 1.0, + "content": "in (10), resulting in a consist ELBO with", + "type": "text" + }, + { + "bbox": [ + 438, + 562, + 496, + 574 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { \\mathbb { D } _ { U } } ( \\mathbf { X } ; \\pmb { \\theta } , \\phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 496, + 554, + 501, + 582 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "interline_equation", + "bbox": [ + 113, + 582, + 482, + 609 + ], + "lines": [ + { + "bbox": [ + 113, + 582, + 482, + 609 + ], + "spans": [ + { + "bbox": [ + 113, + 582, + 482, + 609 + ], + "score": 0.92, + "content": "\\mathbb { E } _ { p ( { \\mathbf { c } } | { \\mathbf { X } } ) } \\log { \\frac { p ( { \\mathbf { c } } ) } { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } } \\approx { \\mathrm { ( } } { \\mathrm { w h e n ~ } } p ( { \\mathbf { c } } | { \\mathbf { X } } ) \\approx q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) { \\mathrm { ) } } \\mathbb { E } _ { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } \\log { \\frac { p ( { \\mathbf { c } } ) } { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } } = D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { c } } ) )", + "type": "interline_equation", + "image_path": "2351c089caa35b04cd2b3ae813eaf36c669d3069f9c420f6e0484be262265240.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 113, + 582, + 482, + 591.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 113, + 591.0, + 482, + 600.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 113, + 600.0, + 482, + 609.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 611, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 610, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 257, + 624 + ], + "score": 1.0, + "content": "Combining the ELBO form (10) of", + "type": "text" + }, + { + "bbox": [ + 257, + 611, + 272, + 622 + ], + "score": 0.89, + "content": "\\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 610, + 506, + 624 + ], + "score": 1.0, + "content": "with the mutual information decomposition (8) and the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 621, + 377, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 377, + 634 + ], + "score": 1.0, + "content": "approximation (11), the new objective for semi-supervised VAE is:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "interline_equation", + "bbox": [ + 121, + 633, + 474, + 665 + ], + "lines": [ + { + "bbox": [ + 121, + 633, + 474, + 665 + ], + "spans": [ + { + "bbox": [ + 121, + 633, + 474, + 665 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { { \\mathcal { L } } _ { \\mathbb { D } _ { L } } ( { \\mathbf { X } } , { \\mathbf { y } } ; \\pmb { \\theta } , \\phi ) = { \\mathbb { E } } _ { q _ { \\phi } ( { \\mathbf { z } } | { \\mathbf { X } } ) , p ( { \\mathbf { c } } | { \\mathbf { X } } ) } [ - \\log p _ { \\theta } ( { \\mathbf { X } } | { \\mathbf { z } } , { \\mathbf { c } } ) ] + \\beta _ { \\mathbf { z } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { z } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { z } } ) ) - { \\mathbf { I } } _ { \\mathbf { z } } | } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad + \\beta _ { \\mathbf { c } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { c } } ) ) - { \\mathbf { I } } _ { \\mathbf { c } } | + D _ { \\mathrm { K L } } ( p ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "e66145bbe8ea7f846663612151d4d0a5731458eddb81a69f9f408a7e32e6c97a.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 121, + 633, + 474, + 643.6666666666666 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 121, + 643.6666666666666, + 474, + 654.3333333333333 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 121, + 654.3333333333333, + 474, + 664.9999999999999 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 664, + 350, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 664, + 349, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 349, + 677 + ], + "score": 1.0, + "content": "With (9) and (12), the objective for the entire dataset is now", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "interline_equation", + "bbox": [ + 151, + 677, + 462, + 697 + ], + "lines": [ + { + "bbox": [ + 151, + 677, + 462, + 697 + ], + "spans": [ + { + "bbox": [ + 151, + 677, + 462, + 697 + ], + "score": 0.9, + "content": "\\operatorname* { m i n } _ { \\phi , \\theta } \\mathbb { E } _ { \\mathbf { X } \\sim p _ { e m p } ( \\mathbf { X } ; \\mathbb { D } _ { U } ) } \\mathcal { L } _ { \\mathbb { D } _ { U } } ( \\mathbf { X } ; \\theta , \\phi ) + \\mathbb { E } _ { ( \\mathbf { X } , \\mathbf { y } ) \\sim p _ { e m p } ( ( \\mathbf { X } , \\mathbf { y } ) ; \\mathbb { D } _ { L } ) } \\mathcal { L } _ { \\mathbb { D } _ { L } } ( \\mathbf { X } , \\mathbf { y } ; \\theta , \\phi )", + "type": "interline_equation", + "image_path": "bb2c07a93d0d6b41efa21c4dd7c9218009935c23ad85692a25d6cc9527ef7399.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 151, + 677, + 462, + 697 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "This one-stage objective with a simple approximate transformation (9) unifies the generation loss as", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "well as the target of SSL and results in improved performance of semi-supervised learning, which", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 721, + 232, + 731 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 232, + 731 + ], + "score": 1.0, + "content": "we demonstrate in Section 4.1.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "4", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 81, + 430, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 432, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 432, + 96 + ], + "score": 1.0, + "content": "3 ONE-STAGE SEMI-SUPERVISED OPTIMAL TRANSPORT VAE", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 104, + 505, + 172 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 117 + ], + "score": 1.0, + "content": "In this section, we introduce our semi-supervised VAE framework, OSPOT-VAE. Firstly, we derive", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 117, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 104, + 117, + 506, + 128 + ], + "score": 1.0, + "content": "a one-stage loss function that unifies the generation and classification loss under one ELBO without", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 128, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 505, + 140 + ], + "score": 1.0, + "content": "introducing any additional auxiliary loss items like (7). Then, we analyze a phenomenon that good", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "score": 1.0, + "content": "ELBO values do not guarantee good semi-supervised performance and propose the optimal transport", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 148, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 162 + ], + "score": 1.0, + "content": "estimation to deal with it. At last, combining the two parts, we give the detailed algorithm of OSPOT-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 159, + 333, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 333, + 173 + ], + "score": 1.0, + "content": "VAE and discuss some problems in model optimization.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 106, + 506, + 173 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 184, + 287, + 196 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 288, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 288, + 196 + ], + "score": 1.0, + "content": "3.1 ONE-STAGE SEMI-SUPERVISED VAE", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 204, + 505, + 255 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 282, + 219 + ], + "score": 1.0, + "content": "Following the notations and assumptions", + "type": "text" + }, + { + "bbox": [ + 282, + 205, + 305, + 217 + ], + "score": 0.88, + "content": "( 3 , 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 203, + 506, + 219 + ], + "score": 1.0, + "content": "in Section 2.2, we derive our one-stage semi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 216, + 507, + 244 + ], + "spans": [ + { + "bbox": [ + 104, + 217, + 307, + 240 + ], + "score": 1.0, + "content": "supervised VAE. With the empirical distribution", + "type": "text" + }, + { + "bbox": [ + 307, + 216, + 441, + 244 + ], + "score": 0.93, + "content": "p _ { e m p } ( \\mathbf { X } ; \\mathbb { D } ) = { \\frac { 1 } { | \\mathbb { D } | } } \\sum _ { \\mathbf { X } ^ { \\prime } \\in \\mathbb { D } } \\mathbf { 1 } _ { \\mathbf { X } = \\mathbf { X } ^ { \\prime } }", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 217, + 507, + 240 + ], + "score": 1.0, + "content": ", we utilize the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 241, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 257 + ], + "score": 1.0, + "content": "decomposition in (Zhao et al., 2017) and rewrite the second part of (5) into (proof in Appendix A.2)", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 104, + 203, + 507, + 257 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 256, + 464, + 271 + ], + "lines": [ + { + "bbox": [ + 146, + 256, + 464, + 271 + ], + "spans": [ + { + "bbox": [ + 146, + 256, + 464, + 271 + ], + "score": 0.76, + "content": "\\begin{array} { r } { \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) \\big ) = \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } ) \\| p ( \\mathbf { z } ) \\big ) \\geq \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) } \\end{array}", + "type": "interline_equation", + "image_path": "e45412571be7347757b52de94b85bc837cd0ab42234dead9a5dca5f5e646be14.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 146, + 256, + 464, + 271 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 272, + 505, + 345 + ], + "lines": [ + { + "bbox": [ + 105, + 272, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 133, + 295 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 272, + 241, + 300 + ], + "score": 0.9, + "content": "q _ { \\phi } ( \\mathbf { z } ) = \\frac { 1 } { | \\mathbb { D } | } \\sum _ { \\mathbf { X } \\in \\mathbb { D } } q _ { \\phi } ( \\mathbf { z } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 275, + 260, + 293 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 261, + 277, + 302, + 291 + ], + "score": 0.9, + "content": "\\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 275, + 445, + 293 + ], + "score": 1.0, + "content": "is the mutual information between", + "type": "text" + }, + { + "bbox": [ + 446, + 278, + 455, + 288 + ], + "score": 0.75, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 275, + 474, + 293 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 475, + 280, + 481, + 288 + ], + "score": 0.68, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 275, + 506, + 293 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 297, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 237, + 313 + ], + "score": 1.0, + "content": "left part of (8) equals to 0 when", + "type": "text" + }, + { + "bbox": [ + 237, + 300, + 247, + 310 + ], + "score": 0.57, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 297, + 265, + 313 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 265, + 302, + 272, + 309 + ], + "score": 0.68, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 297, + 434, + 313 + ], + "score": 1.0, + "content": "are independent. This is undesirable, so", + "type": "text" + }, + { + "bbox": [ + 434, + 299, + 475, + 312 + ], + "score": 0.92, + "content": "\\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 297, + 506, + 313 + ], + "score": 1.0, + "content": "can be", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "score": 1.0, + "content": "regarded as the lower bound of controlled mutual information. The continuous variables in (8) can", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 322, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 256, + 335 + ], + "score": 1.0, + "content": "be easily extend to discrete variables", + "type": "text" + }, + { + "bbox": [ + 257, + 325, + 263, + 333 + ], + "score": 0.34, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 322, + 316, + 335 + ], + "score": 1.0, + "content": ". We can use", + "type": "text" + }, + { + "bbox": [ + 317, + 323, + 327, + 334 + ], + "score": 0.88, + "content": "\\mathbf { I _ { z } }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 322, + 345, + 335 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 345, + 323, + 356, + 334 + ], + "score": 0.88, + "content": "\\mathbf { I _ { c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 322, + 505, + 335 + ], + "score": 1.0, + "content": "to denote the controlled information", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 333, + 354, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 339, + 346 + ], + "score": 1.0, + "content": "capacity and derive the objective for the unlabeled dataset", + "type": "text" + }, + { + "bbox": [ + 339, + 334, + 354, + 345 + ], + "score": 0.87, + "content": "\\mathbb { D } _ { U }", + "type": "inline_equation" + } + ], + "index": 16 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 272, + 506, + 346 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 345, + 465, + 376 + ], + "lines": [ + { + "bbox": [ + 146, + 345, + 465, + 376 + ], + "spans": [ + { + "bbox": [ + 146, + 345, + 465, + 376 + ], + "score": 0.87, + "content": "\\begin{array} { r l } & { { \\mathcal { L } } _ { \\mathbb { D } _ { U } } ( { \\mathbf { X } } ; \\pmb { \\theta } , \\phi ) = { \\mathbb { E } } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] + \\beta _ { \\mathbf { z } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } ) - \\mathbf { I } _ { \\mathbf { z } } | } \\\\ & { \\quad \\quad \\quad \\quad + \\beta _ { \\mathbf { c } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { c } ) ) - \\mathbf { I } _ { \\mathbf { c } } | } \\end{array}", + "type": "interline_equation", + "image_path": "e47bd738aba74b86eb37673f2a87cc738f4e6ff9a0f4d33604e11a454b465a51.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 146, + 345, + 465, + 355.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 146, + 355.3333333333333, + 465, + 365.66666666666663 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 146, + 365.66666666666663, + 465, + 375.99999999999994 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 376, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 375, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 133, + 389 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 376, + 168, + 388 + ], + "score": 0.91, + "content": "\\beta , \\mathbf { I _ { z } } , \\mathbf { I _ { c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 375, + 506, + 389 + ], + "score": 1.0, + "content": "are all hyper-parameters forcing the KL divergence term to match the mutual infor-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 386, + 224, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 189, + 399 + ], + "score": 1.0, + "content": "mation capacities of", + "type": "text" + }, + { + "bbox": [ + 189, + 389, + 196, + 397 + ], + "score": 0.77, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 386, + 213, + 399 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 214, + 389, + 220, + 397 + ], + "score": 0.4, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 386, + 224, + 399 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 375, + 506, + 399 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 403, + 505, + 437 + ], + "lines": [ + { + "bbox": [ + 106, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 194, + 416 + ], + "score": 1.0, + "content": "For the labeled subset", + "type": "text" + }, + { + "bbox": [ + 195, + 405, + 208, + 415 + ], + "score": 0.89, + "content": "\\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 404, + 427, + 416 + ], + "score": 1.0, + "content": ", instead of directly employing class label y as sampled", + "type": "text" + }, + { + "bbox": [ + 428, + 406, + 434, + 414 + ], + "score": 0.42, + "content": "\\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 404, + 505, + 416 + ], + "score": 1.0, + "content": ", we view it as the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 414, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 300, + 429 + ], + "score": 1.0, + "content": "parameter of the true posterior distribution, i.e.", + "type": "text" + }, + { + "bbox": [ + 301, + 415, + 402, + 427 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } | \\mathbf { X } ) = \\mathbf { M u l t } ( \\mathbf { c } ; K , \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 414, + 506, + 429 + ], + "score": 1.0, + "content": "and derive the following", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 425, + 201, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 201, + 438 + ], + "score": 1.0, + "content": "one-stage ELBO form:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 404, + 506, + 438 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 117, + 438, + 477, + 522 + ], + "lines": [ + { + "bbox": [ + 117, + 438, + 477, + 522 + ], + "spans": [ + { + "bbox": [ + 117, + 438, + 477, + 522 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\log \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } [ \\log p ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) + \\log \\frac { p ( \\mathbf { z } ) p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } ] = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\log p ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) } \\\\ & { - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( p ( \\mathbf { c } | \\mathbf { X } ) \\| q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) ) + \\mathbb { E } _ { p ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } } \\end{array}", + "type": "interline_equation", + "image_path": "f80090e70218c819d33a52d82e700b959d22225248b220c0e7f82186a6bafc88.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 117, + 438, + 477, + 466.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 117, + 466.0, + 477, + 494.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 117, + 494.0, + 477, + 522.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 522, + 505, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 155, + 535 + ], + "score": 1.0, + "content": "Notice that", + "type": "text" + }, + { + "bbox": [ + 155, + 522, + 252, + 535 + ], + "score": 0.9, + "content": "D _ { \\mathrm { K L } } ( p ( \\mathbf { c } | \\mathbf { X } ) \\| q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 522, + 506, + 535 + ], + "score": 1.0, + "content": "is equal to the common cross-entropy loss for y is a one-hot", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 285, + 546 + ], + "score": 1.0, + "content": "vector. In this respect, the margin between", + "type": "text" + }, + { + "bbox": [ + 285, + 533, + 316, + 545 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 534, + 335, + 546 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 336, + 533, + 371, + 546 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "can be significantly small when", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 462, + 556 + ], + "score": 1.0, + "content": "the suitable optimization method is chosen. This allows us to utilize the approximation", + "type": "text" + }, + { + "bbox": [ + 462, + 545, + 505, + 556 + ], + "score": 0.91, + "content": "p ( \\mathbf { c } | \\mathbf { X } ) \\approx", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 554, + 501, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 142, + 575 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 554, + 185, + 582 + ], + "score": 1.0, + "content": "to modify", + "type": "text" + }, + { + "bbox": [ + 185, + 555, + 270, + 582 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { p ( \\mathbf { c } | \\mathbf { X } ) } \\log { \\frac { p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 554, + 438, + 582 + ], + "score": 1.0, + "content": "in (10), resulting in a consist ELBO with", + "type": "text" + }, + { + "bbox": [ + 438, + 562, + 496, + 574 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { \\mathbb { D } _ { U } } ( \\mathbf { X } ; \\pmb { \\theta } , \\phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 496, + 554, + 501, + 582 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 522, + 506, + 582 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 113, + 582, + 482, + 609 + ], + "lines": [ + { + "bbox": [ + 113, + 582, + 482, + 609 + ], + "spans": [ + { + "bbox": [ + 113, + 582, + 482, + 609 + ], + "score": 0.92, + "content": "\\mathbb { E } _ { p ( { \\mathbf { c } } | { \\mathbf { X } } ) } \\log { \\frac { p ( { \\mathbf { c } } ) } { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } } \\approx { \\mathrm { ( } } { \\mathrm { w h e n ~ } } p ( { \\mathbf { c } } | { \\mathbf { X } } ) \\approx q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) { \\mathrm { ) } } \\mathbb { E } _ { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } \\log { \\frac { p ( { \\mathbf { c } } ) } { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } } = D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { c } } ) )", + "type": "interline_equation", + "image_path": "2351c089caa35b04cd2b3ae813eaf36c669d3069f9c420f6e0484be262265240.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 113, + 582, + 482, + 591.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 113, + 591.0, + 482, + 600.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 113, + 600.0, + 482, + 609.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 611, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 610, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 257, + 624 + ], + "score": 1.0, + "content": "Combining the ELBO form (10) of", + "type": "text" + }, + { + "bbox": [ + 257, + 611, + 272, + 622 + ], + "score": 0.89, + "content": "\\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 610, + 506, + 624 + ], + "score": 1.0, + "content": "with the mutual information decomposition (8) and the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 621, + 377, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 377, + 634 + ], + "score": 1.0, + "content": "approximation (11), the new objective for semi-supervised VAE is:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 610, + 506, + 634 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 121, + 633, + 474, + 665 + ], + "lines": [ + { + "bbox": [ + 121, + 633, + 474, + 665 + ], + "spans": [ + { + "bbox": [ + 121, + 633, + 474, + 665 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { { \\mathcal { L } } _ { \\mathbb { D } _ { L } } ( { \\mathbf { X } } , { \\mathbf { y } } ; \\pmb { \\theta } , \\phi ) = { \\mathbb { E } } _ { q _ { \\phi } ( { \\mathbf { z } } | { \\mathbf { X } } ) , p ( { \\mathbf { c } } | { \\mathbf { X } } ) } [ - \\log p _ { \\theta } ( { \\mathbf { X } } | { \\mathbf { z } } , { \\mathbf { c } } ) ] + \\beta _ { \\mathbf { z } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { z } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { z } } ) ) - { \\mathbf { I } } _ { \\mathbf { z } } | } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad + \\beta _ { \\mathbf { c } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { c } } ) ) - { \\mathbf { I } } _ { \\mathbf { c } } | + D _ { \\mathrm { K L } } ( p ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "e66145bbe8ea7f846663612151d4d0a5731458eddb81a69f9f408a7e32e6c97a.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 121, + 633, + 474, + 643.6666666666666 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 121, + 643.6666666666666, + 474, + 654.3333333333333 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 121, + 654.3333333333333, + 474, + 664.9999999999999 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 664, + 350, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 664, + 349, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 349, + 677 + ], + "score": 1.0, + "content": "With (9) and (12), the objective for the entire dataset is now", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40, + "bbox_fs": [ + 106, + 664, + 349, + 677 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 151, + 677, + 462, + 697 + ], + "lines": [ + { + "bbox": [ + 151, + 677, + 462, + 697 + ], + "spans": [ + { + "bbox": [ + 151, + 677, + 462, + 697 + ], + "score": 0.9, + "content": "\\operatorname* { m i n } _ { \\phi , \\theta } \\mathbb { E } _ { \\mathbf { X } \\sim p _ { e m p } ( \\mathbf { X } ; \\mathbb { D } _ { U } ) } \\mathcal { L } _ { \\mathbb { D } _ { U } } ( \\mathbf { X } ; \\theta , \\phi ) + \\mathbb { E } _ { ( \\mathbf { X } , \\mathbf { y } ) \\sim p _ { e m p } ( ( \\mathbf { X } , \\mathbf { y } ) ; \\mathbb { D } _ { L } ) } \\mathcal { L } _ { \\mathbb { D } _ { L } } ( \\mathbf { X } , \\mathbf { y } ; \\theta , \\phi )", + "type": "interline_equation", + "image_path": "bb2c07a93d0d6b41efa21c4dd7c9218009935c23ad85692a25d6cc9527ef7399.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 151, + 677, + 462, + 697 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "This one-stage objective with a simple approximate transformation (9) unifies the generation loss as", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "well as the target of SSL and results in improved performance of semi-supervised learning, which", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 721, + 232, + 731 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 232, + 731 + ], + "score": 1.0, + "content": "we demonstrate in Section 4.1.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 697, + 506, + 731 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 81, + 504, + 117 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 398, + 94 + ], + "score": 1.0, + "content": "Algorithm 1 Optimal transport estimation ingests a batch of observation", + "type": "text" + }, + { + "bbox": [ + 399, + 83, + 408, + 92 + ], + "score": 0.49, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "as well as the represen-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 132, + 106 + ], + "score": 1.0, + "content": "tation", + "type": "text" + }, + { + "bbox": [ + 132, + 93, + 168, + 105 + ], + "score": 0.36, + "content": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 93, + 172, + 106 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 172, + 93, + 208, + 105 + ], + "score": 0.6, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "inferred from the original VAE and returns the estimation of the margin", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 104, + 324, + 117 + ], + "spans": [ + { + "bbox": [ + 107, + 105, + 203, + 117 + ], + "score": 0.89, + "content": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 104, + 222, + 117 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 223, + 104, + 319, + 117 + ], + "score": 0.93, + "content": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { \\bar { c } } | \\mathbf { X } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 104, + 324, + 117 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 109, + 121, + 137, + 131 + ], + "lines": [ + { + "bbox": [ + 108, + 119, + 139, + 137 + ], + "spans": [ + { + "bbox": [ + 108, + 119, + 139, + 137 + ], + "score": 1.0, + "content": "Input:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 123, + 131, + 397, + 177 + ], + "lines": [ + { + "bbox": [ + 122, + 130, + 320, + 144 + ], + "spans": [ + { + "bbox": [ + 122, + 130, + 208, + 144 + ], + "score": 1.0, + "content": "Batch of observation", + "type": "text" + }, + { + "bbox": [ + 209, + 131, + 219, + 141 + ], + "score": 0.63, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 130, + 277, + 144 + ], + "score": 1.0, + "content": "sampled from", + "type": "text" + }, + { + "bbox": [ + 277, + 131, + 315, + 143 + ], + "score": 0.91, + "content": "p _ { e m p } ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 130, + 320, + 144 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 122, + 141, + 397, + 156 + ], + "spans": [ + { + "bbox": [ + 122, + 141, + 200, + 156 + ], + "score": 1.0, + "content": "Inferred parameter", + "type": "text" + }, + { + "bbox": [ + 201, + 142, + 257, + 155 + ], + "score": 0.89, + "content": "( \\mu , \\mathrm { d i a g } ( \\sigma ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 141, + 270, + 156 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 270, + 143, + 394, + 155 + ], + "score": 0.86, + "content": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) = \\mathcal { N } ( \\mathbf { z } ; \\pmb { \\mu } , \\mathrm { d i a g } ( \\pmb { \\sigma } ^ { 2 } ) ) ;", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 141, + 397, + 156 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 122, + 152, + 332, + 167 + ], + "spans": [ + { + "bbox": [ + 122, + 152, + 200, + 167 + ], + "score": 1.0, + "content": "Inferred parameter", + "type": "text" + }, + { + "bbox": [ + 200, + 156, + 209, + 164 + ], + "score": 0.76, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 152, + 221, + 167 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 221, + 155, + 327, + 167 + ], + "score": 0.88, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) = \\mathbf { M } \\mathbf { u } \\mathrm { l t } ( \\mathbf { c } ; K , \\pi )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 152, + 332, + 167 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 122, + 163, + 366, + 179 + ], + "spans": [ + { + "bbox": [ + 122, + 163, + 190, + 179 + ], + "score": 1.0, + "content": "Hyperparameter", + "type": "text" + }, + { + "bbox": [ + 191, + 167, + 199, + 175 + ], + "score": 0.79, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 163, + 319, + 179 + ], + "score": 1.0, + "content": "for mixup vicinal distribution", + "type": "text" + }, + { + "bbox": [ + 320, + 165, + 366, + 177 + ], + "score": 0.84, + "content": "p _ { m i x u p } ( \\mathbf { X } )", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 5.5 + }, + { + "type": "title", + "bbox": [ + 110, + 177, + 145, + 188 + ], + "lines": [ + { + "bbox": [ + 107, + 173, + 147, + 192 + ], + "spans": [ + { + "bbox": [ + 107, + 173, + 147, + 192 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 111, + 188, + 346, + 351 + ], + "lines": [ + { + "bbox": [ + 123, + 186, + 243, + 202 + ], + "spans": [ + { + "bbox": [ + 123, + 187, + 133, + 198 + ], + "score": 0.73, + "content": "\\tilde { \\mathbf { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 186, + 192, + 202 + ], + "score": 1.0, + "content": "sampled from", + "type": "text" + }, + { + "bbox": [ + 192, + 188, + 238, + 200 + ], + "score": 0.92, + "content": "p _ { m i x u p } ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 186, + 243, + 202 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 122, + 199, + 344, + 214 + ], + "spans": [ + { + "bbox": [ + 122, + 200, + 169, + 214 + ], + "score": 1.0, + "content": "Estimation", + "type": "text" + }, + { + "bbox": [ + 169, + 201, + 189, + 213 + ], + "score": 0.91, + "content": "L _ { M \\mathbf { z } }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 200, + 246, + 214 + ], + "score": 1.0, + "content": "of the margin", + "type": "text" + }, + { + "bbox": [ + 246, + 199, + 344, + 213 + ], + "score": 0.9, + "content": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) | | p ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) )", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 122, + 213, + 344, + 227 + ], + "spans": [ + { + "bbox": [ + 122, + 213, + 169, + 227 + ], + "score": 1.0, + "content": "Estimation", + "type": "text" + }, + { + "bbox": [ + 169, + 215, + 189, + 226 + ], + "score": 0.89, + "content": "L _ { M _ { \\mathbf { c } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 213, + 246, + 227 + ], + "score": 1.0, + "content": "of the margin", + "type": "text" + }, + { + "bbox": [ + 247, + 214, + 344, + 226 + ], + "score": 0.89, + "content": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) | | p ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) )", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 112, + 225, + 334, + 239 + ], + "spans": [ + { + "bbox": [ + 112, + 227, + 118, + 236 + ], + "score": 1.0, + "content": "1:", + "type": "text" + }, + { + "bbox": [ + 118, + 226, + 200, + 237 + ], + "score": 0.69, + "content": ": \\mathrm { \\bf ~ X } ^ { \\prime } , \\pi ^ { \\prime } , \\mu ^ { \\prime } , \\sigma ^ { \\prime 2 } = \\mathrm { \\bf F }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 225, + 279, + 239 + ], + "score": 1.0, + "content": "andomPermutation", + "type": "text" + }, + { + "bbox": [ + 279, + 227, + 334, + 238 + ], + "score": 0.85, + "content": "( \\mathbf { X } , \\pi , \\mu , \\sigma ^ { 2 } )", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 111, + 236, + 288, + 252 + ], + "spans": [ + { + "bbox": [ + 111, + 236, + 123, + 252 + ], + "score": 1.0, + "content": "2:", + "type": "text" + }, + { + "bbox": [ + 124, + 238, + 235, + 250 + ], + "score": 0.85, + "content": "\\tilde { \\mathbf { X } } = \\lambda * \\mathbf { X } + \\left( 1 - \\lambda \\right) * \\mathbf { X } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 236, + 237, + 252 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 238, + 239, + 288, + 250 + ], + "score": 0.69, + "content": "\\lambda \\in \\beta ( \\alpha , \\alpha )", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 111, + 249, + 262, + 262 + ], + "spans": [ + { + "bbox": [ + 111, + 249, + 221, + 262 + ], + "score": 1.0, + "content": "3: π˜ = OptimalTransportC", + "type": "text" + }, + { + "bbox": [ + 222, + 250, + 262, + 261 + ], + "score": 0.67, + "content": "\\prime ( \\pi , \\pi ^ { \\prime } , \\lambda )", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 111, + 259, + 339, + 274 + ], + "spans": [ + { + "bbox": [ + 111, + 259, + 124, + 274 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 124, + 261, + 168, + 272 + ], + "score": 0.43, + "content": "( \\tilde { \\mu } , \\tilde { \\sigma } ^ { 2 } ) =", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 259, + 339, + 274 + ], + "score": 1.0, + "content": "OptimalTransportZ((µ, σ2), (µ0, σ02), λ)", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 111, + 272, + 248, + 286 + ], + "spans": [ + { + "bbox": [ + 111, + 272, + 123, + 286 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 124, + 272, + 248, + 286 + ], + "score": 0.85, + "content": "q _ { \\phi } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) , q _ { \\phi } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) = \\mathrm { V A E } ( \\tilde { \\mathbf { X } } )", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 111, + 285, + 243, + 299 + ], + "spans": [ + { + "bbox": [ + 111, + 285, + 123, + 299 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 124, + 286, + 243, + 299 + ], + "score": 0.85, + "content": "\\tilde { p } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) = \\mathcal { N } ( z ; \\tilde { \\mu } , \\mathrm { d i a g } ( \\tilde { \\pmb { \\sigma } } ^ { 2 } ) )", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 109, + 297, + 223, + 312 + ], + "spans": [ + { + "bbox": [ + 109, + 297, + 123, + 312 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 124, + 299, + 223, + 311 + ], + "score": 0.86, + "content": "\\tilde { p } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) = \\mathbf { M u l t } ( \\mathbf { c } ; K , \\tilde { \\pi } )", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 109, + 310, + 253, + 325 + ], + "spans": [ + { + "bbox": [ + 109, + 310, + 123, + 325 + ], + "score": 1.0, + "content": "8:", + "type": "text" + }, + { + "bbox": [ + 123, + 311, + 253, + 324 + ], + "score": 0.87, + "content": "{ \\cal L } _ { M _ { \\mathbf { z } } } = { \\cal D } _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { z } } | \\tilde { \\mathbf { X } } ) | | \\tilde { p } ( { \\mathbf { z } } | \\tilde { \\mathbf { X } } ) )", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 109, + 323, + 253, + 339 + ], + "spans": [ + { + "bbox": [ + 109, + 323, + 123, + 339 + ], + "score": 1.0, + "content": "9:", + "type": "text" + }, + { + "bbox": [ + 123, + 324, + 253, + 337 + ], + "score": 0.88, + "content": "{ \\cal L } _ { M _ { \\mathbf { c } } } = { \\cal D } _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | \\tilde { \\mathbf { X } } ) | | \\tilde { p } ( { \\mathbf { c } } | \\tilde { \\mathbf { X } } ) )", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 109, + 335, + 212, + 352 + ], + "spans": [ + { + "bbox": [ + 109, + 335, + 155, + 352 + ], + "score": 1.0, + "content": "10: return", + "type": "text" + }, + { + "bbox": [ + 155, + 337, + 212, + 350 + ], + "score": 0.83, + "content": "\\tilde { \\mathbf { X } } , L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } }", + "type": "inline_equation" + } + ], + "index": 21 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 108, + 374, + 282, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 373, + 284, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 284, + 387 + ], + "score": 1.0, + "content": "3.2 OPTIMAL TRANSPORT ESTIMATION", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 394, + 505, + 441 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 397, + 408 + ], + "score": 1.0, + "content": "To summarize the above, VAE aims to learn the useful representation", + "type": "text" + }, + { + "bbox": [ + 398, + 395, + 433, + 407 + ], + "score": 0.92, + "content": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 393, + 454, + 408 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 454, + 395, + 490, + 407 + ], + "score": 0.92, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 393, + 505, + 408 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 405, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 365, + 420 + ], + "score": 1.0, + "content": "reducing the KL divergence between the empirical distribution", + "type": "text" + }, + { + "bbox": [ + 365, + 406, + 403, + 418 + ], + "score": 0.91, + "content": "p _ { e m p } ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 405, + 506, + 420 + ], + "score": 1.0, + "content": "and the model marginal", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 416, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 271, + 430 + ], + "score": 0.91, + "content": "\\begin{array} { r } { p ( \\mathbf { X } ) \\ { \\stackrel { } { = } } \\ \\int _ { \\mathbf { z } } \\int _ { \\mathbf { c } } p ( \\mathbf { X } ) p ( \\mathbf { \\tilde { z } } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) d \\mathbf { z } d \\mathbf { c } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 417, + 375, + 430 + ], + "score": 1.0, + "content": ". Instead of minimizing", + "type": "text" + }, + { + "bbox": [ + 375, + 417, + 467, + 429 + ], + "score": 0.91, + "content": "\\hat { D _ { \\mathrm { K L } } } ( p _ { e m p } ( \\mathbf { X } ) | | p ( \\mathbf { X } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 417, + 505, + 430 + ], + "score": 1.0, + "content": "directly,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 428, + 474, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 474, + 442 + ], + "score": 1.0, + "content": "VAE models use the expected ELBO mentioned in (5) as target via the following inequality", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "interline_equation", + "bbox": [ + 187, + 446, + 424, + 461 + ], + "lines": [ + { + "bbox": [ + 187, + 446, + 424, + 461 + ], + "spans": [ + { + "bbox": [ + 187, + 446, + 424, + 461 + ], + "score": 0.9, + "content": "D _ { \\mathrm { K L } } ( p _ { e m p } ( \\mathbf { X } ) \\| p ( \\mathbf { X } ) ) \\leq H ( p _ { e m p } ( \\mathbf { X } ) ) - \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\mathrm { E L B O }", + "type": "interline_equation", + "image_path": "6d5acdd29c8437cdc290bb9221d06570314efa5a1c575c569f47582b6d75ef48.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 187, + 446, + 424, + 461 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 464, + 505, + 520 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "However, one phenomenon is that good ELBO values do not imply accurate inference. A typical", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 475, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 505, + 488 + ], + "score": 1.0, + "content": "example has been discussed in (Zhao et al., 2017). Here we mainly focus on the cause of this", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 486, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 505, + 500 + ], + "score": 1.0, + "content": "phenomenon and propose optimal transport estimation to alleviate this problem in semi-supervised", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 497, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 505, + 510 + ], + "score": 1.0, + "content": "learning. Following the work in (Rezende et al., 2014), we write down the closed form of the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 508, + 446, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 446, + 521 + ], + "score": 1.0, + "content": "expected margin between true log-likelihood and ELBO as (proof in Appendix A.3):", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 524, + 496, + 540 + ], + "lines": [ + { + "bbox": [ + 114, + 524, + 496, + 540 + ], + "spans": [ + { + "bbox": [ + 114, + 524, + 496, + 540 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { p _ { \\epsilon m p } ( \\mathbf { X } ) } [ \\log p ( \\mathbf { X } ) - \\mathrm { E L B O } ] = \\mathbb { E } _ { p _ { \\epsilon m p } ( \\mathbf { X } ) } [ D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { c } | \\mathbf { X } ) ) ]", + "type": "interline_equation", + "image_path": "1094a5ca0aeba83fb1bfdd67768ecac528b1fc67dc9103847c2d257a00e6ad3c.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 114, + 524, + 496, + 540 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 547, + 505, + 626 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 505, + 560 + ], + "score": 1.0, + "content": "Combined with the decomposition (8), training the expected ELBO target can only reduce the dif-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 558, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 265, + 572 + ], + "score": 1.0, + "content": "ference between marginal distributions", + "type": "text" + }, + { + "bbox": [ + 265, + 559, + 315, + 571 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { c } ) , \\bar { q } _ { \\phi } ( \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 558, + 334, + 572 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 334, + 559, + 373, + 570 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } ) \\mathbf { , } p ( \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 558, + 505, + 572 + ], + "score": 1.0, + "content": ". It means that even with a good", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 103, + 567, + 507, + 586 + ], + "spans": [ + { + "bbox": [ + 103, + 567, + 183, + 586 + ], + "score": 1.0, + "content": "ELBO, the margin", + "type": "text" + }, + { + "bbox": [ + 183, + 570, + 318, + 583 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\dot { \\mathbf { X } } ) \\big | \\big | p ( \\dot { \\mathbf { z } } | \\mathbf { X } ) \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 567, + 338, + 586 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 338, + 570, + 472, + 583 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 567, + 507, + 586 + ], + "score": 1.0, + "content": "in (15)", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 581, + 506, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 593 + ], + "score": 1.0, + "content": "can still be large. In this scenario, the consistent optimization of ELBO will contribute no more to", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "score": 1.0, + "content": "the semi-supervised classification performance. However, optimizing the margin in (15) directly", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 603, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 181, + 615 + ], + "score": 1.0, + "content": "is impossible, for", + "type": "text" + }, + { + "bbox": [ + 182, + 603, + 212, + 615 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 603, + 232, + 615 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 233, + 603, + 263, + 615 + ], + "score": 0.93, + "content": "p ( \\mathbf { z } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 603, + 505, + 615 + ], + "score": 1.0, + "content": "are unknown. To remedy this, we extend the empirically", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 613, + 446, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 446, + 627 + ], + "score": 1.0, + "content": "effective approximation in (Zhang et al., 2018) to our VAE framework with the form", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37 + }, + { + "type": "interline_equation", + "bbox": [ + 141, + 629, + 469, + 662 + ], + "lines": [ + { + "bbox": [ + 141, + 629, + 469, + 662 + ], + "spans": [ + { + "bbox": [ + 141, + 629, + 469, + 662 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) \\approx \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) ( \\alpha \\to 0 ) } \\\\ & { \\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) \\approx \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) ( \\alpha \\to 0 ) } \\end{array}", + "type": "interline_equation", + "image_path": "cc4ef86cc3bbd69ae4053aa0af711278880e6e00b9f74d53d807949a3f913750.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 141, + 629, + 469, + 640.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 141, + 640.0, + 469, + 651.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 141, + 651.0, + 469, + 662.0 + ], + "spans": [], + "index": 43 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 136, + 679 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 666, + 183, + 678 + ], + "score": 0.92, + "content": "p _ { m i x u p } ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 665, + 448, + 679 + ], + "score": 1.0, + "content": "is the mixup vicinal distribution (Zhang et al., 2018) and", + "type": "text" + }, + { + "bbox": [ + 449, + 668, + 457, + 676 + ], + "score": 0.77, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 665, + 506, + 679 + ], + "score": 1.0, + "content": "is the re-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "lated parameter. Then we propose optimal transport estimation to construct the estimations of", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 683, + 508, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 249, + 701 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\big | \\big | p \\big ( \\mathbf { z } | \\bar { \\mathbf { X } } \\big ) \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 683, + 281, + 712 + ], + "score": 1.0, + "content": "as wellariables", + "type": "text" + }, + { + "bbox": [ + 288, + 683, + 294, + 712 + ], + "score": 1.0, + "content": "s", + "type": "text" + }, + { + "bbox": [ + 294, + 687, + 436, + 701 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\big | \\big | p ( \\mathbf { z } | \\mathbf { X } ) \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 683, + 508, + 712 + ], + "score": 1.0, + "content": "by applying op-ptimal transport", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 281, + 701, + 315, + 709 + ], + "spans": [ + { + "bbox": [ + 281, + 701, + 288, + 709 + ], + "score": 0.57, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 702, + 315, + 709 + ], + "score": 0.27, + "content": "\\mathbf { c }", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "estimation are provided in Algorithm 1, and we present the details of the optimal transport scheme", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 209, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 209, + 732 + ], + "score": 1.0, + "content": "in the rest of this section.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 46.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 81, + 504, + 117 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 398, + 94 + ], + "score": 1.0, + "content": "Algorithm 1 Optimal transport estimation ingests a batch of observation", + "type": "text" + }, + { + "bbox": [ + 399, + 83, + 408, + 92 + ], + "score": 0.49, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "as well as the represen-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 132, + 106 + ], + "score": 1.0, + "content": "tation", + "type": "text" + }, + { + "bbox": [ + 132, + 93, + 168, + 105 + ], + "score": 0.36, + "content": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 93, + 172, + 106 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 172, + 93, + 208, + 105 + ], + "score": 0.6, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "inferred from the original VAE and returns the estimation of the margin", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 104, + 324, + 117 + ], + "spans": [ + { + "bbox": [ + 107, + 105, + 203, + 117 + ], + "score": 0.89, + "content": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 104, + 222, + 117 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 223, + 104, + 319, + 117 + ], + "score": 0.93, + "content": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { \\bar { c } } | \\mathbf { X } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 104, + 324, + 117 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 82, + 506, + 117 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 121, + 137, + 131 + ], + "lines": [ + { + "bbox": [ + 108, + 119, + 139, + 137 + ], + "spans": [ + { + "bbox": [ + 108, + 119, + 139, + 137 + ], + "score": 1.0, + "content": "Input:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 123, + 131, + 397, + 177 + ], + "lines": [ + { + "bbox": [ + 122, + 130, + 320, + 144 + ], + "spans": [ + { + "bbox": [ + 122, + 130, + 208, + 144 + ], + "score": 1.0, + "content": "Batch of observation", + "type": "text" + }, + { + "bbox": [ + 209, + 131, + 219, + 141 + ], + "score": 0.63, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 130, + 277, + 144 + ], + "score": 1.0, + "content": "sampled from", + "type": "text" + }, + { + "bbox": [ + 277, + 131, + 315, + 143 + ], + "score": 0.91, + "content": "p _ { e m p } ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 130, + 320, + 144 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 122, + 141, + 397, + 156 + ], + "spans": [ + { + "bbox": [ + 122, + 141, + 200, + 156 + ], + "score": 1.0, + "content": "Inferred parameter", + "type": "text" + }, + { + "bbox": [ + 201, + 142, + 257, + 155 + ], + "score": 0.89, + "content": "( \\mu , \\mathrm { d i a g } ( \\sigma ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 141, + 270, + 156 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 270, + 143, + 394, + 155 + ], + "score": 0.86, + "content": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) = \\mathcal { N } ( \\mathbf { z } ; \\pmb { \\mu } , \\mathrm { d i a g } ( \\pmb { \\sigma } ^ { 2 } ) ) ;", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 141, + 397, + 156 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 122, + 152, + 332, + 167 + ], + "spans": [ + { + "bbox": [ + 122, + 152, + 200, + 167 + ], + "score": 1.0, + "content": "Inferred parameter", + "type": "text" + }, + { + "bbox": [ + 200, + 156, + 209, + 164 + ], + "score": 0.76, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 152, + 221, + 167 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 221, + 155, + 327, + 167 + ], + "score": 0.88, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) = \\mathbf { M } \\mathbf { u } \\mathrm { l t } ( \\mathbf { c } ; K , \\pi )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 152, + 332, + 167 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 122, + 163, + 366, + 179 + ], + "spans": [ + { + "bbox": [ + 122, + 163, + 190, + 179 + ], + "score": 1.0, + "content": "Hyperparameter", + "type": "text" + }, + { + "bbox": [ + 191, + 167, + 199, + 175 + ], + "score": 0.79, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 163, + 319, + 179 + ], + "score": 1.0, + "content": "for mixup vicinal distribution", + "type": "text" + }, + { + "bbox": [ + 320, + 165, + 366, + 177 + ], + "score": 0.84, + "content": "p _ { m i x u p } ( \\mathbf { X } )", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 5.5, + "bbox_fs": [ + 122, + 130, + 397, + 179 + ] + }, + { + "type": "title", + "bbox": [ + 110, + 177, + 145, + 188 + ], + "lines": [ + { + "bbox": [ + 107, + 173, + 147, + 192 + ], + "spans": [ + { + "bbox": [ + 107, + 173, + 147, + 192 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 111, + 188, + 346, + 351 + ], + "lines": [ + { + "bbox": [ + 123, + 186, + 243, + 202 + ], + "spans": [ + { + "bbox": [ + 123, + 187, + 133, + 198 + ], + "score": 0.73, + "content": "\\tilde { \\mathbf { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 186, + 192, + 202 + ], + "score": 1.0, + "content": "sampled from", + "type": "text" + }, + { + "bbox": [ + 192, + 188, + 238, + 200 + ], + "score": 0.92, + "content": "p _ { m i x u p } ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 186, + 243, + 202 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 122, + 199, + 344, + 214 + ], + "spans": [ + { + "bbox": [ + 122, + 200, + 169, + 214 + ], + "score": 1.0, + "content": "Estimation", + "type": "text" + }, + { + "bbox": [ + 169, + 201, + 189, + 213 + ], + "score": 0.91, + "content": "L _ { M \\mathbf { z } }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 200, + 246, + 214 + ], + "score": 1.0, + "content": "of the margin", + "type": "text" + }, + { + "bbox": [ + 246, + 199, + 344, + 213 + ], + "score": 0.9, + "content": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) | | p ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) )", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 122, + 213, + 344, + 227 + ], + "spans": [ + { + "bbox": [ + 122, + 213, + 169, + 227 + ], + "score": 1.0, + "content": "Estimation", + "type": "text" + }, + { + "bbox": [ + 169, + 215, + 189, + 226 + ], + "score": 0.89, + "content": "L _ { M _ { \\mathbf { c } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 213, + 246, + 227 + ], + "score": 1.0, + "content": "of the margin", + "type": "text" + }, + { + "bbox": [ + 247, + 214, + 344, + 226 + ], + "score": 0.89, + "content": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) | | p ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) )", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 112, + 225, + 334, + 239 + ], + "spans": [ + { + "bbox": [ + 112, + 227, + 118, + 236 + ], + "score": 1.0, + "content": "1:", + "type": "text" + }, + { + "bbox": [ + 118, + 226, + 200, + 237 + ], + "score": 0.69, + "content": ": \\mathrm { \\bf ~ X } ^ { \\prime } , \\pi ^ { \\prime } , \\mu ^ { \\prime } , \\sigma ^ { \\prime 2 } = \\mathrm { \\bf F }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 225, + 279, + 239 + ], + "score": 1.0, + "content": "andomPermutation", + "type": "text" + }, + { + "bbox": [ + 279, + 227, + 334, + 238 + ], + "score": 0.85, + "content": "( \\mathbf { X } , \\pi , \\mu , \\sigma ^ { 2 } )", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 111, + 236, + 288, + 252 + ], + "spans": [ + { + "bbox": [ + 111, + 236, + 123, + 252 + ], + "score": 1.0, + "content": "2:", + "type": "text" + }, + { + "bbox": [ + 124, + 238, + 235, + 250 + ], + "score": 0.85, + "content": "\\tilde { \\mathbf { X } } = \\lambda * \\mathbf { X } + \\left( 1 - \\lambda \\right) * \\mathbf { X } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 236, + 237, + 252 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 238, + 239, + 288, + 250 + ], + "score": 0.69, + "content": "\\lambda \\in \\beta ( \\alpha , \\alpha )", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 111, + 249, + 262, + 262 + ], + "spans": [ + { + "bbox": [ + 111, + 249, + 221, + 262 + ], + "score": 1.0, + "content": "3: π˜ = OptimalTransportC", + "type": "text" + }, + { + "bbox": [ + 222, + 250, + 262, + 261 + ], + "score": 0.67, + "content": "\\prime ( \\pi , \\pi ^ { \\prime } , \\lambda )", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 111, + 259, + 339, + 274 + ], + "spans": [ + { + "bbox": [ + 111, + 259, + 124, + 274 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 124, + 261, + 168, + 272 + ], + "score": 0.43, + "content": "( \\tilde { \\mu } , \\tilde { \\sigma } ^ { 2 } ) =", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 259, + 339, + 274 + ], + "score": 1.0, + "content": "OptimalTransportZ((µ, σ2), (µ0, σ02), λ)", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 111, + 272, + 248, + 286 + ], + "spans": [ + { + "bbox": [ + 111, + 272, + 123, + 286 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 124, + 272, + 248, + 286 + ], + "score": 0.85, + "content": "q _ { \\phi } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) , q _ { \\phi } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) = \\mathrm { V A E } ( \\tilde { \\mathbf { X } } )", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 111, + 285, + 243, + 299 + ], + "spans": [ + { + "bbox": [ + 111, + 285, + 123, + 299 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 124, + 286, + 243, + 299 + ], + "score": 0.85, + "content": "\\tilde { p } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) = \\mathcal { N } ( z ; \\tilde { \\mu } , \\mathrm { d i a g } ( \\tilde { \\pmb { \\sigma } } ^ { 2 } ) )", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 109, + 297, + 223, + 312 + ], + "spans": [ + { + "bbox": [ + 109, + 297, + 123, + 312 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 124, + 299, + 223, + 311 + ], + "score": 0.86, + "content": "\\tilde { p } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) = \\mathbf { M u l t } ( \\mathbf { c } ; K , \\tilde { \\pi } )", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 109, + 310, + 253, + 325 + ], + "spans": [ + { + "bbox": [ + 109, + 310, + 123, + 325 + ], + "score": 1.0, + "content": "8:", + "type": "text" + }, + { + "bbox": [ + 123, + 311, + 253, + 324 + ], + "score": 0.87, + "content": "{ \\cal L } _ { M _ { \\mathbf { z } } } = { \\cal D } _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { z } } | \\tilde { \\mathbf { X } } ) | | \\tilde { p } ( { \\mathbf { z } } | \\tilde { \\mathbf { X } } ) )", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 109, + 323, + 253, + 339 + ], + "spans": [ + { + "bbox": [ + 109, + 323, + 123, + 339 + ], + "score": 1.0, + "content": "9:", + "type": "text" + }, + { + "bbox": [ + 123, + 324, + 253, + 337 + ], + "score": 0.88, + "content": "{ \\cal L } _ { M _ { \\mathbf { c } } } = { \\cal D } _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | \\tilde { \\mathbf { X } } ) | | \\tilde { p } ( { \\mathbf { c } } | \\tilde { \\mathbf { X } } ) )", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 109, + 335, + 212, + 352 + ], + "spans": [ + { + "bbox": [ + 109, + 335, + 155, + 352 + ], + "score": 1.0, + "content": "10: return", + "type": "text" + }, + { + "bbox": [ + 155, + 337, + 212, + 350 + ], + "score": 0.83, + "content": "\\tilde { \\mathbf { X } } , L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } }", + "type": "inline_equation" + } + ], + "index": 21 + } + ], + "index": 15, + "bbox_fs": [ + 109, + 186, + 344, + 352 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 374, + 282, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 373, + 284, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 284, + 387 + ], + "score": 1.0, + "content": "3.2 OPTIMAL TRANSPORT ESTIMATION", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 394, + 505, + 441 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 397, + 408 + ], + "score": 1.0, + "content": "To summarize the above, VAE aims to learn the useful representation", + "type": "text" + }, + { + "bbox": [ + 398, + 395, + 433, + 407 + ], + "score": 0.92, + "content": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 393, + 454, + 408 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 454, + 395, + 490, + 407 + ], + "score": 0.92, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 393, + 505, + 408 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 405, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 365, + 420 + ], + "score": 1.0, + "content": "reducing the KL divergence between the empirical distribution", + "type": "text" + }, + { + "bbox": [ + 365, + 406, + 403, + 418 + ], + "score": 0.91, + "content": "p _ { e m p } ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 405, + 506, + 420 + ], + "score": 1.0, + "content": "and the model marginal", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 416, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 271, + 430 + ], + "score": 0.91, + "content": "\\begin{array} { r } { p ( \\mathbf { X } ) \\ { \\stackrel { } { = } } \\ \\int _ { \\mathbf { z } } \\int _ { \\mathbf { c } } p ( \\mathbf { X } ) p ( \\mathbf { \\tilde { z } } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) d \\mathbf { z } d \\mathbf { c } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 417, + 375, + 430 + ], + "score": 1.0, + "content": ". Instead of minimizing", + "type": "text" + }, + { + "bbox": [ + 375, + 417, + 467, + 429 + ], + "score": 0.91, + "content": "\\hat { D _ { \\mathrm { K L } } } ( p _ { e m p } ( \\mathbf { X } ) | | p ( \\mathbf { X } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 417, + 505, + 430 + ], + "score": 1.0, + "content": "directly,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 428, + 474, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 474, + 442 + ], + "score": 1.0, + "content": "VAE models use the expected ELBO mentioned in (5) as target via the following inequality", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 393, + 506, + 442 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 187, + 446, + 424, + 461 + ], + "lines": [ + { + "bbox": [ + 187, + 446, + 424, + 461 + ], + "spans": [ + { + "bbox": [ + 187, + 446, + 424, + 461 + ], + "score": 0.9, + "content": "D _ { \\mathrm { K L } } ( p _ { e m p } ( \\mathbf { X } ) \\| p ( \\mathbf { X } ) ) \\leq H ( p _ { e m p } ( \\mathbf { X } ) ) - \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\mathrm { E L B O }", + "type": "interline_equation", + "image_path": "6d5acdd29c8437cdc290bb9221d06570314efa5a1c575c569f47582b6d75ef48.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 187, + 446, + 424, + 461 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 464, + 505, + 520 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "However, one phenomenon is that good ELBO values do not imply accurate inference. A typical", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 475, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 505, + 488 + ], + "score": 1.0, + "content": "example has been discussed in (Zhao et al., 2017). Here we mainly focus on the cause of this", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 486, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 505, + 500 + ], + "score": 1.0, + "content": "phenomenon and propose optimal transport estimation to alleviate this problem in semi-supervised", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 497, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 505, + 510 + ], + "score": 1.0, + "content": "learning. Following the work in (Rezende et al., 2014), we write down the closed form of the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 508, + 446, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 446, + 521 + ], + "score": 1.0, + "content": "expected margin between true log-likelihood and ELBO as (proof in Appendix A.3):", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 464, + 505, + 521 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 524, + 496, + 540 + ], + "lines": [ + { + "bbox": [ + 114, + 524, + 496, + 540 + ], + "spans": [ + { + "bbox": [ + 114, + 524, + 496, + 540 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { p _ { \\epsilon m p } ( \\mathbf { X } ) } [ \\log p ( \\mathbf { X } ) - \\mathrm { E L B O } ] = \\mathbb { E } _ { p _ { \\epsilon m p } ( \\mathbf { X } ) } [ D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { c } | \\mathbf { X } ) ) ]", + "type": "interline_equation", + "image_path": "1094a5ca0aeba83fb1bfdd67768ecac528b1fc67dc9103847c2d257a00e6ad3c.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 114, + 524, + 496, + 540 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 547, + 505, + 626 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 505, + 560 + ], + "score": 1.0, + "content": "Combined with the decomposition (8), training the expected ELBO target can only reduce the dif-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 558, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 265, + 572 + ], + "score": 1.0, + "content": "ference between marginal distributions", + "type": "text" + }, + { + "bbox": [ + 265, + 559, + 315, + 571 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { c } ) , \\bar { q } _ { \\phi } ( \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 558, + 334, + 572 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 334, + 559, + 373, + 570 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } ) \\mathbf { , } p ( \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 558, + 505, + 572 + ], + "score": 1.0, + "content": ". It means that even with a good", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 103, + 567, + 507, + 586 + ], + "spans": [ + { + "bbox": [ + 103, + 567, + 183, + 586 + ], + "score": 1.0, + "content": "ELBO, the margin", + "type": "text" + }, + { + "bbox": [ + 183, + 570, + 318, + 583 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\dot { \\mathbf { X } } ) \\big | \\big | p ( \\dot { \\mathbf { z } } | \\mathbf { X } ) \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 567, + 338, + 586 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 338, + 570, + 472, + 583 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 567, + 507, + 586 + ], + "score": 1.0, + "content": "in (15)", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 581, + 506, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 593 + ], + "score": 1.0, + "content": "can still be large. In this scenario, the consistent optimization of ELBO will contribute no more to", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "score": 1.0, + "content": "the semi-supervised classification performance. However, optimizing the margin in (15) directly", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 603, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 181, + 615 + ], + "score": 1.0, + "content": "is impossible, for", + "type": "text" + }, + { + "bbox": [ + 182, + 603, + 212, + 615 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 603, + 232, + 615 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 233, + 603, + 263, + 615 + ], + "score": 0.93, + "content": "p ( \\mathbf { z } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 603, + 505, + 615 + ], + "score": 1.0, + "content": "are unknown. To remedy this, we extend the empirically", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 613, + 446, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 446, + 627 + ], + "score": 1.0, + "content": "effective approximation in (Zhang et al., 2018) to our VAE framework with the form", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37, + "bbox_fs": [ + 103, + 546, + 507, + 627 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 141, + 629, + 469, + 662 + ], + "lines": [ + { + "bbox": [ + 141, + 629, + 469, + 662 + ], + "spans": [ + { + "bbox": [ + 141, + 629, + 469, + 662 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) \\approx \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) ( \\alpha \\to 0 ) } \\\\ & { \\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) \\approx \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) ( \\alpha \\to 0 ) } \\end{array}", + "type": "interline_equation", + "image_path": "cc4ef86cc3bbd69ae4053aa0af711278880e6e00b9f74d53d807949a3f913750.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 141, + 629, + 469, + 640.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 141, + 640.0, + 469, + 651.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 141, + 651.0, + 469, + 662.0 + ], + "spans": [], + "index": 43 + } + ] + }, + { + "type": "list", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 136, + 679 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 666, + 183, + 678 + ], + "score": 0.92, + "content": "p _ { m i x u p } ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 665, + 448, + 679 + ], + "score": 1.0, + "content": "is the mixup vicinal distribution (Zhang et al., 2018) and", + "type": "text" + }, + { + "bbox": [ + 449, + 668, + 457, + 676 + ], + "score": 0.77, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 665, + 506, + 679 + ], + "score": 1.0, + "content": "is the re-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "lated parameter. Then we propose optimal transport estimation to construct the estimations of", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 683, + 508, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 249, + 701 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\big | \\big | p \\big ( \\mathbf { z } | \\bar { \\mathbf { X } } \\big ) \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 683, + 281, + 712 + ], + "score": 1.0, + "content": "as wellariables", + "type": "text" + }, + { + "bbox": [ + 288, + 683, + 294, + 712 + ], + "score": 1.0, + "content": "s", + "type": "text" + }, + { + "bbox": [ + 294, + 687, + 436, + 701 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\big | \\big | p ( \\mathbf { z } | \\mathbf { X } ) \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 683, + 508, + 712 + ], + "score": 1.0, + "content": "by applying op-ptimal transport", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 281, + 701, + 315, + 709 + ], + "spans": [ + { + "bbox": [ + 281, + 701, + 288, + 709 + ], + "score": 0.57, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 702, + 315, + 709 + ], + "score": 0.27, + "content": "\\mathbf { c }", + "type": "inline_equation" + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "estimation are provided in Algorithm 1, and we present the details of the optimal transport scheme", + "type": "text" + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 720, + 209, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 209, + 732 + ], + "score": 1.0, + "content": "in the rest of this section.", + "type": "text" + } + ], + "index": 49, + "is_list_end_line": true + } + ], + "index": 46.5, + "bbox_fs": [ + 105, + 665, + 508, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 335, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 333, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 328, + 96 + ], + "score": 1.0, + "content": "Algorithm 2 OSPOT-VAE training process with epoch", + "type": "text" + }, + { + "bbox": [ + 328, + 84, + 333, + 92 + ], + "score": 0.55, + "content": "t", + "type": "inline_equation" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 109, + 99, + 137, + 109 + ], + "lines": [ + { + "bbox": [ + 109, + 99, + 137, + 109 + ], + "spans": [], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 123, + 108, + 444, + 165 + ], + "lines": [ + { + "bbox": [ + 122, + 106, + 445, + 122 + ], + "spans": [ + { + "bbox": [ + 122, + 106, + 213, + 122 + ], + "score": 1.0, + "content": "Batch of labeled pairs", + "type": "text" + }, + { + "bbox": [ + 214, + 108, + 278, + 120 + ], + "score": 0.9, + "content": "( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ) \\in \\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 106, + 397, + 122 + ], + "score": 1.0, + "content": ",Batch of unlabeled examples", + "type": "text" + }, + { + "bbox": [ + 397, + 108, + 440, + 120 + ], + "score": 0.91, + "content": "\\mathbf { X } _ { U } \\in \\mathbb { D } _ { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 106, + 445, + 122 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 122, + 118, + 372, + 132 + ], + "spans": [ + { + "bbox": [ + 122, + 118, + 224, + 132 + ], + "score": 1.0, + "content": "ELBO hyperparameters:", + "type": "text" + }, + { + "bbox": [ + 224, + 120, + 367, + 131 + ], + "score": 0.78, + "content": "\\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } = \\mathrm { E L B O S c h e d u l e r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 118, + 372, + 132 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 122, + 129, + 419, + 144 + ], + "spans": [ + { + "bbox": [ + 122, + 129, + 277, + 144 + ], + "score": 1.0, + "content": "Optimal transport estimation weights:", + "type": "text" + }, + { + "bbox": [ + 277, + 133, + 297, + 142 + ], + "score": 0.36, + "content": "w _ { M _ { \\mathbf { z } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 131, + 334, + 142 + ], + "score": 0.7, + "content": "M _ { \\mathbf { z } } \\mathbf { , } w _ { } M _ { \\mathbf { c } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 129, + 402, + 144 + ], + "score": 1.0, + "content": "WeightScheduler", + "type": "text" + }, + { + "bbox": [ + 402, + 130, + 414, + 142 + ], + "score": 0.61, + "content": "\\cdot ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 129, + 419, + 144 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 121, + 141, + 265, + 154 + ], + "spans": [ + { + "bbox": [ + 121, + 141, + 201, + 154 + ], + "score": 1.0, + "content": "Model parameters:", + "type": "text" + }, + { + "bbox": [ + 202, + 141, + 260, + 154 + ], + "score": 0.93, + "content": "\\pmb \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 141, + 265, + 154 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 123, + 154, + 218, + 166 + ], + "spans": [ + { + "bbox": [ + 123, + 154, + 218, + 166 + ], + "score": 1.0, + "content": "Model optimizer: SGD", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 109, + 165, + 145, + 175 + ], + "lines": [ + { + "bbox": [ + 107, + 163, + 147, + 180 + ], + "spans": [ + { + "bbox": [ + 107, + 163, + 147, + 180 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 110, + 174, + 411, + 262 + ], + "lines": [ + { + "bbox": [ + 108, + 173, + 411, + 250 + ], + "spans": [ + { + "bbox": [ + 108, + 183, + 123, + 203 + ], + "score": 1.0, + "content": "1:", + "type": "text" + }, + { + "bbox": [ + 122, + 173, + 209, + 187 + ], + "score": 1.0, + "content": "Updated parameters:", + "type": "text" + }, + { + "bbox": [ + 123, + 179, + 411, + 250 + ], + "score": 0.28, + "content": "\\begin{array} { r l } & { \\mathrm { U p d a t e d ~ p a r a m e t e r s : ~ } \\theta ^ { \\mathrm { { t x } ^ { \\prime } } } , \\phi ^ { \\mathrm { { t x } ^ { \\prime } } } , \\phi ^ { \\mathrm { { t x } ^ { \\prime } } } } \\\\ & { L _ { L } = \\mathcal { L } _ { \\mathbb { D } _ { L } } \\big ( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ; \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big ) } \\\\ & { L _ { U } = \\mathcal { L } _ { \\mathbb { D } _ { U } } \\big ( \\mathbf { X } _ { U } ; \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big ) } \\\\ & { L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } } = \\mathrm { O p t i m a l T r a n s p o r t E s t i m a t i o n } \\big ( \\mathbf { X } _ { U } , q _ { \\phi } \\big ( \\mathbf { z } \\big | \\mathbf { X } _ { U } \\big ) , q _ { \\phi } \\big ( \\mathbf { c } | \\mathbf { X } _ { U } \\big ) \\big ) } \\\\ & { L = L _ { L } + L _ { U } + w _ { M _ { \\mathbf { z } } } L _ { M _ { \\mathbf { z } } } + w _ { M _ { \\mathbf { c } } } L _ { M _ { \\mathbf { c } } } } \\\\ & { \\theta ^ { ( t ) } , \\phi ^ { ( t ) } = \\mathrm { S G D } \\big ( \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } , \\frac { \\partial L } { \\partial \\theta } , \\frac { \\partial L } { \\partial \\phi } \\big ) } \\end{array}", + "type": "inline_equation", + "image_path": "fc7cfbb459d2d240293b020376c2b821484d1a5b2a3d57c2da3e5f66a4316e1b.jpg" + }, + { + "bbox": [ + 209, + 174, + 247, + 187 + ], + "score": 0.75, + "content": "\\pmb { \\theta } ^ { ( t ) } , \\phi ^ { ( t ) }", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 108, + 196, + 123, + 214 + ], + "spans": [ + { + "bbox": [ + 108, + 196, + 123, + 214 + ], + "score": 1.0, + "content": "2:", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 108, + 208, + 123, + 225 + ], + "spans": [ + { + "bbox": [ + 108, + 208, + 123, + 225 + ], + "score": 1.0, + "content": "3:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 109, + 219, + 123, + 237 + ], + "spans": [ + { + "bbox": [ + 109, + 219, + 123, + 237 + ], + "score": 1.0, + "content": "4:", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 108, + 231, + 123, + 250 + ], + "spans": [ + { + "bbox": [ + 108, + 231, + 123, + 250 + ], + "score": 1.0, + "content": "5:", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 109, + 246, + 194, + 261 + ], + "spans": [ + { + "bbox": [ + 109, + 246, + 155, + 261 + ], + "score": 1.0, + "content": "6: return", + "type": "text" + }, + { + "bbox": [ + 156, + 247, + 194, + 261 + ], + "score": 0.76, + "content": "\\theta ^ { ( t ) } , \\phi ^ { ( t ) }", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 505, + 363 + ], + "lines": [ + { + "bbox": [ + 106, + 282, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 505, + 295 + ], + "score": 1.0, + "content": "Optimal Transport Scheme: The mixup vicinal distribution can be understood as applying linear", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 223, + 306 + ], + "score": 1.0, + "content": "transport between the points", + "type": "text" + }, + { + "bbox": [ + 223, + 293, + 270, + 305 + ], + "score": 0.9, + "content": "\\mathbf { X } , \\mathbf { X ^ { \\prime } } \\in \\mathbb { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 293, + 505, + 306 + ], + "score": 1.0, + "content": ", extending the original dataset with new points falling on", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 304, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 173, + 318 + ], + "score": 1.0, + "content": "one straight line", + "type": "text" + }, + { + "bbox": [ + 173, + 304, + 325, + 317 + ], + "score": 0.88, + "content": "\\tilde { \\mathbf { X } } = \\lambda * \\mathbf { X } + ( 1 - \\lambda ) * \\mathbf { X ^ { \\prime } } , \\lambda \\in [ 0 , 1 ] .", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 305, + 345, + 318 + ], + "score": 1.0, + "content": ". For", + "type": "text" + }, + { + "bbox": [ + 345, + 304, + 355, + 316 + ], + "score": 0.8, + "content": "\\mathbf { \\tilde { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 305, + 505, + 318 + ], + "score": 1.0, + "content": ", it is a natural thought that this linear", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "score": 1.0, + "content": "transformation could associate with the shortest-path transport in the latent space. Based on this,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 327, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 225, + 342 + ], + "score": 1.0, + "content": "we calculate the distributions", + "type": "text" + }, + { + "bbox": [ + 225, + 327, + 289, + 341 + ], + "score": 0.93, + "content": "\\tilde { p } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) , \\tilde { p } ( \\mathbf { c } | \\tilde { \\mathbf { X } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 329, + 300, + 342 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 301, + 330, + 317, + 340 + ], + "score": 0.26, + "content": "\\mathbf { z } , \\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "and consider them as the estimation of the true", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "posterior distributions. Following the work of (Ambrosio & Gigli, 2013), the norm-2 based optimal", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 351, + 401, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 177, + 364 + ], + "score": 1.0, + "content": "transport scheme", + "type": "text" + }, + { + "bbox": [ + 177, + 351, + 205, + 363 + ], + "score": 0.93, + "content": "\\gamma ( \\mathbf { x } , \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 351, + 312, + 364 + ], + "score": 1.0, + "content": "between two distributions", + "type": "text" + }, + { + "bbox": [ + 312, + 351, + 331, + 363 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 351, + 349, + 364 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 349, + 351, + 368, + 363 + ], + "score": 0.9, + "content": "p ( \\mathsf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 351, + 401, + 364 + ], + "score": 1.0, + "content": "satisfy:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 365, + 398, + 420 + ], + "lines": [ + { + "bbox": [ + 214, + 365, + 398, + 420 + ], + "spans": [ + { + "bbox": [ + 214, + 365, + 398, + 420 + ], + "score": 0.84, + "content": "\\begin{array} { c } { { \\displaystyle \\operatorname* { i n f } _ { \\gamma ( \\mathbf { x } , \\mathbf { y } ) } \\int _ { \\mathbf { x } } \\int _ { \\mathbf { y } } \\| \\mathbf { x } - \\mathbf { y } \\| _ { 2 } ^ { 2 } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { x } d \\mathbf { y } } } \\\\ { { \\mathrm { s . t . } ~ \\displaystyle \\int _ { \\mathbf { y } } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { y } = p ( \\mathbf { x } ) ; \\displaystyle \\int _ { \\mathbf { x } } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { x } = p ( \\mathbf { y } ) } } \\end{array}", + "type": "interline_equation", + "image_path": "2116513f91ae226591325da222f39f13d12ceacb6040877ced75229faef559bb.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 214, + 365, + 398, + 383.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 214, + 383.3333333333333, + 398, + 401.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 214, + 401.66666666666663, + 398, + 419.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 504, + 457 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 221, + 436 + ], + "score": 1.0, + "content": "For the continuous variable", + "type": "text" + }, + { + "bbox": [ + 221, + 423, + 309, + 435 + ], + "score": 0.91, + "content": "\\mathbf z \\sim \\mathcal N ( \\pmb \\mu , \\mathrm { d i a g } ( \\pmb \\sigma ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 423, + 397, + 436 + ], + "score": 1.0, + "content": "and discrete variable", + "type": "text" + }, + { + "bbox": [ + 398, + 423, + 468, + 435 + ], + "score": 0.93, + "content": "\\mathbf { c } \\sim \\mathbf { M } \\mathbf { u } \\mathrm { l t } ( K , \\pi )", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 423, + 505, + 436 + ], + "score": 1.0, + "content": ", the fol-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "score": 1.0, + "content": "lowing 2 propositions are proposed to calculate the shortest-path based on optimal transport scheme", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 445, + 227, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 227, + 458 + ], + "score": 1.0, + "content": "(see Appendix A.4 for proof).", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 105, + 459, + 503, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 457, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 479, + 472 + ], + "score": 1.0, + "content": "Proposition 3.1. The shortest-path derived from optimal transport scheme (17) between", + "type": "text" + }, + { + "bbox": [ + 479, + 461, + 505, + 470 + ], + "score": 0.75, + "content": "\\mathbf { z } _ { 1 } ~ \\sim", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 107, + 469, + 362, + 483 + ], + "spans": [ + { + "bbox": [ + 107, + 469, + 178, + 482 + ], + "score": 0.92, + "content": "\\mathcal { N } ( \\bar { \\mu } _ { 1 } , d i a g ( \\sigma _ { 1 } ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 469, + 196, + 483 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 197, + 470, + 290, + 482 + ], + "score": 0.9, + "content": "\\mathbf { z } _ { 2 } \\sim \\mathcal { N } ( \\bar { \\pmb { \\mu } } _ { 2 } , d i a g ( \\pmb { \\sigma } _ { 2 } ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 469, + 312, + 483 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 312, + 470, + 351, + 482 + ], + "score": 0.92, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 469, + 362, + 483 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 483, + 353, + 513 + ], + "lines": [ + { + "bbox": [ + 257, + 483, + 353, + 513 + ], + "spans": [ + { + "bbox": [ + 257, + 483, + 353, + 513 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\tilde { \\pmb { \\mu } } = \\lambda \\pmb { \\mu } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\mu } _ { 2 } } \\\\ { \\tilde { \\pmb { \\sigma } } = \\lambda \\pmb { \\sigma } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\sigma } _ { 2 } } \\end{array}", + "type": "interline_equation", + "image_path": "c10fbe1c072f505ca83c0c1cb24197e36de3bdeb0bb5ba0043666b4eef2fcefd.jpg" + } + ] + } + ], + "index": 29.5, + "virtual_lines": [ + { + "bbox": [ + 257, + 483, + 353, + 498.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 257, + 498.0, + 353, + 513.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 513, + 506, + 537 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 309, + 527 + ], + "score": 1.0, + "content": "Proposition 3.2. The shortest-path derived from", + "type": "text" + }, + { + "bbox": [ + 309, + 514, + 324, + 524 + ], + "score": 0.69, + "content": "K L", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 513, + 505, + 527 + ], + "score": 1.0, + "content": "divergence based optimal transport scheme", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 524, + 381, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 141, + 537 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 142, + 525, + 216, + 537 + ], + "score": 0.92, + "content": "\\mathbf { c } _ { 1 } \\sim M u l t ( K , \\pmb { \\pi } _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 524, + 235, + 537 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 235, + 525, + 310, + 537 + ], + "score": 0.91, + "content": "\\mathbf { c } _ { 2 } \\sim M u l t ( K , \\pmb { \\pi } _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 524, + 331, + 537 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 331, + 525, + 370, + 537 + ], + "score": 0.92, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 524, + 381, + 537 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 538, + 352, + 552 + ], + "lines": [ + { + "bbox": [ + 258, + 538, + 352, + 552 + ], + "spans": [ + { + "bbox": [ + 258, + 538, + 352, + 552 + ], + "score": 0.88, + "content": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 }", + "type": "interline_equation", + "image_path": "aa7a6f003d5e3a5a5f6228d01b90d7176fa15f0bf5e398fef355eecf2ef68797.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 258, + 538, + 352, + 552 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 560, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "Algorithm 1 yields the optimal transport estimation of the margin in (15), which leads to a tighter", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "score": 1.0, + "content": "ELBO. In Section 4.2, we demonstrate that with this tighter ELBO, the inference performance of", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 583, + 416, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 416, + 594 + ], + "score": 1.0, + "content": "semi-supervised VAE is significantly improved on many benchmark datasets.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 108, + 606, + 268, + 618 + ], + "lines": [ + { + "bbox": [ + 105, + 605, + 270, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 270, + 619 + ], + "score": 1.0, + "content": "3.3 OPTIMIZATION OF OSPOT-VAE", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "Combining one-stage semi-supervised VAE and optimal transport estimation, we can get the com-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "plete OSPOT-VAE model. The full OSPOT-VAE algorithm is provided in Algorithm 2, and a", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "score": 1.0, + "content": "schematic is shown in Figure 1. Note that the conditions for the approximations used in Algo-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 187, + 673 + ], + "score": 1.0, + "content": "rithm 1,2 satisfy (1)", + "type": "text" + }, + { + "bbox": [ + 187, + 660, + 266, + 672 + ], + "score": 0.92, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\approx p ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "and (2) the VAE model has already achieved a good ELBO.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 671, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 434, + 684 + ], + "score": 1.0, + "content": "Therefore, the warm-up schedule (Higgins et al., 2017) is used to set parameters", + "type": "text" + }, + { + "bbox": [ + 434, + 671, + 487, + 682 + ], + "score": 0.92, + "content": "\\mathbf { I } _ { z } , \\mathbf { I } _ { c } , \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 671, + 506, + 684 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 681, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 683, + 150, + 694 + ], + "score": 0.89, + "content": "w _ { M _ { \\mathbf { z } } } , w _ { M _ { \\mathbf { c } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 681, + 309, + 695 + ], + "score": 1.0, + "content": ". We list the details of “ELBOScheduler", + "type": "text" + }, + { + "bbox": [ + 309, + 682, + 321, + 694 + ], + "score": 0.61, + "content": "( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 681, + 416, + 695 + ], + "score": 1.0, + "content": "” and “WeightScheduler", + "type": "text" + }, + { + "bbox": [ + 416, + 682, + 433, + 694 + ], + "score": 0.67, + "content": "( t ) ^ { , }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 681, + 506, + 695 + ], + "score": 1.0, + "content": "in Appendix A.5.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "score": 1.0, + "content": "In Algorithm 2, we apply stochastic gradient descent (SGD) as optimizer, which needs to cal-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 186, + 723 + ], + "score": 1.0, + "content": "culate the gradient", + "type": "text" + }, + { + "bbox": [ + 187, + 710, + 216, + 722 + ], + "score": 0.92, + "content": "\\nabla _ { \\pmb { \\theta } , \\pmb { \\phi } } L", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 709, + 290, + 723 + ], + "score": 1.0, + "content": ". The target loss", + "type": "text" + }, + { + "bbox": [ + 290, + 710, + 299, + 720 + ], + "score": 0.79, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "consists of KL divergence and the expected log-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "likelihood . The derivation of KL divergence part has a closed form, while calculating the gradient", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 335, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 333, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 328, + 96 + ], + "score": 1.0, + "content": "Algorithm 2 OSPOT-VAE training process with epoch", + "type": "text" + }, + { + "bbox": [ + 328, + 84, + 333, + 92 + ], + "score": 0.55, + "content": "t", + "type": "inline_equation" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 109, + 99, + 137, + 109 + ], + "lines": [ + { + "bbox": [ + 109, + 99, + 137, + 109 + ], + "spans": [], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "list", + "bbox": [ + 123, + 108, + 444, + 165 + ], + "lines": [ + { + "bbox": [ + 122, + 106, + 445, + 122 + ], + "spans": [ + { + "bbox": [ + 122, + 106, + 213, + 122 + ], + "score": 1.0, + "content": "Batch of labeled pairs", + "type": "text" + }, + { + "bbox": [ + 214, + 108, + 278, + 120 + ], + "score": 0.9, + "content": "( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ) \\in \\mathbb { D } _ { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 106, + 397, + 122 + ], + "score": 1.0, + "content": ",Batch of unlabeled examples", + "type": "text" + }, + { + "bbox": [ + 397, + 108, + 440, + 120 + ], + "score": 0.91, + "content": "\\mathbf { X } _ { U } \\in \\mathbb { D } _ { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 106, + 445, + 122 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 122, + 118, + 372, + 132 + ], + "spans": [ + { + "bbox": [ + 122, + 118, + 224, + 132 + ], + "score": 1.0, + "content": "ELBO hyperparameters:", + "type": "text" + }, + { + "bbox": [ + 224, + 120, + 367, + 131 + ], + "score": 0.78, + "content": "\\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } = \\mathrm { E L B O S c h e d u l e r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 118, + 372, + 132 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 122, + 129, + 419, + 144 + ], + "spans": [ + { + "bbox": [ + 122, + 129, + 277, + 144 + ], + "score": 1.0, + "content": "Optimal transport estimation weights:", + "type": "text" + }, + { + "bbox": [ + 277, + 133, + 297, + 142 + ], + "score": 0.36, + "content": "w _ { M _ { \\mathbf { z } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 131, + 334, + 142 + ], + "score": 0.7, + "content": "M _ { \\mathbf { z } } \\mathbf { , } w _ { } M _ { \\mathbf { c } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 129, + 402, + 144 + ], + "score": 1.0, + "content": "WeightScheduler", + "type": "text" + }, + { + "bbox": [ + 402, + 130, + 414, + 142 + ], + "score": 0.61, + "content": "\\cdot ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 129, + 419, + 144 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 121, + 141, + 265, + 154 + ], + "spans": [ + { + "bbox": [ + 121, + 141, + 201, + 154 + ], + "score": 1.0, + "content": "Model parameters:", + "type": "text" + }, + { + "bbox": [ + 202, + 141, + 260, + 154 + ], + "score": 0.93, + "content": "\\pmb \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 141, + 265, + 154 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 123, + 154, + 218, + 166 + ], + "spans": [ + { + "bbox": [ + 123, + 154, + 218, + 166 + ], + "score": 1.0, + "content": "Model optimizer: SGD", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + } + ], + "index": 4, + "bbox_fs": [ + 121, + 106, + 445, + 166 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 165, + 145, + 175 + ], + "lines": [ + { + "bbox": [ + 107, + 163, + 147, + 180 + ], + "spans": [ + { + "bbox": [ + 107, + 163, + 147, + 180 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "index", + "bbox": [ + 110, + 174, + 411, + 262 + ], + "lines": [ + { + "bbox": [ + 108, + 173, + 411, + 250 + ], + "spans": [ + { + "bbox": [ + 108, + 183, + 123, + 203 + ], + "score": 1.0, + "content": "1:", + "type": "text" + }, + { + "bbox": [ + 122, + 173, + 209, + 187 + ], + "score": 1.0, + "content": "Updated parameters:", + "type": "text" + }, + { + "bbox": [ + 123, + 179, + 411, + 250 + ], + "score": 0.28, + "content": "\\begin{array} { r l } & { \\mathrm { U p d a t e d ~ p a r a m e t e r s : ~ } \\theta ^ { \\mathrm { { t x } ^ { \\prime } } } , \\phi ^ { \\mathrm { { t x } ^ { \\prime } } } , \\phi ^ { \\mathrm { { t x } ^ { \\prime } } } } \\\\ & { L _ { L } = \\mathcal { L } _ { \\mathbb { D } _ { L } } \\big ( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ; \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big ) } \\\\ & { L _ { U } = \\mathcal { L } _ { \\mathbb { D } _ { U } } \\big ( \\mathbf { X } _ { U } ; \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big ) } \\\\ & { L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } } = \\mathrm { O p t i m a l T r a n s p o r t E s t i m a t i o n } \\big ( \\mathbf { X } _ { U } , q _ { \\phi } \\big ( \\mathbf { z } \\big | \\mathbf { X } _ { U } \\big ) , q _ { \\phi } \\big ( \\mathbf { c } | \\mathbf { X } _ { U } \\big ) \\big ) } \\\\ & { L = L _ { L } + L _ { U } + w _ { M _ { \\mathbf { z } } } L _ { M _ { \\mathbf { z } } } + w _ { M _ { \\mathbf { c } } } L _ { M _ { \\mathbf { c } } } } \\\\ & { \\theta ^ { ( t ) } , \\phi ^ { ( t ) } = \\mathrm { S G D } \\big ( \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } , \\frac { \\partial L } { \\partial \\theta } , \\frac { \\partial L } { \\partial \\phi } \\big ) } \\end{array}", + "type": "inline_equation", + "image_path": "fc7cfbb459d2d240293b020376c2b821484d1a5b2a3d57c2da3e5f66a4316e1b.jpg" + }, + { + "bbox": [ + 209, + 174, + 247, + 187 + ], + "score": 0.75, + "content": "\\pmb { \\theta } ^ { ( t ) } , \\phi ^ { ( t ) }", + "type": "inline_equation" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 108, + 196, + 123, + 214 + ], + "spans": [ + { + "bbox": [ + 108, + 196, + 123, + 214 + ], + "score": 1.0, + "content": "2:", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 108, + 208, + 123, + 225 + ], + "spans": [ + { + "bbox": [ + 108, + 208, + 123, + 225 + ], + "score": 1.0, + "content": "3:", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 219, + 123, + 237 + ], + "spans": [ + { + "bbox": [ + 109, + 219, + 123, + 237 + ], + "score": 1.0, + "content": "4:", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 108, + 231, + 123, + 250 + ], + "spans": [ + { + "bbox": [ + 108, + 231, + 123, + 250 + ], + "score": 1.0, + "content": "5:", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 246, + 194, + 261 + ], + "spans": [ + { + "bbox": [ + 109, + 246, + 155, + 261 + ], + "score": 1.0, + "content": "6: return", + "type": "text" + }, + { + "bbox": [ + 156, + 247, + 194, + 261 + ], + "score": 0.76, + "content": "\\theta ^ { ( t ) } , \\phi ^ { ( t ) }", + "type": "inline_equation" + } + ], + "index": 13, + "is_list_start_line": true + } + ], + "index": 10.5, + "bbox_fs": [ + 108, + 173, + 411, + 261 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 505, + 363 + ], + "lines": [ + { + "bbox": [ + 106, + 282, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 505, + 295 + ], + "score": 1.0, + "content": "Optimal Transport Scheme: The mixup vicinal distribution can be understood as applying linear", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 223, + 306 + ], + "score": 1.0, + "content": "transport between the points", + "type": "text" + }, + { + "bbox": [ + 223, + 293, + 270, + 305 + ], + "score": 0.9, + "content": "\\mathbf { X } , \\mathbf { X ^ { \\prime } } \\in \\mathbb { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 293, + 505, + 306 + ], + "score": 1.0, + "content": ", extending the original dataset with new points falling on", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 304, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 173, + 318 + ], + "score": 1.0, + "content": "one straight line", + "type": "text" + }, + { + "bbox": [ + 173, + 304, + 325, + 317 + ], + "score": 0.88, + "content": "\\tilde { \\mathbf { X } } = \\lambda * \\mathbf { X } + ( 1 - \\lambda ) * \\mathbf { X ^ { \\prime } } , \\lambda \\in [ 0 , 1 ] .", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 305, + 345, + 318 + ], + "score": 1.0, + "content": ". For", + "type": "text" + }, + { + "bbox": [ + 345, + 304, + 355, + 316 + ], + "score": 0.8, + "content": "\\mathbf { \\tilde { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 305, + 505, + 318 + ], + "score": 1.0, + "content": ", it is a natural thought that this linear", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "score": 1.0, + "content": "transformation could associate with the shortest-path transport in the latent space. Based on this,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 327, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 225, + 342 + ], + "score": 1.0, + "content": "we calculate the distributions", + "type": "text" + }, + { + "bbox": [ + 225, + 327, + 289, + 341 + ], + "score": 0.93, + "content": "\\tilde { p } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) , \\tilde { p } ( \\mathbf { c } | \\tilde { \\mathbf { X } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 329, + 300, + 342 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 301, + 330, + 317, + 340 + ], + "score": 0.26, + "content": "\\mathbf { z } , \\mathbf { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "and consider them as the estimation of the true", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "posterior distributions. Following the work of (Ambrosio & Gigli, 2013), the norm-2 based optimal", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 351, + 401, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 177, + 364 + ], + "score": 1.0, + "content": "transport scheme", + "type": "text" + }, + { + "bbox": [ + 177, + 351, + 205, + 363 + ], + "score": 0.93, + "content": "\\gamma ( \\mathbf { x } , \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 351, + 312, + 364 + ], + "score": 1.0, + "content": "between two distributions", + "type": "text" + }, + { + "bbox": [ + 312, + 351, + 331, + 363 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 351, + 349, + 364 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 349, + 351, + 368, + 363 + ], + "score": 0.9, + "content": "p ( \\mathsf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 351, + 401, + 364 + ], + "score": 1.0, + "content": "satisfy:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 282, + 505, + 364 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 365, + 398, + 420 + ], + "lines": [ + { + "bbox": [ + 214, + 365, + 398, + 420 + ], + "spans": [ + { + "bbox": [ + 214, + 365, + 398, + 420 + ], + "score": 0.84, + "content": "\\begin{array} { c } { { \\displaystyle \\operatorname* { i n f } _ { \\gamma ( \\mathbf { x } , \\mathbf { y } ) } \\int _ { \\mathbf { x } } \\int _ { \\mathbf { y } } \\| \\mathbf { x } - \\mathbf { y } \\| _ { 2 } ^ { 2 } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { x } d \\mathbf { y } } } \\\\ { { \\mathrm { s . t . } ~ \\displaystyle \\int _ { \\mathbf { y } } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { y } = p ( \\mathbf { x } ) ; \\displaystyle \\int _ { \\mathbf { x } } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { x } = p ( \\mathbf { y } ) } } \\end{array}", + "type": "interline_equation", + "image_path": "2116513f91ae226591325da222f39f13d12ceacb6040877ced75229faef559bb.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 214, + 365, + 398, + 383.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 214, + 383.3333333333333, + 398, + 401.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 214, + 401.66666666666663, + 398, + 419.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 504, + 457 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 221, + 436 + ], + "score": 1.0, + "content": "For the continuous variable", + "type": "text" + }, + { + "bbox": [ + 221, + 423, + 309, + 435 + ], + "score": 0.91, + "content": "\\mathbf z \\sim \\mathcal N ( \\pmb \\mu , \\mathrm { d i a g } ( \\pmb \\sigma ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 423, + 397, + 436 + ], + "score": 1.0, + "content": "and discrete variable", + "type": "text" + }, + { + "bbox": [ + 398, + 423, + 468, + 435 + ], + "score": 0.93, + "content": "\\mathbf { c } \\sim \\mathbf { M } \\mathbf { u } \\mathrm { l t } ( K , \\pi )", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 423, + 505, + 436 + ], + "score": 1.0, + "content": ", the fol-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "score": 1.0, + "content": "lowing 2 propositions are proposed to calculate the shortest-path based on optimal transport scheme", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 445, + 227, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 227, + 458 + ], + "score": 1.0, + "content": "(see Appendix A.4 for proof).", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 423, + 506, + 458 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 459, + 503, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 457, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 479, + 472 + ], + "score": 1.0, + "content": "Proposition 3.1. The shortest-path derived from optimal transport scheme (17) between", + "type": "text" + }, + { + "bbox": [ + 479, + 461, + 505, + 470 + ], + "score": 0.75, + "content": "\\mathbf { z } _ { 1 } ~ \\sim", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 107, + 469, + 362, + 483 + ], + "spans": [ + { + "bbox": [ + 107, + 469, + 178, + 482 + ], + "score": 0.92, + "content": "\\mathcal { N } ( \\bar { \\mu } _ { 1 } , d i a g ( \\sigma _ { 1 } ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 469, + 196, + 483 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 197, + 470, + 290, + 482 + ], + "score": 0.9, + "content": "\\mathbf { z } _ { 2 } \\sim \\mathcal { N } ( \\bar { \\pmb { \\mu } } _ { 2 } , d i a g ( \\pmb { \\sigma } _ { 2 } ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 469, + 312, + 483 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 312, + 470, + 351, + 482 + ], + "score": 0.92, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 469, + 362, + 483 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 457, + 505, + 483 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 483, + 353, + 513 + ], + "lines": [ + { + "bbox": [ + 257, + 483, + 353, + 513 + ], + "spans": [ + { + "bbox": [ + 257, + 483, + 353, + 513 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\tilde { \\pmb { \\mu } } = \\lambda \\pmb { \\mu } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\mu } _ { 2 } } \\\\ { \\tilde { \\pmb { \\sigma } } = \\lambda \\pmb { \\sigma } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\sigma } _ { 2 } } \\end{array}", + "type": "interline_equation", + "image_path": "c10fbe1c072f505ca83c0c1cb24197e36de3bdeb0bb5ba0043666b4eef2fcefd.jpg" + } + ] + } + ], + "index": 29.5, + "virtual_lines": [ + { + "bbox": [ + 257, + 483, + 353, + 498.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 257, + 498.0, + 353, + 513.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 513, + 506, + 537 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 309, + 527 + ], + "score": 1.0, + "content": "Proposition 3.2. The shortest-path derived from", + "type": "text" + }, + { + "bbox": [ + 309, + 514, + 324, + 524 + ], + "score": 0.69, + "content": "K L", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 513, + 505, + 527 + ], + "score": 1.0, + "content": "divergence based optimal transport scheme", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 524, + 381, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 141, + 537 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 142, + 525, + 216, + 537 + ], + "score": 0.92, + "content": "\\mathbf { c } _ { 1 } \\sim M u l t ( K , \\pmb { \\pi } _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 524, + 235, + 537 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 235, + 525, + 310, + 537 + ], + "score": 0.91, + "content": "\\mathbf { c } _ { 2 } \\sim M u l t ( K , \\pmb { \\pi } _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 524, + 331, + 537 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 331, + 525, + 370, + 537 + ], + "score": 0.92, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 524, + 381, + 537 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 106, + 513, + 505, + 537 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 538, + 352, + 552 + ], + "lines": [ + { + "bbox": [ + 258, + 538, + 352, + 552 + ], + "spans": [ + { + "bbox": [ + 258, + 538, + 352, + 552 + ], + "score": 0.88, + "content": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 }", + "type": "interline_equation", + "image_path": "aa7a6f003d5e3a5a5f6228d01b90d7176fa15f0bf5e398fef355eecf2ef68797.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 258, + 538, + 352, + 552 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 560, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "Algorithm 1 yields the optimal transport estimation of the margin in (15), which leads to a tighter", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "score": 1.0, + "content": "ELBO. In Section 4.2, we demonstrate that with this tighter ELBO, the inference performance of", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 583, + 416, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 416, + 594 + ], + "score": 1.0, + "content": "semi-supervised VAE is significantly improved on many benchmark datasets.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 106, + 560, + 506, + 594 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 606, + 268, + 618 + ], + "lines": [ + { + "bbox": [ + 105, + 605, + 270, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 270, + 619 + ], + "score": 1.0, + "content": "3.3 OPTIMIZATION OF OSPOT-VAE", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "Combining one-stage semi-supervised VAE and optimal transport estimation, we can get the com-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "plete OSPOT-VAE model. The full OSPOT-VAE algorithm is provided in Algorithm 2, and a", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "score": 1.0, + "content": "schematic is shown in Figure 1. Note that the conditions for the approximations used in Algo-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 187, + 673 + ], + "score": 1.0, + "content": "rithm 1,2 satisfy (1)", + "type": "text" + }, + { + "bbox": [ + 187, + 660, + 266, + 672 + ], + "score": 0.92, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\approx p ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "and (2) the VAE model has already achieved a good ELBO.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 671, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 434, + 684 + ], + "score": 1.0, + "content": "Therefore, the warm-up schedule (Higgins et al., 2017) is used to set parameters", + "type": "text" + }, + { + "bbox": [ + 434, + 671, + 487, + 682 + ], + "score": 0.92, + "content": "\\mathbf { I } _ { z } , \\mathbf { I } _ { c } , \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 671, + 506, + 684 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 681, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 683, + 150, + 694 + ], + "score": 0.89, + "content": "w _ { M _ { \\mathbf { z } } } , w _ { M _ { \\mathbf { c } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 681, + 309, + 695 + ], + "score": 1.0, + "content": ". We list the details of “ELBOScheduler", + "type": "text" + }, + { + "bbox": [ + 309, + 682, + 321, + 694 + ], + "score": 0.61, + "content": "( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 681, + 416, + 695 + ], + "score": 1.0, + "content": "” and “WeightScheduler", + "type": "text" + }, + { + "bbox": [ + 416, + 682, + 433, + 694 + ], + "score": 0.67, + "content": "( t ) ^ { , }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 681, + 506, + 695 + ], + "score": 1.0, + "content": "in Appendix A.5.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 626, + 506, + 695 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "score": 1.0, + "content": "In Algorithm 2, we apply stochastic gradient descent (SGD) as optimizer, which needs to cal-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 186, + 723 + ], + "score": 1.0, + "content": "culate the gradient", + "type": "text" + }, + { + "bbox": [ + 187, + 710, + 216, + 722 + ], + "score": 0.92, + "content": "\\nabla _ { \\pmb { \\theta } , \\pmb { \\phi } } L", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 709, + 290, + 723 + ], + "score": 1.0, + "content": ". The target loss", + "type": "text" + }, + { + "bbox": [ + 290, + 710, + 299, + 720 + ], + "score": 0.79, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "consists of KL divergence and the expected log-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "likelihood . The derivation of KL divergence part has a closed form, while calculating the gradient", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 697, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 38, + 507, + 122 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 38, + 507, + 122 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 38, + 507, + 122 + ], + "spans": [ + { + "bbox": [ + 106, + 38, + 507, + 122 + ], + "score": 0.976, + "html": "
MNIST(100 labels)SVHN(1k labels)
BackBoneMethod
Same with M1+M2Disentangled VAE9.71(±0.91)38.91(±1.06)
(Narayanaswamy et al., 2017)11.97(±1.71)54.33(±0.11)
M1(Kingma et al., 2014)3.33(±0.14)36.02(±0.10)
M1+M2(Kingma et al., 2014)
One-stage VAE3.14(±0.19)27.38(±0.78)
", + "type": "table", + "image_path": "0fbc7c4661ac493feb70faa4b6cb7b05b60e960895adf465d9253ef3c448e160.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 38, + 507, + 66.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 66.0, + 507, + 94.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 94.0, + 507, + 122.0 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "table", + "bbox": [ + 106, + 155, + 520, + 318 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 189, + 134, + 421, + 146 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 190, + 134, + 421, + 147 + ], + "spans": [ + { + "bbox": [ + 190, + 134, + 421, + 147 + ], + "score": 1.0, + "content": "Table 1: One-stage VAE error rate in MNIST and SVHN.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "table_body", + "bbox": [ + 106, + 155, + 520, + 318 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 155, + 520, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 520, + 318 + ], + "score": 0.981, + "html": "
BackBoneModel categoryModelCifar10(4k labels)
WRN-28-2DisagreementTemporal Ensembling(TE) (Laine & Aila,2017)16.37
Mean Teacher(Tarvainen & Valpola,2017)15.87
13.13
VAT+EntMin(Miyato et al., 2019) MixMatch(Berthelot et al., 2019)6.37
GS-BadGANt*(Li et al., 2019)17.11
WRN-28-10 GenerativeGenerativeOSPOT-VAE8.51(±0.32)
AutoAugment(Cubuk et al., 2019)14.1
DisagreementTemporal Ensembing(Laine & Aila, 2017)12.16
MixMatch*(Berthelot et al., 2019)4.95
GS-BadGANt*(Li et al., 2019) GAN combine TE‡*(Wei et al., 2018)14.41
", + "type": "table", + "image_path": "dad9f1da926bf83694743684e6abbbe6639a58b1d56565be7fe305c5166c18ee.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 106, + 155, + 520, + 209.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 209.33333333333334, + 520, + 263.6666666666667 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 263.6666666666667, + 520, + 318.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "text", + "bbox": [ + 106, + 325, + 506, + 370 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 236, + 339 + ], + "score": 1.0, + "content": "Table 2: Error rate in Cifar10.", + "type": "text" + }, + { + "bbox": [ + 237, + 326, + 244, + 338 + ], + "score": 0.76, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 325, + 497, + 339 + ], + "score": 1.0, + "content": "denotes the best semi-supervised generative approach result.", + "type": "text" + }, + { + "bbox": [ + 497, + 326, + 505, + 338 + ], + "score": 0.36, + "content": "\\ddagger", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 289, + 349 + ], + "score": 1.0, + "content": "denotes the model ensemble two categories.", + "type": "text" + }, + { + "bbox": [ + 289, + 339, + 297, + 347 + ], + "score": 0.76, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "denotes the corresponding backbone is not exactly", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 347, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 506, + 360 + ], + "score": 1.0, + "content": "WideResNet (Zagoruyko & Komodakis, 2016), but belongs to one kind of its variations with a", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 360, + 248, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 248, + 371 + ], + "score": 1.0, + "content": "comparable amount of parameters.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 106, + 376, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 105, + 375, + 507, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 118, + 392 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 377, + 245, + 390 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 375, + 507, + 392 + ], + "score": 1.0, + "content": "is difficult. To this end, we follow the work of (Rezende et al.,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 387, + 374, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 374, + 401 + ], + "score": 1.0, + "content": "2014) and (Jang et al., 2017), using the reparameterization trick as", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 403, + 480, + 444 + ], + "lines": [ + { + "bbox": [ + 112, + 403, + 480, + 444 + ], + "spans": [ + { + "bbox": [ + 112, + 403, + 480, + 444 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { \\nabla _ { \\theta , \\phi } \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } ) = \\mathbb { E } _ { \\mathcal { N } ( \\epsilon ; \\mathbf { 0 } , \\mathbf { I } ) } \\nabla _ { \\theta , \\phi } \\log p _ { \\theta } ( \\mathbf { X } | \\mu + \\sigma \\cdot \\epsilon ) } \\\\ & { \\mathbb { E } _ { \\mathrm { G u m b e l } ( \\epsilon ; \\mathbf { 0 } , \\mathbf { 1 } ) } ) \\nabla _ { \\theta , \\phi } \\log p _ { \\theta } ( \\mathbf { X } | \\mathrm { S o f t m a x } ( \\frac { \\log \\pi + \\epsilon } { \\tau } ) ) \\to \\nabla _ { \\theta , \\phi } \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } ) ( \\tau \\to 0 ) } \\end{array}", + "type": "interline_equation", + "image_path": "2f16d44efa6c310194e924f6fce7a5608d4e7b7cdb4543c02f19e655fc237739.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 112, + 403, + 480, + 416.6666666666667 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 112, + 416.6666666666667, + 480, + 430.33333333333337 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 112, + 430.33333333333337, + 480, + 444.00000000000006 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 505, + 470 + ], + "lines": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "Note that with (20), the algorithmic complexity of one-stage semi-supervised VAE is independent", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 458, + 451, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 196, + 471 + ], + "score": 1.0, + "content": "with the class number", + "type": "text" + }, + { + "bbox": [ + 196, + 459, + 207, + 469 + ], + "score": 0.82, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 458, + 451, + 471 + ], + "score": 1.0, + "content": ", making it easier to extend to large-scale classification tasks.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 108, + 486, + 200, + 498 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 201, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 201, + 501 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 510, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 511, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 505, + 522 + ], + "score": 1.0, + "content": "In this section, we demonstrate the 3 contributions of our OSPOT-VAE model with sufficient", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 522, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 533 + ], + "score": 1.0, + "content": "experiments on 4 standard SSL benchmark datasets, that is, MNIST, SVHN, Cifar10, and Ci-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "far100. In Section 4.1, we show the validity of our one-stage semi-supervised VAE objective", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 544, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 556 + ], + "score": 1.0, + "content": "(13) by comparing with other one-stage and two-stage VAE models. Then, we evaluate the per-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "formance of OSPOT-VAE under “WideResNet”(Zagoruyko & Komodakis, 2016) backbone and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 566, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 578 + ], + "score": 1.0, + "content": "compare with other state-of-the-art SSL models mentioned in Section 2. Besides, We provide", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "an ablation study to verify the contribution of the optimal transport estimation. As an addi-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 587, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 505, + 600 + ], + "score": 1.0, + "content": "tional application, we show that good generative models and semi-supervised results can be ob-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 598, + 504, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 504, + 611 + ], + "score": 1.0, + "content": "tained at the same time by OSPOT-VAE (Section 4.3). The source code is available at https:", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 609, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 623 + ], + "score": 1.0, + "content": "//github.com/PaperCodeSubmission/OSPOT-VAE; more details are available in Ap-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 620, + 156, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 620, + 156, + 632 + ], + "score": 1.0, + "content": "pendix A.6.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 109, + 645, + 287, + 656 + ], + "lines": [ + { + "bbox": [ + 106, + 644, + 288, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 288, + 658 + ], + "score": 1.0, + "content": "4.1 ONE-STAGE SEMI-SUPERVISED VAE", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "We evaluate the effectiveness of the one-stage semi-supervised VAE objective on 2 standard bench-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "marks, MNIST and SVHN. As for baseline models, we consider two VAE-based SSL models,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 465, + 700 + ], + "score": 1.0, + "content": "which are one-stage disentangled VAE (Narayanaswamy et al., 2017) and two-stage VAE", + "type": "text" + }, + { + "bbox": [ + 465, + 688, + 501, + 699 + ], + "score": 0.67, + "content": "[ \\mathbf { M } 1 + \\mathbf { M } 2 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 687, + 505, + 700 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "(Kingma et al., 2014). For fairness, except the target loss functions, all models use the same struc-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 174, + 722 + ], + "score": 1.0, + "content": "ture as is used in", + "type": "text" + }, + { + "bbox": [ + 174, + 710, + 209, + 721 + ], + "score": 0.8, + "content": "\\mathbf { M } 1 { + } \\mathbf { M } 2", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "(Kingma et al., 2014). The results are presented in Table 1, and our model", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 721, + 232, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 232, + 733 + ], + "score": 1.0, + "content": "achieves the best performance.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 26, + 309, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 38, + 507, + 122 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 38, + 507, + 122 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 38, + 507, + 122 + ], + "spans": [ + { + "bbox": [ + 106, + 38, + 507, + 122 + ], + "score": 0.976, + "html": "
MNIST(100 labels)SVHN(1k labels)
BackBoneMethod
Same with M1+M2Disentangled VAE9.71(±0.91)38.91(±1.06)
(Narayanaswamy et al., 2017)11.97(±1.71)54.33(±0.11)
M1(Kingma et al., 2014)3.33(±0.14)36.02(±0.10)
M1+M2(Kingma et al., 2014)
One-stage VAE3.14(±0.19)27.38(±0.78)
", + "type": "table", + "image_path": "0fbc7c4661ac493feb70faa4b6cb7b05b60e960895adf465d9253ef3c448e160.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 38, + 507, + 66.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 66.0, + 507, + 94.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 94.0, + 507, + 122.0 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "table", + "bbox": [ + 106, + 155, + 520, + 318 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 189, + 134, + 421, + 146 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 190, + 134, + 421, + 147 + ], + "spans": [ + { + "bbox": [ + 190, + 134, + 421, + 147 + ], + "score": 1.0, + "content": "Table 1: One-stage VAE error rate in MNIST and SVHN.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "table_body", + "bbox": [ + 106, + 155, + 520, + 318 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 155, + 520, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 520, + 318 + ], + "score": 0.981, + "html": "
BackBoneModel categoryModelCifar10(4k labels)
WRN-28-2DisagreementTemporal Ensembling(TE) (Laine & Aila,2017)16.37
Mean Teacher(Tarvainen & Valpola,2017)15.87
13.13
VAT+EntMin(Miyato et al., 2019) MixMatch(Berthelot et al., 2019)6.37
GS-BadGANt*(Li et al., 2019)17.11
WRN-28-10 GenerativeGenerativeOSPOT-VAE8.51(±0.32)
AutoAugment(Cubuk et al., 2019)14.1
DisagreementTemporal Ensembing(Laine & Aila, 2017)12.16
MixMatch*(Berthelot et al., 2019)4.95
GS-BadGANt*(Li et al., 2019) GAN combine TE‡*(Wei et al., 2018)14.41
", + "type": "table", + "image_path": "dad9f1da926bf83694743684e6abbbe6639a58b1d56565be7fe305c5166c18ee.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 106, + 155, + 520, + 209.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 209.33333333333334, + 520, + 263.6666666666667 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 263.6666666666667, + 520, + 318.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "text", + "bbox": [ + 106, + 325, + 506, + 370 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 236, + 339 + ], + "score": 1.0, + "content": "Table 2: Error rate in Cifar10.", + "type": "text" + }, + { + "bbox": [ + 237, + 326, + 244, + 338 + ], + "score": 0.76, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 325, + 497, + 339 + ], + "score": 1.0, + "content": "denotes the best semi-supervised generative approach result.", + "type": "text" + }, + { + "bbox": [ + 497, + 326, + 505, + 338 + ], + "score": 0.36, + "content": "\\ddagger", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 289, + 349 + ], + "score": 1.0, + "content": "denotes the model ensemble two categories.", + "type": "text" + }, + { + "bbox": [ + 289, + 339, + 297, + 347 + ], + "score": 0.76, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "denotes the corresponding backbone is not exactly", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 347, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 506, + 360 + ], + "score": 1.0, + "content": "WideResNet (Zagoruyko & Komodakis, 2016), but belongs to one kind of its variations with a", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 360, + 248, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 248, + 371 + ], + "score": 1.0, + "content": "comparable amount of parameters.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 325, + 506, + 371 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 376, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 105, + 375, + 507, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 118, + 392 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 377, + 245, + 390 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 375, + 507, + 392 + ], + "score": 1.0, + "content": "is difficult. To this end, we follow the work of (Rezende et al.,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 387, + 374, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 374, + 401 + ], + "score": 1.0, + "content": "2014) and (Jang et al., 2017), using the reparameterization trick as", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 375, + 507, + 401 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 403, + 480, + 444 + ], + "lines": [ + { + "bbox": [ + 112, + 403, + 480, + 444 + ], + "spans": [ + { + "bbox": [ + 112, + 403, + 480, + 444 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { \\nabla _ { \\theta , \\phi } \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } ) = \\mathbb { E } _ { \\mathcal { N } ( \\epsilon ; \\mathbf { 0 } , \\mathbf { I } ) } \\nabla _ { \\theta , \\phi } \\log p _ { \\theta } ( \\mathbf { X } | \\mu + \\sigma \\cdot \\epsilon ) } \\\\ & { \\mathbb { E } _ { \\mathrm { G u m b e l } ( \\epsilon ; \\mathbf { 0 } , \\mathbf { 1 } ) } ) \\nabla _ { \\theta , \\phi } \\log p _ { \\theta } ( \\mathbf { X } | \\mathrm { S o f t m a x } ( \\frac { \\log \\pi + \\epsilon } { \\tau } ) ) \\to \\nabla _ { \\theta , \\phi } \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } ) ( \\tau \\to 0 ) } \\end{array}", + "type": "interline_equation", + "image_path": "2f16d44efa6c310194e924f6fce7a5608d4e7b7cdb4543c02f19e655fc237739.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 112, + 403, + 480, + 416.6666666666667 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 112, + 416.6666666666667, + 480, + 430.33333333333337 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 112, + 430.33333333333337, + 480, + 444.00000000000006 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 505, + 470 + ], + "lines": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "Note that with (20), the algorithmic complexity of one-stage semi-supervised VAE is independent", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 458, + 451, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 196, + 471 + ], + "score": 1.0, + "content": "with the class number", + "type": "text" + }, + { + "bbox": [ + 196, + 459, + 207, + 469 + ], + "score": 0.82, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 458, + 451, + 471 + ], + "score": 1.0, + "content": ", making it easier to extend to large-scale classification tasks.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 447, + 505, + 471 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 486, + 200, + 498 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 201, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 201, + 501 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 510, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 511, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 505, + 522 + ], + "score": 1.0, + "content": "In this section, we demonstrate the 3 contributions of our OSPOT-VAE model with sufficient", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 522, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 533 + ], + "score": 1.0, + "content": "experiments on 4 standard SSL benchmark datasets, that is, MNIST, SVHN, Cifar10, and Ci-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "far100. In Section 4.1, we show the validity of our one-stage semi-supervised VAE objective", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 544, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 556 + ], + "score": 1.0, + "content": "(13) by comparing with other one-stage and two-stage VAE models. Then, we evaluate the per-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "formance of OSPOT-VAE under “WideResNet”(Zagoruyko & Komodakis, 2016) backbone and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 566, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 578 + ], + "score": 1.0, + "content": "compare with other state-of-the-art SSL models mentioned in Section 2. Besides, We provide", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "an ablation study to verify the contribution of the optimal transport estimation. As an addi-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 587, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 505, + 600 + ], + "score": 1.0, + "content": "tional application, we show that good generative models and semi-supervised results can be ob-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 598, + 504, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 504, + 611 + ], + "score": 1.0, + "content": "tained at the same time by OSPOT-VAE (Section 4.3). The source code is available at https:", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 609, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 623 + ], + "score": 1.0, + "content": "//github.com/PaperCodeSubmission/OSPOT-VAE; more details are available in Ap-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 620, + 156, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 620, + 156, + 632 + ], + "score": 1.0, + "content": "pendix A.6.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 511, + 505, + 632 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 645, + 287, + 656 + ], + "lines": [ + { + "bbox": [ + 106, + 644, + 288, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 288, + 658 + ], + "score": 1.0, + "content": "4.1 ONE-STAGE SEMI-SUPERVISED VAE", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "We evaluate the effectiveness of the one-stage semi-supervised VAE objective on 2 standard bench-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "marks, MNIST and SVHN. As for baseline models, we consider two VAE-based SSL models,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 465, + 700 + ], + "score": 1.0, + "content": "which are one-stage disentangled VAE (Narayanaswamy et al., 2017) and two-stage VAE", + "type": "text" + }, + { + "bbox": [ + 465, + 688, + 501, + 699 + ], + "score": 0.67, + "content": "[ \\mathbf { M } 1 + \\mathbf { M } 2 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 687, + 505, + 700 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "(Kingma et al., 2014). For fairness, except the target loss functions, all models use the same struc-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 174, + 722 + ], + "score": 1.0, + "content": "ture as is used in", + "type": "text" + }, + { + "bbox": [ + 174, + 710, + 209, + 721 + ], + "score": 0.8, + "content": "\\mathbf { M } 1 { + } \\mathbf { M } 2", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "(Kingma et al., 2014). The results are presented in Table 1, and our model", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 721, + 232, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 232, + 733 + ], + "score": 1.0, + "content": "achieves the best performance.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 666, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 69, + 514, + 155 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 137, + 51, + 471, + 64 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 138, + 51, + 472, + 65 + ], + "spans": [ + { + "bbox": [ + 138, + 51, + 249, + 65 + ], + "score": 1.0, + "content": "Table 3: Error in Cifar100.", + "type": "text" + }, + { + "bbox": [ + 249, + 52, + 255, + 64 + ], + "score": 0.67, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 51, + 272, + 65 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 273, + 54, + 280, + 62 + ], + "score": 0.76, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 51, + 472, + 65 + ], + "score": 1.0, + "content": "have the same meaning as described in Table 2.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 106, + 69, + 514, + 155 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 69, + 514, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 69, + 514, + 155 + ], + "score": 0.977, + "html": "
BackBoneModelCifar100(4k labels)Cifar100(10k labels)
WRN-28-2II - Model(Laine & Aila, 2017)39.19
GS-BadGANt*(Li et al., 2019)45.1137.16
LP*(Iscen et al., 2019)43.7335.92
OSPOT-VAE40.58(±0.48)31.41(±0.21)
WRN-28-10MixMatch* (Berthelot et al., 2019)25.88
OSPOT-VAE33.76(±0.53)25.30(±0.31)
", + "type": "table", + "image_path": "df43f7645c30e3e246c8d138916fed7f8b94841ff7bba869abee8e5251a4c8b1.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 69, + 514, + 97.66666666666667 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 97.66666666666667, + 514, + 126.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 126.33333333333334, + 514, + 155.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "table", + "bbox": [ + 107, + 183, + 506, + 243 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 184, + 164, + 425, + 176 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 185, + 163, + 425, + 176 + ], + "spans": [ + { + "bbox": [ + 185, + 163, + 425, + 176 + ], + "score": 1.0, + "content": "Table 4: Ablation study with SVHN, Cifar10, and Cifar100.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "table_body", + "bbox": [ + 107, + 183, + 506, + 243 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 183, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 107, + 183, + 506, + 243 + ], + "score": 0.975, + "html": "
MethodsSVHN(1k labels)Cifar10(4k labels)Cifar100(10k labels)
One-stage VAE10.53(±0.17)18.26(±0.51)38.62(±0.67)
Optimal transport estimation (with encoder only)6.54(±0.62)10.71(±0.44)36.21(±0.29)
OSPOT-VAE5.79(±0.15)8.51(±0.32)31.41(±0.21)
", + "type": "table", + "image_path": "4a6868c1c37140527432e50a5891d56788d6ea26e5419e19e71f238f827f019e.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 107, + 183, + 506, + 203.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 107, + 203.0, + 506, + 223.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 107, + 223.0, + 506, + 243.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "title", + "bbox": [ + 107, + 251, + 189, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 249, + 191, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 191, + 265 + ], + "score": 1.0, + "content": "4.2 OSPOT-VAE", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 271, + 505, + 371 + ], + "lines": [ + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "score": 1.0, + "content": "We compare the results of OSPOT-VAE with two categories of state-of-the-art models mentioned", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 282, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 505, + 295 + ], + "score": 1.0, + "content": "in Section2. In all experiments, we use the “WideResNet-28” model or other deep models with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 294, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 104, + 294, + 506, + 306 + ], + "score": 1.0, + "content": "a comparable amount of parameters as the backbone. The results in Table 2,3 demonstrate that", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 304, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 317 + ], + "score": 1.0, + "content": "our model outperforms most of the existing methods and surpasses state-of-the-art semi-supervised", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 315, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 505, + 328 + ], + "score": 1.0, + "content": "generative models (Dai et al., 2017) by a large margin. Notice that recently, data-augmentation", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 327, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 505, + 339 + ], + "score": 1.0, + "content": "based method, MixMatch, (Berthelot et al., 2019) achieves the absolute state-of-the-art results in all", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 337, + 505, + 350 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 350 + ], + "score": 1.0, + "content": "benchmarks. It uses pre-designed sophisticated data augmentation strategies for different datasets", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 348, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 505, + 362 + ], + "score": 1.0, + "content": "and outperforms our model. We list its results fairly as a comparison, while OSPOT-VAE surpasses", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 358, + 195, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 195, + 371 + ], + "score": 1.0, + "content": "it in Cifar100 dataset.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 375, + 505, + 465 + ], + "lines": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "Ablation Study: The OSPOT-VAE model consists of two parts: (1) a one-stage VAE objective and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "(2) an optimal transport estimation. In ablation study, we analyze the effect of each component in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "score": 1.0, + "content": "our model with the backbone “WideResNet-28-2”. To study the independent effects of transport es-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 409, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 422 + ], + "score": 1.0, + "content": "timation, we combine it with the encoder part of OSPOT-VAE to build a classifier with loss function", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 126, + 432 + ], + "score": 0.9, + "content": "L _ { M _ { \\mathbf { c } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 420, + 505, + 433 + ], + "score": 1.0, + "content": ". The improved classification error rates in Table 4 show that, with optimal transport estima-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 218, + 444 + ], + "score": 1.0, + "content": "tion, the posterior inference", + "type": "text" + }, + { + "bbox": [ + 219, + 431, + 254, + 444 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 431, + 390, + 444 + ], + "score": 1.0, + "content": "gets closer to the true distribution", + "type": "text" + }, + { + "bbox": [ + 391, + 431, + 421, + 443 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 431, + 506, + 444 + ], + "score": 1.0, + "content": ". It indicates that our", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 442, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 104, + 442, + 506, + 455 + ], + "score": 1.0, + "content": "optimal transport estimation does reduce the gap between ELBO and the log-likelihood of the input", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 453, + 446, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 446, + 467 + ], + "score": 1.0, + "content": "data and yield a tighter ELBO, which leads to a better semi-supervised performance.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5 + }, + { + "type": "table", + "bbox": [ + 181, + 488, + 426, + 526 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 149, + 469, + 462, + 480 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 147, + 467, + 462, + 483 + ], + "spans": [ + { + "bbox": [ + 147, + 467, + 396, + 483 + ], + "score": 1.0, + "content": "Table 5: Generative performance measured by ELBO with EL", + "type": "text" + }, + { + "bbox": [ + 397, + 468, + 462, + 481 + ], + "score": 0.42, + "content": "\\mathbf { B O } \\leq \\log p ( \\mathbf { X } )", + "type": "inline_equation" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "table_body", + "bbox": [ + 181, + 488, + 426, + 526 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 181, + 488, + 426, + 526 + ], + "spans": [ + { + "bbox": [ + 181, + 488, + 426, + 526 + ], + "score": 0.96, + "html": "
ModelCifar10Cifar100
Pure VAE-226.25(±14.25)−1292.91(±1.10)
OSPOT-VAE-237.62(±6.27)-1271.82(±24.15)
", + "type": "table", + "image_path": "9ce60c9faf8f59fa59253a8299b0174ce2fbbee584b15e9749f147ca0247b678.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 181, + 488, + 426, + 500.6666666666667 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 181, + 500.6666666666667, + 426, + 513.3333333333334 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 181, + 513.3333333333334, + 426, + 526.0 + ], + "spans": [], + "index": 29 + } + ] + } + ], + "index": 27.0 + }, + { + "type": "title", + "bbox": [ + 107, + 538, + 253, + 549 + ], + "lines": [ + { + "bbox": [ + 106, + 538, + 254, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 254, + 550 + ], + "score": 1.0, + "content": "4.3 GENERATIVE PERFORMANCE", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 558, + 505, + 625 + ], + "lines": [ + { + "bbox": [ + 107, + 555, + 509, + 577 + ], + "spans": [ + { + "bbox": [ + 107, + 559, + 172, + 572 + ], + "score": 0.85, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\mathrm { E L B O }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 555, + 509, + 577 + ], + "score": 1.0, + "content": "measures the margin between the true data distribution and the distribution learned", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 568, + 507, + 587 + ], + "spans": [ + { + "bbox": [ + 104, + 568, + 335, + 587 + ], + "score": 1.0, + "content": "by generation models (Doersch, 2016). By comparing the", + "type": "text" + }, + { + "bbox": [ + 335, + 570, + 374, + 583 + ], + "score": 0.92, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 568, + 507, + 587 + ], + "score": 1.0, + "content": "value of pure unsupervised VAE", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 580, + 506, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 506, + 593 + ], + "score": 1.0, + "content": "and our semi-supervised VAE model under the same “WideResNet-28-2” backbone, we demonstrate", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 591, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 505, + 604 + ], + "score": 1.0, + "content": "that good generative models and semi-supervised results can be obtained at the same time in OSPOT-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "VAE. The results in Table 5 show that the data generative distribution learned by our OSPOT-VAE", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 613, + 501, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 501, + 626 + ], + "score": 1.0, + "content": "model is as good as the pure VAE model. Further generated results are available in Appendix A.7.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 107, + 641, + 195, + 654 + ], + "lines": [ + { + "bbox": [ + 104, + 639, + 198, + 657 + ], + "spans": [ + { + "bbox": [ + 104, + 639, + 198, + 657 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "In this work, we pointed out that it was the large margin between ELBO and the true log-likelihood", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "of the raw data that limits the performance of semi-supervised VAE. To this end, we introduced", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "OSPOT-VAE, a one-stage generative model that unified the classification and generation objec-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "tive and achieved a tighter ELBO by optimal transport estimation. We demonstrated our asser-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "tion through extensive experiments, and our semi-supervised results significantly outperform former", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 720, + 442, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 442, + 733 + ], + "score": 1.0, + "content": "state-of-the-art generative SSL methods by a large margin on Cifar10 and Cifar100.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 69, + 514, + 155 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 137, + 51, + 471, + 64 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 138, + 51, + 472, + 65 + ], + "spans": [ + { + "bbox": [ + 138, + 51, + 249, + 65 + ], + "score": 1.0, + "content": "Table 3: Error in Cifar100.", + "type": "text" + }, + { + "bbox": [ + 249, + 52, + 255, + 64 + ], + "score": 0.67, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 51, + 272, + 65 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 273, + 54, + 280, + 62 + ], + "score": 0.76, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 51, + 472, + 65 + ], + "score": 1.0, + "content": "have the same meaning as described in Table 2.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 106, + 69, + 514, + 155 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 69, + 514, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 69, + 514, + 155 + ], + "score": 0.977, + "html": "
BackBoneModelCifar100(4k labels)Cifar100(10k labels)
WRN-28-2II - Model(Laine & Aila, 2017)39.19
GS-BadGANt*(Li et al., 2019)45.1137.16
LP*(Iscen et al., 2019)43.7335.92
OSPOT-VAE40.58(±0.48)31.41(±0.21)
WRN-28-10MixMatch* (Berthelot et al., 2019)25.88
OSPOT-VAE33.76(±0.53)25.30(±0.31)
", + "type": "table", + "image_path": "df43f7645c30e3e246c8d138916fed7f8b94841ff7bba869abee8e5251a4c8b1.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 69, + 514, + 97.66666666666667 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 97.66666666666667, + 514, + 126.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 126.33333333333334, + 514, + 155.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "table", + "bbox": [ + 107, + 183, + 506, + 243 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 184, + 164, + 425, + 176 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 185, + 163, + 425, + 176 + ], + "spans": [ + { + "bbox": [ + 185, + 163, + 425, + 176 + ], + "score": 1.0, + "content": "Table 4: Ablation study with SVHN, Cifar10, and Cifar100.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "table_body", + "bbox": [ + 107, + 183, + 506, + 243 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 183, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 107, + 183, + 506, + 243 + ], + "score": 0.975, + "html": "
MethodsSVHN(1k labels)Cifar10(4k labels)Cifar100(10k labels)
One-stage VAE10.53(±0.17)18.26(±0.51)38.62(±0.67)
Optimal transport estimation (with encoder only)6.54(±0.62)10.71(±0.44)36.21(±0.29)
OSPOT-VAE5.79(±0.15)8.51(±0.32)31.41(±0.21)
", + "type": "table", + "image_path": "4a6868c1c37140527432e50a5891d56788d6ea26e5419e19e71f238f827f019e.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 107, + 183, + 506, + 203.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 107, + 203.0, + 506, + 223.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 107, + 223.0, + 506, + 243.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "title", + "bbox": [ + 107, + 251, + 189, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 249, + 191, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 191, + 265 + ], + "score": 1.0, + "content": "4.2 OSPOT-VAE", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 271, + 505, + 371 + ], + "lines": [ + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "score": 1.0, + "content": "We compare the results of OSPOT-VAE with two categories of state-of-the-art models mentioned", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 282, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 505, + 295 + ], + "score": 1.0, + "content": "in Section2. In all experiments, we use the “WideResNet-28” model or other deep models with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 294, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 104, + 294, + 506, + 306 + ], + "score": 1.0, + "content": "a comparable amount of parameters as the backbone. The results in Table 2,3 demonstrate that", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 304, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 317 + ], + "score": 1.0, + "content": "our model outperforms most of the existing methods and surpasses state-of-the-art semi-supervised", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 315, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 505, + 328 + ], + "score": 1.0, + "content": "generative models (Dai et al., 2017) by a large margin. Notice that recently, data-augmentation", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 327, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 505, + 339 + ], + "score": 1.0, + "content": "based method, MixMatch, (Berthelot et al., 2019) achieves the absolute state-of-the-art results in all", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 337, + 505, + 350 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 350 + ], + "score": 1.0, + "content": "benchmarks. It uses pre-designed sophisticated data augmentation strategies for different datasets", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 348, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 505, + 362 + ], + "score": 1.0, + "content": "and outperforms our model. We list its results fairly as a comparison, while OSPOT-VAE surpasses", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 358, + 195, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 195, + 371 + ], + "score": 1.0, + "content": "it in Cifar100 dataset.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 13, + "bbox_fs": [ + 104, + 272, + 506, + 371 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 375, + 505, + 465 + ], + "lines": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "Ablation Study: The OSPOT-VAE model consists of two parts: (1) a one-stage VAE objective and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "(2) an optimal transport estimation. In ablation study, we analyze the effect of each component in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "score": 1.0, + "content": "our model with the backbone “WideResNet-28-2”. To study the independent effects of transport es-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 409, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 422 + ], + "score": 1.0, + "content": "timation, we combine it with the encoder part of OSPOT-VAE to build a classifier with loss function", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 126, + 432 + ], + "score": 0.9, + "content": "L _ { M _ { \\mathbf { c } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 420, + 505, + 433 + ], + "score": 1.0, + "content": ". The improved classification error rates in Table 4 show that, with optimal transport estima-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 218, + 444 + ], + "score": 1.0, + "content": "tion, the posterior inference", + "type": "text" + }, + { + "bbox": [ + 219, + 431, + 254, + 444 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 431, + 390, + 444 + ], + "score": 1.0, + "content": "gets closer to the true distribution", + "type": "text" + }, + { + "bbox": [ + 391, + 431, + 421, + 443 + ], + "score": 0.92, + "content": "p ( \\mathbf { c } | \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 431, + 506, + 444 + ], + "score": 1.0, + "content": ". It indicates that our", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 442, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 104, + 442, + 506, + 455 + ], + "score": 1.0, + "content": "optimal transport estimation does reduce the gap between ELBO and the log-likelihood of the input", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 453, + 446, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 446, + 467 + ], + "score": 1.0, + "content": "data and yield a tighter ELBO, which leads to a better semi-supervised performance.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5, + "bbox_fs": [ + 104, + 376, + 506, + 467 + ] + }, + { + "type": "table", + "bbox": [ + 181, + 488, + 426, + 526 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 149, + 469, + 462, + 480 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 147, + 467, + 462, + 483 + ], + "spans": [ + { + "bbox": [ + 147, + 467, + 396, + 483 + ], + "score": 1.0, + "content": "Table 5: Generative performance measured by ELBO with EL", + "type": "text" + }, + { + "bbox": [ + 397, + 468, + 462, + 481 + ], + "score": 0.42, + "content": "\\mathbf { B O } \\leq \\log p ( \\mathbf { X } )", + "type": "inline_equation" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "table_body", + "bbox": [ + 181, + 488, + 426, + 526 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 181, + 488, + 426, + 526 + ], + "spans": [ + { + "bbox": [ + 181, + 488, + 426, + 526 + ], + "score": 0.96, + "html": "
ModelCifar10Cifar100
Pure VAE-226.25(±14.25)−1292.91(±1.10)
OSPOT-VAE-237.62(±6.27)-1271.82(±24.15)
", + "type": "table", + "image_path": "9ce60c9faf8f59fa59253a8299b0174ce2fbbee584b15e9749f147ca0247b678.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 181, + 488, + 426, + 500.6666666666667 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 181, + 500.6666666666667, + 426, + 513.3333333333334 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 181, + 513.3333333333334, + 426, + 526.0 + ], + "spans": [], + "index": 29 + } + ] + } + ], + "index": 27.0 + }, + { + "type": "title", + "bbox": [ + 107, + 538, + 253, + 549 + ], + "lines": [ + { + "bbox": [ + 106, + 538, + 254, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 254, + 550 + ], + "score": 1.0, + "content": "4.3 GENERATIVE PERFORMANCE", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 558, + 505, + 625 + ], + "lines": [ + { + "bbox": [ + 107, + 555, + 509, + 577 + ], + "spans": [ + { + "bbox": [ + 107, + 559, + 172, + 572 + ], + "score": 0.85, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\mathrm { E L B O }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 555, + 509, + 577 + ], + "score": 1.0, + "content": "measures the margin between the true data distribution and the distribution learned", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 568, + 507, + 587 + ], + "spans": [ + { + "bbox": [ + 104, + 568, + 335, + 587 + ], + "score": 1.0, + "content": "by generation models (Doersch, 2016). By comparing the", + "type": "text" + }, + { + "bbox": [ + 335, + 570, + 374, + 583 + ], + "score": 0.92, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 568, + 507, + 587 + ], + "score": 1.0, + "content": "value of pure unsupervised VAE", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 580, + 506, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 506, + 593 + ], + "score": 1.0, + "content": "and our semi-supervised VAE model under the same “WideResNet-28-2” backbone, we demonstrate", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 591, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 505, + 604 + ], + "score": 1.0, + "content": "that good generative models and semi-supervised results can be obtained at the same time in OSPOT-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "VAE. The results in Table 5 show that the data generative distribution learned by our OSPOT-VAE", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 613, + 501, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 501, + 626 + ], + "score": 1.0, + "content": "model is as good as the pure VAE model. Further generated results are available in Appendix A.7.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 555, + 509, + 626 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 641, + 195, + 654 + ], + "lines": [ + { + "bbox": [ + 104, + 639, + 198, + 657 + ], + "spans": [ + { + "bbox": [ + 104, + 639, + 198, + 657 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "In this work, we pointed out that it was the large margin between ELBO and the true log-likelihood", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "of the raw data that limits the performance of semi-supervised VAE. To this end, we introduced", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "OSPOT-VAE, a one-stage generative model that unified the classification and generation objec-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "tive and achieved a tighter ELBO by optimal transport estimation. We demonstrated our asser-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "tion through extensive experiments, and our semi-supervised results significantly outperform former", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 720, + 442, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 442, + 733 + ], + "score": 1.0, + "content": "state-of-the-art generative SSL methods by a large margin on Cifar10 and Cifar100.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 666, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 176, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 100, + 411, + 113 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 412, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 412, + 114 + ], + "score": 1.0, + "content": "Luigi Ambrosio and Nicola Gigli. A users guide to optimal transport. 2013.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 109, + 121, + 504, + 155 + ], + "lines": [ + { + "bbox": [ + 107, + 121, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 107, + 121, + 505, + 133 + ], + "score": 1.0, + "content": "David Berthelot, Nicholas Carlini, Ian J. Goodfellow, Nicolas Papernot, Avital Oliver, and Colin", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 132, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 116, + 132, + 506, + 145 + ], + "score": 1.0, + "content": "Raffel. Mixmatch: A holistic approach to semi-supervised learning. CoRR, abs/1905.02249,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 143, + 356, + 155 + ], + "spans": [ + { + "bbox": [ + 116, + 143, + 356, + 155 + ], + "score": 1.0, + "content": "2019. URL http://arxiv.org/abs/1905.02249.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 163, + 505, + 208 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 176 + ], + "score": 1.0, + "content": "Grigorios G. Chrysos, Jean Kossaifi, and Stefanos Zafeiriou. Robust conditional generative adver-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 116, + 175, + 505, + 187 + ], + "score": 1.0, + "content": "sarial networks. In 7th International Conference on Learning Representations, ICLR 2019, New", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 115, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "Orleans, LA, USA, May 6-9, 2019, 2019. URL https://openreview.net/forum?id=", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 197, + 181, + 210 + ], + "spans": [ + { + "bbox": [ + 115, + 197, + 181, + 210 + ], + "score": 1.0, + "content": "Byg0DsCqYQ.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 217, + 506, + 250 + ], + "lines": [ + { + "bbox": [ + 106, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "Ekin D. Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasudevan, and Quoc V. Le. Autoaugment:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 228, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 116, + 228, + 505, + 240 + ], + "score": 1.0, + "content": "Learning augmentation strategies from data. In The IEEE Conference on Computer Vision and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 239, + 281, + 252 + ], + "spans": [ + { + "bbox": [ + 116, + 239, + 281, + 252 + ], + "score": 1.0, + "content": "Pattern Recognition (CVPR), June 2019.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 259, + 505, + 326 + ], + "lines": [ + { + "bbox": [ + 105, + 258, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 506, + 272 + ], + "score": 1.0, + "content": "Zihang Dai, Zhilin Yang, Fan Yang, William W. Cohen, and Ruslan Salakhutdi-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 269, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 115, + 271, + 137, + 282 + ], + "score": 1.0, + "content": "nov.", + "type": "text" + }, + { + "bbox": [ + 154, + 269, + 429, + 283 + ], + "score": 1.0, + "content": "Good semi-supervised learning that requires a bad GAN.", + "type": "text" + }, + { + "bbox": [ + 445, + 269, + 506, + 283 + ], + "score": 1.0, + "content": "In Advances", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 280, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 114, + 281, + 282, + 295 + ], + "score": 1.0, + "content": "in Neural Information Processing", + "type": "text" + }, + { + "bbox": [ + 283, + 280, + 506, + 295 + ], + "score": 1.0, + "content": "Systems 30: Annual Conference on Neu-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 113, + 290, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 113, + 290, + 505, + 306 + ], + "score": 1.0, + "content": "ral Information Processing Systems 2017, 4-9 December 2017, Long Beach,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 301, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 115, + 302, + 281, + 316 + ], + "score": 1.0, + "content": "CA, USA, pp. 6510–6520, 2017.", + "type": "text" + }, + { + "bbox": [ + 302, + 301, + 505, + 317 + ], + "score": 1.0, + "content": "URL http://papers.nips.cc/paper/", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 313, + 473, + 328 + ], + "spans": [ + { + "bbox": [ + 115, + 313, + 473, + 328 + ], + "score": 1.0, + "content": "7229-good-semi-supervised-learning-that-requires-a-bad-gan.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 107, + 335, + 502, + 358 + ], + "lines": [ + { + "bbox": [ + 105, + 334, + 504, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 504, + 348 + ], + "score": 1.0, + "content": "Carl Doersch. Tutorial on variational autoencoders. CoRR, abs/1606.05908, 2016. URL http:", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 117, + 347, + 276, + 358 + ], + "spans": [ + { + "bbox": [ + 117, + 347, + 276, + 358 + ], + "score": 1.0, + "content": "//arxiv.org/abs/1606.05908.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 366, + 505, + 423 + ], + "lines": [ + { + "bbox": [ + 106, + 367, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 505, + 379 + ], + "score": 1.0, + "content": "Irina Higgins, Lo¨ıc Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 377, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 115, + 377, + 506, + 390 + ], + "score": 1.0, + "content": "Shakir Mohamed, and Alexander Lerchner. beta-vae: Learning basic visual concepts with a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 388, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 115, + 388, + 505, + 402 + ], + "score": 1.0, + "content": "constrained variational framework. In 5th International Conference on Learning Representa-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 400, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 115, + 400, + 505, + 412 + ], + "score": 1.0, + "content": "tions, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017. URL", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 410, + 366, + 424 + ], + "spans": [ + { + "bbox": [ + 115, + 410, + 302, + 424 + ], + "score": 1.0, + "content": "https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 302, + 412, + 309, + 420 + ], + "score": 0.56, + "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 410, + 366, + 424 + ], + "score": 1.0, + "content": "Sy2fzU9gl.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 431, + 505, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 431, + 504, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 504, + 443 + ], + "score": 1.0, + "content": "Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, and Ondrej Chum. Label propagation for deep semi-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 442, + 504, + 454 + ], + "spans": [ + { + "bbox": [ + 116, + 442, + 504, + 454 + ], + "score": 1.0, + "content": "supervised learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 453, + 259, + 465 + ], + "spans": [ + { + "bbox": [ + 116, + 453, + 259, + 465 + ], + "score": 1.0, + "content": "Recognition, pp. 5070–5079, 2019.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 505, + 518 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 505, + 487 + ], + "score": 1.0, + "content": "Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. In", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 116, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 115, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "26, 2017, Conference Track Proceedings, 2017. URL https://openreview.net/forum?", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 506, + 193, + 519 + ], + "spans": [ + { + "bbox": [ + 116, + 506, + 129, + 519 + ], + "score": 1.0, + "content": "id", + "type": "text" + }, + { + "bbox": [ + 129, + 508, + 136, + 516 + ], + "score": 0.39, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 506, + 193, + 519 + ], + "score": 1.0, + "content": "rkE3y85ee.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 527, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 106, + 527, + 504, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 504, + 539 + ], + "score": 1.0, + "content": "Diederik P. Kingma, Shakir Mohamed, Danilo Jimenez Rezende, and Max Welling. Semi-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 537, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 115, + 537, + 506, + 552 + ], + "score": 1.0, + "content": "supervised learning with deep generative models. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 116, + 549, + 505, + 562 + ], + "score": 1.0, + "content": "Systems 27: Annual Conference on Neural Information Processing Systems 2014, December 8-13", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 559, + 504, + 573 + ], + "spans": [ + { + "bbox": [ + 115, + 559, + 504, + 573 + ], + "score": 1.0, + "content": "2014, Montreal, Quebec, Canada, pp. 3581–3589, 2014. URL http://papers.nips.cc/", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 114, + 571, + 502, + 584 + ], + "spans": [ + { + "bbox": [ + 114, + 571, + 502, + 584 + ], + "score": 1.0, + "content": "paper/5352-semi-supervised-learning-with-deep-generative-models.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 505, + 625 + ], + "lines": [ + { + "bbox": [ + 105, + 591, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 505, + 604 + ], + "score": 1.0, + "content": "Max Kuang and Esteban G. Tabak. Preconditioning of optimal transport. SIAM J. Scientific", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 603, + 504, + 615 + ], + "spans": [ + { + "bbox": [ + 116, + 603, + 504, + 615 + ], + "score": 1.0, + "content": "Computing, 39(4), 2017. doi: 10.1137/16M1074953. URL https://doi.org/10.1137/", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 614, + 181, + 624 + ], + "spans": [ + { + "bbox": [ + 115, + 614, + 181, + 624 + ], + "score": 1.0, + "content": "16M1074953.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 106, + 634, + 505, + 678 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 647 + ], + "score": 1.0, + "content": "Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. In 5th Inter-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 645, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 115, + 645, + 505, + 657 + ], + "score": 1.0, + "content": "national Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 655, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 115, + 655, + 505, + 669 + ], + "score": 1.0, + "content": "2017, Conference Track Proceedings, 2017. URL https://openreview.net/forum?", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 666, + 194, + 681 + ], + "spans": [ + { + "bbox": [ + 115, + 666, + 129, + 681 + ], + "score": 1.0, + "content": "id", + "type": "text" + }, + { + "bbox": [ + 129, + 669, + 136, + 677 + ], + "score": 0.39, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 666, + 194, + 681 + ], + "score": 1.0, + "content": "BJ6oOfqge.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "Wenyuan Li, Zichen Wang, Jiayun Li, Jennifer Polson, William Speier, and Corey W. Arnold. Semi-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 115, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "supervised learning based on generative adversarial network: a comparison between good GAN", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "and bad GAN approach. CoRR, abs/1905.06484, 2019. URL http://arxiv.org/abs/", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 721, + 182, + 731 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 182, + 731 + ], + "score": 1.0, + "content": "1905.06484.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 176, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 100, + 411, + 113 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 412, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 412, + 114 + ], + "score": 1.0, + "content": "Luigi Ambrosio and Nicola Gigli. A users guide to optimal transport. 2013.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 100, + 412, + 114 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 121, + 504, + 155 + ], + "lines": [ + { + "bbox": [ + 107, + 121, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 107, + 121, + 505, + 133 + ], + "score": 1.0, + "content": "David Berthelot, Nicholas Carlini, Ian J. Goodfellow, Nicolas Papernot, Avital Oliver, and Colin", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 132, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 116, + 132, + 506, + 145 + ], + "score": 1.0, + "content": "Raffel. Mixmatch: A holistic approach to semi-supervised learning. CoRR, abs/1905.02249,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 143, + 356, + 155 + ], + "spans": [ + { + "bbox": [ + 116, + 143, + 356, + 155 + ], + "score": 1.0, + "content": "2019. URL http://arxiv.org/abs/1905.02249.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 107, + 121, + 506, + 155 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 163, + 505, + 208 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 176 + ], + "score": 1.0, + "content": "Grigorios G. Chrysos, Jean Kossaifi, and Stefanos Zafeiriou. Robust conditional generative adver-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 116, + 175, + 505, + 187 + ], + "score": 1.0, + "content": "sarial networks. In 7th International Conference on Learning Representations, ICLR 2019, New", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 115, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "Orleans, LA, USA, May 6-9, 2019, 2019. URL https://openreview.net/forum?id=", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 197, + 181, + 210 + ], + "spans": [ + { + "bbox": [ + 115, + 197, + 181, + 210 + ], + "score": 1.0, + "content": "Byg0DsCqYQ.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 164, + 505, + 210 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 217, + 506, + 250 + ], + "lines": [ + { + "bbox": [ + 106, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "Ekin D. Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasudevan, and Quoc V. Le. Autoaugment:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 228, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 116, + 228, + 505, + 240 + ], + "score": 1.0, + "content": "Learning augmentation strategies from data. In The IEEE Conference on Computer Vision and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 239, + 281, + 252 + ], + "spans": [ + { + "bbox": [ + 116, + 239, + 281, + 252 + ], + "score": 1.0, + "content": "Pattern Recognition (CVPR), June 2019.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 217, + 505, + 252 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 259, + 505, + 326 + ], + "lines": [ + { + "bbox": [ + 105, + 258, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 506, + 272 + ], + "score": 1.0, + "content": "Zihang Dai, Zhilin Yang, Fan Yang, William W. Cohen, and Ruslan Salakhutdi-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 269, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 115, + 271, + 137, + 282 + ], + "score": 1.0, + "content": "nov.", + "type": "text" + }, + { + "bbox": [ + 154, + 269, + 429, + 283 + ], + "score": 1.0, + "content": "Good semi-supervised learning that requires a bad GAN.", + "type": "text" + }, + { + "bbox": [ + 445, + 269, + 506, + 283 + ], + "score": 1.0, + "content": "In Advances", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 280, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 114, + 281, + 282, + 295 + ], + "score": 1.0, + "content": "in Neural Information Processing", + "type": "text" + }, + { + "bbox": [ + 283, + 280, + 506, + 295 + ], + "score": 1.0, + "content": "Systems 30: Annual Conference on Neu-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 113, + 290, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 113, + 290, + 505, + 306 + ], + "score": 1.0, + "content": "ral Information Processing Systems 2017, 4-9 December 2017, Long Beach,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 301, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 115, + 302, + 281, + 316 + ], + "score": 1.0, + "content": "CA, USA, pp. 6510–6520, 2017.", + "type": "text" + }, + { + "bbox": [ + 302, + 301, + 505, + 317 + ], + "score": 1.0, + "content": "URL http://papers.nips.cc/paper/", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 313, + 473, + 328 + ], + "spans": [ + { + "bbox": [ + 115, + 313, + 473, + 328 + ], + "score": 1.0, + "content": "7229-good-semi-supervised-learning-that-requires-a-bad-gan.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 258, + 506, + 328 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 335, + 502, + 358 + ], + "lines": [ + { + "bbox": [ + 105, + 334, + 504, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 504, + 348 + ], + "score": 1.0, + "content": "Carl Doersch. Tutorial on variational autoencoders. CoRR, abs/1606.05908, 2016. URL http:", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 117, + 347, + 276, + 358 + ], + "spans": [ + { + "bbox": [ + 117, + 347, + 276, + 358 + ], + "score": 1.0, + "content": "//arxiv.org/abs/1606.05908.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 334, + 504, + 358 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 366, + 505, + 423 + ], + "lines": [ + { + "bbox": [ + 106, + 367, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 505, + 379 + ], + "score": 1.0, + "content": "Irina Higgins, Lo¨ıc Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 377, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 115, + 377, + 506, + 390 + ], + "score": 1.0, + "content": "Shakir Mohamed, and Alexander Lerchner. beta-vae: Learning basic visual concepts with a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 388, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 115, + 388, + 505, + 402 + ], + "score": 1.0, + "content": "constrained variational framework. In 5th International Conference on Learning Representa-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 400, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 115, + 400, + 505, + 412 + ], + "score": 1.0, + "content": "tions, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017. URL", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 410, + 366, + 424 + ], + "spans": [ + { + "bbox": [ + 115, + 410, + 302, + 424 + ], + "score": 1.0, + "content": "https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 302, + 412, + 309, + 420 + ], + "score": 0.56, + "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 410, + 366, + 424 + ], + "score": 1.0, + "content": "Sy2fzU9gl.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 367, + 506, + 424 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 431, + 505, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 431, + 504, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 504, + 443 + ], + "score": 1.0, + "content": "Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, and Ondrej Chum. Label propagation for deep semi-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 442, + 504, + 454 + ], + "spans": [ + { + "bbox": [ + 116, + 442, + 504, + 454 + ], + "score": 1.0, + "content": "supervised learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 453, + 259, + 465 + ], + "spans": [ + { + "bbox": [ + 116, + 453, + 259, + 465 + ], + "score": 1.0, + "content": "Recognition, pp. 5070–5079, 2019.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 431, + 504, + 465 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 505, + 518 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 505, + 487 + ], + "score": 1.0, + "content": "Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. In", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 116, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 115, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "26, 2017, Conference Track Proceedings, 2017. URL https://openreview.net/forum?", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 506, + 193, + 519 + ], + "spans": [ + { + "bbox": [ + 116, + 506, + 129, + 519 + ], + "score": 1.0, + "content": "id", + "type": "text" + }, + { + "bbox": [ + 129, + 508, + 136, + 516 + ], + "score": 0.39, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 506, + 193, + 519 + ], + "score": 1.0, + "content": "rkE3y85ee.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 473, + 505, + 519 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 527, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 106, + 527, + 504, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 504, + 539 + ], + "score": 1.0, + "content": "Diederik P. Kingma, Shakir Mohamed, Danilo Jimenez Rezende, and Max Welling. Semi-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 537, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 115, + 537, + 506, + 552 + ], + "score": 1.0, + "content": "supervised learning with deep generative models. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 116, + 549, + 505, + 562 + ], + "score": 1.0, + "content": "Systems 27: Annual Conference on Neural Information Processing Systems 2014, December 8-13", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 559, + 504, + 573 + ], + "spans": [ + { + "bbox": [ + 115, + 559, + 504, + 573 + ], + "score": 1.0, + "content": "2014, Montreal, Quebec, Canada, pp. 3581–3589, 2014. URL http://papers.nips.cc/", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 114, + 571, + 502, + 584 + ], + "spans": [ + { + "bbox": [ + 114, + 571, + 502, + 584 + ], + "score": 1.0, + "content": "paper/5352-semi-supervised-learning-with-deep-generative-models.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34, + "bbox_fs": [ + 106, + 527, + 506, + 584 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 505, + 625 + ], + "lines": [ + { + "bbox": [ + 105, + 591, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 505, + 604 + ], + "score": 1.0, + "content": "Max Kuang and Esteban G. Tabak. Preconditioning of optimal transport. SIAM J. Scientific", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 603, + 504, + 615 + ], + "spans": [ + { + "bbox": [ + 116, + 603, + 504, + 615 + ], + "score": 1.0, + "content": "Computing, 39(4), 2017. doi: 10.1137/16M1074953. URL https://doi.org/10.1137/", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 614, + 181, + 624 + ], + "spans": [ + { + "bbox": [ + 115, + 614, + 181, + 624 + ], + "score": 1.0, + "content": "16M1074953.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 591, + 505, + 624 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 634, + 505, + 678 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 647 + ], + "score": 1.0, + "content": "Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. In 5th Inter-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 645, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 115, + 645, + 505, + 657 + ], + "score": 1.0, + "content": "national Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 655, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 115, + 655, + 505, + 669 + ], + "score": 1.0, + "content": "2017, Conference Track Proceedings, 2017. URL https://openreview.net/forum?", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 666, + 194, + 681 + ], + "spans": [ + { + "bbox": [ + 115, + 666, + 129, + 681 + ], + "score": 1.0, + "content": "id", + "type": "text" + }, + { + "bbox": [ + 129, + 669, + 136, + 677 + ], + "score": 0.39, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 666, + 194, + 681 + ], + "score": 1.0, + "content": "BJ6oOfqge.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 633, + 505, + 681 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "Wenyuan Li, Zichen Wang, Jiayun Li, Jennifer Polson, William Speier, and Corey W. Arnold. Semi-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 115, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "supervised learning based on generative adversarial network: a comparison between good GAN", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "and bad GAN approach. CoRR, abs/1905.06484, 2019. URL http://arxiv.org/abs/", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 721, + 182, + 731 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 182, + 731 + ], + "score": 1.0, + "content": "1905.06484.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5, + "bbox_fs": [ + 106, + 687, + 505, + 731 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 107, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 107, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "Yucen Luo, Jun Zhu, Mengxi Li, Yong Ren, and Bo Zhang. Smooth neighbors on teacher graphs for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 107 + ], + "score": 1.0, + "content": "semi-supervised learning. In 2018 IEEE Conference on Computer Vision and Pattern Recognition,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 116, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018, pp. 8896–8905, 2018. doi: 10.1109/", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 116, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "CVPR.2018.00927. URL http://openaccess.thecvf.com/content_cvpr_2018/", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 126, + 414, + 138 + ], + "spans": [ + { + "bbox": [ + 116, + 126, + 414, + 138 + ], + "score": 1.0, + "content": "html/Luo_Smooth_Neighbors_on_CVPR_2018_paper.html.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 145, + 505, + 190 + ], + "lines": [ + { + "bbox": [ + 106, + 145, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 505, + 157 + ], + "score": 1.0, + "content": "Takeru Miyato, Shin-ichi Maeda, Masanori Koyama, and Shin Ishii. Virtual adversarial training: A", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 156, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 115, + 156, + 505, + 168 + ], + "score": 1.0, + "content": "regularization method for supervised and semi-supervised learning. IEEE Trans. Pattern Anal.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 167, + 504, + 180 + ], + "spans": [ + { + "bbox": [ + 115, + 167, + 504, + 180 + ], + "score": 1.0, + "content": "Mach. Intell., 41(8):1979–1993, 2019. doi: 10.1109/TPAMI.2018.2858821. URL https://", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 179, + 325, + 190 + ], + "spans": [ + { + "bbox": [ + 116, + 179, + 325, + 190 + ], + "score": 1.0, + "content": "doi.org/10.1109/TPAMI.2018.2858821.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 197, + 505, + 253 + ], + "lines": [ + { + "bbox": [ + 106, + 198, + 504, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 504, + 210 + ], + "score": 1.0, + "content": "Siddharth Narayanaswamy, Brooks Paige, Jan-Willem van de Meent, Alban Desmaison, Noah D.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 207, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 115, + 207, + 505, + 223 + ], + "score": 1.0, + "content": "Goodman, Pushmeet Kohli, Frank D. Wood, and Philip H. S. Torr. Learning disentangled rep-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 220, + 506, + 232 + ], + "spans": [ + { + "bbox": [ + 116, + 220, + 506, + 232 + ], + "score": 1.0, + "content": "resentations with semi-supervised deep generative models. In Advances in Neural Information", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 230, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 116, + 230, + 506, + 243 + ], + "score": 1.0, + "content": "Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, 4-9", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 241, + 368, + 253 + ], + "spans": [ + { + "bbox": [ + 116, + 241, + 368, + 253 + ], + "score": 1.0, + "content": "December 2017, Long Beach, CA, USA, pp. 5925–5935, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 505, + 305 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 506, + 273 + ], + "score": 1.0, + "content": "Avital Oliver, Augustus Odena, Colin Raffel, Ekin D. Cubuk, and Ian J. Goodfellow. Realistic", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 270, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 115, + 270, + 505, + 285 + ], + "score": 1.0, + "content": "evaluation of semi-supervised learning algorithms. In 6th International Conference on Learning", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 283, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 116, + 283, + 505, + 295 + ], + "score": 1.0, + "content": "Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Workshop Track", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 294, + 469, + 307 + ], + "spans": [ + { + "bbox": [ + 116, + 294, + 405, + 307 + ], + "score": 1.0, + "content": "Proceedings, 2018. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 405, + 295, + 412, + 303 + ], + "score": 0.56, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 294, + 469, + 307 + ], + "score": 1.0, + "content": "ByCZsFyPf.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 312, + 505, + 357 + ], + "lines": [ + { + "bbox": [ + 106, + 313, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 324 + ], + "score": 1.0, + "content": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 324, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 115, + 324, + 505, + 336 + ], + "score": 1.0, + "content": "approximate inference in deep generative models. In Proceedings of the 31th International Con-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 114, + 335, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 114, + 335, + 505, + 347 + ], + "score": 1.0, + "content": "ference on Machine Learning, ICML 2014, Beijing, China, 21-26 June 2014, pp. 1278–1286,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 345, + 451, + 358 + ], + "spans": [ + { + "bbox": [ + 115, + 345, + 451, + 358 + ], + "score": 1.0, + "content": "2014. URL http://proceedings.mlr.press/v32/rezende14.html.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 364, + 505, + 420 + ], + "lines": [ + { + "bbox": [ + 106, + 364, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 377 + ], + "score": 1.0, + "content": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, Xi Chen, and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 373, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 115, + 373, + 506, + 390 + ], + "score": 1.0, + "content": "Xi Chen. Improved techniques for training gans. In D. D. Lee, M. Sugiyama, U. V. Luxburg,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 114, + 384, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 114, + 384, + 505, + 401 + ], + "score": 1.0, + "content": "I. Guyon, and R. Garnett (eds.), Advances in Neural Information Processing Systems 29, pp.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 505, + 410 + ], + "score": 1.0, + "content": "2234–2242. Curran Associates, Inc., 2016. URL http://papers.nips.cc/paper/", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 408, + 399, + 421 + ], + "spans": [ + { + "bbox": [ + 116, + 408, + 399, + 421 + ], + "score": 1.0, + "content": "6125-improved-techniques-for-training-gans.pdf.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 427, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 105, + 427, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 505, + 440 + ], + "score": 1.0, + "content": "Jost Tobias Springenberg. Unsupervised and semi-supervised learning with categorical generative", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 439, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 116, + 439, + 504, + 451 + ], + "score": 1.0, + "content": "adversarial networks. In 4th International Conference on Learning Representations, ICLR 2016,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 448, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 115, + 448, + 505, + 463 + ], + "score": 1.0, + "content": "San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings, 2016. URL http:", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 117, + 461, + 278, + 473 + ], + "spans": [ + { + "bbox": [ + 117, + 461, + 278, + 473 + ], + "score": 1.0, + "content": "//arxiv.org/abs/1511.06390.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 479, + 505, + 524 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 492 + ], + "score": 1.0, + "content": "Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged con-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 491, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 115, + 491, + 505, + 502 + ], + "score": 1.0, + "content": "sistency targets improve semi-supervised deep learning results. In 5th International Conference", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 502, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 116, + 502, + 506, + 514 + ], + "score": 1.0, + "content": "on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Workshop Track", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 512, + 470, + 525 + ], + "spans": [ + { + "bbox": [ + 115, + 512, + 405, + 525 + ], + "score": 1.0, + "content": "Proceedings, 2017. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 405, + 514, + 412, + 522 + ], + "score": 0.53, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 512, + 470, + 525 + ], + "score": 1.0, + "content": "ry8u21rtl.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "score": 1.0, + "content": "Philippe Thomas. Semi-supervised learning by olivier chapelle, bernhard scholkopf, and alexander ¨", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 115, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "zien (review). IEEE Trans. Neural Networks, 20(3):542, 2009. doi: 10.1109/TNN.2009.2015974.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 552, + 384, + 566 + ], + "spans": [ + { + "bbox": [ + 115, + 552, + 384, + 566 + ], + "score": 1.0, + "content": "URL https://doi.org/10.1109/TNN.2009.2015974.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 106, + 573, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 505, + 585 + ], + "score": 1.0, + "content": "Vikas Verma, Alex Lamb, Christopher Beckham, Aaron C. Courville, Ioannis Mitliagkas, and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 117, + 583, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 117, + 583, + 505, + 596 + ], + "score": 1.0, + "content": "Yoshua Bengio. Manifold mixup: Encouraging meaningful on-manifold interpolation as a regu-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 595, + 484, + 607 + ], + "spans": [ + { + "bbox": [ + 116, + 595, + 484, + 607 + ], + "score": 1.0, + "content": "larizer. CoRR, abs/1806.05236, 2018. URL http://arxiv.org/abs/1806.05236.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 505, + 669 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "Vikas Verma, Alex Lamb, Juho Kannala, Yoshua Bengio, and David Lopez-Paz. Interpolation", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 624, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 115, + 624, + 505, + 638 + ], + "score": 1.0, + "content": "consistency training for semi-supervised learning. In Proceedings of the Twenty-Eighth Inter-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 636, + 504, + 648 + ], + "spans": [ + { + "bbox": [ + 115, + 636, + 504, + 648 + ], + "score": 1.0, + "content": "national Joint Conference on Artificial Intelligence, IJCAI 2019, Macao, China, August 10-16,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 645, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 645, + 504, + 660 + ], + "score": 1.0, + "content": "2019, pp. 3635–3641, 2019. doi: 10.24963/ijcai.2019/504. URL https://doi.org/10.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 657, + 242, + 671 + ], + "spans": [ + { + "bbox": [ + 115, + 657, + 242, + 671 + ], + "score": 1.0, + "content": "24963/ijcai.2019/504.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 676, + 504, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 504, + 689 + ], + "score": 1.0, + "content": "Xiang Wei, Boqing Gong, Zixia Liu, Wei Lu, and Liqiang Wang. Improving the improved train-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 115, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "ing of wasserstein gans: A consistency term and its dual effect. In 6th International Con-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 114, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 114, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "ference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 114, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 114, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "2018, Conference Track Proceedings, 2018. URL https://openreview.net/forum?", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 115, + 720, + 193, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 193, + 733 + ], + "score": 1.0, + "content": "id=SJx9GQb0-.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 107, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 107, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "Yucen Luo, Jun Zhu, Mengxi Li, Yong Ren, and Bo Zhang. Smooth neighbors on teacher graphs for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 107 + ], + "score": 1.0, + "content": "semi-supervised learning. In 2018 IEEE Conference on Computer Vision and Pattern Recognition,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 116, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018, pp. 8896–8905, 2018. doi: 10.1109/", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 116, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "CVPR.2018.00927. URL http://openaccess.thecvf.com/content_cvpr_2018/", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 126, + 414, + 138 + ], + "spans": [ + { + "bbox": [ + 116, + 126, + 414, + 138 + ], + "score": 1.0, + "content": "html/Luo_Smooth_Neighbors_on_CVPR_2018_paper.html.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 107, + 83, + 505, + 138 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 145, + 505, + 190 + ], + "lines": [ + { + "bbox": [ + 106, + 145, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 505, + 157 + ], + "score": 1.0, + "content": "Takeru Miyato, Shin-ichi Maeda, Masanori Koyama, and Shin Ishii. Virtual adversarial training: A", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 156, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 115, + 156, + 505, + 168 + ], + "score": 1.0, + "content": "regularization method for supervised and semi-supervised learning. IEEE Trans. Pattern Anal.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 167, + 504, + 180 + ], + "spans": [ + { + "bbox": [ + 115, + 167, + 504, + 180 + ], + "score": 1.0, + "content": "Mach. Intell., 41(8):1979–1993, 2019. doi: 10.1109/TPAMI.2018.2858821. URL https://", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 179, + 325, + 190 + ], + "spans": [ + { + "bbox": [ + 116, + 179, + 325, + 190 + ], + "score": 1.0, + "content": "doi.org/10.1109/TPAMI.2018.2858821.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 145, + 505, + 190 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 197, + 505, + 253 + ], + "lines": [ + { + "bbox": [ + 106, + 198, + 504, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 504, + 210 + ], + "score": 1.0, + "content": "Siddharth Narayanaswamy, Brooks Paige, Jan-Willem van de Meent, Alban Desmaison, Noah D.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 207, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 115, + 207, + 505, + 223 + ], + "score": 1.0, + "content": "Goodman, Pushmeet Kohli, Frank D. Wood, and Philip H. S. Torr. Learning disentangled rep-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 220, + 506, + 232 + ], + "spans": [ + { + "bbox": [ + 116, + 220, + 506, + 232 + ], + "score": 1.0, + "content": "resentations with semi-supervised deep generative models. In Advances in Neural Information", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 230, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 116, + 230, + 506, + 243 + ], + "score": 1.0, + "content": "Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, 4-9", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 241, + 368, + 253 + ], + "spans": [ + { + "bbox": [ + 116, + 241, + 368, + 253 + ], + "score": 1.0, + "content": "December 2017, Long Beach, CA, USA, pp. 5925–5935, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 198, + 506, + 253 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 505, + 305 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 506, + 273 + ], + "score": 1.0, + "content": "Avital Oliver, Augustus Odena, Colin Raffel, Ekin D. Cubuk, and Ian J. Goodfellow. Realistic", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 270, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 115, + 270, + 505, + 285 + ], + "score": 1.0, + "content": "evaluation of semi-supervised learning algorithms. In 6th International Conference on Learning", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 283, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 116, + 283, + 505, + 295 + ], + "score": 1.0, + "content": "Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Workshop Track", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 294, + 469, + 307 + ], + "spans": [ + { + "bbox": [ + 116, + 294, + 405, + 307 + ], + "score": 1.0, + "content": "Proceedings, 2018. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 405, + 295, + 412, + 303 + ], + "score": 0.56, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 294, + 469, + 307 + ], + "score": 1.0, + "content": "ByCZsFyPf.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 259, + 506, + 307 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 312, + 505, + 357 + ], + "lines": [ + { + "bbox": [ + 106, + 313, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 324 + ], + "score": 1.0, + "content": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 324, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 115, + 324, + 505, + 336 + ], + "score": 1.0, + "content": "approximate inference in deep generative models. In Proceedings of the 31th International Con-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 114, + 335, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 114, + 335, + 505, + 347 + ], + "score": 1.0, + "content": "ference on Machine Learning, ICML 2014, Beijing, China, 21-26 June 2014, pp. 1278–1286,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 345, + 451, + 358 + ], + "spans": [ + { + "bbox": [ + 115, + 345, + 451, + 358 + ], + "score": 1.0, + "content": "2014. URL http://proceedings.mlr.press/v32/rezende14.html.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 313, + 505, + 358 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 364, + 505, + 420 + ], + "lines": [ + { + "bbox": [ + 106, + 364, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 377 + ], + "score": 1.0, + "content": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, Xi Chen, and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 373, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 115, + 373, + 506, + 390 + ], + "score": 1.0, + "content": "Xi Chen. Improved techniques for training gans. In D. D. Lee, M. Sugiyama, U. V. Luxburg,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 114, + 384, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 114, + 384, + 505, + 401 + ], + "score": 1.0, + "content": "I. Guyon, and R. Garnett (eds.), Advances in Neural Information Processing Systems 29, pp.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 505, + 410 + ], + "score": 1.0, + "content": "2234–2242. Curran Associates, Inc., 2016. URL http://papers.nips.cc/paper/", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 408, + 399, + 421 + ], + "spans": [ + { + "bbox": [ + 116, + 408, + 399, + 421 + ], + "score": 1.0, + "content": "6125-improved-techniques-for-training-gans.pdf.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 364, + 506, + 421 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 427, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 105, + 427, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 505, + 440 + ], + "score": 1.0, + "content": "Jost Tobias Springenberg. Unsupervised and semi-supervised learning with categorical generative", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 439, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 116, + 439, + 504, + 451 + ], + "score": 1.0, + "content": "adversarial networks. In 4th International Conference on Learning Representations, ICLR 2016,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 448, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 115, + 448, + 505, + 463 + ], + "score": 1.0, + "content": "San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings, 2016. URL http:", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 117, + 461, + 278, + 473 + ], + "spans": [ + { + "bbox": [ + 117, + 461, + 278, + 473 + ], + "score": 1.0, + "content": "//arxiv.org/abs/1511.06390.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 427, + 505, + 473 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 479, + 505, + 524 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 492 + ], + "score": 1.0, + "content": "Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged con-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 491, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 115, + 491, + 505, + 502 + ], + "score": 1.0, + "content": "sistency targets improve semi-supervised deep learning results. In 5th International Conference", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 502, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 116, + 502, + 506, + 514 + ], + "score": 1.0, + "content": "on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Workshop Track", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 512, + 470, + 525 + ], + "spans": [ + { + "bbox": [ + 115, + 512, + 405, + 525 + ], + "score": 1.0, + "content": "Proceedings, 2017. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 405, + 514, + 412, + 522 + ], + "score": 0.53, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 512, + 470, + 525 + ], + "score": 1.0, + "content": "ry8u21rtl.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 479, + 506, + 525 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "score": 1.0, + "content": "Philippe Thomas. Semi-supervised learning by olivier chapelle, bernhard scholkopf, and alexander ¨", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 115, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "zien (review). IEEE Trans. Neural Networks, 20(3):542, 2009. doi: 10.1109/TNN.2009.2015974.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 552, + 384, + 566 + ], + "spans": [ + { + "bbox": [ + 115, + 552, + 384, + 566 + ], + "score": 1.0, + "content": "URL https://doi.org/10.1109/TNN.2009.2015974.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 531, + 506, + 566 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 106, + 573, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 505, + 585 + ], + "score": 1.0, + "content": "Vikas Verma, Alex Lamb, Christopher Beckham, Aaron C. Courville, Ioannis Mitliagkas, and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 117, + 583, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 117, + 583, + 505, + 596 + ], + "score": 1.0, + "content": "Yoshua Bengio. Manifold mixup: Encouraging meaningful on-manifold interpolation as a regu-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 595, + 484, + 607 + ], + "spans": [ + { + "bbox": [ + 116, + 595, + 484, + 607 + ], + "score": 1.0, + "content": "larizer. CoRR, abs/1806.05236, 2018. URL http://arxiv.org/abs/1806.05236.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 106, + 573, + 505, + 607 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 505, + 669 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "Vikas Verma, Alex Lamb, Juho Kannala, Yoshua Bengio, and David Lopez-Paz. Interpolation", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 624, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 115, + 624, + 505, + 638 + ], + "score": 1.0, + "content": "consistency training for semi-supervised learning. In Proceedings of the Twenty-Eighth Inter-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 636, + 504, + 648 + ], + "spans": [ + { + "bbox": [ + 115, + 636, + 504, + 648 + ], + "score": 1.0, + "content": "national Joint Conference on Artificial Intelligence, IJCAI 2019, Macao, China, August 10-16,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 645, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 645, + 504, + 660 + ], + "score": 1.0, + "content": "2019, pp. 3635–3641, 2019. doi: 10.24963/ijcai.2019/504. URL https://doi.org/10.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 657, + 242, + 671 + ], + "spans": [ + { + "bbox": [ + 115, + 657, + 242, + 671 + ], + "score": 1.0, + "content": "24963/ijcai.2019/504.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43, + "bbox_fs": [ + 106, + 613, + 505, + 671 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 676, + 504, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 504, + 689 + ], + "score": 1.0, + "content": "Xiang Wei, Boqing Gong, Zixia Liu, Wei Lu, and Liqiang Wang. Improving the improved train-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 115, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "ing of wasserstein gans: A consistency term and its dual effect. In 6th International Con-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 114, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 114, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "ference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 114, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 114, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "2018, Conference Track Proceedings, 2018. URL https://openreview.net/forum?", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 115, + 720, + 193, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 193, + 733 + ], + "score": 1.0, + "content": "id=SJx9GQb0-.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48, + "bbox_fs": [ + 106, + 676, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 404, + 95 + ], + "score": 1.0, + "content": "Zhirong Wu, Yuanjun Xiong, Stella X. Yu, and Dahua Lin.", + "type": "text" + }, + { + "bbox": [ + 423, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Unsupervised fea-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 376, + 106 + ], + "score": 1.0, + "content": "ture learning via non-parametric instance discrimination.", + "type": "text" + }, + { + "bbox": [ + 394, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "In 2018 IEEE Confer-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 102, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 102, + 505, + 117 + ], + "score": 1.0, + "content": "ence on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 115, + 114, + 367, + 128 + ], + "score": 1.0, + "content": "UT, USA, June 18-22, 2018, pp. 3733–3742, 2018.", + "type": "text" + }, + { + "bbox": [ + 384, + 114, + 505, + 127 + ], + "score": 1.0, + "content": "doi: 10.1109/CVPR.2018.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 125, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 115, + 125, + 506, + 140 + ], + "score": 1.0, + "content": "00393. URL http://openaccess.thecvf.com/content_cvpr_2018/html/Wu_", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 117, + 137, + 421, + 149 + ], + "spans": [ + { + "bbox": [ + 117, + 137, + 421, + 149 + ], + "score": 1.0, + "content": "Unsupervised_Feature_Learning_CVPR_2018_paper.html.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 155, + 503, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 155, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 505, + 167 + ], + "score": 1.0, + "content": "Qizhe Xie, Zihang Dai, Eduard H. Hovy, Minh-Thang Luong, and Quoc V. Le. Unsupervised", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 166, + 504, + 178 + ], + "spans": [ + { + "bbox": [ + 116, + 166, + 504, + 178 + ], + "score": 1.0, + "content": "data augmentation. CoRR, abs/1904.12848, 2019. URL http://arxiv.org/abs/1904.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 177, + 152, + 189 + ], + "spans": [ + { + "bbox": [ + 116, + 177, + 152, + 189 + ], + "score": 1.0, + "content": "12848.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 194, + 502, + 217 + ], + "lines": [ + { + "bbox": [ + 105, + 194, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 505, + 207 + ], + "score": 1.0, + "content": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. CoRR, abs/1605.07146, 2016.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 205, + 329, + 217 + ], + "spans": [ + { + "bbox": [ + 117, + 205, + 329, + 217 + ], + "score": 1.0, + "content": "URL http://arxiv.org/abs/1605.07146.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 222, + 505, + 268 + ], + "lines": [ + { + "bbox": [ + 105, + 223, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 505, + 236 + ], + "score": 1.0, + "content": "Hongyi Zhang, Moustapha Cisse, Yann N. Dauphin, and David Lopez-Paz. mixup: Beyond empiri- ´", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 235, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 116, + 235, + 505, + 246 + ], + "score": 1.0, + "content": "cal risk minimization. In 6th International Conference on Learning Representations, ICLR 2018,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 117, + 246, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 117, + 246, + 506, + 258 + ], + "score": 1.0, + "content": "Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings, 2018. URL", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 257, + 367, + 269 + ], + "spans": [ + { + "bbox": [ + 116, + 257, + 302, + 269 + ], + "score": 1.0, + "content": "https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 302, + 258, + 309, + 266 + ], + "score": 0.28, + "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 257, + 367, + 269 + ], + "score": 1.0, + "content": "r1Ddp1-Rb.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 274, + 502, + 296 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 504, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 504, + 286 + ], + "score": 1.0, + "content": "Shengjia Zhao, Jiaming Song, and Stefano Ermon. Infovae: Information maximizing variational au-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 285, + 502, + 297 + ], + "spans": [ + { + "bbox": [ + 116, + 285, + 502, + 297 + ], + "score": 1.0, + "content": "toencoders. CoRR, abs/1706.02262, 2017. URL http://arxiv.org/abs/1706.02262.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 108, + 316, + 183, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 314, + 184, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 184, + 330 + ], + "score": 1.0, + "content": "A APPENDIX", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 108, + 339, + 262, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 262, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 262, + 352 + ], + "score": 1.0, + "content": "A.1 BASIC INEQUALITY OF ELBO", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 360, + 310, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 358, + 312, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 312, + 375 + ], + "score": 1.0, + "content": "Proposition A.1. The Basic inequality of ELBO is", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 129, + 375, + 482, + 389 + ], + "lines": [ + { + "bbox": [ + 129, + 375, + 482, + 389 + ], + "spans": [ + { + "bbox": [ + 129, + 375, + 482, + 389 + ], + "score": 0.83, + "content": "\\log p ( \\mathbf { X } ) \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } ) )", + "type": "interline_equation", + "image_path": "612d808a01d0d89ef90412fd6c19d21ab0340f377dd2441e51d376bbb90036ee.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 129, + 375, + 482, + 389 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 397, + 132, + 408 + ], + "lines": [ + { + "bbox": [ + 104, + 396, + 133, + 409 + ], + "spans": [ + { + "bbox": [ + 104, + 396, + 133, + 409 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 425, + 486, + 544 + ], + "lines": [ + { + "bbox": [ + 111, + 425, + 486, + 544 + ], + "spans": [ + { + "bbox": [ + 111, + 425, + 486, + 544 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\log \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } = \\log \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\log \\frac { p ( \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } + \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) } \\\\ & { = - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) | p ( \\mathbf { z } , \\mathbf { c } ) ) + \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] } \\\\ & { = \\mathrm { w i t h ~ a s s u m p t i o n } \\left( 3 . 4 \\right) \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | p ( \\mathbf { c } ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "7b06614e163105f312f4bdf8734e569068cbd46e884e30ae0efe770b28018490.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 111, + 425, + 486, + 464.6666666666667 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 111, + 464.6666666666667, + 486, + 504.33333333333337 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 111, + 504.33333333333337, + 486, + 544.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 552, + 312, + 564 + ], + "lines": [ + { + "bbox": [ + 106, + 552, + 312, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 312, + 565 + ], + "score": 1.0, + "content": "A.2 DECOMPOSITION OF ELBO IN INFO-VAE", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 108, + 572, + 406, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 406, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 406, + 587 + ], + "score": 1.0, + "content": "Proposition A.2. The expected ELBO with empirical distribution satisfies", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "interline_equation", + "bbox": [ + 173, + 588, + 437, + 602 + ], + "lines": [ + { + "bbox": [ + 173, + 588, + 437, + 602 + ], + "spans": [ + { + "bbox": [ + 173, + 588, + 437, + 602 + ], + "score": 0.87, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) \\big ) = \\mathbf { I } _ { q _ { \\phi } } \\big ( \\mathbf { X } ; \\mathbf { z } \\big ) + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } ) \\| p ( \\mathbf { z } ) \\big )", + "type": "interline_equation", + "image_path": "b1363b4acbb9e8505530319cfe147d33224841d2d1effe447c6b47d4d9ee17b1.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 173, + 588, + 437, + 602 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 609, + 132, + 621 + ], + "lines": [ + { + "bbox": [ + 103, + 609, + 133, + 622 + ], + "spans": [ + { + "bbox": [ + 103, + 609, + 133, + 622 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "interline_equation", + "bbox": [ + 127, + 639, + 484, + 738 + ], + "lines": [ + { + "bbox": [ + 127, + 639, + 484, + 738 + ], + "spans": [ + { + "bbox": [ + 127, + 639, + 484, + 738 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\displaystyle \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } ) ) = \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } { p ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } } \\\\ & { \\displaystyle = \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } { q _ { \\phi } ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } + \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } ) } { p ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } } \\\\ & { \\displaystyle = \\int _ { \\mathbf { X } } \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { X } ) \\log \\frac { q _ { \\phi } ( \\mathbf { z } , \\mathbf { X } ) } { q _ { \\phi } ( \\mathbf { z } ) p _ { e m p } ( \\mathbf { X } ) } d \\mathbf { z } d \\mathbf { X } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } ) | | p ( \\mathbf { z } ) ) } \\\\ & { \\displaystyle = \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } ) | | p ( \\mathbf { z } ) ) } \\end{array} \\overset { \\mathrm { U L } } { \\mathop : }", + "type": "interline_equation", + "image_path": "f78348420b8b221d1f87f4a036fd463c20b56ccf333cfc0d107bdef15cdc595f.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 127, + 639, + 484, + 672.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 127, + 672.0, + 484, + 705.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 127, + 705.0, + 484, + 738.0 + ], + "spans": [], + "index": 31 + } + ] + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 492, + 477, + 504, + 488 + ], + "lines": [ + { + "bbox": [ + 493, + 478, + 504, + 490 + ], + "spans": [ + { + "bbox": [ + 493, + 478, + 504, + 490 + ], + "score": 0.981, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 473, + 681, + 483, + 691 + ], + "lines": [] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 761 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "11", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 404, + 95 + ], + "score": 1.0, + "content": "Zhirong Wu, Yuanjun Xiong, Stella X. Yu, and Dahua Lin.", + "type": "text" + }, + { + "bbox": [ + 423, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Unsupervised fea-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 376, + 106 + ], + "score": 1.0, + "content": "ture learning via non-parametric instance discrimination.", + "type": "text" + }, + { + "bbox": [ + 394, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "In 2018 IEEE Confer-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 102, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 102, + 505, + 117 + ], + "score": 1.0, + "content": "ence on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 115, + 114, + 367, + 128 + ], + "score": 1.0, + "content": "UT, USA, June 18-22, 2018, pp. 3733–3742, 2018.", + "type": "text" + }, + { + "bbox": [ + 384, + 114, + 505, + 127 + ], + "score": 1.0, + "content": "doi: 10.1109/CVPR.2018.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 125, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 115, + 125, + 506, + 140 + ], + "score": 1.0, + "content": "00393. URL http://openaccess.thecvf.com/content_cvpr_2018/html/Wu_", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 117, + 137, + 421, + 149 + ], + "spans": [ + { + "bbox": [ + 117, + 137, + 421, + 149 + ], + "score": 1.0, + "content": "Unsupervised_Feature_Learning_CVPR_2018_paper.html.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 106, + 81, + 506, + 149 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 155, + 503, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 155, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 505, + 167 + ], + "score": 1.0, + "content": "Qizhe Xie, Zihang Dai, Eduard H. Hovy, Minh-Thang Luong, and Quoc V. Le. Unsupervised", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 166, + 504, + 178 + ], + "spans": [ + { + "bbox": [ + 116, + 166, + 504, + 178 + ], + "score": 1.0, + "content": "data augmentation. CoRR, abs/1904.12848, 2019. URL http://arxiv.org/abs/1904.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 177, + 152, + 189 + ], + "spans": [ + { + "bbox": [ + 116, + 177, + 152, + 189 + ], + "score": 1.0, + "content": "12848.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 155, + 505, + 189 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 194, + 502, + 217 + ], + "lines": [ + { + "bbox": [ + 105, + 194, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 505, + 207 + ], + "score": 1.0, + "content": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. CoRR, abs/1605.07146, 2016.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 205, + 329, + 217 + ], + "spans": [ + { + "bbox": [ + 117, + 205, + 329, + 217 + ], + "score": 1.0, + "content": "URL http://arxiv.org/abs/1605.07146.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 194, + 505, + 217 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 222, + 505, + 268 + ], + "lines": [ + { + "bbox": [ + 105, + 223, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 505, + 236 + ], + "score": 1.0, + "content": "Hongyi Zhang, Moustapha Cisse, Yann N. Dauphin, and David Lopez-Paz. mixup: Beyond empiri- ´", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 235, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 116, + 235, + 505, + 246 + ], + "score": 1.0, + "content": "cal risk minimization. In 6th International Conference on Learning Representations, ICLR 2018,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 117, + 246, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 117, + 246, + 506, + 258 + ], + "score": 1.0, + "content": "Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings, 2018. URL", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 257, + 367, + 269 + ], + "spans": [ + { + "bbox": [ + 116, + 257, + 302, + 269 + ], + "score": 1.0, + "content": "https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 302, + 258, + 309, + 266 + ], + "score": 0.28, + "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 257, + 367, + 269 + ], + "score": 1.0, + "content": "r1Ddp1-Rb.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 223, + 506, + 269 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 274, + 502, + 296 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 504, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 504, + 286 + ], + "score": 1.0, + "content": "Shengjia Zhao, Jiaming Song, and Stefano Ermon. Infovae: Information maximizing variational au-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 285, + 502, + 297 + ], + "spans": [ + { + "bbox": [ + 116, + 285, + 502, + 297 + ], + "score": 1.0, + "content": "toencoders. CoRR, abs/1706.02262, 2017. URL http://arxiv.org/abs/1706.02262.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 273, + 504, + 297 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 316, + 183, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 314, + 184, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 184, + 330 + ], + "score": 1.0, + "content": "A APPENDIX", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 108, + 339, + 262, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 262, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 262, + 352 + ], + "score": 1.0, + "content": "A.1 BASIC INEQUALITY OF ELBO", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 339, + 262, + 352 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 360, + 310, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 358, + 312, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 312, + 375 + ], + "score": 1.0, + "content": "Proposition A.1. The Basic inequality of ELBO is", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 358, + 312, + 375 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 129, + 375, + 482, + 389 + ], + "lines": [ + { + "bbox": [ + 129, + 375, + 482, + 389 + ], + "spans": [ + { + "bbox": [ + 129, + 375, + 482, + 389 + ], + "score": 0.83, + "content": "\\log p ( \\mathbf { X } ) \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } ) )", + "type": "interline_equation", + "image_path": "612d808a01d0d89ef90412fd6c19d21ab0340f377dd2441e51d376bbb90036ee.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 129, + 375, + 482, + 389 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 397, + 132, + 408 + ], + "lines": [ + { + "bbox": [ + 104, + 396, + 133, + 409 + ], + "spans": [ + { + "bbox": [ + 104, + 396, + 133, + 409 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 104, + 396, + 133, + 409 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 425, + 486, + 544 + ], + "lines": [ + { + "bbox": [ + 111, + 425, + 486, + 544 + ], + "spans": [ + { + "bbox": [ + 111, + 425, + 486, + 544 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\log \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } = \\log \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\log \\frac { p ( \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } + \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) } \\\\ & { = - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) | p ( \\mathbf { z } , \\mathbf { c } ) ) + \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] } \\\\ & { = \\mathrm { w i t h ~ a s s u m p t i o n } \\left( 3 . 4 \\right) \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | p ( \\mathbf { c } ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "7b06614e163105f312f4bdf8734e569068cbd46e884e30ae0efe770b28018490.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 111, + 425, + 486, + 464.6666666666667 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 111, + 464.6666666666667, + 486, + 504.33333333333337 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 111, + 504.33333333333337, + 486, + 544.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 552, + 312, + 564 + ], + "lines": [ + { + "bbox": [ + 106, + 552, + 312, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 312, + 565 + ], + "score": 1.0, + "content": "A.2 DECOMPOSITION OF ELBO IN INFO-VAE", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 108, + 572, + 406, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 406, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 406, + 587 + ], + "score": 1.0, + "content": "Proposition A.2. The expected ELBO with empirical distribution satisfies", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 572, + 406, + 587 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 173, + 588, + 437, + 602 + ], + "lines": [ + { + "bbox": [ + 173, + 588, + 437, + 602 + ], + "spans": [ + { + "bbox": [ + 173, + 588, + 437, + 602 + ], + "score": 0.87, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) \\big ) = \\mathbf { I } _ { q _ { \\phi } } \\big ( \\mathbf { X } ; \\mathbf { z } \\big ) + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } ) \\| p ( \\mathbf { z } ) \\big )", + "type": "interline_equation", + "image_path": "b1363b4acbb9e8505530319cfe147d33224841d2d1effe447c6b47d4d9ee17b1.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 173, + 588, + 437, + 602 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 609, + 132, + 621 + ], + "lines": [ + { + "bbox": [ + 103, + 609, + 133, + 622 + ], + "spans": [ + { + "bbox": [ + 103, + 609, + 133, + 622 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 103, + 609, + 133, + 622 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 127, + 639, + 484, + 738 + ], + "lines": [ + { + "bbox": [ + 127, + 639, + 484, + 738 + ], + "spans": [ + { + "bbox": [ + 127, + 639, + 484, + 738 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\displaystyle \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } ) ) = \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } { p ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } } \\\\ & { \\displaystyle = \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } { q _ { \\phi } ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } + \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } ) } { p ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } } \\\\ & { \\displaystyle = \\int _ { \\mathbf { X } } \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { X } ) \\log \\frac { q _ { \\phi } ( \\mathbf { z } , \\mathbf { X } ) } { q _ { \\phi } ( \\mathbf { z } ) p _ { e m p } ( \\mathbf { X } ) } d \\mathbf { z } d \\mathbf { X } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } ) | | p ( \\mathbf { z } ) ) } \\\\ & { \\displaystyle = \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } ) | | p ( \\mathbf { z } ) ) } \\end{array} \\overset { \\mathrm { U L } } { \\mathop : }", + "type": "interline_equation", + "image_path": "f78348420b8b221d1f87f4a036fd463c20b56ccf333cfc0d107bdef15cdc595f.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 127, + 639, + 484, + 672.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 127, + 672.0, + 484, + 705.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 127, + 705.0, + 484, + 738.0 + ], + "spans": [], + "index": 31 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 271, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 271, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 271, + 95 + ], + "score": 1.0, + "content": "A.3 THE EQUATION FORM OF ELBO", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 104, + 102, + 505, + 126 + ], + "lines": [ + { + "bbox": [ + 105, + 101, + 507, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 101, + 407, + 118 + ], + "score": 1.0, + "content": "Proposition A.3. The expected margin between the true log-likelihood", + "type": "text" + }, + { + "bbox": [ + 408, + 103, + 484, + 117 + ], + "score": 0.92, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\log p ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 101, + 507, + 118 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 113, + 183, + 129 + ], + "spans": [ + { + "bbox": [ + 107, + 114, + 171, + 127 + ], + "score": 0.86, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } E L B O", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 113, + 183, + 129 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 131, + 495, + 146 + ], + "lines": [ + { + "bbox": [ + 114, + 131, + 495, + 146 + ], + "spans": [ + { + "bbox": [ + 114, + 131, + 495, + 146 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { p _ { c m p } ( \\mathbf { X } ) } [ \\log p ( \\mathbf { X } ) - E L B O ] = \\mathbb { E } _ { p _ { c m p } ( \\mathbf { X } ) } [ D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) ]", + "type": "interline_equation", + "image_path": "cb6a8d2fae314051384231eb0da47cce2013ff7800ca46f50afbfd817c45a02d.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 114, + 131, + 495, + 146 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 154, + 132, + 165 + ], + "lines": [ + { + "bbox": [ + 104, + 155, + 133, + 167 + ], + "spans": [ + { + "bbox": [ + 104, + 155, + 133, + 167 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 171, + 287, + 183 + ], + "lines": [ + { + "bbox": [ + 106, + 170, + 288, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 170, + 288, + 185 + ], + "score": 1.0, + "content": "We just need to prove the following equation", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 157, + 186, + 453, + 200 + ], + "lines": [ + { + "bbox": [ + 157, + 186, + 453, + 200 + ], + "spans": [ + { + "bbox": [ + 157, + 186, + 453, + 200 + ], + "score": 0.87, + "content": "\\log p ( \\mathbf { X } ) - \\mathrm { E L B O } = D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) )", + "type": "interline_equation", + "image_path": "dab21f179631ac39d23d8f914db2bb86f9def4cf8a35c232b1b5f0dc8b4bd497.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 157, + 186, + 453, + 200 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 268, + 216 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 270, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 235, + 217 + ], + "score": 1.0, + "content": "and the proof under assumption", + "type": "text" + }, + { + "bbox": [ + 236, + 204, + 258, + 216 + ], + "score": 0.87, + "content": "( 3 , 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 203, + 270, + 217 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 220, + 511, + 293 + ], + "lines": [ + { + "bbox": [ + 111, + 220, + 511, + 293 + ], + "spans": [ + { + "bbox": [ + 111, + 220, + 511, + 293 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\displaystyle \\int _ { z , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log p ( \\mathbf { X } ) d \\mathbf { z } d \\mathbf { c } = \\displaystyle \\int _ { z , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } d \\mathbf { z } d \\mathbf { c } } \\\\ & { = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } + \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } { p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\mathrm { E L B O } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "80740ddc57f57e984002db29dfb772d961712b7adeb1e2bbb08fa8b680584e08.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 111, + 220, + 511, + 244.33333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 111, + 244.33333333333334, + 511, + 268.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 111, + 268.6666666666667, + 511, + 293.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 302, + 268, + 314 + ], + "lines": [ + { + "bbox": [ + 106, + 302, + 268, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 268, + 315 + ], + "score": 1.0, + "content": "A.4 OPTIMAL TRANSPORT SCHEME", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 106, + 323, + 177, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 322, + 178, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 178, + 336 + ], + "score": 1.0, + "content": "Proposition A.4.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 343, + 503, + 367 + ], + "lines": [ + { + "bbox": [ + 106, + 342, + 503, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 407, + 358 + ], + "score": 1.0, + "content": "1. The shortest-path derived from optimal transport scheme (17) between", + "type": "text" + }, + { + "bbox": [ + 408, + 343, + 503, + 356 + ], + "score": 0.9, + "content": "\\mathbf z _ { 1 } \\sim \\mathcal N ( \\pmb { \\mu } _ { 1 } , \\mathrm { d i a g } ( \\pmb { \\sigma } _ { 1 } ^ { 2 } ) )", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 354, + 289, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 123, + 367 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 354, + 217, + 367 + ], + "score": 0.92, + "content": "\\mathbf z _ { 2 } \\sim \\mathcal N ( \\pmb { \\mu } _ { 2 } , \\mathrm { d i a g } ( \\pmb { \\sigma } _ { 2 } ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 354, + 239, + 367 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 239, + 355, + 278, + 367 + ], + "score": 0.93, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 354, + 289, + 367 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 370, + 353, + 399 + ], + "lines": [ + { + "bbox": [ + 257, + 370, + 353, + 399 + ], + "spans": [ + { + "bbox": [ + 257, + 370, + 353, + 399 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\tilde { \\pmb { \\mu } } = \\lambda \\pmb { \\mu } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\mu } _ { 2 } } \\\\ { \\tilde { \\pmb { \\sigma } } = \\lambda \\pmb { \\sigma } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\sigma } _ { 2 } } \\end{array}", + "type": "interline_equation", + "image_path": "1a5a90aae83b2c000b08123e680b76b3be8bde46513ca2ad4be17032da3680d2.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 257, + 370, + 353, + 384.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 257, + 384.5, + 353, + 399.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 402, + 132, + 414 + ], + "lines": [ + { + "bbox": [ + 103, + 401, + 134, + 416 + ], + "spans": [ + { + "bbox": [ + 103, + 401, + 134, + 416 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 419, + 504, + 442 + ], + "lines": [ + { + "bbox": [ + 106, + 418, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 432 + ], + "score": 1.0, + "content": "Utilizing the conclusions in Kuang & Tabak (2017), the closed form of optimal transport from one", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 429, + 466, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 210, + 442 + ], + "score": 1.0, + "content": "multi-normal distribution", + "type": "text" + }, + { + "bbox": [ + 210, + 430, + 271, + 442 + ], + "score": 0.93, + "content": "\\mathcal { N } ( \\mathbf { z } _ { 1 } ; \\mu _ { 1 } , \\pmb { \\Sigma } _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 429, + 394, + 442 + ], + "score": 1.0, + "content": "to another normal distribution", + "type": "text" + }, + { + "bbox": [ + 394, + 430, + 455, + 442 + ], + "score": 0.92, + "content": "\\bar { \\mathcal { N } } ( \\mathbf { z } _ { 2 } ; \\mu _ { 2 } , \\Sigma _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 429, + 466, + 442 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 170, + 445, + 439, + 464 + ], + "lines": [ + { + "bbox": [ + 170, + 445, + 439, + 464 + ], + "spans": [ + { + "bbox": [ + 170, + 445, + 439, + 464 + ], + "score": 0.89, + "content": "\\begin{array} { r } { { \\mathbf z } \\to { \\mathcal T } ( { \\mathbf z } ) = \\mu _ { 2 } + { \\mathbf T } ( { \\mathbf z } - { \\boldsymbol \\mu _ { 1 } } ) ; \\qquad { \\mathbf T } = { \\mathbf { \\Sigma } } _ { 1 } ^ { - \\frac { 1 } { 2 } } \\bigr ( { \\mathbf { \\Sigma } } _ { 1 } ^ { \\frac { 1 } { 2 } } { \\mathbf { \\Sigma } } _ { 2 } { \\mathbf { \\Sigma } } _ { 1 } ^ { \\frac { 1 } { 2 } } \\bigr ) ^ { \\frac { 1 } { 2 } } { \\mathbf { \\Sigma } } _ { 1 } ^ { - \\frac { 1 } { 2 } } } \\end{array}", + "type": "interline_equation", + "image_path": "06967874ef6391478d9c20495ad2985d13efc1f8f4a3f5904488e6273f878eb5.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 170, + 445, + 439, + 464 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 467, + 405, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 405, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 387, + 480 + ], + "score": 1.0, + "content": "Utilize the diag matrix assumption, the optimal transport scheme with", + "type": "text" + }, + { + "bbox": [ + 388, + 468, + 394, + 477 + ], + "score": 0.84, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 466, + 405, + 480 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 250, + 482, + 361, + 511 + ], + "lines": [ + { + "bbox": [ + 250, + 482, + 361, + 511 + ], + "spans": [ + { + "bbox": [ + 250, + 482, + 361, + 511 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\mathbf { z } _ { \\lambda } = ( 1 - \\lambda ) \\mathbf { z } _ { 1 } + \\lambda T ( \\mathbf { z } _ { 1 } ) } \\\\ { \\mathbf { T } = d i a g ( \\pmb { \\sigma } _ { 1 } / \\pmb { \\sigma } _ { 2 } ) } \\end{array}", + "type": "interline_equation", + "image_path": "8ed6d6dfe07b316d50a012889236cdfb64cad97d1f007d67edca2f2219b7905b.jpg" + } + ] + } + ], + "index": 22.5, + "virtual_lines": [ + { + "bbox": [ + 250, + 482, + 361, + 496.5 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 250, + 496.5, + 361, + 511.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 245, + 527 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 246, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 246, + 529 + ], + "score": 1.0, + "content": "Utilize (A.2), we can get (A.1) as", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 531, + 471, + 545 + ], + "lines": [ + { + "bbox": [ + 133, + 531, + 471, + 545 + ], + "spans": [ + { + "bbox": [ + 133, + 531, + 471, + 545 + ], + "score": 0.89, + "content": "{ \\bf z } _ { \\lambda } \\sim \\mathcal { N } ( \\tilde { \\pmb { \\mu } } , d i a g ( \\tilde { \\pmb { \\sigma } } ^ { 2 } ) ) ; \\qquad \\tilde { \\pmb { \\mu } } = \\lambda { \\pmb { \\mu } } _ { 1 } + ( 1 - \\lambda ) { \\pmb { \\mu } } _ { 2 } ; \\qquad \\tilde { \\pmb { \\sigma } } = \\lambda { \\pmb { \\sigma } } _ { 1 } + ( 1 - \\lambda ) { \\pmb { \\sigma } } _ { 2 } \\quad [ \\lambda \\in \\mathcal { N } ^ { \\pmb { \\mu } } ] .", + "type": "interline_equation", + "image_path": "fe26ce5355e26a16d568164dd51d1262af9e361cbae73fb249d507ba3db905f5.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 133, + 531, + 471, + 545 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 548, + 504, + 572 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 480, + 563 + ], + "score": 1.0, + "content": "2. The shortest-path derived from KL divergence based optimal transport scheme between", + "type": "text" + }, + { + "bbox": [ + 480, + 551, + 505, + 560 + ], + "score": 0.81, + "content": "\\mathbf { c } _ { 1 } \\sim", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 560, + 324, + 573 + ], + "spans": [ + { + "bbox": [ + 107, + 560, + 159, + 572 + ], + "score": 0.92, + "content": "\\mathrm { M u l t } ( K , \\pi _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 560, + 177, + 573 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 178, + 560, + 253, + 572 + ], + "score": 0.93, + "content": "{ \\bf c } _ { 2 } \\sim \\mathrm { M u l t } ( K , \\pi _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 560, + 274, + 573 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 274, + 560, + 313, + 572 + ], + "score": 0.93, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 560, + 324, + 573 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 575, + 353, + 589 + ], + "lines": [ + { + "bbox": [ + 258, + 575, + 353, + 589 + ], + "spans": [ + { + "bbox": [ + 258, + 575, + 353, + 589 + ], + "score": 0.89, + "content": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 }", + "type": "interline_equation", + "image_path": "0f5495ae1d02266c38058a36e99eea685a61a09835dd32f0f0a2b01da853e839.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 258, + 575, + 353, + 589 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 593, + 132, + 604 + ], + "lines": [ + { + "bbox": [ + 104, + 594, + 133, + 605 + ], + "spans": [ + { + "bbox": [ + 104, + 594, + 133, + 605 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 609, + 505, + 644 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 462, + 622 + ], + "score": 1.0, + "content": "As the definition (17) has no closed-form solution for multinomial distribution, we use", + "type": "text" + }, + { + "bbox": [ + 463, + 610, + 478, + 620 + ], + "score": 0.28, + "content": "\\mathrm { K L }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 608, + 505, + 622 + ], + "score": 1.0, + "content": "diver-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 621, + 506, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 375, + 633 + ], + "score": 1.0, + "content": "gence instead. The KL divergence based optimal transport target", + "type": "text" + }, + { + "bbox": [ + 375, + 621, + 449, + 632 + ], + "score": 0.9, + "content": "\\mathbf { c } _ { \\lambda } \\sim \\mathbf { M u l t } ( K , \\tilde { \\pi } )", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 621, + 506, + 633 + ], + "score": 1.0, + "content": "between two", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 632, + 465, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 205, + 644 + ], + "score": 1.0, + "content": "multinomial distribution", + "type": "text" + }, + { + "bbox": [ + 206, + 632, + 281, + 644 + ], + "score": 0.92, + "content": "{ \\bf c } _ { 1 } \\sim \\mathrm { M u l t } ( K , \\pi _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 632, + 299, + 644 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 300, + 632, + 375, + 644 + ], + "score": 0.92, + "content": "{ \\bf c } _ { 2 } \\sim \\bf { M u l t } ( { \\cal K } , \\pi _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 632, + 396, + 644 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 397, + 632, + 435, + 644 + ], + "score": 0.91, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 632, + 465, + 644 + ], + "score": 1.0, + "content": "satisfy", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 647, + 432, + 682 + ], + "lines": [ + { + "bbox": [ + 178, + 647, + 432, + 682 + ], + "spans": [ + { + "bbox": [ + 178, + 647, + 432, + 682 + ], + "score": 0.92, + "content": "\\operatorname* { m i n } _ { \\tilde { \\pi } } \\lambda D _ { \\mathrm { K L } } ( \\pi _ { 1 } \\| \\tilde { \\pi } ) + ( 1 - \\lambda ) D _ { \\mathrm { K L } } ( \\pi _ { 2 } \\| \\tilde { \\pi } ) \\qquad s . t . \\sum _ { i = 1 } ^ { K } \\tilde { \\pi } _ { i } = 1", + "type": "interline_equation", + "image_path": "91abaf2591281c43c4b97376ffcbedc9d914c9e77d16d2eab169e9f3adc31089.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 178, + 647, + 432, + 658.6666666666666 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 178, + 658.6666666666666, + 432, + 670.3333333333333 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 178, + 670.3333333333333, + 432, + 681.9999999999999 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 685, + 273, + 697 + ], + "lines": [ + { + "bbox": [ + 106, + 685, + 273, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 685, + 273, + 699 + ], + "score": 1.0, + "content": "The Lagrange multiplier form of (A.4) is", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "interline_equation", + "bbox": [ + 169, + 701, + 442, + 735 + ], + "lines": [ + { + "bbox": [ + 169, + 701, + 442, + 735 + ], + "spans": [ + { + "bbox": [ + 169, + 701, + 442, + 735 + ], + "score": 0.94, + "content": "\\mathcal { L } ( \\tilde { \\pi } , t ) = \\lambda D _ { \\mathrm { K L } } ( \\pi _ { 1 } \\| \\tilde { \\pi } ) + ( 1 - \\lambda ) D _ { \\mathrm { K L } } ( \\pi _ { 2 } \\| \\tilde { \\pi } ) + t * ( \\sum _ { i = 1 } ^ { K } \\tilde { \\pi } _ { i } - 1 )", + "type": "interline_equation", + "image_path": "1937c5d3482a04a0b865cadf84482e4f7fc8637a676bcb628f7b4bd463552f6b.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 169, + 701, + 442, + 712.3333333333334 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 169, + 712.3333333333334, + 442, + 723.6666666666667 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 169, + 723.6666666666667, + 442, + 735.0000000000001 + ], + "spans": [], + "index": 39 + } + ] + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 309, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 511, + 250, + 522, + 260 + ], + "lines": [ + { + "bbox": [ + 513, + 252, + 522, + 261 + ], + "spans": [ + { + "bbox": [ + 513, + 252, + 522, + 261 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 468, + 533, + 477, + 542 + ], + "lines": [ + { + "bbox": [ + 472, + 534, + 477, + 544 + ], + "spans": [ + { + "bbox": [ + 472, + 534, + 477, + 544 + ], + "score": 0.917, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 271, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 271, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 271, + 95 + ], + "score": 1.0, + "content": "A.3 THE EQUATION FORM OF ELBO", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 104, + 102, + 505, + 126 + ], + "lines": [ + { + "bbox": [ + 105, + 101, + 507, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 101, + 407, + 118 + ], + "score": 1.0, + "content": "Proposition A.3. The expected margin between the true log-likelihood", + "type": "text" + }, + { + "bbox": [ + 408, + 103, + 484, + 117 + ], + "score": 0.92, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\log p ( \\mathbf { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 101, + 507, + 118 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 113, + 183, + 129 + ], + "spans": [ + { + "bbox": [ + 107, + 114, + 171, + 127 + ], + "score": 0.86, + "content": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } E L B O", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 113, + 183, + 129 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 101, + 507, + 129 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 131, + 495, + 146 + ], + "lines": [ + { + "bbox": [ + 114, + 131, + 495, + 146 + ], + "spans": [ + { + "bbox": [ + 114, + 131, + 495, + 146 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { p _ { c m p } ( \\mathbf { X } ) } [ \\log p ( \\mathbf { X } ) - E L B O ] = \\mathbb { E } _ { p _ { c m p } ( \\mathbf { X } ) } [ D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) ]", + "type": "interline_equation", + "image_path": "cb6a8d2fae314051384231eb0da47cce2013ff7800ca46f50afbfd817c45a02d.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 114, + 131, + 495, + 146 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 154, + 132, + 165 + ], + "lines": [ + { + "bbox": [ + 104, + 155, + 133, + 167 + ], + "spans": [ + { + "bbox": [ + 104, + 155, + 133, + 167 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 171, + 287, + 183 + ], + "lines": [ + { + "bbox": [ + 106, + 170, + 288, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 170, + 288, + 185 + ], + "score": 1.0, + "content": "We just need to prove the following equation", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 106, + 170, + 288, + 185 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 157, + 186, + 453, + 200 + ], + "lines": [ + { + "bbox": [ + 157, + 186, + 453, + 200 + ], + "spans": [ + { + "bbox": [ + 157, + 186, + 453, + 200 + ], + "score": 0.87, + "content": "\\log p ( \\mathbf { X } ) - \\mathrm { E L B O } = D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) )", + "type": "interline_equation", + "image_path": "dab21f179631ac39d23d8f914db2bb86f9def4cf8a35c232b1b5f0dc8b4bd497.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 157, + 186, + 453, + 200 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 268, + 216 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 270, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 235, + 217 + ], + "score": 1.0, + "content": "and the proof under assumption", + "type": "text" + }, + { + "bbox": [ + 236, + 204, + 258, + 216 + ], + "score": 0.87, + "content": "( 3 , 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 203, + 270, + 217 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 203, + 270, + 217 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 220, + 511, + 293 + ], + "lines": [ + { + "bbox": [ + 111, + 220, + 511, + 293 + ], + "spans": [ + { + "bbox": [ + 111, + 220, + 511, + 293 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\displaystyle \\int _ { z , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log p ( \\mathbf { X } ) d \\mathbf { z } d \\mathbf { c } = \\displaystyle \\int _ { z , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } d \\mathbf { z } d \\mathbf { c } } \\\\ & { = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } + \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } { p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\mathrm { E L B O } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "80740ddc57f57e984002db29dfb772d961712b7adeb1e2bbb08fa8b680584e08.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 111, + 220, + 511, + 244.33333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 111, + 244.33333333333334, + 511, + 268.6666666666667 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 111, + 268.6666666666667, + 511, + 293.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 302, + 268, + 314 + ], + "lines": [ + { + "bbox": [ + 106, + 302, + 268, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 268, + 315 + ], + "score": 1.0, + "content": "A.4 OPTIMAL TRANSPORT SCHEME", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 106, + 323, + 177, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 322, + 178, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 178, + 336 + ], + "score": 1.0, + "content": "Proposition A.4.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 343, + 503, + 367 + ], + "lines": [ + { + "bbox": [ + 106, + 342, + 503, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 407, + 358 + ], + "score": 1.0, + "content": "1. The shortest-path derived from optimal transport scheme (17) between", + "type": "text" + }, + { + "bbox": [ + 408, + 343, + 503, + 356 + ], + "score": 0.9, + "content": "\\mathbf z _ { 1 } \\sim \\mathcal N ( \\pmb { \\mu } _ { 1 } , \\mathrm { d i a g } ( \\pmb { \\sigma } _ { 1 } ^ { 2 } ) )", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 354, + 289, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 123, + 367 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 354, + 217, + 367 + ], + "score": 0.92, + "content": "\\mathbf z _ { 2 } \\sim \\mathcal N ( \\pmb { \\mu } _ { 2 } , \\mathrm { d i a g } ( \\pmb { \\sigma } _ { 2 } ^ { 2 } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 354, + 239, + 367 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 239, + 355, + 278, + 367 + ], + "score": 0.93, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 354, + 289, + 367 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 342, + 503, + 367 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 370, + 353, + 399 + ], + "lines": [ + { + "bbox": [ + 257, + 370, + 353, + 399 + ], + "spans": [ + { + "bbox": [ + 257, + 370, + 353, + 399 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\tilde { \\pmb { \\mu } } = \\lambda \\pmb { \\mu } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\mu } _ { 2 } } \\\\ { \\tilde { \\pmb { \\sigma } } = \\lambda \\pmb { \\sigma } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\sigma } _ { 2 } } \\end{array}", + "type": "interline_equation", + "image_path": "1a5a90aae83b2c000b08123e680b76b3be8bde46513ca2ad4be17032da3680d2.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 257, + 370, + 353, + 384.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 257, + 384.5, + 353, + 399.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 402, + 132, + 414 + ], + "lines": [ + { + "bbox": [ + 103, + 401, + 134, + 416 + ], + "spans": [ + { + "bbox": [ + 103, + 401, + 134, + 416 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 419, + 504, + 442 + ], + "lines": [ + { + "bbox": [ + 106, + 418, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 432 + ], + "score": 1.0, + "content": "Utilizing the conclusions in Kuang & Tabak (2017), the closed form of optimal transport from one", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 429, + 466, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 210, + 442 + ], + "score": 1.0, + "content": "multi-normal distribution", + "type": "text" + }, + { + "bbox": [ + 210, + 430, + 271, + 442 + ], + "score": 0.93, + "content": "\\mathcal { N } ( \\mathbf { z } _ { 1 } ; \\mu _ { 1 } , \\pmb { \\Sigma } _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 429, + 394, + 442 + ], + "score": 1.0, + "content": "to another normal distribution", + "type": "text" + }, + { + "bbox": [ + 394, + 430, + 455, + 442 + ], + "score": 0.92, + "content": "\\bar { \\mathcal { N } } ( \\mathbf { z } _ { 2 } ; \\mu _ { 2 } , \\Sigma _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 429, + 466, + 442 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 106, + 418, + 505, + 442 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 170, + 445, + 439, + 464 + ], + "lines": [ + { + "bbox": [ + 170, + 445, + 439, + 464 + ], + "spans": [ + { + "bbox": [ + 170, + 445, + 439, + 464 + ], + "score": 0.89, + "content": "\\begin{array} { r } { { \\mathbf z } \\to { \\mathcal T } ( { \\mathbf z } ) = \\mu _ { 2 } + { \\mathbf T } ( { \\mathbf z } - { \\boldsymbol \\mu _ { 1 } } ) ; \\qquad { \\mathbf T } = { \\mathbf { \\Sigma } } _ { 1 } ^ { - \\frac { 1 } { 2 } } \\bigr ( { \\mathbf { \\Sigma } } _ { 1 } ^ { \\frac { 1 } { 2 } } { \\mathbf { \\Sigma } } _ { 2 } { \\mathbf { \\Sigma } } _ { 1 } ^ { \\frac { 1 } { 2 } } \\bigr ) ^ { \\frac { 1 } { 2 } } { \\mathbf { \\Sigma } } _ { 1 } ^ { - \\frac { 1 } { 2 } } } \\end{array}", + "type": "interline_equation", + "image_path": "06967874ef6391478d9c20495ad2985d13efc1f8f4a3f5904488e6273f878eb5.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 170, + 445, + 439, + 464 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 467, + 405, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 405, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 387, + 480 + ], + "score": 1.0, + "content": "Utilize the diag matrix assumption, the optimal transport scheme with", + "type": "text" + }, + { + "bbox": [ + 388, + 468, + 394, + 477 + ], + "score": 0.84, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 466, + 405, + 480 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 466, + 405, + 480 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 250, + 482, + 361, + 511 + ], + "lines": [ + { + "bbox": [ + 250, + 482, + 361, + 511 + ], + "spans": [ + { + "bbox": [ + 250, + 482, + 361, + 511 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\mathbf { z } _ { \\lambda } = ( 1 - \\lambda ) \\mathbf { z } _ { 1 } + \\lambda T ( \\mathbf { z } _ { 1 } ) } \\\\ { \\mathbf { T } = d i a g ( \\pmb { \\sigma } _ { 1 } / \\pmb { \\sigma } _ { 2 } ) } \\end{array}", + "type": "interline_equation", + "image_path": "8ed6d6dfe07b316d50a012889236cdfb64cad97d1f007d67edca2f2219b7905b.jpg" + } + ] + } + ], + "index": 22.5, + "virtual_lines": [ + { + "bbox": [ + 250, + 482, + 361, + 496.5 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 250, + 496.5, + 361, + 511.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 245, + 527 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 246, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 246, + 529 + ], + "score": 1.0, + "content": "Utilize (A.2), we can get (A.1) as", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 513, + 246, + 529 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 531, + 471, + 545 + ], + "lines": [ + { + "bbox": [ + 133, + 531, + 471, + 545 + ], + "spans": [ + { + "bbox": [ + 133, + 531, + 471, + 545 + ], + "score": 0.89, + "content": "{ \\bf z } _ { \\lambda } \\sim \\mathcal { N } ( \\tilde { \\pmb { \\mu } } , d i a g ( \\tilde { \\pmb { \\sigma } } ^ { 2 } ) ) ; \\qquad \\tilde { \\pmb { \\mu } } = \\lambda { \\pmb { \\mu } } _ { 1 } + ( 1 - \\lambda ) { \\pmb { \\mu } } _ { 2 } ; \\qquad \\tilde { \\pmb { \\sigma } } = \\lambda { \\pmb { \\sigma } } _ { 1 } + ( 1 - \\lambda ) { \\pmb { \\sigma } } _ { 2 } \\quad [ \\lambda \\in \\mathcal { N } ^ { \\pmb { \\mu } } ] .", + "type": "interline_equation", + "image_path": "fe26ce5355e26a16d568164dd51d1262af9e361cbae73fb249d507ba3db905f5.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 133, + 531, + 471, + 545 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 548, + 504, + 572 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 480, + 563 + ], + "score": 1.0, + "content": "2. The shortest-path derived from KL divergence based optimal transport scheme between", + "type": "text" + }, + { + "bbox": [ + 480, + 551, + 505, + 560 + ], + "score": 0.81, + "content": "\\mathbf { c } _ { 1 } \\sim", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 560, + 324, + 573 + ], + "spans": [ + { + "bbox": [ + 107, + 560, + 159, + 572 + ], + "score": 0.92, + "content": "\\mathrm { M u l t } ( K , \\pi _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 560, + 177, + 573 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 178, + 560, + 253, + 572 + ], + "score": 0.93, + "content": "{ \\bf c } _ { 2 } \\sim \\mathrm { M u l t } ( K , \\pi _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 560, + 274, + 573 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 274, + 560, + 313, + 572 + ], + "score": 0.93, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 560, + 324, + 573 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 546, + 505, + 573 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 575, + 353, + 589 + ], + "lines": [ + { + "bbox": [ + 258, + 575, + 353, + 589 + ], + "spans": [ + { + "bbox": [ + 258, + 575, + 353, + 589 + ], + "score": 0.89, + "content": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 }", + "type": "interline_equation", + "image_path": "0f5495ae1d02266c38058a36e99eea685a61a09835dd32f0f0a2b01da853e839.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 258, + 575, + 353, + 589 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 593, + 132, + 604 + ], + "lines": [ + { + "bbox": [ + 104, + 594, + 133, + 605 + ], + "spans": [ + { + "bbox": [ + 104, + 594, + 133, + 605 + ], + "score": 1.0, + "content": "proof", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 609, + 505, + 644 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 462, + 622 + ], + "score": 1.0, + "content": "As the definition (17) has no closed-form solution for multinomial distribution, we use", + "type": "text" + }, + { + "bbox": [ + 463, + 610, + 478, + 620 + ], + "score": 0.28, + "content": "\\mathrm { K L }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 608, + 505, + 622 + ], + "score": 1.0, + "content": "diver-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 621, + 506, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 375, + 633 + ], + "score": 1.0, + "content": "gence instead. The KL divergence based optimal transport target", + "type": "text" + }, + { + "bbox": [ + 375, + 621, + 449, + 632 + ], + "score": 0.9, + "content": "\\mathbf { c } _ { \\lambda } \\sim \\mathbf { M u l t } ( K , \\tilde { \\pi } )", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 621, + 506, + 633 + ], + "score": 1.0, + "content": "between two", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 632, + 465, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 205, + 644 + ], + "score": 1.0, + "content": "multinomial distribution", + "type": "text" + }, + { + "bbox": [ + 206, + 632, + 281, + 644 + ], + "score": 0.92, + "content": "{ \\bf c } _ { 1 } \\sim \\mathrm { M u l t } ( K , \\pi _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 632, + 299, + 644 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 300, + 632, + 375, + 644 + ], + "score": 0.92, + "content": "{ \\bf c } _ { 2 } \\sim \\bf { M u l t } ( { \\cal K } , \\pi _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 632, + 396, + 644 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 397, + 632, + 435, + 644 + ], + "score": 0.91, + "content": "\\lambda \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 632, + 465, + 644 + ], + "score": 1.0, + "content": "satisfy", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 608, + 506, + 644 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 647, + 432, + 682 + ], + "lines": [ + { + "bbox": [ + 178, + 647, + 432, + 682 + ], + "spans": [ + { + "bbox": [ + 178, + 647, + 432, + 682 + ], + "score": 0.92, + "content": "\\operatorname* { m i n } _ { \\tilde { \\pi } } \\lambda D _ { \\mathrm { K L } } ( \\pi _ { 1 } \\| \\tilde { \\pi } ) + ( 1 - \\lambda ) D _ { \\mathrm { K L } } ( \\pi _ { 2 } \\| \\tilde { \\pi } ) \\qquad s . t . \\sum _ { i = 1 } ^ { K } \\tilde { \\pi } _ { i } = 1", + "type": "interline_equation", + "image_path": "91abaf2591281c43c4b97376ffcbedc9d914c9e77d16d2eab169e9f3adc31089.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 178, + 647, + 432, + 658.6666666666666 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 178, + 658.6666666666666, + 432, + 670.3333333333333 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 178, + 670.3333333333333, + 432, + 681.9999999999999 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 685, + 273, + 697 + ], + "lines": [ + { + "bbox": [ + 106, + 685, + 273, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 685, + 273, + 699 + ], + "score": 1.0, + "content": "The Lagrange multiplier form of (A.4) is", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36, + "bbox_fs": [ + 106, + 685, + 273, + 699 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 169, + 701, + 442, + 735 + ], + "lines": [ + { + "bbox": [ + 169, + 701, + 442, + 735 + ], + "spans": [ + { + "bbox": [ + 169, + 701, + 442, + 735 + ], + "score": 0.94, + "content": "\\mathcal { L } ( \\tilde { \\pi } , t ) = \\lambda D _ { \\mathrm { K L } } ( \\pi _ { 1 } \\| \\tilde { \\pi } ) + ( 1 - \\lambda ) D _ { \\mathrm { K L } } ( \\pi _ { 2 } \\| \\tilde { \\pi } ) + t * ( \\sum _ { i = 1 } ^ { K } \\tilde { \\pi } _ { i } - 1 )", + "type": "interline_equation", + "image_path": "1937c5d3482a04a0b865cadf84482e4f7fc8637a676bcb628f7b4bd463552f6b.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 169, + 701, + 442, + 712.3333333333334 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 169, + 712.3333333333334, + 442, + 723.6666666666667 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 169, + 723.6666666666667, + 442, + 735.0000000000001 + ], + "spans": [], + "index": 39 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 114, + 99, + 497, + 191 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 245, + 80, + 366, + 92 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 245, + 79, + 366, + 93 + ], + "spans": [ + { + "bbox": [ + 245, + 79, + 366, + 93 + ], + "score": 1.0, + "content": "Table 6: Schedule Parameters", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 114, + 99, + 497, + 191 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 99, + 497, + 191 + ], + "spans": [ + { + "bbox": [ + 114, + 99, + 497, + 191 + ], + "score": 0.983, + "html": "
MNISTSVHN(one-stage)SVHNCifar10Cifar100
hmaxtmaxhmaxtmaxhmaxtmaxhmaxtmaxhmaxtmax
305011751e-31501e-31501e-1150
B3050117511501e-31501e-3150
117.5505017512801502001501280150
Ic1750501752.31502.31504.6150
WMz///一1e-31501e-31501e-1150
WMc/1140012801280
", + "type": "table", + "image_path": "391bdb23dc0c933375929b262309ab1307a56f3b9a9bb2b19986783d65f23084.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 114, + 99, + 497, + 129.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 114, + 129.66666666666666, + 497, + 160.33333333333331 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 114, + 160.33333333333331, + 497, + 190.99999999999997 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 106, + 214, + 249, + 226 + ], + "lines": [ + { + "bbox": [ + 107, + 215, + 249, + 227 + ], + "spans": [ + { + "bbox": [ + 107, + 215, + 249, + 227 + ], + "score": 1.0, + "content": "and the related KKT conditions are", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 232, + 387, + 293 + ], + "lines": [ + { + "bbox": [ + 224, + 232, + 387, + 293 + ], + "spans": [ + { + "bbox": [ + 224, + 232, + 387, + 293 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\frac { \\partial \\mathcal { L } ( \\tilde { \\boldsymbol { \\pi } } , t ) } { \\partial \\tilde { \\boldsymbol { \\pi } } } = t - \\frac { \\lambda \\boldsymbol { \\pi } _ { 1 } + ( 1 - \\lambda ) \\boldsymbol { \\pi } _ { 2 } } { \\tilde { \\boldsymbol { \\pi } } } = 0 } \\\\ { t \\ast ( \\displaystyle \\sum _ { i = 1 } ^ { K } \\tilde { \\boldsymbol { \\pi } } _ { i } - 1 ) = 0 } \\end{array}", + "type": "interline_equation", + "image_path": "a47c8579fdd46832ed1ce705a4d8e646e3b49efb82a32523535c1f7b79b1e4c4.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 224, + 232, + 387, + 247.25 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 224, + 247.25, + 387, + 262.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 224, + 262.5, + 387, + 277.75 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 224, + 277.75, + 387, + 293.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 301, + 352, + 314 + ], + "lines": [ + { + "bbox": [ + 106, + 301, + 352, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 331, + 315 + ], + "score": 1.0, + "content": "Solve the equation (A.5), we can get the closed form of", + "type": "text" + }, + { + "bbox": [ + 331, + 302, + 339, + 312 + ], + "score": 0.83, + "content": "\\tilde { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 301, + 352, + 315 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "interline_equation", + "bbox": [ + 249, + 322, + 362, + 336 + ], + "lines": [ + { + "bbox": [ + 249, + 322, + 362, + 336 + ], + "spans": [ + { + "bbox": [ + 249, + 322, + 362, + 336 + ], + "score": 0.84, + "content": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 } \\quad \\bigsqcup", + "type": "interline_equation", + "image_path": "70cd9fe91d1124c77a6c6ac100b6a430b601579ae84870e3978ef0332e5c4e3f.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 249, + 322, + 362, + 336 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 353, + 227, + 364 + ], + "lines": [ + { + "bbox": [ + 106, + 351, + 228, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 228, + 366 + ], + "score": 1.0, + "content": "A.5 SCHEDULE ANALYSIS", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 152, + 405, + 435, + 596 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 152, + 405, + 435, + 596 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 152, + 405, + 435, + 596 + ], + "spans": [ + { + "bbox": [ + 152, + 405, + 435, + 596 + ], + "score": 0.967, + "type": "image", + "image_path": "712ae7b375b430e62be3a30c858dcdfec56c3e221e02585de9e0516814126163.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 152, + 405, + 435, + 468.6666666666667 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 152, + 468.6666666666667, + 435, + 532.3333333333334 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 152, + 532.3333333333334, + 435, + 596.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 197, + 604, + 414, + 616 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 197, + 603, + 414, + 617 + ], + "spans": [ + { + "bbox": [ + 197, + 603, + 414, + 617 + ], + "score": 1.0, + "content": "Figure 2: The Exponential Function-Based Scheduler", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 14.0 + }, + { + "type": "text", + "bbox": [ + 106, + 639, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 652 + ], + "score": 1.0, + "content": "The warm-up scheduler aims to slowly increase the parameters until they reach their maximum. For", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 211, + 663 + ], + "score": 1.0, + "content": "a certain hyperparameter", + "type": "text" + }, + { + "bbox": [ + 211, + 651, + 218, + 661 + ], + "score": 0.8, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 651, + 505, + 663 + ], + "score": 1.0, + "content": ", there are 2 parameters control its warm-up process, the target value", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 661, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 107, + 662, + 130, + 673 + ], + "score": 0.9, + "content": "h _ { m a x }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 661, + 209, + 674 + ], + "score": 1.0, + "content": "and the total epoch", + "type": "text" + }, + { + "bbox": [ + 210, + 662, + 231, + 672 + ], + "score": 0.91, + "content": "t _ { m a x }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 661, + 505, + 674 + ], + "score": 1.0, + "content": "to reach the target value. We use the exponential function to get the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 672, + 183, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 160, + 685 + ], + "score": 1.0, + "content": "middle value", + "type": "text" + }, + { + "bbox": [ + 160, + 673, + 171, + 684 + ], + "score": 0.88, + "content": "h _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 672, + 183, + 685 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 687, + 403, + 702 + ], + "lines": [ + { + "bbox": [ + 207, + 687, + 403, + 702 + ], + "spans": [ + { + "bbox": [ + 207, + 687, + 403, + 702 + ], + "score": 0.88, + "content": "h _ { t } = h _ { m a x } \\times \\exp { [ - 5 * ( 1 - \\operatorname* { m i n } ( 1 , t / t _ { m a x } ) ) ^ { 2 } ] }", + "type": "interline_equation", + "image_path": "9ae375ce93609156d7fd38df51dd5de71145d94170619f98c18b778c8534f02b.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 207, + 687, + 403, + 702 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "The curve of (A.6) is shown in Figure 2, and we list the scheduler parameters of 4 benchmark", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 721, + 345, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 345, + 732 + ], + "score": 1.0, + "content": "datasets, i.e. MNIST, SVHN, Cifar10, Cifar100, in Table 6.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 309, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 351, + 324, + 361, + 334 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 114, + 99, + 497, + 191 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 245, + 80, + 366, + 92 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 245, + 79, + 366, + 93 + ], + "spans": [ + { + "bbox": [ + 245, + 79, + 366, + 93 + ], + "score": 1.0, + "content": "Table 6: Schedule Parameters", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 114, + 99, + 497, + 191 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 99, + 497, + 191 + ], + "spans": [ + { + "bbox": [ + 114, + 99, + 497, + 191 + ], + "score": 0.983, + "html": "
MNISTSVHN(one-stage)SVHNCifar10Cifar100
hmaxtmaxhmaxtmaxhmaxtmaxhmaxtmaxhmaxtmax
305011751e-31501e-31501e-1150
B3050117511501e-31501e-3150
117.5505017512801502001501280150
Ic1750501752.31502.31504.6150
WMz///一1e-31501e-31501e-1150
WMc/1140012801280
", + "type": "table", + "image_path": "391bdb23dc0c933375929b262309ab1307a56f3b9a9bb2b19986783d65f23084.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 114, + 99, + 497, + 129.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 114, + 129.66666666666666, + 497, + 160.33333333333331 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 114, + 160.33333333333331, + 497, + 190.99999999999997 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 106, + 214, + 249, + 226 + ], + "lines": [ + { + "bbox": [ + 107, + 215, + 249, + 227 + ], + "spans": [ + { + "bbox": [ + 107, + 215, + 249, + 227 + ], + "score": 1.0, + "content": "and the related KKT conditions are", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 107, + 215, + 249, + 227 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 232, + 387, + 293 + ], + "lines": [ + { + "bbox": [ + 224, + 232, + 387, + 293 + ], + "spans": [ + { + "bbox": [ + 224, + 232, + 387, + 293 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\frac { \\partial \\mathcal { L } ( \\tilde { \\boldsymbol { \\pi } } , t ) } { \\partial \\tilde { \\boldsymbol { \\pi } } } = t - \\frac { \\lambda \\boldsymbol { \\pi } _ { 1 } + ( 1 - \\lambda ) \\boldsymbol { \\pi } _ { 2 } } { \\tilde { \\boldsymbol { \\pi } } } = 0 } \\\\ { t \\ast ( \\displaystyle \\sum _ { i = 1 } ^ { K } \\tilde { \\boldsymbol { \\pi } } _ { i } - 1 ) = 0 } \\end{array}", + "type": "interline_equation", + "image_path": "a47c8579fdd46832ed1ce705a4d8e646e3b49efb82a32523535c1f7b79b1e4c4.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 224, + 232, + 387, + 247.25 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 224, + 247.25, + 387, + 262.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 224, + 262.5, + 387, + 277.75 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 224, + 277.75, + 387, + 293.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 301, + 352, + 314 + ], + "lines": [ + { + "bbox": [ + 106, + 301, + 352, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 331, + 315 + ], + "score": 1.0, + "content": "Solve the equation (A.5), we can get the closed form of", + "type": "text" + }, + { + "bbox": [ + 331, + 302, + 339, + 312 + ], + "score": 0.83, + "content": "\\tilde { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 301, + 352, + 315 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 106, + 301, + 352, + 315 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 249, + 322, + 362, + 336 + ], + "lines": [ + { + "bbox": [ + 249, + 322, + 362, + 336 + ], + "spans": [ + { + "bbox": [ + 249, + 322, + 362, + 336 + ], + "score": 0.84, + "content": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 } \\quad \\bigsqcup", + "type": "interline_equation", + "image_path": "70cd9fe91d1124c77a6c6ac100b6a430b601579ae84870e3978ef0332e5c4e3f.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 249, + 322, + 362, + 336 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 353, + 227, + 364 + ], + "lines": [ + { + "bbox": [ + 106, + 351, + 228, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 228, + 366 + ], + "score": 1.0, + "content": "A.5 SCHEDULE ANALYSIS", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 152, + 405, + 435, + 596 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 152, + 405, + 435, + 596 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 152, + 405, + 435, + 596 + ], + "spans": [ + { + "bbox": [ + 152, + 405, + 435, + 596 + ], + "score": 0.967, + "type": "image", + "image_path": "712ae7b375b430e62be3a30c858dcdfec56c3e221e02585de9e0516814126163.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 152, + 405, + 435, + 468.6666666666667 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 152, + 468.6666666666667, + 435, + 532.3333333333334 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 152, + 532.3333333333334, + 435, + 596.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 197, + 604, + 414, + 616 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 197, + 603, + 414, + 617 + ], + "spans": [ + { + "bbox": [ + 197, + 603, + 414, + 617 + ], + "score": 1.0, + "content": "Figure 2: The Exponential Function-Based Scheduler", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 14.0 + }, + { + "type": "text", + "bbox": [ + 106, + 639, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 652 + ], + "score": 1.0, + "content": "The warm-up scheduler aims to slowly increase the parameters until they reach their maximum. For", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 211, + 663 + ], + "score": 1.0, + "content": "a certain hyperparameter", + "type": "text" + }, + { + "bbox": [ + 211, + 651, + 218, + 661 + ], + "score": 0.8, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 651, + 505, + 663 + ], + "score": 1.0, + "content": ", there are 2 parameters control its warm-up process, the target value", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 661, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 107, + 662, + 130, + 673 + ], + "score": 0.9, + "content": "h _ { m a x }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 661, + 209, + 674 + ], + "score": 1.0, + "content": "and the total epoch", + "type": "text" + }, + { + "bbox": [ + 210, + 662, + 231, + 672 + ], + "score": 0.91, + "content": "t _ { m a x }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 661, + 505, + 674 + ], + "score": 1.0, + "content": "to reach the target value. We use the exponential function to get the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 672, + 183, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 160, + 685 + ], + "score": 1.0, + "content": "middle value", + "type": "text" + }, + { + "bbox": [ + 160, + 673, + 171, + 684 + ], + "score": 0.88, + "content": "h _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 672, + 183, + 685 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 639, + 505, + 685 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 687, + 403, + 702 + ], + "lines": [ + { + "bbox": [ + 207, + 687, + 403, + 702 + ], + "spans": [ + { + "bbox": [ + 207, + 687, + 403, + 702 + ], + "score": 0.88, + "content": "h _ { t } = h _ { m a x } \\times \\exp { [ - 5 * ( 1 - \\operatorname* { m i n } ( 1 , t / t _ { m a x } ) ) ^ { 2 } ] }", + "type": "interline_equation", + "image_path": "9ae375ce93609156d7fd38df51dd5de71145d94170619f98c18b778c8534f02b.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 207, + 687, + 403, + 702 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "The curve of (A.6) is shown in Figure 2, and we list the scheduler parameters of 4 benchmark", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 721, + 345, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 345, + 732 + ], + "score": 1.0, + "content": "datasets, i.e. MNIST, SVHN, Cifar10, Cifar100, in Table 6.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 106, + 709, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 99, + 525, + 214 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 233, + 80, + 378, + 92 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 232, + 79, + 379, + 93 + ], + "spans": [ + { + "bbox": [ + 232, + 79, + 379, + 93 + ], + "score": 1.0, + "content": "Table 7: Details of Training Process", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 106, + 99, + 525, + 214 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 99, + 525, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 99, + 525, + 214 + ], + "score": 0.981, + "html": "
MNISTSVHN(one-stage)SVHNCifar10Cifar100
Latent Dim(z/c)10/1032/10128/10128/10128/100
Mutual Info(z/c)17.5/17.050/501280/2.3200/2.31280/4.6
loss of -log pe(X|c,z)BCEBCEMSEMSEBCE
a of pmixup(X)一一222
optimizerAdamSGD
learning rate5e-41e-30.1
lr scheduler(decay ratio)一一every 50 epoch after 200-th(0.5)[500,600,650](0.2)
weight decay005e-4
", + "type": "table", + "image_path": "6570e9dfa5eec3f1555477b23cafd74049c581d87a93e5e2dcffad761cee71c9.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 99, + 525, + 137.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 137.33333333333334, + 525, + 175.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 175.66666666666669, + 525, + 214.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "title", + "bbox": [ + 108, + 243, + 272, + 255 + ], + "lines": [ + { + "bbox": [ + 106, + 243, + 273, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 273, + 256 + ], + "score": 1.0, + "content": "A.6 DETAILS OF TRAINING PROCESS", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 268, + 505, + 302 + ], + "lines": [ + { + "bbox": [ + 106, + 268, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 505, + 280 + ], + "score": 1.0, + "content": "Here we list some important items need to set in the training process for different process. We", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 279, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 279, + 505, + 291 + ], + "score": 1.0, + "content": "classify these items into 2 categories: (1) items related to the loss function and (2) items related to", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 290, + 467, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 467, + 302 + ], + "score": 1.0, + "content": "the optimization strategy. The details are as follows and we list the exact value in Table 7:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 131, + 316, + 274, + 327 + ], + "lines": [ + { + "bbox": [ + 131, + 316, + 274, + 327 + ], + "spans": [ + { + "bbox": [ + 131, + 316, + 274, + 327 + ], + "score": 1.0, + "content": "1. Items related to the loss function", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 151, + 340, + 506, + 549 + ], + "lines": [ + { + "bbox": [ + 152, + 341, + 208, + 351 + ], + "spans": [ + { + "bbox": [ + 152, + 341, + 208, + 351 + ], + "score": 1.0, + "content": "• Latent dim", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 161, + 350, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 161, + 350, + 506, + 363 + ], + "score": 1.0, + "content": "The latent dim of discrete variable c is the same as the number of classifications, that", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 162, + 362, + 475, + 374 + ], + "spans": [ + { + "bbox": [ + 162, + 363, + 174, + 374 + ], + "score": 1.0, + "content": "is,", + "type": "text" + }, + { + "bbox": [ + 174, + 362, + 184, + 372 + ], + "score": 0.73, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 363, + 285, + 374 + ], + "score": 1.0, + "content": ". For continuous variable", + "type": "text" + }, + { + "bbox": [ + 285, + 364, + 292, + 372 + ], + "score": 0.65, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 363, + 475, + 374 + ], + "score": 1.0, + "content": ", the latent dim is determined by experiments.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 153, + 380, + 243, + 392 + ], + "spans": [ + { + "bbox": [ + 153, + 380, + 243, + 392 + ], + "score": 1.0, + "content": "• Mutual information", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 162, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 162, + 390, + 505, + 403 + ], + "score": 1.0, + "content": "We find the value of continuous mutual information will affect the generative per-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 162, + 402, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 162, + 402, + 506, + 415 + ], + "score": 1.0, + "content": "formance, but have little impact on semi-supervised learning results, so we choose a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 162, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 162, + 414, + 505, + 425 + ], + "score": 1.0, + "content": "suitable value to get the best generative performance. For discrete information, we", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 162, + 424, + 390, + 436 + ], + "spans": [ + { + "bbox": [ + 162, + 425, + 334, + 436 + ], + "score": 1.0, + "content": "choose the value in the ideal scene, that is,", + "type": "text" + }, + { + "bbox": [ + 334, + 424, + 387, + 436 + ], + "score": 0.93, + "content": "\\mathbf { I } _ { \\mathbf { c } } = \\log ( K )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 425, + 390, + 436 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 150, + 439, + 358, + 457 + ], + "spans": [ + { + "bbox": [ + 150, + 439, + 222, + 457 + ], + "score": 1.0, + "content": "• Calculation of", + "type": "text" + }, + { + "bbox": [ + 222, + 442, + 358, + 455 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 162, + 453, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 162, + 454, + 207, + 466 + ], + "score": 0.91, + "content": "p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 453, + 386, + 467 + ], + "score": 1.0, + "content": "has two forms: (1) normal distribution with", + "type": "text" + }, + { + "bbox": [ + 387, + 454, + 445, + 466 + ], + "score": 0.91, + "content": "\\mathcal { N } ( f _ { \\boldsymbol { \\theta } } ( \\mathbf { c } , \\mathbf { z } ) , \\mathbf { I } )", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 453, + 506, + 467 + ], + "score": 1.0, + "content": "and (2) multi-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 161, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 161, + 465, + 262, + 478 + ], + "score": 1.0, + "content": "nomial distribution with", + "type": "text" + }, + { + "bbox": [ + 263, + 465, + 360, + 477 + ], + "score": 0.89, + "content": "\\mathbf { M u l t } ( d i m ( \\mathbf { X } ) , f _ { \\theta } ( \\mathbf { c } , \\mathbf { z } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 465, + 505, + 478 + ], + "score": 1.0, + "content": ". For the two forms, the loss func-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 161, + 475, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 161, + 475, + 192, + 488 + ], + "score": 1.0, + "content": "tion of", + "type": "text" + }, + { + "bbox": [ + 192, + 476, + 261, + 488 + ], + "score": 0.93, + "content": "- \\log p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 475, + 505, + 488 + ], + "score": 1.0, + "content": "is mean square error(MSE) and binary cross entropy(BCE)", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 162, + 487, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 162, + 487, + 505, + 499 + ], + "score": 1.0, + "content": "respectively. We use reparameterization trick in (20) to approximate expectation, and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 162, + 498, + 277, + 510 + ], + "spans": [ + { + "bbox": [ + 162, + 498, + 277, + 510 + ], + "score": 1.0, + "content": "the sampling frequency is 1.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 152, + 515, + 228, + 529 + ], + "spans": [ + { + "bbox": [ + 152, + 515, + 162, + 529 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 162, + 518, + 170, + 525 + ], + "score": 0.73, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 515, + 182, + 529 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 182, + 516, + 228, + 527 + ], + "score": 0.9, + "content": "p _ { m i x u p } ( \\mathbf { X } )", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 161, + 525, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 161, + 525, + 181, + 540 + ], + "score": 1.0, + "content": "The", + "type": "text" + }, + { + "bbox": [ + 181, + 527, + 213, + 538 + ], + "score": 0.92, + "content": "\\beta ( \\alpha , \\alpha )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 525, + 505, + 540 + ], + "score": 1.0, + "content": "for mixup vicinal distribution will strongly affect SSL performance. We", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 162, + 538, + 278, + 550 + ], + "spans": [ + { + "bbox": [ + 162, + 538, + 278, + 550 + ], + "score": 1.0, + "content": "set it to 2 in all experiments.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 131, + 562, + 307, + 573 + ], + "lines": [ + { + "bbox": [ + 128, + 559, + 308, + 577 + ], + "spans": [ + { + "bbox": [ + 128, + 559, + 308, + 577 + ], + "score": 1.0, + "content": "2. Items related to the optimization strategy", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 152, + 586, + 505, + 706 + ], + "lines": [ + { + "bbox": [ + 152, + 587, + 205, + 599 + ], + "spans": [ + { + "bbox": [ + 152, + 587, + 205, + 599 + ], + "score": 1.0, + "content": "• Optimizer", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 160, + 597, + 504, + 610 + ], + "spans": [ + { + "bbox": [ + 160, + 597, + 504, + 610 + ], + "score": 1.0, + "content": "In one-stage VAE, we use Adam. In OSPOT-VAE, we use SGD with momentum 0.9.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 153, + 614, + 218, + 628 + ], + "spans": [ + { + "bbox": [ + 153, + 614, + 218, + 628 + ], + "score": 1.0, + "content": "• Learning rate", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 161, + 626, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 161, + 626, + 504, + 639 + ], + "score": 1.0, + "content": "We set the initial learning rate to 0.1 in OSPOT-VAE, which obtains best SSL perfor-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 162, + 638, + 193, + 650 + ], + "spans": [ + { + "bbox": [ + 162, + 638, + 193, + 650 + ], + "score": 1.0, + "content": "mance.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 159, + 654, + 324, + 668 + ], + "spans": [ + { + "bbox": [ + 159, + 654, + 324, + 668 + ], + "score": 1.0, + "content": "The scheduler of adjusting learning rate", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 161, + 665, + 471, + 679 + ], + "spans": [ + { + "bbox": [ + 161, + 665, + 471, + 679 + ], + "score": 1.0, + "content": "We decay the learning ratio in some milestones with the specified decay rate.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 152, + 682, + 219, + 698 + ], + "spans": [ + { + "bbox": [ + 152, + 682, + 219, + 698 + ], + "score": 1.0, + "content": "• Weight decay", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 161, + 694, + 388, + 708 + ], + "spans": [ + { + "bbox": [ + 161, + 694, + 314, + 708 + ], + "score": 1.0, + "content": "Weight decay controls the strength of", + "type": "text" + }, + { + "bbox": [ + 314, + 697, + 326, + 706 + ], + "score": 0.89, + "content": "L _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 694, + 388, + 708 + ], + "score": 1.0, + "content": "regularization.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 442, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 443, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 443, + 733 + ], + "score": 1.0, + "content": "We also list some standard training curves on benchmark datasets in Figure 3 and 4.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 99, + 525, + 214 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 233, + 80, + 378, + 92 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 232, + 79, + 379, + 93 + ], + "spans": [ + { + "bbox": [ + 232, + 79, + 379, + 93 + ], + "score": 1.0, + "content": "Table 7: Details of Training Process", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 106, + 99, + 525, + 214 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 99, + 525, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 99, + 525, + 214 + ], + "score": 0.981, + "html": "
MNISTSVHN(one-stage)SVHNCifar10Cifar100
Latent Dim(z/c)10/1032/10128/10128/10128/100
Mutual Info(z/c)17.5/17.050/501280/2.3200/2.31280/4.6
loss of -log pe(X|c,z)BCEBCEMSEMSEBCE
a of pmixup(X)一一222
optimizerAdamSGD
learning rate5e-41e-30.1
lr scheduler(decay ratio)一一every 50 epoch after 200-th(0.5)[500,600,650](0.2)
weight decay005e-4
", + "type": "table", + "image_path": "6570e9dfa5eec3f1555477b23cafd74049c581d87a93e5e2dcffad761cee71c9.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 99, + 525, + 137.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 137.33333333333334, + 525, + 175.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 175.66666666666669, + 525, + 214.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "title", + "bbox": [ + 108, + 243, + 272, + 255 + ], + "lines": [ + { + "bbox": [ + 106, + 243, + 273, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 273, + 256 + ], + "score": 1.0, + "content": "A.6 DETAILS OF TRAINING PROCESS", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 268, + 505, + 302 + ], + "lines": [ + { + "bbox": [ + 106, + 268, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 505, + 280 + ], + "score": 1.0, + "content": "Here we list some important items need to set in the training process for different process. We", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 279, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 279, + 505, + 291 + ], + "score": 1.0, + "content": "classify these items into 2 categories: (1) items related to the loss function and (2) items related to", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 290, + 467, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 467, + 302 + ], + "score": 1.0, + "content": "the optimization strategy. The details are as follows and we list the exact value in Table 7:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 106, + 268, + 505, + 302 + ] + }, + { + "type": "text", + "bbox": [ + 131, + 316, + 274, + 327 + ], + "lines": [ + { + "bbox": [ + 131, + 316, + 274, + 327 + ], + "spans": [ + { + "bbox": [ + 131, + 316, + 274, + 327 + ], + "score": 1.0, + "content": "1. Items related to the loss function", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 131, + 316, + 274, + 327 + ] + }, + { + "type": "list", + "bbox": [ + 151, + 340, + 506, + 549 + ], + "lines": [ + { + "bbox": [ + 152, + 341, + 208, + 351 + ], + "spans": [ + { + "bbox": [ + 152, + 341, + 208, + 351 + ], + "score": 1.0, + "content": "• Latent dim", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 161, + 350, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 161, + 350, + 506, + 363 + ], + "score": 1.0, + "content": "The latent dim of discrete variable c is the same as the number of classifications, that", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 162, + 362, + 475, + 374 + ], + "spans": [ + { + "bbox": [ + 162, + 363, + 174, + 374 + ], + "score": 1.0, + "content": "is,", + "type": "text" + }, + { + "bbox": [ + 174, + 362, + 184, + 372 + ], + "score": 0.73, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 363, + 285, + 374 + ], + "score": 1.0, + "content": ". For continuous variable", + "type": "text" + }, + { + "bbox": [ + 285, + 364, + 292, + 372 + ], + "score": 0.65, + "content": "\\mathbf { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 363, + 475, + 374 + ], + "score": 1.0, + "content": ", the latent dim is determined by experiments.", + "type": "text" + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 153, + 380, + 243, + 392 + ], + "spans": [ + { + "bbox": [ + 153, + 380, + 243, + 392 + ], + "score": 1.0, + "content": "• Mutual information", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 162, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 162, + 390, + 505, + 403 + ], + "score": 1.0, + "content": "We find the value of continuous mutual information will affect the generative per-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 162, + 402, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 162, + 402, + 506, + 415 + ], + "score": 1.0, + "content": "formance, but have little impact on semi-supervised learning results, so we choose a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 162, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 162, + 414, + 505, + 425 + ], + "score": 1.0, + "content": "suitable value to get the best generative performance. For discrete information, we", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 162, + 424, + 390, + 436 + ], + "spans": [ + { + "bbox": [ + 162, + 425, + 334, + 436 + ], + "score": 1.0, + "content": "choose the value in the ideal scene, that is,", + "type": "text" + }, + { + "bbox": [ + 334, + 424, + 387, + 436 + ], + "score": 0.93, + "content": "\\mathbf { I } _ { \\mathbf { c } } = \\log ( K )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 425, + 390, + 436 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 150, + 439, + 358, + 457 + ], + "spans": [ + { + "bbox": [ + 150, + 439, + 222, + 457 + ], + "score": 1.0, + "content": "• Calculation of", + "type": "text" + }, + { + "bbox": [ + 222, + 442, + 358, + 455 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + } + ], + "index": 17, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 162, + 453, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 162, + 454, + 207, + 466 + ], + "score": 0.91, + "content": "p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 453, + 386, + 467 + ], + "score": 1.0, + "content": "has two forms: (1) normal distribution with", + "type": "text" + }, + { + "bbox": [ + 387, + 454, + 445, + 466 + ], + "score": 0.91, + "content": "\\mathcal { N } ( f _ { \\boldsymbol { \\theta } } ( \\mathbf { c } , \\mathbf { z } ) , \\mathbf { I } )", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 453, + 506, + 467 + ], + "score": 1.0, + "content": "and (2) multi-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 161, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 161, + 465, + 262, + 478 + ], + "score": 1.0, + "content": "nomial distribution with", + "type": "text" + }, + { + "bbox": [ + 263, + 465, + 360, + 477 + ], + "score": 0.89, + "content": "\\mathbf { M u l t } ( d i m ( \\mathbf { X } ) , f _ { \\theta } ( \\mathbf { c } , \\mathbf { z } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 465, + 505, + 478 + ], + "score": 1.0, + "content": ". For the two forms, the loss func-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 161, + 475, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 161, + 475, + 192, + 488 + ], + "score": 1.0, + "content": "tion of", + "type": "text" + }, + { + "bbox": [ + 192, + 476, + 261, + 488 + ], + "score": 0.93, + "content": "- \\log p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 475, + 505, + 488 + ], + "score": 1.0, + "content": "is mean square error(MSE) and binary cross entropy(BCE)", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 162, + 487, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 162, + 487, + 505, + 499 + ], + "score": 1.0, + "content": "respectively. We use reparameterization trick in (20) to approximate expectation, and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 162, + 498, + 277, + 510 + ], + "spans": [ + { + "bbox": [ + 162, + 498, + 277, + 510 + ], + "score": 1.0, + "content": "the sampling frequency is 1.", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 152, + 515, + 228, + 529 + ], + "spans": [ + { + "bbox": [ + 152, + 515, + 162, + 529 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 162, + 518, + 170, + 525 + ], + "score": 0.73, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 515, + 182, + 529 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 182, + 516, + 228, + 527 + ], + "score": 0.9, + "content": "p _ { m i x u p } ( \\mathbf { X } )", + "type": "inline_equation" + } + ], + "index": 23, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 161, + 525, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 161, + 525, + 181, + 540 + ], + "score": 1.0, + "content": "The", + "type": "text" + }, + { + "bbox": [ + 181, + 527, + 213, + 538 + ], + "score": 0.92, + "content": "\\beta ( \\alpha , \\alpha )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 525, + 505, + 540 + ], + "score": 1.0, + "content": "for mixup vicinal distribution will strongly affect SSL performance. We", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 162, + 538, + 278, + 550 + ], + "spans": [ + { + "bbox": [ + 162, + 538, + 278, + 550 + ], + "score": 1.0, + "content": "set it to 2 in all experiments.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + } + ], + "index": 17, + "bbox_fs": [ + 150, + 341, + 506, + 550 + ] + }, + { + "type": "text", + "bbox": [ + 131, + 562, + 307, + 573 + ], + "lines": [ + { + "bbox": [ + 128, + 559, + 308, + 577 + ], + "spans": [ + { + "bbox": [ + 128, + 559, + 308, + 577 + ], + "score": 1.0, + "content": "2. Items related to the optimization strategy", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 128, + 559, + 308, + 577 + ] + }, + { + "type": "list", + "bbox": [ + 152, + 586, + 505, + 706 + ], + "lines": [ + { + "bbox": [ + 152, + 587, + 205, + 599 + ], + "spans": [ + { + "bbox": [ + 152, + 587, + 205, + 599 + ], + "score": 1.0, + "content": "• Optimizer", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 160, + 597, + 504, + 610 + ], + "spans": [ + { + "bbox": [ + 160, + 597, + 504, + 610 + ], + "score": 1.0, + "content": "In one-stage VAE, we use Adam. In OSPOT-VAE, we use SGD with momentum 0.9.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 153, + 614, + 218, + 628 + ], + "spans": [ + { + "bbox": [ + 153, + 614, + 218, + 628 + ], + "score": 1.0, + "content": "• Learning rate", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 161, + 626, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 161, + 626, + 504, + 639 + ], + "score": 1.0, + "content": "We set the initial learning rate to 0.1 in OSPOT-VAE, which obtains best SSL perfor-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 162, + 638, + 193, + 650 + ], + "spans": [ + { + "bbox": [ + 162, + 638, + 193, + 650 + ], + "score": 1.0, + "content": "mance.", + "type": "text" + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 159, + 654, + 324, + 668 + ], + "spans": [ + { + "bbox": [ + 159, + 654, + 324, + 668 + ], + "score": 1.0, + "content": "The scheduler of adjusting learning rate", + "type": "text" + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 161, + 665, + 471, + 679 + ], + "spans": [ + { + "bbox": [ + 161, + 665, + 471, + 679 + ], + "score": 1.0, + "content": "We decay the learning ratio in some milestones with the specified decay rate.", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 152, + 682, + 219, + 698 + ], + "spans": [ + { + "bbox": [ + 152, + 682, + 219, + 698 + ], + "score": 1.0, + "content": "• Weight decay", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 161, + 694, + 388, + 708 + ], + "spans": [ + { + "bbox": [ + 161, + 694, + 314, + 708 + ], + "score": 1.0, + "content": "Weight decay controls the strength of", + "type": "text" + }, + { + "bbox": [ + 314, + 697, + 326, + 706 + ], + "score": 0.89, + "content": "L _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 694, + 388, + 708 + ], + "score": 1.0, + "content": "regularization.", + "type": "text" + } + ], + "index": 35, + "is_list_end_line": true + } + ], + "index": 31, + "bbox_fs": [ + 152, + 587, + 504, + 708 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 442, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 443, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 443, + 733 + ], + "score": 1.0, + "content": "We also list some standard training curves on benchmark datasets in Figure 3 and 4.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36, + "bbox_fs": [ + 106, + 720, + 443, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 82, + 501, + 209 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 82, + 501, + 209 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 82, + 501, + 209 + ], + "spans": [ + { + "bbox": [ + 108, + 82, + 501, + 209 + ], + "score": 0.958, + "type": "image", + "image_path": "cb00a1218b54c04419c02f3df09a2f7602585690b55a8620fed2b95acb2b231b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 82, + 501, + 124.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 124.33333333333334, + 501, + 166.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 166.66666666666669, + 501, + 209.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 155, + 219, + 454, + 244 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 221, + 218, + 388, + 231 + ], + "spans": [ + { + "bbox": [ + 221, + 218, + 388, + 231 + ], + "score": 1.0, + "content": "Figure 3: The training curves of Cifar10.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 152, + 227, + 453, + 247 + ], + "spans": [ + { + "bbox": [ + 152, + 227, + 316, + 247 + ], + "score": 1.0, + "content": "Left: classification performance. Right:", + "type": "text" + }, + { + "bbox": [ + 316, + 230, + 453, + 244 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "image", + "bbox": [ + 107, + 288, + 500, + 414 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 288, + 500, + 414 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 288, + 500, + 414 + ], + "spans": [ + { + "bbox": [ + 107, + 288, + 500, + 414 + ], + "score": 0.959, + "type": "image", + "image_path": "c99910939dd2e58faf73c25c385af2b0e5d3daf4d34a0192b639b515b048f86f.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 107, + 288, + 500, + 330.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 107, + 330.0, + 500, + 372.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 107, + 372.0, + 500, + 414.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 150, + 423, + 457, + 448 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 219, + 422, + 391, + 435 + ], + "spans": [ + { + "bbox": [ + 219, + 422, + 391, + 435 + ], + "score": 1.0, + "content": "Figure 4: The training curves of Cifar100.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 153, + 431, + 453, + 451 + ], + "spans": [ + { + "bbox": [ + 153, + 431, + 316, + 451 + ], + "score": 1.0, + "content": "Left: classification performance. Right:", + "type": "text" + }, + { + "bbox": [ + 316, + 434, + 453, + 448 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + } + ], + "index": 9 + } + ], + "index": 8.5 + } + ], + "index": 7.25 + }, + { + "type": "title", + "bbox": [ + 108, + 476, + 258, + 487 + ], + "lines": [ + { + "bbox": [ + 106, + 476, + 259, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 259, + 488 + ], + "score": 1.0, + "content": "A.7 GENERATIVE PERFORMANCE", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 411, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 412, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 412, + 514 + ], + "score": 1.0, + "content": "Following figures5-7 show the generation performance of our OSPOT-VAE.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 126, + 532, + 484, + 708 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 532, + 484, + 708 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 126, + 532, + 484, + 708 + ], + "spans": [ + { + "bbox": [ + 126, + 532, + 484, + 708 + ], + "score": 0.976, + "type": "image", + "image_path": "25e8427bc09e3e5a473bfd67e0a4ebd896f3b330cc776a7003253b54f424ad84.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 126, + 532, + 484, + 590.6666666666666 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 126, + 590.6666666666666, + 484, + 649.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 126, + 649.3333333333333, + 484, + 707.9999999999999 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 150, + 717, + 460, + 729 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 149, + 716, + 461, + 730 + ], + "spans": [ + { + "bbox": [ + 149, + 716, + 461, + 730 + ], + "score": 1.0, + "content": "Figure 5: The generative performance of OSPOT-VAE in MNIST and SVHN", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 14.0 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 82, + 501, + 209 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 82, + 501, + 209 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 82, + 501, + 209 + ], + "spans": [ + { + "bbox": [ + 108, + 82, + 501, + 209 + ], + "score": 0.958, + "type": "image", + "image_path": "cb00a1218b54c04419c02f3df09a2f7602585690b55a8620fed2b95acb2b231b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 82, + 501, + 124.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 124.33333333333334, + 501, + 166.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 166.66666666666669, + 501, + 209.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 155, + 219, + 454, + 244 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 221, + 218, + 388, + 231 + ], + "spans": [ + { + "bbox": [ + 221, + 218, + 388, + 231 + ], + "score": 1.0, + "content": "Figure 3: The training curves of Cifar10.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 152, + 227, + 453, + 247 + ], + "spans": [ + { + "bbox": [ + 152, + 227, + 316, + 247 + ], + "score": 1.0, + "content": "Left: classification performance. Right:", + "type": "text" + }, + { + "bbox": [ + 316, + 230, + 453, + 244 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "image", + "bbox": [ + 107, + 288, + 500, + 414 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 288, + 500, + 414 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 288, + 500, + 414 + ], + "spans": [ + { + "bbox": [ + 107, + 288, + 500, + 414 + ], + "score": 0.959, + "type": "image", + "image_path": "c99910939dd2e58faf73c25c385af2b0e5d3daf4d34a0192b639b515b048f86f.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 107, + 288, + 500, + 330.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 107, + 330.0, + 500, + 372.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 107, + 372.0, + 500, + 414.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 150, + 423, + 457, + 448 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 219, + 422, + 391, + 435 + ], + "spans": [ + { + "bbox": [ + 219, + 422, + 391, + 435 + ], + "score": 1.0, + "content": "Figure 4: The training curves of Cifar100.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 153, + 431, + 453, + 451 + ], + "spans": [ + { + "bbox": [ + 153, + 431, + 316, + 451 + ], + "score": 1.0, + "content": "Left: classification performance. Right:", + "type": "text" + }, + { + "bbox": [ + 316, + 434, + 453, + 448 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )", + "type": "inline_equation" + } + ], + "index": 9 + } + ], + "index": 8.5 + } + ], + "index": 7.25 + }, + { + "type": "title", + "bbox": [ + 108, + 476, + 258, + 487 + ], + "lines": [ + { + "bbox": [ + 106, + 476, + 259, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 259, + 488 + ], + "score": 1.0, + "content": "A.7 GENERATIVE PERFORMANCE", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 411, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 412, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 412, + 514 + ], + "score": 1.0, + "content": "Following figures5-7 show the generation performance of our OSPOT-VAE.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 500, + 412, + 514 + ] + }, + { + "type": "image", + "bbox": [ + 126, + 532, + 484, + 708 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 532, + 484, + 708 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 126, + 532, + 484, + 708 + ], + "spans": [ + { + "bbox": [ + 126, + 532, + 484, + 708 + ], + "score": 0.976, + "type": "image", + "image_path": "25e8427bc09e3e5a473bfd67e0a4ebd896f3b330cc776a7003253b54f424ad84.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 126, + 532, + 484, + 590.6666666666666 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 126, + 590.6666666666666, + 484, + 649.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 126, + 649.3333333333333, + 484, + 707.9999999999999 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 150, + 717, + 460, + 729 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 149, + 716, + 461, + 730 + ], + "spans": [ + { + "bbox": [ + 149, + 716, + 461, + 730 + ], + "score": 1.0, + "content": "Figure 5: The generative performance of OSPOT-VAE in MNIST and SVHN", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 14.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 126, + 80, + 484, + 255 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 80, + 484, + 255 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 80, + 484, + 255 + ], + "spans": [ + { + "bbox": [ + 126, + 80, + 484, + 255 + ], + "score": 0.969, + "type": "image", + "image_path": "ee3780017d783d4bf7dd38dea93a2d2af64e16334a08785719316d9fd5e1961d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 126, + 80, + 484, + 138.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 126, + 138.33333333333334, + 484, + 196.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 126, + 196.66666666666669, + 484, + 255.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 123, + 264, + 484, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 264, + 485, + 278 + ], + "spans": [ + { + "bbox": [ + 126, + 264, + 485, + 278 + ], + "score": 1.0, + "content": "Figure 6: Compare the generative performance of pure VAE and OSPOT-VAE in Cifar10", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "image", + "bbox": [ + 127, + 299, + 484, + 475 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 127, + 299, + 484, + 475 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 127, + 299, + 484, + 475 + ], + "spans": [ + { + "bbox": [ + 127, + 299, + 484, + 475 + ], + "score": 0.968, + "type": "image", + "image_path": "5871b177cdb29994688d651de1f8c9d3eec78a3fb02e32f4b5033317a4d4ac6b.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 127, + 299, + 484, + 357.6666666666667 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 127, + 357.6666666666667, + 484, + 416.33333333333337 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 127, + 416.33333333333337, + 484, + 475.00000000000006 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 121, + 484, + 487, + 497 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 123, + 484, + 487, + 497 + ], + "spans": [ + { + "bbox": [ + 123, + 484, + 487, + 497 + ], + "score": 1.0, + "content": "Figure 7: Compare the generative performance of pure VAE and OSPOT-VAE in Cifar100", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + } + ], + "index": 6.0 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 126, + 80, + 484, + 255 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 80, + 484, + 255 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 80, + 484, + 255 + ], + "spans": [ + { + "bbox": [ + 126, + 80, + 484, + 255 + ], + "score": 0.969, + "type": "image", + "image_path": "ee3780017d783d4bf7dd38dea93a2d2af64e16334a08785719316d9fd5e1961d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 126, + 80, + 484, + 138.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 126, + 138.33333333333334, + 484, + 196.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 126, + 196.66666666666669, + 484, + 255.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 123, + 264, + 484, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 264, + 485, + 278 + ], + "spans": [ + { + "bbox": [ + 126, + 264, + 485, + 278 + ], + "score": 1.0, + "content": "Figure 6: Compare the generative performance of pure VAE and OSPOT-VAE in Cifar10", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "image", + "bbox": [ + 127, + 299, + 484, + 475 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 127, + 299, + 484, + 475 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 127, + 299, + 484, + 475 + ], + "spans": [ + { + "bbox": [ + 127, + 299, + 484, + 475 + ], + "score": 0.968, + "type": "image", + "image_path": "5871b177cdb29994688d651de1f8c9d3eec78a3fb02e32f4b5033317a4d4ac6b.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 127, + 299, + 484, + 357.6666666666667 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 127, + 357.6666666666667, + 484, + 416.33333333333337 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 127, + 416.33333333333337, + 484, + 475.00000000000006 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 121, + 484, + 487, + 497 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 123, + 484, + 487, + 497 + ], + "spans": [ + { + "bbox": [ + 123, + 484, + 487, + 497 + ], + "score": 1.0, + "content": "Figure 7: Compare the generative performance of pure VAE and OSPOT-VAE in Cifar100", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + } + ], + "index": 6.0 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/S1ejj64YvS/S1ejj64YvS_model.json b/parse/train/S1ejj64YvS/S1ejj64YvS_model.json new file mode 100644 index 0000000000000000000000000000000000000000..c171cb5d85f248fee44565e46475a0382be006e5 --- /dev/null +++ b/parse/train/S1ejj64YvS/S1ejj64YvS_model.json @@ -0,0 +1,23666 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 589, + 1302, + 589, + 1302, + 1046, + 398, + 1046 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1377, + 1403, + 1377, + 1403, + 1713, + 298, + 1713 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1728, + 1403, + 1728, + 1403, + 2034, + 298, + 2034 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1177, + 1403, + 1177, + 1403, + 1360, + 299, + 1360 + ], + "score": 0.98 + }, + { + "category_id": 0, + "poly": [ + 300, + 218, + 1399, + 218, + 1399, + 324, + 300, + 324 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 313, + 378, + 680, + 378, + 680, + 438, + 313, + 438 + ], + "score": 0.931 + }, + { + "category_id": 0, + "poly": [ + 302, + 1107, + 573, + 1107, + 573, + 1142, + 302, + 1142 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 773, + 519, + 927, + 519, + 927, + 553, + 773, + 553 + ], + "score": 0.877 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 857, + 2088, + 857, + 2112, + 842, + 2112 + ], + "score": 0.661 + }, + { + "category_id": 13, + "poly": [ + 1190, + 925, + 1265, + 925, + 1265, + 954, + 1190, + 954 + ], + "score": 0.88, + "latex": "6 . 1 1 \\%" + }, + { + "category_id": 13, + "poly": [ + 1177, + 956, + 1266, + 956, + 1266, + 985, + 1177, + 985 + ], + "score": 0.88, + "latex": "2 5 . 3 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1067, + 925, + 1155, + 925, + 1155, + 954, + 1067, + 954 + ], + "score": 0.87, + "latex": "1 4 . 4 1 \\%" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 214.0, + 1403.0, + 214.0, + 1403.0, + 275.0, + 294.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 272.0, + 851.0, + 272.0, + 851.0, + 329.0, + 294.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1104.0, + 579.0, + 1104.0, + 579.0, + 1151.0, + 294.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 518.0, + 932.0, + 518.0, + 932.0, + 556.0, + 770.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 2088.0, + 858.0, + 2088.0, + 858.0, + 2116.0, + 841.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 589.0, + 1306.0, + 589.0, + 1306.0, + 624.0, + 393.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 620.0, + 1306.0, + 620.0, + 1306.0, + 655.0, + 393.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 650.0, + 1307.0, + 650.0, + 1307.0, + 687.0, + 393.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 681.0, + 1304.0, + 681.0, + 1304.0, + 716.0, + 394.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 712.0, + 1306.0, + 712.0, + 1306.0, + 744.0, + 392.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 743.0, + 1307.0, + 743.0, + 1307.0, + 778.0, + 394.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 770.0, + 1306.0, + 770.0, + 1306.0, + 807.0, + 393.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 802.0, + 1307.0, + 802.0, + 1307.0, + 837.0, + 393.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 833.0, + 1308.0, + 833.0, + 1308.0, + 868.0, + 392.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 863.0, + 1305.0, + 863.0, + 1305.0, + 897.0, + 393.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 894.0, + 1306.0, + 894.0, + 1306.0, + 929.0, + 394.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 925.0, + 1066.0, + 925.0, + 1066.0, + 957.0, + 395.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 925.0, + 1189.0, + 925.0, + 1189.0, + 957.0, + 1156.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 925.0, + 1305.0, + 925.0, + 1305.0, + 957.0, + 1266.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 955.0, + 1176.0, + 955.0, + 1176.0, + 990.0, + 395.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 955.0, + 1306.0, + 955.0, + 1306.0, + 990.0, + 1267.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 984.0, + 1305.0, + 984.0, + 1305.0, + 1018.0, + 395.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1017.0, + 1213.0, + 1017.0, + 1213.0, + 1049.0, + 394.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1378.0, + 1405.0, + 1378.0, + 1405.0, + 1412.0, + 294.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1406.0, + 1404.0, + 1406.0, + 1404.0, + 1443.0, + 293.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1435.0, + 1404.0, + 1435.0, + 1404.0, + 1477.0, + 292.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1466.0, + 1404.0, + 1466.0, + 1404.0, + 1505.0, + 292.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1498.0, + 1404.0, + 1498.0, + 1404.0, + 1536.0, + 293.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1530.0, + 1405.0, + 1530.0, + 1405.0, + 1565.0, + 294.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1562.0, + 1404.0, + 1562.0, + 1404.0, + 1593.0, + 296.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1591.0, + 1403.0, + 1591.0, + 1403.0, + 1626.0, + 294.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1621.0, + 1405.0, + 1621.0, + 1405.0, + 1655.0, + 294.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1653.0, + 1405.0, + 1653.0, + 1405.0, + 1687.0, + 294.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1682.0, + 1054.0, + 1682.0, + 1054.0, + 1717.0, + 294.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1726.0, + 1404.0, + 1726.0, + 1404.0, + 1766.0, + 293.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1759.0, + 1404.0, + 1759.0, + 1404.0, + 1795.0, + 294.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1785.0, + 1407.0, + 1785.0, + 1407.0, + 1828.0, + 292.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1820.0, + 1405.0, + 1820.0, + 1405.0, + 1856.0, + 293.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1849.0, + 1405.0, + 1849.0, + 1405.0, + 1885.0, + 293.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1882.0, + 1405.0, + 1882.0, + 1405.0, + 1915.0, + 293.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1910.0, + 1405.0, + 1910.0, + 1405.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1403.0, + 1942.0, + 1403.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1975.0, + 1404.0, + 1975.0, + 1404.0, + 2007.0, + 296.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 1403.0, + 2002.0, + 1403.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1174.0, + 1407.0, + 1174.0, + 1407.0, + 1217.0, + 293.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1208.0, + 1407.0, + 1208.0, + 1407.0, + 1243.0, + 293.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1239.0, + 1405.0, + 1239.0, + 1405.0, + 1275.0, + 295.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1269.0, + 1403.0, + 1269.0, + 1403.0, + 1304.0, + 293.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1300.0, + 1405.0, + 1300.0, + 1405.0, + 1333.0, + 293.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1330.0, + 974.0, + 1330.0, + 974.0, + 1363.0, + 295.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 379.0, + 560.0, + 379.0, + 560.0, + 411.0, + 315.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 407.0, + 681.0, + 407.0, + 681.0, + 442.0, + 312.0, + 442.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1749, + 1403, + 1749, + 1403, + 1903, + 298, + 1903 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1331, + 1403, + 1331, + 1403, + 1516, + 298, + 1516 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 673, + 1405, + 673, + 1405, + 858, + 298, + 858 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1531, + 1403, + 1531, + 1403, + 1654, + 298, + 1654 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 297, + 535, + 1405, + 535, + 1405, + 659, + 297, + 659 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 296, + 145, + 1404, + 145, + 1404, + 458, + 296, + 458 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 298, + 1972, + 1405, + 1972, + 1405, + 2035, + 298, + 2035 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 368, + 934, + 1404, + 934, + 1404, + 1204, + 368, + 1204 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 655, + 1918, + 1047, + 1918, + 1047, + 1958, + 655, + 1958 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 300, + 874, + 778, + 874, + 778, + 906, + 300, + 906 + ], + "score": 0.927 + }, + { + "category_id": 4, + "poly": [ + 620, + 482, + 1081, + 482, + 1081, + 516, + 620, + 516 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 299, + 1261, + 933, + 1261, + 933, + 1297, + 299, + 1297 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 302, + 1691, + 784, + 1691, + 784, + 1723, + 302, + 1723 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1924, + 1399, + 1924, + 1399, + 1953, + 1366, + 1953 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 296, + 75, + 858, + 75, + 858, + 104, + 296, + 104 + ], + "score": 0.825 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.727 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.193 + }, + { + "category_id": 2, + "poly": [ + 296, + 75, + 858, + 75, + 858, + 104, + 296, + 104 + ], + "score": 0.107 + }, + { + "category_id": 13, + "poly": [ + 924, + 1363, + 1016, + 1363, + 1016, + 1396, + 924, + 1396 + ], + "score": 0.93, + "latex": "p ( \\mathbf { X } , \\mathbf { y } )" + }, + { + "category_id": 13, + "poly": [ + 820, + 1454, + 882, + 1454, + 882, + 1487, + 820, + 1487 + ], + "score": 0.91, + "latex": "p ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 935, + 1872, + 1033, + 1872, + 1033, + 1903, + 935, + 1903 + ], + "score": 0.91, + "latex": "\\mathbf { X } \\in \\mathbb { D } _ { U }" + }, + { + "category_id": 13, + "poly": [ + 1331, + 1393, + 1393, + 1393, + 1393, + 1427, + 1331, + 1427 + ], + "score": 0.91, + "latex": "p ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1395, + 394, + 1395, + 394, + 1426, + 298, + 1426 + ], + "score": 0.91, + "latex": "f ( \\mathbf { X } ; \\phi )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1454, + 404, + 1454, + 404, + 1485, + 297, + 1485 + ], + "score": 0.9, + "latex": "\\mathbf { X } \\in \\mathbb { D } _ { U }" + }, + { + "category_id": 14, + "poly": [ + 652, + 1917, + 1047, + 1917, + 1047, + 1958, + 652, + 1958 + ], + "score": 0.9, + "latex": "\\| f ( \\operatorname { A u g m e n t } ( \\mathbf { X } ) ; \\phi ) - f ( \\mathbf { X } ; \\phi ) \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1362, + 461, + 1362, + 461, + 1395, + 298, + 1395 + ], + "score": 0.89, + "latex": "( \\mathbf { X } , \\bar { \\mathbf { y } } ) \\ \\in \\ \\mathbb { D } _ { L }" + }, + { + "category_id": 13, + "poly": [ + 1334, + 1486, + 1374, + 1486, + 1374, + 1516, + 1334, + 1516 + ], + "score": 0.87, + "latex": "\\mathbb { D } _ { L }" + }, + { + "category_id": 13, + "poly": [ + 344, + 1486, + 385, + 1486, + 385, + 1515, + 344, + 1515 + ], + "score": 0.87, + "latex": "\\mathbb { D } _ { U }" + }, + { + "category_id": 13, + "poly": [ + 1361, + 1455, + 1400, + 1455, + 1400, + 1485, + 1361, + 1485 + ], + "score": 0.87, + "latex": "\\mathbb { D } _ { L }" + }, + { + "category_id": 13, + "poly": [ + 606, + 1395, + 630, + 1395, + 630, + 1425, + 606, + 1425 + ], + "score": 0.84, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 516, + 1973, + 562, + 1973, + 562, + 2005, + 516, + 2005 + ], + "score": 0.66, + "latex": "( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1010, + 1399, + 1033, + 1399, + 1033, + 1425, + 1010, + 1425 + ], + "score": 0.37, + "latex": "\\mathbf { y }" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 158.0, + 364.0, + 158.0, + 364.0, + 228.0, + 332.0, + 228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 160.0, + 563.0, + 160.0, + 563.0, + 192.0, + 389.0, + 192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 150.0, + 904.0, + 150.0, + 904.0, + 226.0, + 609.0, + 226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 163.0, + 1025.0, + 163.0, + 1025.0, + 240.0, + 931.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 164.0, + 1101.0, + 164.0, + 1101.0, + 191.0, + 1026.0, + 191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 207.0, + 1255.0, + 207.0, + 1255.0, + 240.0, + 1120.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 207.0, + 1339.0, + 207.0, + 1339.0, + 237.0, + 1273.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 264.0, + 359.0, + 264.0, + 359.0, + 298.0, + 329.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 247.0, + 466.0, + 247.0, + 466.0, + 277.0, + 382.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 247.0, + 564.0, + 247.0, + 564.0, + 277.0, + 494.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 243.0, + 597.0, + 243.0, + 597.0, + 309.0, + 568.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 269.0, + 641.0, + 269.0, + 641.0, + 281.0, + 631.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 259.0, + 702.0, + 259.0, + 702.0, + 289.0, + 678.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 231.0, + 773.0, + 231.0, + 773.0, + 251.0, + 759.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 231.0, + 831.0, + 231.0, + 831.0, + 255.0, + 812.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 229.0, + 890.0, + 229.0, + 890.0, + 256.0, + 870.0, + 256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 249.0, + 1006.0, + 249.0, + 1006.0, + 277.0, + 933.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 247.0, + 1103.0, + 247.0, + 1103.0, + 272.0, + 1028.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 228.0, + 1193.0, + 228.0, + 1193.0, + 257.0, + 1117.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 237.0, + 1269.0, + 237.0, + 1269.0, + 282.0, + 1243.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 247.0, + 1352.0, + 247.0, + 1352.0, + 271.0, + 1333.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 291.0, + 819.0, + 291.0, + 819.0, + 327.0, + 788.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 276.0, + 912.0, + 276.0, + 912.0, + 314.0, + 839.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 281.0, + 1025.0, + 281.0, + 1025.0, + 307.0, + 996.0, + 307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 285.0, + 1118.0, + 285.0, + 1118.0, + 314.0, + 1085.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 295.0, + 1392.0, + 295.0, + 1392.0, + 323.0, + 1255.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 330.0, + 558.0, + 330.0, + 558.0, + 359.0, + 387.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 328.0, + 1009.0, + 328.0, + 1009.0, + 354.0, + 935.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 323.0, + 1104.0, + 323.0, + 1104.0, + 354.0, + 1031.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 322.0, + 1405.0, + 322.0, + 1405.0, + 354.0, + 1171.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 379.0, + 510.0, + 379.0, + 510.0, + 407.0, + 389.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 354.0, + 534.0, + 354.0, + 534.0, + 370.0, + 518.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 358.0, + 1026.0, + 358.0, + 1026.0, + 402.0, + 993.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 367.0, + 1121.0, + 367.0, + 1121.0, + 392.0, + 1090.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 351.0, + 1400.0, + 351.0, + 1400.0, + 379.0, + 1168.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 414.0, + 645.0, + 414.0, + 645.0, + 447.0, + 393.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 427.0, + 1241.0, + 427.0, + 1241.0, + 459.0, + 899.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 205.0, + 1114.0, + 205.0, + 1114.0, + 230.5, + 1088.0, + 230.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 298.0, + 1022.0, + 298.0, + 1022.0, + 322.5, + 998.0, + 322.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 349.5, + 1177.0, + 349.5, + 1177.0, + 370.0, + 1144.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 480.0, + 1081.0, + 480.0, + 1081.0, + 518.0, + 618.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1258.0, + 936.0, + 1258.0, + 936.0, + 1303.0, + 291.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1691.0, + 786.0, + 1691.0, + 786.0, + 1727.0, + 295.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 859.0, + 72.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 859.0, + 72.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1747.0, + 1407.0, + 1747.0, + 1407.0, + 1790.0, + 293.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1780.0, + 1404.0, + 1780.0, + 1404.0, + 1817.0, + 294.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1810.0, + 1405.0, + 1810.0, + 1405.0, + 1849.0, + 293.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1839.0, + 1407.0, + 1839.0, + 1407.0, + 1878.0, + 293.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1870.0, + 934.0, + 1870.0, + 934.0, + 1907.0, + 294.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1870.0, + 1349.0, + 1870.0, + 1349.0, + 1907.0, + 1034.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1329.0, + 1405.0, + 1329.0, + 1405.0, + 1369.0, + 293.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1362.0, + 923.0, + 1362.0, + 923.0, + 1398.0, + 462.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1362.0, + 1405.0, + 1362.0, + 1405.0, + 1398.0, + 1017.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1393.0, + 605.0, + 1393.0, + 605.0, + 1429.0, + 395.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 1393.0, + 1009.0, + 1393.0, + 1009.0, + 1429.0, + 631.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1393.0, + 1330.0, + 1393.0, + 1330.0, + 1429.0, + 1034.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1393.0, + 1404.0, + 1393.0, + 1404.0, + 1429.0, + 1394.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1422.0, + 1407.0, + 1422.0, + 1407.0, + 1458.0, + 294.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1451.0, + 296.0, + 1451.0, + 296.0, + 1490.0, + 293.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1451.0, + 819.0, + 1451.0, + 819.0, + 1490.0, + 405.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1451.0, + 1360.0, + 1451.0, + 1360.0, + 1490.0, + 883.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1451.0, + 1404.0, + 1451.0, + 1404.0, + 1490.0, + 1401.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1479.0, + 343.0, + 1479.0, + 343.0, + 1523.0, + 292.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 1479.0, + 1333.0, + 1479.0, + 1333.0, + 1523.0, + 386.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1479.0, + 1389.0, + 1479.0, + 1389.0, + 1523.0, + 1375.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 673.0, + 1406.0, + 673.0, + 1406.0, + 710.0, + 292.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 704.0, + 1405.0, + 704.0, + 1405.0, + 740.0, + 295.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 731.0, + 1406.0, + 731.0, + 1406.0, + 773.0, + 293.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 768.0, + 1403.0, + 768.0, + 1403.0, + 800.0, + 296.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 795.0, + 1408.0, + 795.0, + 1408.0, + 834.0, + 293.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 824.0, + 478.0, + 824.0, + 478.0, + 860.0, + 294.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1530.0, + 1407.0, + 1530.0, + 1407.0, + 1566.0, + 293.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1562.0, + 1404.0, + 1562.0, + 1404.0, + 1595.0, + 294.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1592.0, + 1407.0, + 1592.0, + 1407.0, + 1629.0, + 293.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1623.0, + 528.0, + 1623.0, + 528.0, + 1656.0, + 296.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 536.0, + 1403.0, + 536.0, + 1403.0, + 569.0, + 296.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 568.0, + 1403.0, + 568.0, + 1403.0, + 601.0, + 295.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 597.0, + 1405.0, + 597.0, + 1405.0, + 633.0, + 292.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 625.0, + 415.0, + 625.0, + 415.0, + 664.0, + 289.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1972.0, + 515.0, + 1972.0, + 515.0, + 2007.0, + 296.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 1972.0, + 1403.0, + 1972.0, + 1403.0, + 2007.0, + 563.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 1403.0, + 2002.0, + 1403.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 931.0, + 1280.0, + 931.0, + 1280.0, + 973.0, + 365.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 976.0, + 1405.0, + 976.0, + 1405.0, + 1012.0, + 368.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1009.0, + 1259.0, + 1009.0, + 1259.0, + 1042.0, + 393.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1050.0, + 1404.0, + 1050.0, + 1404.0, + 1083.0, + 370.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1079.0, + 1407.0, + 1079.0, + 1407.0, + 1115.0, + 394.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1112.0, + 1404.0, + 1112.0, + 1404.0, + 1144.0, + 395.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1142.0, + 1405.0, + 1142.0, + 1405.0, + 1174.0, + 395.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1172.0, + 1146.0, + 1172.0, + 1146.0, + 1204.0, + 394.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 872.0, + 781.0, + 872.0, + 781.0, + 910.0, + 296.0, + 910.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1726, + 1405, + 1726, + 1405, + 2036, + 296, + 2036 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 296, + 602, + 1403, + 602, + 1403, + 790, + 296, + 790 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1478, + 1404, + 1478, + 1404, + 1662, + 297, + 1662 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 945, + 1404, + 945, + 1404, + 1069, + 298, + 1069 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1249, + 1404, + 1249, + 1404, + 1374, + 298, + 1374 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 341, + 1407, + 341, + 1407, + 435, + 299, + 435 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 296, + 803, + 1400, + 803, + 1400, + 897, + 296, + 897 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 572, + 1385, + 1125, + 1385, + 1125, + 1473, + 572, + 1473 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 293, + 229, + 1401, + 229, + 1401, + 292, + 293, + 292 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 703, + 297, + 998, + 297, + 998, + 337, + 703, + 337 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 295, + 1117, + 1401, + 1117, + 1401, + 1180, + 295, + 1180 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 296, + 526, + 1400, + 526, + 1400, + 590, + 296, + 590 + ], + "score": 0.928 + }, + { + "category_id": 8, + "poly": [ + 387, + 903, + 1296, + 903, + 1296, + 940, + 387, + 940 + ], + "score": 0.926 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 858, + 74, + 858, + 105, + 297, + 105 + ], + "score": 0.919 + }, + { + "category_id": 8, + "poly": [ + 440, + 1669, + 1262, + 1669, + 1262, + 1713, + 440, + 1713 + ], + "score": 0.917 + }, + { + "category_id": 0, + "poly": [ + 301, + 469, + 688, + 469, + 688, + 501, + 301, + 501 + ], + "score": 0.911 + }, + { + "category_id": 8, + "poly": [ + 308, + 1074, + 1331, + 1074, + 1331, + 1113, + 308, + 1113 + ], + "score": 0.909 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1414, + 1400, + 1414, + 1400, + 1443, + 1366, + 1443 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1674, + 1400, + 1674, + 1400, + 1704, + 1366, + 1704 + ], + "score": 0.882 + }, + { + "category_id": 9, + "poly": [ + 1366, + 303, + 1400, + 303, + 1400, + 332, + 1366, + 332 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1222, + 1399, + 1222, + 1399, + 1248, + 1365, + 1248 + ], + "score": 0.875 + }, + { + "category_id": 9, + "poly": [ + 1366, + 907, + 1400, + 907, + 1400, + 936, + 1366, + 936 + ], + "score": 0.87 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1079, + 1400, + 1079, + 1400, + 1108, + 1366, + 1108 + ], + "score": 0.837 + }, + { + "category_id": 8, + "poly": [ + 302, + 1185, + 1361, + 1185, + 1361, + 1225, + 302, + 1225 + ], + "score": 0.698 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.662 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.37 + }, + { + "category_id": 1, + "poly": [ + 302, + 1185, + 1361, + 1185, + 1361, + 1225, + 302, + 1225 + ], + "score": 0.163 + }, + { + "category_id": 14, + "poly": [ + 573, + 1379, + 1126, + 1379, + 1126, + 1476, + 573, + 1476 + ], + "score": 0.93, + "latex": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } f ( \\mathbf { X } ; \\mathbf { c } , \\theta ) = \\sum _ { k = 1 } ^ { K } q _ { \\phi } ( \\mathbf { y } _ { k } | \\mathbf { X } ) f ( \\mathbf { X } ; \\mathbf { y } _ { k } , \\theta )" + }, + { + "category_id": 13, + "poly": [ + 629, + 1602, + 728, + 1602, + 728, + 1636, + 629, + 1636 + ], + "score": 0.93, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 909, + 1540, + 1008, + 1540, + 1008, + 1575, + 909, + 1575 + ], + "score": 0.93, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1911, + 657, + 1911, + 657, + 1945, + 297, + 1945 + ], + "score": 0.92, + "latex": "p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { z } _ { 1 } ) p _ { \\pmb { \\theta } } ( \\mathbf { z } _ { 1 } | \\mathbf { z } _ { 2 } , \\bar { \\mathbf { c } } ) p ( \\mathbf { z } _ { 2 } ) p ( \\mathbf { c } )" + }, + { + "category_id": 13, + "poly": [ + 372, + 945, + 506, + 945, + 506, + 979, + 372, + 979 + ], + "score": 0.92, + "latex": "\\operatorname { M u l t } ( K , \\pi )" + }, + { + "category_id": 13, + "poly": [ + 504, + 1249, + 607, + 1249, + 607, + 1282, + 504, + 1282 + ], + "score": 0.92, + "latex": "\\log p ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1049, + 1343, + 1147, + 1343, + 1147, + 1376, + 1049, + 1376 + ], + "score": 0.92, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1944, + 396, + 1944, + 396, + 1977, + 298, + 1977 + ], + "score": 0.91, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1209, + 945, + 1354, + 945, + 1354, + 979, + 1209, + 979 + ], + "score": 0.91, + "latex": "f ( \\mathbf { X } ; \\mathbf { z } , \\mathbf { c } , \\theta )" + }, + { + "category_id": 14, + "poly": [ + 702, + 295, + 999, + 295, + 999, + 337, + 702, + 337 + ], + "score": 0.91, + "latex": "\\| f ( \\mathbf { X } ; \\boldsymbol { \\phi } _ { 1 } ) - f ( \\mathbf { X } ; \\boldsymbol { \\phi } _ { 2 } ) \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1048, + 1311, + 1147, + 1311, + 1147, + 1344, + 1048, + 1344 + ], + "score": 0.91, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1188, + 1881, + 1405, + 1881, + 1405, + 1915, + 1188, + 1915 + ], + "score": 0.9, + "latex": "\\begin{array} { r l r } { \\mathrm { \\nabla } p _ { \\pmb \\theta } ( \\mathbf { X } , \\mathbf { z } _ { 1 } , \\mathbf { z } _ { 2 } , \\mathbf { c } ) } & { { } = } & { } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 754, + 1280, + 1012, + 1280, + 1012, + 1314, + 754, + 1314 + ], + "score": 0.9, + "latex": "\\mathcal { L } ( \\mathbf { X } ; \\boldsymbol { \\phi } , \\pmb { \\theta } ) = - \\mathrm { E L B O }" + }, + { + "category_id": 14, + "poly": [ + 311, + 1184, + 1400, + 1184, + 1400, + 1226, + 311, + 1226 + ], + "score": 0.9, + "latex": "\\log p ( \\mathbf { X } ) \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } ) ) = \\mathrm { E L B O } ( \\phi ) ," + }, + { + "category_id": 14, + "poly": [ + 440, + 1669, + 1262, + 1669, + 1262, + 1723, + 440, + 1723 + ], + "score": 0.89, + "latex": "\\operatorname* { m i n } _ { \\phi , \\theta } \\mathbb { E } _ { \\mathbf { X } \\sim \\mathbb { D } _ { U } } \\mathcal { L } \\big ( \\mathbf { X } ; \\phi , \\pmb { \\theta } ) + \\mathbb { E } _ { ( \\mathbf { X } , \\mathbf { y } ) \\sim \\mathbb { D } _ { L } } \\big [ \\mathcal { L } \\big ( \\mathbf { X } , \\mathbf { c } = \\mathbf { y } ; \\phi , \\pmb { \\theta } \\big ) - \\log q _ { \\phi } ( \\mathbf { y } | \\mathbf { X } ) \\big ]" + }, + { + "category_id": 14, + "poly": [ + 400, + 902, + 1301, + 902, + 1301, + 940, + 400, + 940 + ], + "score": 0.88, + "latex": "p ( \\mathbf { z } ) = { \\mathcal { N } } ( z ; \\mathbf { 0 } , I ) ; \\qquad p ( \\mathbf { c } ) = \\mathbf { M } \\mathbf { u } \\mathbf { l } \\mathbf { t } ( \\mathbf { c } ; K , \\pi ) ; \\qquad p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) = f ( \\mathbf { X } ; \\mathbf { z } , \\mathbf { c } , \\theta )" + }, + { + "category_id": 14, + "poly": [ + 326, + 1074, + 1342, + 1074, + 1342, + 1113, + 326, + 1113 + ], + "score": 0.86, + "latex": "\\begin{array} { r } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) = q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) ; \\qquad p ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) = p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) ; \\qquad p ( \\mathbf { z } , \\mathbf { c } ) = p ( \\mathbf { z } ) p ( \\mathbf { c } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1130, + 1825, + 1161, + 1825, + 1161, + 1851, + 1130, + 1851 + ], + "score": 0.86, + "latex": "\\mathbf { z } _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 656, + 836, + 681, + 836, + 681, + 866, + 656, + 866 + ], + "score": 0.86, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 360, + 1484, + 395, + 1484, + 395, + 1512, + 360, + 1512 + ], + "score": 0.85, + "latex": "\\mathbf { y } _ { k }" + }, + { + "category_id": 13, + "poly": [ + 952, + 1854, + 983, + 1854, + 983, + 1881, + 952, + 1881 + ], + "score": 0.85, + "latex": "\\mathbf { z } _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 973, + 946, + 1002, + 946, + 1002, + 973, + 973, + 973 + ], + "score": 0.84, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 353, + 1882, + 451, + 1882, + 451, + 1910, + 353, + 1910 + ], + "score": 0.83, + "latex": "\\mathbf { M } 1 { + } \\mathbf { M } 2" + }, + { + "category_id": 13, + "poly": [ + 979, + 637, + 1049, + 637, + 1049, + 667, + 979, + 667 + ], + "score": 0.83, + "latex": "( \\mathsf { K } { + } 1 )" + }, + { + "category_id": 13, + "poly": [ + 817, + 697, + 886, + 697, + 886, + 728, + 817, + 728 + ], + "score": 0.8, + "latex": "( \\mathsf { K } { + } 1 )" + }, + { + "category_id": 13, + "poly": [ + 580, + 836, + 601, + 836, + 601, + 862, + 580, + 862 + ], + "score": 0.76, + "latex": "\\pmb \\theta" + }, + { + "category_id": 13, + "poly": [ + 837, + 1482, + 856, + 1482, + 856, + 1508, + 837, + 1508 + ], + "score": 0.71, + "latex": "\\mathbf { k }" + }, + { + "category_id": 13, + "poly": [ + 1065, + 871, + 1084, + 871, + 1084, + 893, + 1065, + 893 + ], + "score": 0.68, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 1011, + 1011, + 1034, + 1011, + 1034, + 1038, + 1011, + 1038 + ], + "score": 0.68, + "latex": "\\mathbf { y }" + }, + { + "category_id": 13, + "poly": [ + 723, + 1012, + 743, + 1012, + 743, + 1034, + 723, + 1034 + ], + "score": 0.65, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 1171, + 951, + 1194, + 951, + 1194, + 973, + 1171, + 973 + ], + "score": 0.65, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1355, + 1315, + 1375, + 1315, + 1375, + 1343, + 1355, + 1343 + ], + "score": 0.63, + "latex": "\\mathbf { y }" + }, + { + "category_id": 13, + "poly": [ + 1001, + 867, + 1028, + 867, + 1028, + 893, + 1001, + 893 + ], + "score": 0.6, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 397, + 866, + 424, + 866, + 424, + 893, + 397, + 893 + ], + "score": 0.58, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 650, + 529, + 678, + 529, + 678, + 556, + 650, + 556 + ], + "score": 0.55, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 1134, + 871, + 1152, + 871, + 1152, + 893, + 1134, + 893 + ], + "score": 0.54, + "latex": "\\mathbf { c }" + }, + { + "category_id": 13, + "poly": [ + 1367, + 1851, + 1393, + 1851, + 1393, + 1878, + 1367, + 1878 + ], + "score": 0.54, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 639, + 563, + 658, + 563, + 658, + 585, + 639, + 585 + ], + "score": 0.49, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 648, + 872, + 666, + 872, + 666, + 893, + 648, + 893 + ], + "score": 0.46, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 797, + 1012, + 814, + 1012, + 814, + 1034, + 797, + 1034 + ], + "score": 0.4, + "latex": "\\mathbf { c }" + }, + { + "category_id": 13, + "poly": [ + 907, + 1512, + 932, + 1512, + 932, + 1538, + 907, + 1538 + ], + "score": 0.35, + "latex": "\\mathrm { K }" + }, + { + "category_id": 13, + "poly": [ + 408, + 1315, + 429, + 1315, + 429, + 1343, + 408, + 1343 + ], + "score": 0.25, + "latex": "\\mathbf { y }" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 466.0, + 692.0, + 466.0, + 692.0, + 506.0, + 294.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 860.0, + 2085.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1727.0, + 1405.0, + 1727.0, + 1405.0, + 1763.0, + 296.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1755.0, + 1406.0, + 1755.0, + 1406.0, + 1799.0, + 293.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1790.0, + 1405.0, + 1790.0, + 1405.0, + 1826.0, + 294.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1818.0, + 1129.0, + 1818.0, + 1129.0, + 1855.0, + 293.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 1818.0, + 1405.0, + 1818.0, + 1405.0, + 1855.0, + 1162.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 951.0, + 1849.0, + 951.0, + 1884.0, + 294.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1849.0, + 1366.0, + 1849.0, + 1366.0, + 1884.0, + 984.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1849.0, + 1402.0, + 1849.0, + 1402.0, + 1884.0, + 1394.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1881.0, + 352.0, + 1881.0, + 352.0, + 1917.0, + 295.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1881.0, + 1187.0, + 1881.0, + 1187.0, + 1917.0, + 452.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1910.0, + 296.0, + 1910.0, + 296.0, + 1948.0, + 291.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 1910.0, + 1406.0, + 1910.0, + 1406.0, + 1948.0, + 658.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1940.0, + 297.0, + 1940.0, + 297.0, + 1980.0, + 291.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1940.0, + 1406.0, + 1940.0, + 1406.0, + 1980.0, + 397.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2007.0, + 293.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 698.0, + 2001.0, + 698.0, + 2041.0, + 294.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 605.0, + 1403.0, + 605.0, + 1403.0, + 637.0, + 295.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 636.0, + 978.0, + 636.0, + 978.0, + 669.0, + 292.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 636.0, + 1404.0, + 636.0, + 1404.0, + 669.0, + 1050.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 663.0, + 1405.0, + 663.0, + 1405.0, + 701.0, + 293.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 693.0, + 816.0, + 693.0, + 816.0, + 733.0, + 293.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 693.0, + 1405.0, + 693.0, + 1405.0, + 733.0, + 887.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 727.0, + 1403.0, + 727.0, + 1403.0, + 759.0, + 296.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 758.0, + 789.0, + 758.0, + 789.0, + 794.0, + 294.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1479.0, + 359.0, + 1479.0, + 359.0, + 1514.0, + 295.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1479.0, + 836.0, + 1479.0, + 836.0, + 1514.0, + 396.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 1479.0, + 1404.0, + 1479.0, + 1404.0, + 1514.0, + 857.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1510.0, + 906.0, + 1510.0, + 906.0, + 1546.0, + 295.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1510.0, + 1405.0, + 1510.0, + 1405.0, + 1546.0, + 933.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1538.0, + 908.0, + 1538.0, + 908.0, + 1576.0, + 294.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 1538.0, + 1406.0, + 1538.0, + 1406.0, + 1576.0, + 1009.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1571.0, + 1405.0, + 1571.0, + 1405.0, + 1605.0, + 292.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1600.0, + 628.0, + 1600.0, + 628.0, + 1636.0, + 292.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 1600.0, + 1406.0, + 1600.0, + 1406.0, + 1636.0, + 729.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1632.0, + 589.0, + 1632.0, + 589.0, + 1668.0, + 295.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 943.0, + 371.0, + 943.0, + 371.0, + 980.0, + 295.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 943.0, + 972.0, + 943.0, + 972.0, + 980.0, + 507.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 943.0, + 1170.0, + 943.0, + 1170.0, + 980.0, + 1003.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 943.0, + 1208.0, + 943.0, + 1208.0, + 980.0, + 1195.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 943.0, + 1406.0, + 943.0, + 1406.0, + 980.0, + 1355.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 975.0, + 1405.0, + 975.0, + 1405.0, + 1011.0, + 295.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1006.0, + 722.0, + 1006.0, + 722.0, + 1042.0, + 294.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1006.0, + 796.0, + 1006.0, + 796.0, + 1042.0, + 744.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 1006.0, + 1010.0, + 1006.0, + 1010.0, + 1042.0, + 815.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1006.0, + 1405.0, + 1006.0, + 1405.0, + 1042.0, + 1035.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1039.0, + 841.0, + 1039.0, + 841.0, + 1071.0, + 296.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1249.0, + 503.0, + 1249.0, + 503.0, + 1282.0, + 294.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 1249.0, + 1404.0, + 1249.0, + 1404.0, + 1282.0, + 608.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1279.0, + 753.0, + 1279.0, + 753.0, + 1315.0, + 294.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1279.0, + 1405.0, + 1279.0, + 1405.0, + 1315.0, + 1013.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1310.0, + 407.0, + 1310.0, + 407.0, + 1347.0, + 293.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1310.0, + 1047.0, + 1310.0, + 1047.0, + 1347.0, + 430.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 1310.0, + 1354.0, + 1310.0, + 1354.0, + 1347.0, + 1148.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1310.0, + 1406.0, + 1310.0, + 1406.0, + 1347.0, + 1376.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1341.0, + 1048.0, + 1341.0, + 1048.0, + 1377.0, + 293.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 1341.0, + 1183.0, + 1341.0, + 1183.0, + 1377.0, + 1148.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 341.0, + 1404.0, + 341.0, + 1404.0, + 375.0, + 296.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 371.0, + 1403.0, + 371.0, + 1403.0, + 405.0, + 296.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 403.0, + 416.0, + 403.0, + 416.0, + 434.0, + 296.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 803.0, + 1405.0, + 803.0, + 1405.0, + 840.0, + 294.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 835.0, + 579.0, + 835.0, + 579.0, + 869.0, + 295.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 835.0, + 655.0, + 835.0, + 655.0, + 869.0, + 602.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 835.0, + 1405.0, + 835.0, + 1405.0, + 869.0, + 682.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 863.0, + 396.0, + 863.0, + 396.0, + 900.0, + 294.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 863.0, + 647.0, + 863.0, + 647.0, + 900.0, + 425.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 863.0, + 1000.0, + 863.0, + 1000.0, + 900.0, + 667.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 863.0, + 1064.0, + 863.0, + 1064.0, + 900.0, + 1029.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 863.0, + 1133.0, + 863.0, + 1133.0, + 900.0, + 1085.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 863.0, + 1195.0, + 863.0, + 1195.0, + 900.0, + 1153.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 1404.0, + 227.0, + 1404.0, + 266.0, + 293.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 262.0, + 699.0, + 262.0, + 699.0, + 294.0, + 296.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1117.0, + 1406.0, + 1117.0, + 1406.0, + 1152.0, + 295.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1149.0, + 861.0, + 1149.0, + 861.0, + 1184.0, + 297.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 525.0, + 649.0, + 525.0, + 649.0, + 563.0, + 292.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 525.0, + 1406.0, + 525.0, + 1406.0, + 563.0, + 679.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 554.0, + 638.0, + 554.0, + 638.0, + 594.0, + 295.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 554.0, + 872.0, + 554.0, + 872.0, + 594.0, + 659.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1182.0, + 310.0, + 1182.0, + 310.0, + 1230.0, + 303.0, + 1230.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 291, + 1405, + 291, + 1405, + 479, + 297, + 479 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1122, + 1403, + 1122, + 1403, + 1216, + 299, + 1216 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 299, + 1941, + 1401, + 1941, + 1401, + 2034, + 299, + 2034 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 297, + 567, + 1405, + 567, + 1405, + 710, + 297, + 710 + ], + "score": 0.966 + }, + { + "category_id": 8, + "poly": [ + 330, + 1221, + 1321, + 1221, + 1321, + 1448, + 330, + 1448 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 296, + 757, + 1405, + 757, + 1405, + 959, + 296, + 959 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 296, + 1451, + 1404, + 1451, + 1404, + 1617, + 296, + 1617 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 297, + 1046, + 1404, + 1046, + 1404, + 1108, + 297, + 1108 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 406, + 963, + 1286, + 963, + 1286, + 1043, + 406, + 1043 + ], + "score": 0.937 + }, + { + "category_id": 8, + "poly": [ + 415, + 1884, + 1283, + 1884, + 1283, + 1931, + 415, + 1931 + ], + "score": 0.914 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 296, + 1847, + 973, + 1847, + 973, + 1880, + 296, + 1880 + ], + "score": 0.895 + }, + { + "category_id": 0, + "poly": [ + 303, + 512, + 799, + 512, + 799, + 545, + 303, + 545 + ], + "score": 0.886 + }, + { + "category_id": 0, + "poly": [ + 295, + 225, + 1197, + 225, + 1197, + 263, + 295, + 263 + ], + "score": 0.874 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1887, + 1400, + 1887, + 1400, + 1918, + 1351, + 1918 + ], + "score": 0.872 + }, + { + "category_id": 9, + "poly": [ + 1365, + 989, + 1400, + 989, + 1400, + 1018, + 1365, + 1018 + ], + "score": 0.872 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1319, + 1401, + 1319, + 1401, + 1350, + 1352, + 1350 + ], + "score": 0.867 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1789, + 1400, + 1789, + 1400, + 1819, + 1352, + 1819 + ], + "score": 0.861 + }, + { + "category_id": 8, + "poly": [ + 406, + 713, + 1288, + 713, + 1288, + 752, + 406, + 752 + ], + "score": 0.856 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1640, + 1401, + 1640, + 1401, + 1671, + 1352, + 1671 + ], + "score": 0.831 + }, + { + "category_id": 9, + "poly": [ + 1365, + 718, + 1400, + 718, + 1400, + 746, + 1365, + 746 + ], + "score": 0.813 + }, + { + "category_id": 1, + "poly": [ + 298, + 1698, + 1402, + 1698, + 1402, + 1758, + 298, + 1758 + ], + "score": 0.808 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2112, + 840, + 2112 + ], + "score": 0.775 + }, + { + "category_id": 8, + "poly": [ + 319, + 1763, + 1318, + 1763, + 1318, + 1844, + 319, + 1844 + ], + "score": 0.571 + }, + { + "category_id": 8, + "poly": [ + 311, + 1620, + 1328, + 1620, + 1328, + 1694, + 311, + 1694 + ], + "score": 0.487 + }, + { + "category_id": 8, + "poly": [ + 339, + 1763, + 1302, + 1763, + 1302, + 1844, + 339, + 1844 + ], + "score": 0.314 + }, + { + "category_id": 1, + "poly": [ + 311, + 1620, + 1328, + 1620, + 1328, + 1694, + 311, + 1694 + ], + "score": 0.197 + }, + { + "category_id": 13, + "poly": [ + 855, + 601, + 1227, + 601, + 1227, + 680, + 855, + 680 + ], + "score": 0.93, + "latex": "p _ { e m p } ( \\mathbf { X } ; \\mathbb { D } ) = { \\frac { 1 } { | \\mathbb { D } | } } \\sum _ { \\mathbf { X } ^ { \\prime } \\in \\mathbb { D } } \\mathbf { 1 } _ { \\mathbf { X } = \\mathbf { X } ^ { \\prime } }" + }, + { + "category_id": 13, + "poly": [ + 934, + 1483, + 1033, + 1483, + 1033, + 1518, + 934, + 1518 + ], + "score": 0.93, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 794, + 1483, + 878, + 1483, + 878, + 1516, + 794, + 1516 + ], + "score": 0.92, + "latex": "p ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1219, + 1562, + 1378, + 1562, + 1378, + 1597, + 1219, + 1597 + ], + "score": 0.92, + "latex": "\\mathcal { L } _ { \\mathbb { D } _ { U } } ( \\mathbf { X } ; \\pmb { \\theta } , \\phi )" + }, + { + "category_id": 13, + "poly": [ + 1207, + 833, + 1321, + 833, + 1321, + 869, + 1207, + 869 + ], + "score": 0.92, + "latex": "\\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 837, + 1154, + 1118, + 1154, + 1118, + 1188, + 837, + 1188 + ], + "score": 0.92, + "latex": "p ( \\mathbf { c } | \\mathbf { X } ) = \\mathbf { M u l t } ( \\mathbf { c } ; K , \\mathbf { y } )" + }, + { + "category_id": 14, + "poly": [ + 314, + 1617, + 1339, + 1617, + 1339, + 1694, + 314, + 1694 + ], + "score": 0.92, + "latex": "\\mathbb { E } _ { p ( { \\mathbf { c } } | { \\mathbf { X } } ) } \\log { \\frac { p ( { \\mathbf { c } } ) } { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } } \\approx { \\mathrm { ( } } { \\mathrm { w h e n ~ } } p ( { \\mathbf { c } } | { \\mathbf { X } } ) \\approx q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) { \\mathrm { ) } } \\mathbb { E } _ { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } \\log { \\frac { p ( { \\mathbf { c } } ) } { q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) } } = D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { c } } ) )" + }, + { + "category_id": 13, + "poly": [ + 373, + 1047, + 467, + 1047, + 467, + 1078, + 373, + 1078 + ], + "score": 0.91, + "latex": "\\beta , \\mathbf { I _ { z } } , \\mathbf { I _ { c } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1562, + 396, + 1562, + 396, + 1598, + 297, + 1598 + ], + "score": 0.91, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1285, + 1514, + 1403, + 1514, + 1403, + 1547, + 1285, + 1547 + ], + "score": 0.91, + "latex": "p ( \\mathbf { c } | \\mathbf { X } ) \\approx" + }, + { + "category_id": 14, + "poly": [ + 326, + 1218, + 1325, + 1218, + 1325, + 1452, + 326, + 1452 + ], + "score": 0.91, + "latex": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\log \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } [ \\log p ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) + \\log \\frac { p ( \\mathbf { z } ) p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } ] = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , p ( \\mathbf { c } | \\mathbf { X } ) } \\log p ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) } \\\\ & { - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( p ( \\mathbf { c } | \\mathbf { X } ) \\| q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) ) + \\mathbb { E } _ { p ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 725, + 771, + 839, + 771, + 839, + 809, + 725, + 809 + ], + "score": 0.9, + "latex": "\\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 516, + 1544, + 752, + 1544, + 752, + 1618, + 516, + 1618 + ], + "score": 0.9, + "latex": "\\mathbb { E } _ { p ( \\mathbf { c } | \\mathbf { X } ) } \\log { \\frac { p ( \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } }" + }, + { + "category_id": 14, + "poly": [ + 337, + 1761, + 1317, + 1761, + 1317, + 1848, + 337, + 1848 + ], + "score": 0.9, + "latex": "\\begin{array} { r l } & { { \\mathcal { L } } _ { \\mathbb { D } _ { L } } ( { \\mathbf { X } } , { \\mathbf { y } } ; \\pmb { \\theta } , \\phi ) = { \\mathbb { E } } _ { q _ { \\phi } ( { \\mathbf { z } } | { \\mathbf { X } } ) , p ( { \\mathbf { c } } | { \\mathbf { X } } ) } [ - \\log p _ { \\theta } ( { \\mathbf { X } } | { \\mathbf { z } } , { \\mathbf { c } } ) ] + \\beta _ { \\mathbf { z } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { z } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { z } } ) ) - { \\mathbf { I } } _ { \\mathbf { z } } | } \\\\ & { \\quad \\quad \\quad \\quad \\quad \\quad + \\beta _ { \\mathbf { c } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| p ( { \\mathbf { c } } ) ) - { \\mathbf { I } } _ { \\mathbf { c } } | + D _ { \\mathrm { K L } } ( p ( { \\mathbf { c } } | { \\mathbf { X } } ) \\| q _ { \\phi } ( { \\mathbf { c } } | { \\mathbf { X } } ) ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 432, + 1451, + 701, + 1451, + 701, + 1488, + 432, + 1488 + ], + "score": 0.9, + "latex": "D _ { \\mathrm { K L } } ( p ( \\mathbf { c } | \\mathbf { X } ) \\| q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) )" + }, + { + "category_id": 14, + "poly": [ + 420, + 1883, + 1284, + 1883, + 1284, + 1937, + 420, + 1937 + ], + "score": 0.9, + "latex": "\\operatorname* { m i n } _ { \\phi , \\theta } \\mathbb { E } _ { \\mathbf { X } \\sim p _ { e m p } ( \\mathbf { X } ; \\mathbb { D } _ { U } ) } \\mathcal { L } _ { \\mathbb { D } _ { U } } ( \\mathbf { X } ; \\theta , \\phi ) + \\mathbb { E } _ { ( \\mathbf { X } , \\mathbf { y } ) \\sim p _ { e m p } ( ( \\mathbf { X } , \\mathbf { y } ) ; \\mathbb { D } _ { L } ) } \\mathcal { L } _ { \\mathbb { D } _ { L } } ( \\mathbf { X } , \\mathbf { y } ; \\theta , \\phi )" + }, + { + "category_id": 13, + "poly": [ + 373, + 756, + 670, + 756, + 670, + 835, + 373, + 835 + ], + "score": 0.9, + "latex": "q _ { \\phi } ( \\mathbf { z } ) = \\frac { 1 } { | \\mathbb { D } | } \\sum _ { \\mathbf { X } \\in \\mathbb { D } } q _ { \\phi } ( \\mathbf { z } | x )" + }, + { + "category_id": 13, + "poly": [ + 542, + 1125, + 580, + 1125, + 580, + 1155, + 542, + 1155 + ], + "score": 0.89, + "latex": "\\mathbb { D } _ { L }" + }, + { + "category_id": 13, + "poly": [ + 715, + 1699, + 756, + 1699, + 756, + 1728, + 715, + 1728 + ], + "score": 0.89, + "latex": "\\mathbb { D } _ { L }" + }, + { + "category_id": 13, + "poly": [ + 786, + 570, + 848, + 570, + 848, + 604, + 786, + 604 + ], + "score": 0.88, + "latex": "( 3 , 4 )" + }, + { + "category_id": 13, + "poly": [ + 881, + 899, + 910, + 899, + 910, + 928, + 881, + 928 + ], + "score": 0.88, + "latex": "\\mathbf { I _ { z } }" + }, + { + "category_id": 13, + "poly": [ + 961, + 899, + 990, + 899, + 990, + 928, + 961, + 928 + ], + "score": 0.88, + "latex": "\\mathbf { I _ { c } }" + }, + { + "category_id": 14, + "poly": [ + 406, + 961, + 1292, + 961, + 1292, + 1047, + 406, + 1047 + ], + "score": 0.87, + "latex": "\\begin{array} { r l } & { { \\mathcal { L } } _ { \\mathbb { D } _ { U } } ( { \\mathbf { X } } ; \\pmb { \\theta } , \\phi ) = { \\mathbb { E } } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] + \\beta _ { \\mathbf { z } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } ) - \\mathbf { I } _ { \\mathbf { z } } | } \\\\ & { \\quad \\quad \\quad \\quad + \\beta _ { \\mathbf { c } } | D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { c } ) ) - \\mathbf { I } _ { \\mathbf { c } } | } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 944, + 929, + 986, + 929, + 986, + 959, + 944, + 959 + ], + "score": 0.87, + "latex": "\\mathbb { D } _ { U }" + }, + { + "category_id": 13, + "poly": [ + 527, + 1082, + 545, + 1082, + 545, + 1104, + 527, + 1104 + ], + "score": 0.77, + "latex": "\\mathbf { z }" + }, + { + "category_id": 14, + "poly": [ + 407, + 712, + 1290, + 712, + 1290, + 753, + 407, + 753 + ], + "score": 0.76, + "latex": "\\begin{array} { r } { \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) \\big ) = \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } ) \\| p ( \\mathbf { z } ) \\big ) \\geq \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1239, + 773, + 1266, + 773, + 1266, + 800, + 1239, + 800 + ], + "score": 0.75, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 738, + 839, + 756, + 839, + 756, + 861, + 738, + 861 + ], + "score": 0.68, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 1320, + 779, + 1337, + 779, + 1337, + 800, + 1320, + 800 + ], + "score": 0.68, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 660, + 835, + 688, + 835, + 688, + 862, + 660, + 862 + ], + "score": 0.57, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 1189, + 1130, + 1206, + 1130, + 1206, + 1151, + 1189, + 1151 + ], + "score": 0.42, + "latex": "\\mathbf { c }" + }, + { + "category_id": 13, + "poly": [ + 595, + 1082, + 612, + 1082, + 612, + 1104, + 595, + 1104 + ], + "score": 0.4, + "latex": "\\mathbf { c }" + }, + { + "category_id": 13, + "poly": [ + 714, + 903, + 732, + 903, + 732, + 925, + 714, + 925 + ], + "score": 0.34, + "latex": "\\mathbf { c }" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 511.0, + 802.0, + 511.0, + 802.0, + 547.0, + 295.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 222.0, + 1202.0, + 222.0, + 1202.0, + 267.0, + 291.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 863.0, + 2085.0, + 863.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 295.0, + 1405.0, + 295.0, + 1405.0, + 327.0, + 295.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 325.0, + 1406.0, + 325.0, + 1406.0, + 358.0, + 291.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 357.0, + 1405.0, + 357.0, + 1405.0, + 389.0, + 295.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 384.0, + 1405.0, + 384.0, + 1405.0, + 421.0, + 294.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 413.0, + 1403.0, + 413.0, + 1403.0, + 452.0, + 294.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 444.0, + 925.0, + 444.0, + 925.0, + 481.0, + 295.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1123.0, + 541.0, + 1123.0, + 541.0, + 1157.0, + 295.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1123.0, + 1188.0, + 1123.0, + 1188.0, + 1157.0, + 581.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 1123.0, + 1404.0, + 1123.0, + 1404.0, + 1157.0, + 1207.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1151.0, + 836.0, + 1151.0, + 836.0, + 1193.0, + 293.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1151.0, + 1407.0, + 1151.0, + 1407.0, + 1193.0, + 1119.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1182.0, + 560.0, + 1182.0, + 560.0, + 1219.0, + 293.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1938.0, + 1406.0, + 1938.0, + 1406.0, + 1979.0, + 293.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 1403.0, + 1973.0, + 1403.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2003.0, + 646.0, + 2003.0, + 646.0, + 2033.0, + 295.0, + 2033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 566.0, + 785.0, + 566.0, + 785.0, + 609.0, + 292.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 566.0, + 1406.0, + 566.0, + 1406.0, + 609.0, + 849.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 604.0, + 854.0, + 604.0, + 854.0, + 669.0, + 291.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 604.0, + 1410.0, + 604.0, + 1410.0, + 669.0, + 1228.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 671.0, + 1403.0, + 671.0, + 1403.0, + 716.0, + 295.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 759.0, + 372.0, + 759.0, + 372.0, + 822.0, + 294.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 766.0, + 724.0, + 766.0, + 724.0, + 815.0, + 671.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 766.0, + 1238.0, + 766.0, + 1238.0, + 815.0, + 840.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 766.0, + 1319.0, + 766.0, + 1319.0, + 815.0, + 1267.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 766.0, + 1406.0, + 766.0, + 1406.0, + 815.0, + 1338.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 827.0, + 659.0, + 827.0, + 659.0, + 871.0, + 293.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 827.0, + 737.0, + 827.0, + 737.0, + 871.0, + 689.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 827.0, + 1206.0, + 827.0, + 1206.0, + 871.0, + 757.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 827.0, + 1406.0, + 827.0, + 1406.0, + 871.0, + 1322.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 865.0, + 1405.0, + 865.0, + 1405.0, + 902.0, + 294.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 897.0, + 713.0, + 897.0, + 713.0, + 931.0, + 295.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 897.0, + 880.0, + 897.0, + 880.0, + 931.0, + 733.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 897.0, + 960.0, + 897.0, + 960.0, + 931.0, + 911.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 897.0, + 1403.0, + 897.0, + 1403.0, + 931.0, + 991.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 927.0, + 943.0, + 927.0, + 943.0, + 963.0, + 294.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1451.0, + 431.0, + 1451.0, + 431.0, + 1488.0, + 294.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1451.0, + 1406.0, + 1451.0, + 1406.0, + 1488.0, + 702.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1484.0, + 793.0, + 1484.0, + 793.0, + 1517.0, + 294.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 1484.0, + 933.0, + 1484.0, + 933.0, + 1517.0, + 879.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1484.0, + 1405.0, + 1484.0, + 1405.0, + 1517.0, + 1034.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1514.0, + 1284.0, + 1514.0, + 1284.0, + 1547.0, + 296.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1540.0, + 296.0, + 1540.0, + 296.0, + 1619.0, + 291.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1540.0, + 515.0, + 1540.0, + 515.0, + 1619.0, + 397.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1540.0, + 1218.0, + 1540.0, + 1218.0, + 1619.0, + 753.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 1540.0, + 1392.0, + 1540.0, + 1392.0, + 1619.0, + 1379.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1042.0, + 372.0, + 1042.0, + 372.0, + 1082.0, + 294.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 1042.0, + 1406.0, + 1042.0, + 1406.0, + 1082.0, + 468.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1074.0, + 526.0, + 1074.0, + 526.0, + 1111.0, + 294.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 1074.0, + 594.0, + 1074.0, + 594.0, + 1111.0, + 546.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 1074.0, + 624.0, + 1074.0, + 624.0, + 1111.0, + 613.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1847.0, + 971.0, + 1847.0, + 971.0, + 1883.0, + 297.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1696.0, + 714.0, + 1696.0, + 714.0, + 1734.0, + 294.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1696.0, + 1406.0, + 1696.0, + 1406.0, + 1734.0, + 757.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1727.0, + 1048.0, + 1727.0, + 1048.0, + 1763.0, + 293.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1633.0, + 313.0, + 1633.0, + 313.0, + 1683.0, + 308.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.75, + 1623.5, + 966.75, + 1623.5, + 966.75, + 1682.0, + 547.75, + 1682.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1521, + 1404, + 1521, + 1404, + 1739, + 296, + 1739 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1290, + 1403, + 1290, + 1403, + 1446, + 297, + 1446 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 295, + 1849, + 1405, + 1849, + 1405, + 2034, + 295, + 2034 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 296, + 1096, + 1404, + 1096, + 1404, + 1225, + 296, + 1225 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 227, + 1401, + 227, + 1401, + 325, + 298, + 325 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 392, + 1751, + 1305, + 1751, + 1305, + 1837, + 392, + 1837 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 520, + 1238, + 1179, + 1238, + 1179, + 1281, + 520, + 1281 + ], + "score": 0.941 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 858, + 74, + 858, + 105, + 298, + 105 + ], + "score": 0.915 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1243, + 1399, + 1243, + 1399, + 1273, + 1352, + 1273 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 306, + 492, + 403, + 492, + 403, + 523, + 306, + 523 + ], + "score": 0.873 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1497, + 1399, + 1497, + 1399, + 1523, + 1352, + 1523 + ], + "score": 0.87 + }, + { + "category_id": 8, + "poly": [ + 311, + 1458, + 1369, + 1458, + 1369, + 1500, + 311, + 1500 + ], + "score": 0.867 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1778, + 1400, + 1778, + 1400, + 1808, + 1352, + 1808 + ], + "score": 0.861 + }, + { + "category_id": 1, + "poly": [ + 343, + 366, + 1103, + 366, + 1103, + 493, + 343, + 493 + ], + "score": 0.857 + }, + { + "category_id": 0, + "poly": [ + 304, + 337, + 383, + 337, + 383, + 365, + 304, + 365 + ], + "score": 0.845 + }, + { + "category_id": 0, + "poly": [ + 300, + 1040, + 786, + 1040, + 786, + 1072, + 300, + 1072 + ], + "score": 0.82 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.754 + }, + { + "category_id": 1, + "poly": [ + 309, + 523, + 962, + 523, + 962, + 977, + 309, + 977 + ], + "score": 0.673 + }, + { + "category_id": 0, + "poly": [ + 300, + 1040, + 786, + 1040, + 786, + 1072, + 300, + 1072 + ], + "score": 0.166 + }, + { + "category_id": 8, + "poly": [ + 309, + 523, + 962, + 523, + 962, + 977, + 309, + 977 + ], + "score": 0.112 + }, + { + "category_id": 13, + "poly": [ + 620, + 290, + 888, + 290, + 888, + 325, + 620, + 325 + ], + "score": 0.93, + "latex": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { \\bar { c } } | \\mathbf { X } ) )" + }, + { + "category_id": 13, + "poly": [ + 648, + 1676, + 732, + 1676, + 732, + 1709, + 648, + 1709 + ], + "score": 0.93, + "latex": "p ( \\mathbf { z } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1263, + 1099, + 1362, + 1099, + 1362, + 1133, + 1263, + 1133 + ], + "score": 0.92, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 506, + 1676, + 590, + 1676, + 590, + 1709, + 506, + 1709 + ], + "score": 0.92, + "latex": "p ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1106, + 1099, + 1205, + 1099, + 1205, + 1132, + 1106, + 1132 + ], + "score": 0.92, + "latex": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 535, + 523, + 663, + 523, + 663, + 558, + 535, + 558 + ], + "score": 0.92, + "latex": "p _ { m i x u p } ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 930, + 1553, + 1038, + 1553, + 1038, + 1586, + 930, + 1586 + ], + "score": 0.92, + "latex": "p ( \\mathbf { c } ) \\mathbf { , } p ( \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 380, + 1850, + 509, + 1850, + 509, + 1885, + 380, + 1885 + ], + "score": 0.92, + "latex": "p _ { m i x u p } ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 772, + 365, + 877, + 365, + 877, + 399, + 772, + 399 + ], + "score": 0.91, + "latex": "p _ { e m p } ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1158, + 754, + 1158, + 754, + 1197, + 297, + 1197 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { p ( \\mathbf { X } ) \\ { \\stackrel { } { = } } \\ \\int _ { \\mathbf { z } } \\int _ { \\mathbf { c } } p ( \\mathbf { X } ) p ( \\mathbf { \\tilde { z } } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) d \\mathbf { z } d \\mathbf { c } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 738, + 1553, + 877, + 1553, + 877, + 1587, + 738, + 1587 + ], + "score": 0.91, + "latex": "q _ { \\phi } ( \\mathbf { c } ) , \\bar { q } _ { \\phi } ( \\mathbf { z } )" + }, + { + "category_id": 14, + "poly": [ + 393, + 1749, + 1305, + 1749, + 1305, + 1841, + 393, + 1841 + ], + "score": 0.91, + "latex": "\\begin{array} { r l } & { \\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) \\approx \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) ( \\alpha \\to 0 ) } \\\\ & { \\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) \\approx \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) ( \\alpha \\to 0 ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 511, + 1585, + 884, + 1585, + 884, + 1621, + 511, + 1621 + ], + "score": 0.91, + "latex": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\dot { \\mathbf { X } } ) \\big | \\big | p ( \\dot { \\mathbf { z } } | \\mathbf { X } ) \\big )" + }, + { + "category_id": 13, + "poly": [ + 471, + 561, + 525, + 561, + 525, + 593, + 471, + 593 + ], + "score": 0.91, + "latex": "L _ { M \\mathbf { z } }" + }, + { + "category_id": 13, + "poly": [ + 818, + 1911, + 1213, + 1911, + 1213, + 1948, + 818, + 1948 + ], + "score": 0.91, + "latex": "\\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\big | \\big | p ( \\mathbf { z } | \\mathbf { X } ) \\big )" + }, + { + "category_id": 13, + "poly": [ + 1016, + 1130, + 1122, + 1130, + 1122, + 1162, + 1016, + 1162 + ], + "score": 0.91, + "latex": "p _ { e m p } ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1044, + 1160, + 1298, + 1160, + 1298, + 1194, + 1044, + 1194 + ], + "score": 0.91, + "latex": "\\hat { D _ { \\mathrm { K L } } } ( p _ { e m p } ( \\mathbf { X } ) | | p ( \\mathbf { X } ) )" + }, + { + "category_id": 14, + "poly": [ + 318, + 1458, + 1380, + 1458, + 1380, + 1501, + 318, + 1501 + ], + "score": 0.9, + "latex": "\\mathbb { E } _ { p _ { \\epsilon m p } ( \\mathbf { X } ) } [ \\log p ( \\mathbf { X } ) - \\mathrm { E L B O } ] = \\mathbb { E } _ { p _ { \\epsilon m p } ( \\mathbf { X } ) } [ D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | | p ( \\mathbf { c } | \\mathbf { X } ) ) ]" + }, + { + "category_id": 13, + "poly": [ + 297, + 1911, + 693, + 1911, + 693, + 1948, + 297, + 1948 + ], + "score": 0.9, + "latex": "\\mathbb { E } _ { p _ { m i x u p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\big | \\big | p \\big ( \\mathbf { z } | \\bar { \\mathbf { X } } \\big ) \\big )" + }, + { + "category_id": 14, + "poly": [ + 520, + 1239, + 1180, + 1239, + 1180, + 1281, + 520, + 1281 + ], + "score": 0.9, + "latex": "D _ { \\mathrm { K L } } ( p _ { e m p } ( \\mathbf { X } ) \\| p ( \\mathbf { X } ) ) \\leq H ( p _ { e m p } ( \\mathbf { X } ) ) - \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\mathrm { E L B O }" + }, + { + "category_id": 13, + "poly": [ + 686, + 555, + 956, + 555, + 956, + 594, + 686, + 594 + ], + "score": 0.9, + "latex": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) | | p ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) )" + }, + { + "category_id": 13, + "poly": [ + 298, + 293, + 566, + 293, + 566, + 325, + 298, + 325 + ], + "score": 0.89, + "latex": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) )" + }, + { + "category_id": 13, + "poly": [ + 559, + 397, + 715, + 397, + 715, + 431, + 559, + 431 + ], + "score": 0.89, + "latex": "( \\mu , \\mathrm { d i a g } ( \\sigma ^ { 2 } ) )" + }, + { + "category_id": 13, + "poly": [ + 940, + 1584, + 1313, + 1584, + 1313, + 1620, + 940, + 1620 + ], + "score": 0.89, + "latex": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) \\big )" + }, + { + "category_id": 13, + "poly": [ + 687, + 595, + 956, + 595, + 956, + 630, + 687, + 630 + ], + "score": 0.89, + "latex": "D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) | | p ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) )" + }, + { + "category_id": 13, + "poly": [ + 471, + 598, + 525, + 598, + 525, + 629, + 471, + 629 + ], + "score": 0.89, + "latex": "L _ { M _ { \\mathbf { c } } }" + }, + { + "category_id": 13, + "poly": [ + 616, + 431, + 910, + 431, + 910, + 464, + 616, + 464 + ], + "score": 0.88, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) = \\mathbf { M } \\mathbf { u } \\mathrm { l t } ( \\mathbf { c } ; K , \\pi )" + }, + { + "category_id": 13, + "poly": [ + 344, + 902, + 703, + 902, + 703, + 937, + 344, + 937 + ], + "score": 0.88, + "latex": "{ \\cal L } _ { M _ { \\mathbf { c } } } = { \\cal D } _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { c } } | \\tilde { \\mathbf { X } } ) | | \\tilde { p } ( { \\mathbf { c } } | \\tilde { \\mathbf { X } } ) )" + }, + { + "category_id": 13, + "poly": [ + 344, + 865, + 704, + 865, + 704, + 901, + 344, + 901 + ], + "score": 0.87, + "latex": "{ \\cal L } _ { M _ { \\mathbf { z } } } = { \\cal D } _ { \\mathrm { K L } } ( q _ { \\phi } ( { \\mathbf { z } } | \\tilde { \\mathbf { X } } ) | | \\tilde { p } ( { \\mathbf { z } } | \\tilde { \\mathbf { X } } ) )" + }, + { + "category_id": 13, + "poly": [ + 751, + 398, + 1097, + 398, + 1097, + 432, + 751, + 432 + ], + "score": 0.86, + "latex": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) = \\mathcal { N } ( \\mathbf { z } ; \\pmb { \\mu } , \\mathrm { d i a g } ( \\pmb { \\sigma } ^ { 2 } ) ) ;" + }, + { + "category_id": 13, + "poly": [ + 345, + 832, + 622, + 832, + 622, + 865, + 345, + 865 + ], + "score": 0.86, + "latex": "\\tilde { p } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) = \\mathbf { M u l t } ( \\mathbf { c } ; K , \\tilde { \\pi } )" + }, + { + "category_id": 13, + "poly": [ + 345, + 797, + 677, + 797, + 677, + 831, + 345, + 831 + ], + "score": 0.85, + "latex": "\\tilde { p } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) = \\mathcal { N } ( z ; \\tilde { \\mu } , \\mathrm { d i a g } ( \\tilde { \\pmb { \\sigma } } ^ { 2 } ) )" + }, + { + "category_id": 13, + "poly": [ + 776, + 631, + 930, + 631, + 930, + 662, + 776, + 662 + ], + "score": 0.85, + "latex": "( \\mathbf { X } , \\pi , \\mu , \\sigma ^ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 345, + 756, + 689, + 756, + 689, + 795, + 345, + 795 + ], + "score": 0.85, + "latex": "q _ { \\phi } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) , q _ { \\phi } ( \\mathbf { c } | \\tilde { \\mathbf { X } } ) = \\mathrm { V A E } ( \\tilde { \\mathbf { X } } )" + }, + { + "category_id": 13, + "poly": [ + 345, + 662, + 653, + 662, + 653, + 697, + 345, + 697 + ], + "score": 0.85, + "latex": "\\tilde { \\mathbf { X } } = \\lambda * \\mathbf { X } + \\left( 1 - \\lambda \\right) * \\mathbf { X } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 889, + 461, + 1017, + 461, + 1017, + 494, + 889, + 494 + ], + "score": 0.84, + "latex": "p _ { m i x u p } ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 433, + 938, + 589, + 938, + 589, + 974, + 433, + 974 + ], + "score": 0.83, + "latex": "\\tilde { \\mathbf { X } } , L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } }" + }, + { + "category_id": 13, + "poly": [ + 531, + 465, + 553, + 465, + 553, + 487, + 531, + 487 + ], + "score": 0.79, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 1248, + 1857, + 1270, + 1857, + 1270, + 1878, + 1248, + 1878 + ], + "score": 0.77, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 558, + 435, + 581, + 435, + 581, + 457, + 558, + 457 + ], + "score": 0.76, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 344, + 520, + 372, + 520, + 372, + 552, + 344, + 552 + ], + "score": 0.73, + "latex": "\\tilde { \\mathbf { X } }" + }, + { + "category_id": 13, + "poly": [ + 662, + 664, + 800, + 664, + 800, + 697, + 662, + 697 + ], + "score": 0.69, + "latex": "\\lambda \\in \\beta ( \\alpha , \\alpha )" + }, + { + "category_id": 13, + "poly": [ + 329, + 628, + 558, + 628, + 558, + 661, + 329, + 661 + ], + "score": 0.69, + "latex": ": \\mathrm { \\bf ~ X } ^ { \\prime } , \\pi ^ { \\prime } , \\mu ^ { \\prime } , \\sigma ^ { \\prime 2 } = \\mathrm { \\bf F }" + }, + { + "category_id": 13, + "poly": [ + 617, + 697, + 729, + 697, + 729, + 727, + 617, + 727 + ], + "score": 0.67, + "latex": "\\prime ( \\pi , \\pi ^ { \\prime } , \\lambda )" + }, + { + "category_id": 13, + "poly": [ + 581, + 366, + 609, + 366, + 609, + 393, + 581, + 393 + ], + "score": 0.63, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 479, + 260, + 578, + 260, + 578, + 293, + 479, + 293 + ], + "score": 0.6, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 782, + 1949, + 801, + 1949, + 801, + 1970, + 782, + 1970 + ], + "score": 0.57, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 1109, + 231, + 1136, + 231, + 1136, + 258, + 1109, + 258 + ], + "score": 0.49, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 346, + 725, + 467, + 725, + 467, + 758, + 346, + 758 + ], + "score": 0.43, + "latex": "( \\tilde { \\mu } , \\tilde { \\sigma } ^ { 2 } ) =" + }, + { + "category_id": 13, + "poly": [ + 369, + 260, + 468, + 260, + 468, + 293, + 369, + 293 + ], + "score": 0.36, + "latex": "q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 859, + 1951, + 876, + 1951, + 876, + 1970, + 859, + 1970 + ], + "score": 0.27, + "latex": "\\mathbf { c }" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 483.0, + 410.0, + 483.0, + 410.0, + 535.0, + 298.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 332.0, + 388.0, + 332.0, + 388.0, + 381.0, + 300.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1037.0, + 790.0, + 1037.0, + 790.0, + 1077.0, + 293.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2119.0, + 838.0, + 2119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1037.0, + 790.0, + 1037.0, + 790.0, + 1077.0, + 293.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1519.0, + 1405.0, + 1519.0, + 1405.0, + 1558.0, + 294.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1551.0, + 737.0, + 1551.0, + 737.0, + 1589.0, + 293.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1551.0, + 929.0, + 1551.0, + 929.0, + 1589.0, + 878.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1551.0, + 1405.0, + 1551.0, + 1405.0, + 1589.0, + 1039.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1577.0, + 510.0, + 1577.0, + 510.0, + 1630.0, + 288.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 1577.0, + 939.0, + 1577.0, + 939.0, + 1630.0, + 885.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 1577.0, + 1410.0, + 1577.0, + 1410.0, + 1630.0, + 1314.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1614.0, + 1406.0, + 1614.0, + 1406.0, + 1649.0, + 293.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1645.0, + 1404.0, + 1645.0, + 1404.0, + 1680.0, + 295.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1676.0, + 505.0, + 1676.0, + 505.0, + 1711.0, + 294.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 1676.0, + 647.0, + 1676.0, + 647.0, + 1711.0, + 591.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 1676.0, + 1404.0, + 1676.0, + 1404.0, + 1711.0, + 733.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1705.0, + 1241.0, + 1705.0, + 1241.0, + 1743.0, + 294.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1291.0, + 1405.0, + 1291.0, + 1405.0, + 1328.0, + 295.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1320.0, + 1405.0, + 1320.0, + 1405.0, + 1357.0, + 295.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1352.0, + 1405.0, + 1352.0, + 1405.0, + 1389.0, + 295.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1381.0, + 1405.0, + 1381.0, + 1405.0, + 1419.0, + 295.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1412.0, + 1239.0, + 1412.0, + 1239.0, + 1449.0, + 295.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 379.0, + 1849.0, + 379.0, + 1888.0, + 294.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1849.0, + 1247.0, + 1849.0, + 1247.0, + 1888.0, + 510.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1849.0, + 1407.0, + 1849.0, + 1407.0, + 1888.0, + 1271.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1879.0, + 1408.0, + 1879.0, + 1408.0, + 1919.0, + 292.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1899.0, + 296.0, + 1899.0, + 296.0, + 1978.0, + 283.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1899.0, + 781.0, + 1899.0, + 781.0, + 1978.0, + 694.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 1899.0, + 817.0, + 1899.0, + 817.0, + 1978.0, + 802.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 1899.0, + 1412.0, + 1899.0, + 1412.0, + 1978.0, + 1214.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2000.0, + 582.0, + 2000.0, + 582.0, + 2036.0, + 293.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1094.0, + 1105.0, + 1094.0, + 1105.0, + 1134.0, + 294.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1094.0, + 1262.0, + 1094.0, + 1262.0, + 1134.0, + 1206.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 1094.0, + 1404.0, + 1094.0, + 1404.0, + 1134.0, + 1363.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1127.0, + 1015.0, + 1127.0, + 1015.0, + 1168.0, + 293.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1127.0, + 1407.0, + 1127.0, + 1407.0, + 1168.0, + 1123.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1159.0, + 1043.0, + 1159.0, + 1043.0, + 1196.0, + 755.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1159.0, + 1405.0, + 1159.0, + 1405.0, + 1196.0, + 1299.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1189.0, + 1318.0, + 1189.0, + 1318.0, + 1229.0, + 294.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 228.0, + 1108.0, + 228.0, + 1108.0, + 262.0, + 296.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 262.0, + 1137.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 259.0, + 368.0, + 259.0, + 368.0, + 297.0, + 294.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 259.0, + 478.0, + 259.0, + 478.0, + 297.0, + 469.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 259.0, + 1406.0, + 259.0, + 1406.0, + 297.0, + 579.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 291.0, + 619.0, + 291.0, + 619.0, + 325.0, + 567.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 291.0, + 901.0, + 291.0, + 901.0, + 325.0, + 889.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 363.0, + 580.0, + 363.0, + 580.0, + 401.0, + 341.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 363.0, + 771.0, + 363.0, + 771.0, + 401.0, + 610.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 363.0, + 890.0, + 363.0, + 890.0, + 401.0, + 878.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 394.0, + 558.0, + 394.0, + 558.0, + 436.0, + 340.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 394.0, + 750.0, + 394.0, + 750.0, + 436.0, + 716.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 394.0, + 1105.0, + 394.0, + 1105.0, + 436.0, + 1098.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 424.0, + 557.0, + 424.0, + 557.0, + 465.0, + 340.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 424.0, + 615.0, + 424.0, + 615.0, + 465.0, + 582.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 424.0, + 924.0, + 424.0, + 924.0, + 465.0, + 911.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 455.0, + 530.0, + 455.0, + 530.0, + 499.0, + 340.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 455.0, + 888.0, + 455.0, + 888.0, + 499.0, + 554.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 455.0, + 1022.0, + 455.0, + 1022.0, + 499.0, + 1018.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 517.0, + 534.0, + 517.0, + 534.0, + 562.0, + 373.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 517.0, + 677.0, + 517.0, + 677.0, + 562.0, + 664.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 556.0, + 470.0, + 556.0, + 470.0, + 595.0, + 339.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 556.0, + 685.0, + 556.0, + 685.0, + 595.0, + 526.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 556.0, + 968.0, + 556.0, + 968.0, + 595.0, + 957.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 593.0, + 470.0, + 593.0, + 470.0, + 632.0, + 341.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 593.0, + 686.0, + 593.0, + 686.0, + 632.0, + 526.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 593.0, + 960.0, + 593.0, + 960.0, + 632.0, + 957.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 633.0, + 328.0, + 633.0, + 328.0, + 658.0, + 313.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 626.0, + 775.0, + 626.0, + 775.0, + 664.0, + 559.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 626.0, + 935.0, + 626.0, + 935.0, + 664.0, + 931.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 658.0, + 344.0, + 658.0, + 344.0, + 701.0, + 311.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 658.0, + 661.0, + 658.0, + 661.0, + 701.0, + 654.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 658.0, + 805.0, + 658.0, + 805.0, + 701.0, + 801.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 694.0, + 616.0, + 694.0, + 616.0, + 730.0, + 309.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 694.0, + 735.0, + 694.0, + 735.0, + 730.0, + 730.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 720.0, + 345.0, + 720.0, + 345.0, + 762.0, + 311.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 720.0, + 944.0, + 720.0, + 944.0, + 762.0, + 468.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 757.0, + 344.0, + 757.0, + 344.0, + 796.0, + 310.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 794.0, + 344.0, + 794.0, + 344.0, + 832.0, + 310.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 827.0, + 344.0, + 827.0, + 344.0, + 868.0, + 305.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 827.0, + 628.0, + 827.0, + 628.0, + 868.0, + 623.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 862.0, + 343.0, + 862.0, + 343.0, + 904.0, + 305.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 862.0, + 708.0, + 862.0, + 708.0, + 904.0, + 705.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 898.0, + 343.0, + 898.0, + 343.0, + 942.0, + 304.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 898.0, + 708.0, + 898.0, + 708.0, + 942.0, + 704.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 932.0, + 432.0, + 932.0, + 432.0, + 980.0, + 304.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 932.0, + 595.0, + 932.0, + 595.0, + 980.0, + 590.0, + 980.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 783, + 1405, + 783, + 1405, + 1010, + 296, + 1010 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1556, + 1405, + 1556, + 1405, + 1650, + 298, + 1650 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1176, + 1402, + 1176, + 1402, + 1271, + 298, + 1271 + ], + "score": 0.966 + }, + { + "category_id": 8, + "poly": [ + 717, + 1346, + 980, + 1346, + 980, + 1420, + 717, + 1420 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 297, + 1740, + 1404, + 1740, + 1404, + 1928, + 297, + 1928 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 300, + 1941, + 1402, + 1941, + 1402, + 2035, + 300, + 2035 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 294, + 1427, + 1406, + 1427, + 1406, + 1492, + 294, + 1492 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 293, + 1276, + 1399, + 1276, + 1399, + 1340, + 293, + 1340 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 720, + 1498, + 978, + 1498, + 978, + 1534, + 720, + 1534 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 342, + 302, + 1234, + 302, + 1234, + 459, + 342, + 459 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 591, + 1017, + 1106, + 1017, + 1106, + 1168, + 591, + 1168 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 301, + 1684, + 747, + 1684, + 747, + 1717, + 301, + 1717 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1075, + 1400, + 1075, + 1400, + 1106, + 1352, + 1106 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1369, + 1399, + 1369, + 1399, + 1399, + 1352, + 1399 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1500, + 1399, + 1500, + 1399, + 1530, + 1352, + 1530 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.794 + }, + { + "category_id": 0, + "poly": [ + 305, + 276, + 383, + 276, + 383, + 303, + 305, + 303 + ], + "score": 0.788 + }, + { + "category_id": 0, + "poly": [ + 304, + 461, + 403, + 461, + 403, + 488, + 304, + 488 + ], + "score": 0.783 + }, + { + "category_id": 1, + "poly": [ + 308, + 486, + 1141, + 486, + 1141, + 728, + 308, + 728 + ], + "score": 0.596 + }, + { + "category_id": 0, + "poly": [ + 298, + 228, + 933, + 228, + 933, + 262, + 298, + 262 + ], + "score": 0.492 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 933, + 228, + 933, + 262, + 298, + 262 + ], + "score": 0.387 + }, + { + "category_id": 8, + "poly": [ + 308, + 486, + 1141, + 486, + 1141, + 728, + 308, + 728 + ], + "score": 0.218 + }, + { + "category_id": 13, + "poly": [ + 1106, + 1177, + 1300, + 1177, + 1300, + 1211, + 1106, + 1211 + ], + "score": 0.93, + "latex": "\\mathbf { c } \\sim \\mathbf { M } \\mathbf { u } \\mathrm { l t } ( K , \\pi )" + }, + { + "category_id": 13, + "poly": [ + 562, + 394, + 723, + 394, + 723, + 430, + 562, + 430 + ], + "score": 0.93, + "latex": "\\pmb \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 493, + 976, + 572, + 976, + 572, + 1010, + 493, + 1010 + ], + "score": 0.93, + "latex": "\\gamma ( \\mathbf { x } , \\mathbf { y } )" + }, + { + "category_id": 13, + "poly": [ + 626, + 911, + 803, + 911, + 803, + 949, + 626, + 949 + ], + "score": 0.93, + "latex": "\\tilde { p } ( \\mathbf { z } | \\tilde { \\mathbf { X } } ) , \\tilde { p } ( \\mathbf { c } | \\tilde { \\mathbf { X } } )" + }, + { + "category_id": 13, + "poly": [ + 1208, + 1865, + 1353, + 1865, + 1353, + 1896, + 1208, + 1896 + ], + "score": 0.92, + "latex": "\\mathbf { I } _ { z } , \\mathbf { I } _ { c } , \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } }" + }, + { + "category_id": 13, + "poly": [ + 922, + 1459, + 1029, + 1459, + 1029, + 1492, + 922, + 1492 + ], + "score": 0.92, + "latex": "\\lambda \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 522, + 1834, + 739, + 1834, + 739, + 1868, + 522, + 1868 + ], + "score": 0.92, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\approx p ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 869, + 1307, + 976, + 1307, + 976, + 1340, + 869, + 1340 + ], + "score": 0.92, + "latex": "\\lambda \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 298, + 1305, + 495, + 1305, + 495, + 1340, + 298, + 1340 + ], + "score": 0.92, + "latex": "\\mathcal { N } ( \\bar { \\mu } _ { 1 } , d i a g ( \\sigma _ { 1 } ^ { 2 } ) )" + }, + { + "category_id": 13, + "poly": [ + 395, + 1459, + 602, + 1459, + 602, + 1492, + 395, + 1492 + ], + "score": 0.92, + "latex": "\\mathbf { c } _ { 1 } \\sim M u l t ( K , \\pmb { \\pi } _ { 1 } )" + }, + { + "category_id": 13, + "poly": [ + 869, + 977, + 921, + 977, + 921, + 1009, + 869, + 1009 + ], + "score": 0.92, + "latex": "p ( \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 520, + 1973, + 601, + 1973, + 601, + 2007, + 520, + 2007 + ], + "score": 0.92, + "latex": "\\nabla _ { \\pmb { \\theta } , \\pmb { \\phi } } L" + }, + { + "category_id": 13, + "poly": [ + 615, + 1176, + 859, + 1176, + 859, + 1211, + 615, + 1211 + ], + "score": 0.91, + "latex": "\\mathbf z \\sim \\mathcal N ( \\pmb \\mu , \\mathrm { d i a g } ( \\pmb \\sigma ^ { 2 } ) )" + }, + { + "category_id": 13, + "poly": [ + 1105, + 302, + 1223, + 302, + 1223, + 334, + 1105, + 334 + ], + "score": 0.91, + "latex": "\\mathbf { X } _ { U } \\in \\mathbb { D } _ { U }" + }, + { + "category_id": 13, + "poly": [ + 655, + 1459, + 863, + 1459, + 863, + 1492, + 655, + 1492 + ], + "score": 0.91, + "latex": "\\mathbf { c } _ { 2 } \\sim M u l t ( K , \\pmb { \\pi } _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 595, + 302, + 773, + 302, + 773, + 334, + 595, + 334 + ], + "score": 0.9, + "latex": "( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ) \\in \\mathbb { D } _ { L }" + }, + { + "category_id": 13, + "poly": [ + 622, + 816, + 752, + 816, + 752, + 848, + 622, + 848 + ], + "score": 0.9, + "latex": "\\mathbf { X } , \\mathbf { X ^ { \\prime } } \\in \\mathbb { D }" + }, + { + "category_id": 13, + "poly": [ + 972, + 977, + 1024, + 977, + 1024, + 1010, + 972, + 1010 + ], + "score": 0.9, + "latex": "p ( \\mathsf { y } )" + }, + { + "category_id": 13, + "poly": [ + 548, + 1306, + 808, + 1306, + 808, + 1340, + 548, + 1340 + ], + "score": 0.9, + "latex": "\\mathbf { z } _ { 2 } \\sim \\mathcal { N } ( \\bar { \\pmb { \\mu } } _ { 2 } , d i a g ( \\pmb { \\sigma } _ { 2 } ^ { 2 } ) )" + }, + { + "category_id": 14, + "poly": [ + 715, + 1344, + 982, + 1344, + 982, + 1425, + 715, + 1425 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\tilde { \\pmb { \\mu } } = \\lambda \\pmb { \\mu } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\mu } _ { 2 } } \\\\ { \\tilde { \\pmb { \\sigma } } = \\lambda \\pmb { \\sigma } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\sigma } _ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1898, + 418, + 1898, + 418, + 1928, + 297, + 1928 + ], + "score": 0.89, + "latex": "w _ { M _ { \\mathbf { z } } } , w _ { M _ { \\mathbf { c } } }" + }, + { + "category_id": 14, + "poly": [ + 719, + 1497, + 980, + 1497, + 980, + 1534, + 719, + 1534 + ], + "score": 0.88, + "latex": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 482, + 847, + 905, + 847, + 905, + 883, + 482, + 883 + ], + "score": 0.88, + "latex": "\\tilde { \\mathbf { X } } = \\lambda * \\mathbf { X } + ( 1 - \\lambda ) * \\mathbf { X ^ { \\prime } } , \\lambda \\in [ 0 , 1 ] ." + }, + { + "category_id": 14, + "poly": [ + 597, + 1015, + 1106, + 1015, + 1106, + 1169, + 597, + 1169 + ], + "score": 0.84, + "latex": "\\begin{array} { c } { { \\displaystyle \\operatorname* { i n f } _ { \\gamma ( \\mathbf { x } , \\mathbf { y } ) } \\int _ { \\mathbf { x } } \\int _ { \\mathbf { y } } \\| \\mathbf { x } - \\mathbf { y } \\| _ { 2 } ^ { 2 } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { x } d \\mathbf { y } } } \\\\ { { \\mathrm { s . t . } ~ \\displaystyle \\int _ { \\mathbf { y } } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { y } = p ( \\mathbf { x } ) ; \\displaystyle \\int _ { \\mathbf { x } } \\gamma ( \\mathbf { x } , \\mathbf { y } ) d \\mathbf { x } = p ( \\mathbf { y } ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 345, + 522, + 899, + 522, + 899, + 553, + 345, + 553 + ], + "score": 0.84, + "latex": "\\begin{array} { r } { L _ { L } = \\mathcal { L } _ { \\mathbb { D } _ { L } } ( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ; \\pmb { \\theta } ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 343, + 651, + 820, + 651, + 820, + 690, + 343, + 690 + ], + "score": 0.8, + "latex": "\\begin{array} { r } { \\pmb { \\theta } ^ { ( t ) } , \\pmb { \\phi } ^ { ( t ) } = \\mathrm { S G D } ( \\pmb { \\theta } ^ { ( t - 1 ) } , \\pmb { \\phi } ^ { ( t - 1 ) } , \\frac { \\partial L } { \\partial \\pmb { \\theta } } , \\frac { \\partial L } { \\partial \\phi } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 960, + 846, + 987, + 846, + 987, + 879, + 960, + 879 + ], + "score": 0.8, + "latex": "\\mathbf { \\tilde { X } }" + }, + { + "category_id": 13, + "poly": [ + 347, + 555, + 864, + 555, + 864, + 589, + 347, + 589 + ], + "score": 0.8, + "latex": "L _ { U } = { \\mathcal { L } } _ { { \\mathbb { D } } _ { U } } \\big ( \\mathbf { X } _ { U } ; \\mathbf { \\pmb { \\theta } } ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big )" + }, + { + "category_id": 13, + "poly": [ + 807, + 1974, + 831, + 1974, + 831, + 2000, + 807, + 2000 + ], + "score": 0.79, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 345, + 589, + 505, + 589, + 505, + 619, + 345, + 619 + ], + "score": 0.78, + "latex": "L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } } = " + }, + { + "category_id": 13, + "poly": [ + 624, + 334, + 1021, + 334, + 1021, + 365, + 624, + 365 + ], + "score": 0.78, + "latex": "\\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } = \\mathrm { E L B O S c h e d u l e r } ( t )" + }, + { + "category_id": 13, + "poly": [ + 434, + 688, + 539, + 688, + 539, + 725, + 434, + 725 + ], + "score": 0.76, + "latex": "\\theta ^ { ( t ) } , \\phi ^ { ( t ) }" + }, + { + "category_id": 13, + "poly": [ + 1333, + 1281, + 1404, + 1281, + 1404, + 1308, + 1333, + 1308 + ], + "score": 0.75, + "latex": "\\mathbf { z } _ { 1 } ~ \\sim" + }, + { + "category_id": 13, + "poly": [ + 583, + 484, + 687, + 484, + 687, + 520, + 583, + 520 + ], + "score": 0.75, + "latex": "\\pmb { \\theta } ^ { ( t ) } , \\phi ^ { ( t ) }" + }, + { + "category_id": 13, + "poly": [ + 797, + 366, + 929, + 366, + 929, + 397, + 797, + 397 + ], + "score": 0.7, + "latex": "M _ { \\mathbf { z } } \\mathbf { , } w _ { } M _ { \\mathbf { c } } =" + }, + { + "category_id": 13, + "poly": [ + 861, + 1430, + 901, + 1430, + 901, + 1456, + 861, + 1456 + ], + "score": 0.69, + "latex": "K L" + }, + { + "category_id": 13, + "poly": [ + 348, + 621, + 804, + 621, + 804, + 650, + 348, + 650 + ], + "score": 0.67, + "latex": "L = L _ { L } + L _ { U } + w _ { M _ { \\mathbf { z } } } L _ { M _ { \\mathbf { z } } } + w _ { M _ { \\mathbf { c } } } L _ { M _ { \\mathbf { c } } }" + }, + { + "category_id": 13, + "poly": [ + 1158, + 1895, + 1204, + 1895, + 1204, + 1928, + 1158, + 1928 + ], + "score": 0.67, + "latex": "( t ) ^ { , }" + }, + { + "category_id": 13, + "poly": [ + 860, + 1895, + 892, + 1895, + 892, + 1928, + 860, + 1928 + ], + "score": 0.61, + "latex": "( t )" + }, + { + "category_id": 13, + "poly": [ + 1119, + 363, + 1152, + 363, + 1152, + 397, + 1119, + 397 + ], + "score": 0.61, + "latex": "\\cdot ( t )" + }, + { + "category_id": 13, + "poly": [ + 913, + 234, + 926, + 234, + 926, + 257, + 913, + 257 + ], + "score": 0.55, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 989, + 332, + 1020, + 332, + 1020, + 365, + 989, + 365 + ], + "score": 0.52, + "latex": "( t )" + }, + { + "category_id": 13, + "poly": [ + 837, + 922, + 854, + 922, + 854, + 943, + 837, + 943 + ], + "score": 0.47, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 772, + 370, + 826, + 370, + 826, + 397, + 772, + 397 + ], + "score": 0.36, + "latex": "w _ { M _ { \\mathbf { z } } }" + }, + { + "category_id": 14, + "poly": [ + 343, + 499, + 1143, + 499, + 1143, + 696, + 343, + 696 + ], + "score": 0.28, + "latex": "\\begin{array} { r l } & { \\mathrm { U p d a t e d ~ p a r a m e t e r s : ~ } \\theta ^ { \\mathrm { { t x } ^ { \\prime } } } , \\phi ^ { \\mathrm { { t x } ^ { \\prime } } } , \\phi ^ { \\mathrm { { t x } ^ { \\prime } } } } \\\\ & { L _ { L } = \\mathcal { L } _ { \\mathbb { D } _ { L } } \\big ( \\mathbf { X } _ { L } , \\mathbf { y } _ { L } ; \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big ) } \\\\ & { L _ { U } = \\mathcal { L } _ { \\mathbb { D } _ { U } } \\big ( \\mathbf { X } _ { U } ; \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } ; \\beta _ { \\mathbf { z } } , \\beta _ { \\mathbf { c } } , \\mathbf { I } _ { \\mathbf { z } } , \\mathbf { I } _ { \\mathbf { c } } \\big ) } \\\\ & { L _ { M _ { \\mathbf { z } } } , L _ { M _ { \\mathbf { c } } } = \\mathrm { O p t i m a l T r a n s p o r t E s t i m a t i o n } \\big ( \\mathbf { X } _ { U } , q _ { \\phi } \\big ( \\mathbf { z } \\big | \\mathbf { X } _ { U } \\big ) , q _ { \\phi } \\big ( \\mathbf { c } | \\mathbf { X } _ { U } \\big ) \\big ) } \\\\ & { L = L _ { L } + L _ { U } + w _ { M _ { \\mathbf { z } } } L _ { M _ { \\mathbf { z } } } + w _ { M _ { \\mathbf { c } } } L _ { M _ { \\mathbf { c } } } } \\\\ & { \\theta ^ { ( t ) } , \\phi ^ { ( t ) } = \\mathrm { S G D } \\big ( \\theta ^ { ( t - 1 ) } , \\phi ^ { ( t - 1 ) } , \\frac { \\partial L } { \\partial \\theta } , \\frac { \\partial L } { \\partial \\phi } \\big ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 837, + 918, + 883, + 918, + 883, + 945, + 837, + 945 + ], + "score": 0.26, + "latex": "\\mathbf { z } , \\mathbf { c }" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1682.0, + 750.0, + 1682.0, + 750.0, + 1721.0, + 293.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 268.0, + 389.0, + 268.0, + 389.0, + 320.0, + 299.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 454.0, + 411.0, + 454.0, + 411.0, + 501.0, + 299.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 225.0, + 912.0, + 225.0, + 912.0, + 268.0, + 296.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 225.0, + 938.0, + 225.0, + 938.0, + 268.0, + 927.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 784.0, + 1405.0, + 784.0, + 1405.0, + 820.0, + 296.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 816.0, + 621.0, + 816.0, + 621.0, + 852.0, + 295.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 816.0, + 1405.0, + 816.0, + 1405.0, + 852.0, + 753.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 849.0, + 481.0, + 849.0, + 481.0, + 885.0, + 294.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 849.0, + 959.0, + 849.0, + 959.0, + 885.0, + 906.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 849.0, + 1405.0, + 849.0, + 1405.0, + 885.0, + 988.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 879.0, + 1405.0, + 879.0, + 1405.0, + 916.0, + 294.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 914.0, + 625.0, + 914.0, + 625.0, + 950.0, + 295.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 914.0, + 836.0, + 914.0, + 836.0, + 950.0, + 804.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 884.0, + 914.0, + 1405.0, + 914.0, + 1405.0, + 950.0, + 884.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 945.0, + 1405.0, + 945.0, + 1405.0, + 981.0, + 293.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 977.0, + 492.0, + 977.0, + 492.0, + 1013.0, + 295.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 977.0, + 868.0, + 977.0, + 868.0, + 1013.0, + 573.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 977.0, + 971.0, + 977.0, + 971.0, + 1013.0, + 922.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 977.0, + 1115.0, + 977.0, + 1115.0, + 1013.0, + 1025.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1557.0, + 1405.0, + 1557.0, + 1405.0, + 1591.0, + 296.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1587.0, + 1406.0, + 1587.0, + 1406.0, + 1621.0, + 295.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1620.0, + 1158.0, + 1620.0, + 1158.0, + 1650.0, + 296.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1175.0, + 614.0, + 1175.0, + 614.0, + 1213.0, + 293.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1175.0, + 1105.0, + 1175.0, + 1105.0, + 1213.0, + 860.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1175.0, + 1403.0, + 1175.0, + 1403.0, + 1213.0, + 1301.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1204.0, + 1406.0, + 1204.0, + 1406.0, + 1244.0, + 292.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1238.0, + 633.0, + 1238.0, + 633.0, + 1273.0, + 296.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1740.0, + 1404.0, + 1740.0, + 1404.0, + 1779.0, + 294.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1771.0, + 1407.0, + 1771.0, + 1407.0, + 1810.0, + 292.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1801.0, + 1404.0, + 1801.0, + 1404.0, + 1839.0, + 292.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1834.0, + 521.0, + 1834.0, + 521.0, + 1870.0, + 294.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1834.0, + 1404.0, + 1834.0, + 1404.0, + 1870.0, + 740.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1864.0, + 1207.0, + 1864.0, + 1207.0, + 1900.0, + 295.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1864.0, + 1406.0, + 1864.0, + 1406.0, + 1900.0, + 1354.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1893.0, + 296.0, + 1893.0, + 296.0, + 1932.0, + 291.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 1893.0, + 859.0, + 1893.0, + 859.0, + 1932.0, + 419.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1893.0, + 1157.0, + 1893.0, + 1157.0, + 1932.0, + 893.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1893.0, + 1406.0, + 1893.0, + 1406.0, + 1932.0, + 1205.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1938.0, + 1404.0, + 1938.0, + 1404.0, + 1978.0, + 293.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1970.0, + 519.0, + 1970.0, + 519.0, + 2010.0, + 293.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 1970.0, + 806.0, + 1970.0, + 806.0, + 2010.0, + 602.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1970.0, + 1403.0, + 1970.0, + 1403.0, + 2010.0, + 832.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 1404.0, + 2000.0, + 1404.0, + 2037.0, + 294.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1425.0, + 860.0, + 1425.0, + 860.0, + 1465.0, + 295.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 1425.0, + 1403.0, + 1425.0, + 1403.0, + 1465.0, + 902.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1458.0, + 394.0, + 1458.0, + 394.0, + 1494.0, + 295.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 1458.0, + 654.0, + 1458.0, + 654.0, + 1494.0, + 603.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 1458.0, + 921.0, + 1458.0, + 921.0, + 1494.0, + 864.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1458.0, + 1061.0, + 1458.0, + 1061.0, + 1494.0, + 1030.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1271.0, + 1332.0, + 1271.0, + 1332.0, + 1313.0, + 292.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1303.0, + 297.0, + 1303.0, + 297.0, + 1343.0, + 292.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1303.0, + 547.0, + 1303.0, + 547.0, + 1343.0, + 496.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1303.0, + 868.0, + 1303.0, + 868.0, + 1343.0, + 809.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 1303.0, + 1008.0, + 1303.0, + 1008.0, + 1343.0, + 977.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 296.0, + 594.0, + 296.0, + 594.0, + 340.0, + 339.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 296.0, + 1104.0, + 296.0, + 1104.0, + 340.0, + 774.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 296.0, + 1237.0, + 296.0, + 1237.0, + 340.0, + 1224.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 330.0, + 623.0, + 330.0, + 623.0, + 368.0, + 341.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 330.0, + 1034.0, + 330.0, + 1034.0, + 368.0, + 1022.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 360.0, + 771.0, + 360.0, + 771.0, + 401.0, + 340.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 360.0, + 1118.0, + 360.0, + 1118.0, + 401.0, + 930.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 360.0, + 1164.0, + 360.0, + 1164.0, + 401.0, + 1153.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 394.0, + 561.0, + 394.0, + 561.0, + 430.0, + 338.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 394.0, + 738.0, + 394.0, + 738.0, + 430.0, + 724.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 428.0, + 608.0, + 428.0, + 608.0, + 463.0, + 342.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 481.0, + 582.0, + 481.0, + 582.0, + 522.0, + 339.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 481.0, + 693.0, + 481.0, + 693.0, + 522.0, + 688.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 509.0, + 342.0, + 509.0, + 342.0, + 564.0, + 302.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 547.0, + 342.0, + 547.0, + 342.0, + 595.0, + 302.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 579.0, + 342.0, + 579.0, + 342.0, + 626.0, + 302.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 610.0, + 342.0, + 610.0, + 342.0, + 659.0, + 304.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 643.0, + 342.0, + 643.0, + 342.0, + 695.0, + 301.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 686.0, + 433.0, + 686.0, + 433.0, + 725.0, + 304.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 686.0, + 544.0, + 686.0, + 544.0, + 725.0, + 540.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 225.0, + 912.0, + 225.0, + 912.0, + 268.0, + 296.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 225.0, + 938.0, + 225.0, + 938.0, + 268.0, + 927.0, + 268.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 296, + 433, + 1445, + 433, + 1445, + 884, + 296, + 884 + ], + "score": 0.981, + "html": "
BackBoneModel categoryModelCifar10(4k labels)
WRN-28-2DisagreementTemporal Ensembling(TE) (Laine & Aila,2017)16.37
Mean Teacher(Tarvainen & Valpola,2017)15.87
13.13
VAT+EntMin(Miyato et al., 2019) MixMatch(Berthelot et al., 2019)6.37
GS-BadGANt*(Li et al., 2019)17.11
WRN-28-10 GenerativeGenerativeOSPOT-VAE8.51(±0.32)
AutoAugment(Cubuk et al., 2019)14.1
DisagreementTemporal Ensembing(Laine & Aila, 2017)12.16
MixMatch*(Berthelot et al., 2019)4.95
GS-BadGANt*(Li et al., 2019) GAN combine TE‡*(Wei et al., 2018)14.41
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1419, + 1404, + 1419, + 1404, + 1756, + 297, + 1756 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1849, + 1403, + 1849, + 1403, + 2034, + 298, + 2034 + ], + "score": 0.979 + }, + { + "category_id": 5, + "poly": [ + 297, + 107, + 1409, + 107, + 1409, + 340, + 297, + 340 + ], + "score": 0.976, + "html": "
MNIST(100 labels)SVHN(1k labels)
BackBoneMethod
Same with M1+M2Disentangled VAE9.71(±0.91)38.91(±1.06)
(Narayanaswamy et al., 2017)11.97(±1.71)54.33(±0.11)
M1(Kingma et al., 2014)3.33(±0.14)36.02(±0.10)
M1+M2(Kingma et al., 2014)
One-stage VAE3.14(±0.19)27.38(±0.78)
" + }, + { + "category_id": 1, + "poly": [ + 295, + 1243, + 1404, + 1243, + 1404, + 1308, + 295, + 1308 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 297, + 1046, + 1404, + 1046, + 1404, + 1111, + 297, + 1111 + ], + "score": 0.92 + }, + { + "category_id": 6, + "poly": [ + 526, + 374, + 1171, + 374, + 1171, + 407, + 526, + 407 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 300, + 1352, + 557, + 1352, + 557, + 1386, + 300, + 1386 + ], + "score": 0.909 + }, + { + "category_id": 8, + "poly": [ + 313, + 1122, + 1333, + 1122, + 1333, + 1231, + 313, + 1231 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 303, + 1792, + 798, + 1792, + 798, + 1824, + 303, + 1824 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1162, + 1399, + 1162, + 1399, + 1191, + 1353, + 1191 + ], + "score": 0.842 + }, + { + "category_id": 6, + "poly": [ + 298, + 74, + 859, + 74, + 859, + 105, + 298, + 105 + ], + "score": 0.759 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.668 + }, + { + "category_id": 1, + "poly": [ + 297, + 905, + 1408, + 905, + 1408, + 1030, + 297, + 1030 + ], + "score": 0.586 + }, + { + "category_id": 6, + "poly": [ + 297, + 905, + 1408, + 905, + 1408, + 1030, + 297, + 1030 + ], + "score": 0.385 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 859, + 2087, + 859, + 2111, + 841, + 2111 + ], + "score": 0.271 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 859, + 74, + 859, + 105, + 298, + 105 + ], + "score": 0.22 + }, + { + "category_id": 14, + "poly": [ + 313, + 1121, + 1336, + 1121, + 1336, + 1236, + 313, + 1236 + ], + "score": 0.9, + "latex": "\\begin{array} { r l } & { \\nabla _ { \\theta , \\phi } \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } ) = \\mathbb { E } _ { \\mathcal { N } ( \\epsilon ; \\mathbf { 0 } , \\mathbf { I } ) } \\nabla _ { \\theta , \\phi } \\log p _ { \\theta } ( \\mathbf { X } | \\mu + \\sigma \\cdot \\epsilon ) } \\\\ & { \\mathbb { E } _ { \\mathrm { G u m b e l } ( \\epsilon ; \\mathbf { 0 } , \\mathbf { 1 } ) } ) \\nabla _ { \\theta , \\phi } \\log p _ { \\theta } ( \\mathbf { X } | \\mathrm { S o f t m a x } ( \\frac { \\log \\pi + \\epsilon } { \\tau } ) ) \\to \\nabla _ { \\theta , \\phi } \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } ) ( \\tau \\to 0 ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 329, + 1049, + 682, + 1049, + 682, + 1085, + 329, + 1085 + ], + "score": 0.89, + "latex": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } )" + }, + { + "category_id": 13, + "poly": [ + 547, + 1277, + 575, + 1277, + 575, + 1303, + 547, + 1303 + ], + "score": 0.82, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 485, + 1973, + 582, + 1973, + 582, + 2003, + 485, + 2003 + ], + "score": 0.8, + "latex": "\\mathbf { M } 1 { + } \\mathbf { M } 2" + }, + { + "category_id": 13, + "poly": [ + 805, + 942, + 826, + 942, + 826, + 965, + 805, + 965 + ], + "score": 0.76, + "latex": "^ *" + }, + { + "category_id": 13, + "poly": [ + 659, + 908, + 680, + 908, + 680, + 940, + 659, + 940 + ], + "score": 0.76, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 1293, + 1912, + 1394, + 1912, + 1394, + 1942, + 1293, + 1942 + ], + "score": 0.67, + "latex": "[ \\mathbf { M } 1 + \\mathbf { M } 2 ]" + }, + { + "category_id": 13, + "poly": [ + 437, + 241, + 535, + 241, + 535, + 271, + 437, + 271 + ], + "score": 0.44, + "latex": "\\mathbf { M } 1 { + } \\mathbf { M } 2" + }, + { + "category_id": 13, + "poly": [ + 564, + 272, + 661, + 272, + 661, + 301, + 564, + 301 + ], + "score": 0.39, + "latex": "\\mathbf { M } 1 { + } \\mathbf { M } 2" + }, + { + "category_id": 13, + "poly": [ + 898, + 814, + 959, + 814, + 959, + 847, + 898, + 847 + ], + "score": 0.39, + "latex": "\\mathrm { T E ^ { \\ddagger * } }" + }, + { + "category_id": 13, + "poly": [ + 1382, + 907, + 1403, + 907, + 1403, + 939, + 1382, + 939 + ], + "score": 0.36, + "latex": "\\ddagger" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 373.0, + 1171.0, + 373.0, + 1171.0, + 409.0, + 529.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1349.0, + 561.0, + 1349.0, + 561.0, + 1392.0, + 293.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1791.0, + 802.0, + 1791.0, + 802.0, + 1828.0, + 296.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 903.0, + 658.0, + 903.0, + 658.0, + 942.0, + 292.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 903.0, + 1381.0, + 903.0, + 1381.0, + 942.0, + 681.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 903.0, + 1408.0, + 903.0, + 1408.0, + 942.0, + 1404.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 938.0, + 804.0, + 938.0, + 804.0, + 971.0, + 295.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 938.0, + 1404.0, + 938.0, + 1404.0, + 971.0, + 827.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 966.0, + 1408.0, + 966.0, + 1408.0, + 1001.0, + 295.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1000.0, + 689.0, + 1000.0, + 689.0, + 1033.0, + 296.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1421.0, + 1404.0, + 1421.0, + 1404.0, + 1452.0, + 295.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1450.0, + 1405.0, + 1450.0, + 1405.0, + 1483.0, + 292.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1479.0, + 1405.0, + 1479.0, + 1405.0, + 1517.0, + 292.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1512.0, + 1405.0, + 1512.0, + 1405.0, + 1547.0, + 296.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1540.0, + 1404.0, + 1540.0, + 1404.0, + 1575.0, + 294.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1573.0, + 1405.0, + 1573.0, + 1405.0, + 1606.0, + 292.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1602.0, + 1404.0, + 1602.0, + 1404.0, + 1637.0, + 294.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1632.0, + 1404.0, + 1632.0, + 1404.0, + 1667.0, + 292.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1662.0, + 1402.0, + 1662.0, + 1402.0, + 1698.0, + 292.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1692.0, + 1404.0, + 1692.0, + 1404.0, + 1732.0, + 294.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1724.0, + 435.0, + 1724.0, + 435.0, + 1757.0, + 292.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1851.0, + 1403.0, + 1851.0, + 1403.0, + 1883.0, + 297.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1880.0, + 1405.0, + 1880.0, + 1405.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1911.0, + 1292.0, + 1911.0, + 1292.0, + 1947.0, + 296.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1911.0, + 1405.0, + 1911.0, + 1405.0, + 1947.0, + 1395.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1941.0, + 1405.0, + 1941.0, + 1405.0, + 1977.0, + 296.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1971.0, + 484.0, + 1971.0, + 484.0, + 2006.0, + 293.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2006.0, + 583.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 645.0, + 2003.0, + 645.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1242.0, + 1404.0, + 1242.0, + 1404.0, + 1278.0, + 295.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1273.0, + 546.0, + 1273.0, + 546.0, + 1310.0, + 294.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1273.0, + 1255.0, + 1273.0, + 1255.0, + 1310.0, + 576.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1042.0, + 328.0, + 1042.0, + 328.0, + 1091.0, + 292.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1042.0, + 1409.0, + 1042.0, + 1409.0, + 1091.0, + 683.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1076.0, + 1040.0, + 1076.0, + 1040.0, + 1115.0, + 295.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 903.0, + 658.0, + 903.0, + 658.0, + 942.0, + 292.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 903.0, + 1381.0, + 903.0, + 1381.0, + 942.0, + 681.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 903.0, + 1408.0, + 903.0, + 1408.0, + 942.0, + 1404.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 938.0, + 804.0, + 938.0, + 804.0, + 971.0, + 295.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 938.0, + 1404.0, + 938.0, + 1404.0, + 971.0, + 827.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 966.0, + 1408.0, + 966.0, + 1408.0, + 1001.0, + 295.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1000.0, + 689.0, + 1000.0, + 689.0, + 1033.0, + 296.0, + 1033.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 754, + 1405, + 754, + 1405, + 1031, + 297, + 1031 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1849, + 1404, + 1849, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 1044, + 1405, + 1044, + 1405, + 1292, + 296, + 1292 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1552, + 1405, + 1552, + 1405, + 1738, + 297, + 1738 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 296, + 194, + 1429, + 194, + 1429, + 432, + 296, + 432 + ], + "score": 0.977, + "html": "
BackBoneModelCifar100(4k labels)Cifar100(10k labels)
WRN-28-2II - Model(Laine & Aila, 2017)39.19
GS-BadGANt*(Li et al., 2019)45.1137.16
LP*(Iscen et al., 2019)43.7335.92
OSPOT-VAE40.58(±0.48)31.41(±0.21)
WRN-28-10MixMatch* (Berthelot et al., 2019)25.88
OSPOT-VAE33.76(±0.53)25.30(±0.31)
" + }, + { + "category_id": 5, + "poly": [ + 298, + 510, + 1406, + 510, + 1406, + 676, + 298, + 676 + ], + "score": 0.975, + "html": "
MethodsSVHN(1k labels)Cifar10(4k labels)Cifar100(10k labels)
One-stage VAE10.53(±0.17)18.26(±0.51)38.62(±0.67)
Optimal transport estimation (with encoder only)6.54(±0.62)10.71(±0.44)36.21(±0.29)
OSPOT-VAE5.79(±0.15)8.51(±0.32)31.41(±0.21)
" + }, + { + "category_id": 5, + "poly": [ + 505, + 1357, + 1186, + 1357, + 1186, + 1462, + 505, + 1462 + ], + "score": 0.96, + "html": "
ModelCifar10Cifar100
Pure VAE-226.25(±14.25)−1292.91(±1.10)
OSPOT-VAE-237.62(±6.27)-1271.82(±24.15)
" + }, + { + "category_id": 0, + "poly": [ + 299, + 698, + 527, + 698, + 527, + 730, + 299, + 730 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 299, + 1781, + 544, + 1781, + 544, + 1817, + 299, + 1817 + ], + "score": 0.907 + }, + { + "category_id": 6, + "poly": [ + 383, + 144, + 1309, + 144, + 1309, + 179, + 383, + 179 + ], + "score": 0.902 + }, + { + "category_id": 0, + "poly": [ + 299, + 1497, + 705, + 1497, + 705, + 1527, + 299, + 1527 + ], + "score": 0.895 + }, + { + "category_id": 6, + "poly": [ + 415, + 1303, + 1286, + 1303, + 1286, + 1336, + 415, + 1336 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.892 + }, + { + "category_id": 6, + "poly": [ + 513, + 456, + 1183, + 456, + 1183, + 489, + 513, + 489 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.799 + }, + { + "category_id": 13, + "poly": [ + 609, + 1199, + 708, + 1199, + 708, + 1234, + 609, + 1234 + ], + "score": 0.93, + "latex": "q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 1087, + 1199, + 1171, + 1199, + 1171, + 1233, + 1087, + 1233 + ], + "score": 0.92, + "latex": "p ( \\mathbf { c } | \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 933, + 1585, + 1040, + 1585, + 1040, + 1620, + 933, + 1620 + ], + "score": 0.92, + "latex": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1169, + 352, + 1169, + 352, + 1202, + 297, + 1202 + ], + "score": 0.9, + "latex": "L _ { M _ { \\mathbf { c } } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1554, + 480, + 1554, + 480, + 1590, + 298, + 1590 + ], + "score": 0.85, + "latex": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\mathrm { E L B O }" + }, + { + "category_id": 13, + "poly": [ + 759, + 150, + 779, + 150, + 779, + 173, + 759, + 173 + ], + "score": 0.76, + "latex": "^ *" + }, + { + "category_id": 13, + "poly": [ + 693, + 147, + 711, + 147, + 711, + 178, + 693, + 178 + ], + "score": 0.67, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 1103, + 1302, + 1284, + 1302, + 1284, + 1337, + 1103, + 1337 + ], + "score": 0.42, + "latex": "\\mathbf { B O } \\leq \\log p ( \\mathbf { X } )" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 694.0, + 531.0, + 694.0, + 531.0, + 737.0, + 292.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1775.0, + 550.0, + 1775.0, + 550.0, + 1826.0, + 290.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 143.0, + 692.0, + 143.0, + 692.0, + 182.0, + 385.0, + 182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 143.0, + 758.0, + 143.0, + 758.0, + 182.0, + 712.0, + 182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 143.0, + 1313.0, + 143.0, + 1313.0, + 182.0, + 780.0, + 182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1497.0, + 708.0, + 1497.0, + 708.0, + 1529.0, + 295.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1298.0, + 1102.0, + 1298.0, + 1102.0, + 1342.0, + 411.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 455.0, + 1183.0, + 455.0, + 1183.0, + 491.0, + 514.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 756.0, + 1405.0, + 756.0, + 1405.0, + 789.0, + 297.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 785.0, + 1405.0, + 785.0, + 1405.0, + 820.0, + 292.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 818.0, + 1407.0, + 818.0, + 1407.0, + 852.0, + 291.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 847.0, + 1406.0, + 847.0, + 1406.0, + 882.0, + 294.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 877.0, + 1405.0, + 877.0, + 1405.0, + 913.0, + 295.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 910.0, + 1403.0, + 910.0, + 1403.0, + 942.0, + 296.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 937.0, + 1403.0, + 937.0, + 1403.0, + 974.0, + 295.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 967.0, + 1405.0, + 967.0, + 1405.0, + 1006.0, + 294.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 997.0, + 543.0, + 997.0, + 543.0, + 1033.0, + 293.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1850.0, + 1405.0, + 1850.0, + 1405.0, + 1886.0, + 293.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1883.0, + 1404.0, + 1883.0, + 1404.0, + 1915.0, + 296.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1404.0, + 1911.0, + 1404.0, + 1947.0, + 294.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1943.0, + 1404.0, + 1943.0, + 1404.0, + 1978.0, + 294.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1971.0, + 1406.0, + 1971.0, + 1406.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 1230.0, + 2001.0, + 1230.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1045.0, + 1405.0, + 1045.0, + 1405.0, + 1081.0, + 294.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1079.0, + 1405.0, + 1079.0, + 1405.0, + 1112.0, + 295.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1106.0, + 1403.0, + 1106.0, + 1403.0, + 1143.0, + 293.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1138.0, + 1405.0, + 1138.0, + 1405.0, + 1173.0, + 294.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1168.0, + 296.0, + 1168.0, + 296.0, + 1205.0, + 293.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1168.0, + 1405.0, + 1168.0, + 1405.0, + 1205.0, + 353.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1198.0, + 608.0, + 1198.0, + 608.0, + 1236.0, + 293.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 1198.0, + 1086.0, + 1198.0, + 1086.0, + 1236.0, + 709.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 1198.0, + 1406.0, + 1198.0, + 1406.0, + 1236.0, + 1172.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1230.0, + 1406.0, + 1230.0, + 1406.0, + 1266.0, + 291.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1259.0, + 1241.0, + 1259.0, + 1241.0, + 1299.0, + 293.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1542.0, + 297.0, + 1542.0, + 297.0, + 1604.0, + 285.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 1542.0, + 1414.0, + 1542.0, + 1414.0, + 1604.0, + 481.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1579.0, + 932.0, + 1579.0, + 932.0, + 1631.0, + 289.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1579.0, + 1411.0, + 1579.0, + 1411.0, + 1631.0, + 1041.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1613.0, + 1407.0, + 1613.0, + 1407.0, + 1649.0, + 294.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1644.0, + 1405.0, + 1644.0, + 1405.0, + 1679.0, + 294.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1674.0, + 1405.0, + 1674.0, + 1405.0, + 1710.0, + 296.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1705.0, + 1393.0, + 1705.0, + 1393.0, + 1739.0, + 294.0, + 1739.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 722, + 1403, + 722, + 1403, + 908, + 297, + 908 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 297, + 1762, + 1403, + 1762, + 1403, + 1886, + 297, + 1886 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 298, + 1911, + 1403, + 1911, + 1403, + 2033, + 298, + 2033 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 298, + 455, + 1405, + 455, + 1405, + 580, + 298, + 580 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 297, + 1464, + 1404, + 1464, + 1404, + 1620, + 297, + 1620 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 297, + 1643, + 1403, + 1643, + 1403, + 1737, + 297, + 1737 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 297, + 604, + 1406, + 604, + 1406, + 697, + 297, + 697 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 299, + 932, + 1396, + 932, + 1396, + 995, + 299, + 995 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 296, + 1316, + 1404, + 1316, + 1404, + 1441, + 296, + 1441 + ], + "score": 0.909 + }, + { + "category_id": 1, + "poly": [ + 297, + 1019, + 1404, + 1019, + 1404, + 1175, + 297, + 1175 + ], + "score": 0.901 + }, + { + "category_id": 1, + "poly": [ + 298, + 1198, + 1404, + 1198, + 1404, + 1292, + 298, + 1292 + ], + "score": 0.897 + }, + { + "category_id": 1, + "poly": [ + 303, + 337, + 1402, + 337, + 1402, + 432, + 303, + 432 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.889 + }, + { + "category_id": 1, + "poly": [ + 297, + 280, + 1142, + 280, + 1142, + 314, + 297, + 314 + ], + "score": 0.886 + }, + { + "category_id": 0, + "poly": [ + 299, + 227, + 489, + 227, + 489, + 262, + 299, + 262 + ], + "score": 0.833 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2111, + 840, + 2111 + ], + "score": 0.77 + }, + { + "category_id": 13, + "poly": [ + 841, + 1146, + 859, + 1146, + 859, + 1169, + 841, + 1169 + ], + "score": 0.56, + "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =" + }, + { + "category_id": 13, + "poly": [ + 360, + 1413, + 378, + 1413, + 378, + 1434, + 360, + 1434 + ], + "score": 0.39, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 360, + 1859, + 378, + 1859, + 378, + 1881, + 360, + 1881 + ], + "score": 0.39, + "latex": "=" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 490.0, + 230.0, + 490.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 718.0, + 1407.0, + 718.0, + 1407.0, + 758.0, + 292.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 755.0, + 381.0, + 755.0, + 381.0, + 786.0, + 320.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 749.0, + 1192.0, + 749.0, + 1192.0, + 787.0, + 429.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 748.0, + 1406.0, + 748.0, + 1406.0, + 787.0, + 1238.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 781.0, + 785.0, + 781.0, + 785.0, + 820.0, + 317.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 778.0, + 1408.0, + 778.0, + 1408.0, + 820.0, + 788.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 807.0, + 1404.0, + 807.0, + 1404.0, + 852.0, + 316.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 839.0, + 783.0, + 839.0, + 783.0, + 880.0, + 320.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 838.0, + 1404.0, + 838.0, + 1404.0, + 883.0, + 841.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 871.0, + 1316.0, + 871.0, + 1316.0, + 912.0, + 321.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1761.0, + 1404.0, + 1761.0, + 1404.0, + 1798.0, + 294.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1793.0, + 1404.0, + 1793.0, + 1404.0, + 1827.0, + 320.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1822.0, + 1405.0, + 1822.0, + 1405.0, + 1860.0, + 321.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1851.0, + 359.0, + 1851.0, + 359.0, + 1893.0, + 322.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1851.0, + 540.0, + 1851.0, + 540.0, + 1893.0, + 379.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1910.0, + 1404.0, + 1910.0, + 1404.0, + 1945.0, + 296.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1944.0, + 1404.0, + 1944.0, + 1404.0, + 1974.0, + 322.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2008.0, + 321.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 2004.0, + 507.0, + 2004.0, + 507.0, + 2033.0, + 323.0, + 2033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 457.0, + 1403.0, + 457.0, + 1403.0, + 489.0, + 297.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 488.0, + 1405.0, + 488.0, + 1405.0, + 521.0, + 323.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 516.0, + 1405.0, + 516.0, + 1405.0, + 551.0, + 322.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 548.0, + 504.0, + 548.0, + 504.0, + 585.0, + 322.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1466.0, + 1402.0, + 1466.0, + 1402.0, + 1499.0, + 296.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1493.0, + 1407.0, + 1493.0, + 1407.0, + 1535.0, + 321.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1526.0, + 1405.0, + 1526.0, + 1405.0, + 1563.0, + 324.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1554.0, + 1401.0, + 1554.0, + 1401.0, + 1594.0, + 320.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1587.0, + 1397.0, + 1587.0, + 1397.0, + 1623.0, + 319.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1643.0, + 1405.0, + 1643.0, + 1405.0, + 1680.0, + 293.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1675.0, + 1401.0, + 1675.0, + 1401.0, + 1709.0, + 324.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1706.0, + 505.0, + 1706.0, + 505.0, + 1736.0, + 322.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 603.0, + 1404.0, + 603.0, + 1404.0, + 641.0, + 295.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 635.0, + 1404.0, + 635.0, + 1404.0, + 669.0, + 323.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 666.0, + 782.0, + 666.0, + 782.0, + 700.0, + 323.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 928.0, + 1400.0, + 928.0, + 1400.0, + 969.0, + 294.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 964.0, + 769.0, + 964.0, + 769.0, + 995.0, + 326.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1316.0, + 1405.0, + 1316.0, + 1405.0, + 1353.0, + 294.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1347.0, + 1404.0, + 1347.0, + 1404.0, + 1383.0, + 323.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1376.0, + 1405.0, + 1376.0, + 1405.0, + 1414.0, + 320.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1408.0, + 359.0, + 1408.0, + 359.0, + 1444.0, + 323.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1408.0, + 538.0, + 1408.0, + 538.0, + 1444.0, + 379.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1021.0, + 1404.0, + 1021.0, + 1404.0, + 1054.0, + 296.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1049.0, + 1406.0, + 1049.0, + 1406.0, + 1085.0, + 322.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1080.0, + 1405.0, + 1080.0, + 1405.0, + 1118.0, + 320.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1112.0, + 1405.0, + 1112.0, + 1405.0, + 1145.0, + 322.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1140.0, + 840.0, + 1140.0, + 840.0, + 1180.0, + 320.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1140.0, + 1019.0, + 1140.0, + 1019.0, + 1180.0, + 860.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1199.0, + 1401.0, + 1199.0, + 1401.0, + 1233.0, + 296.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1229.0, + 1402.0, + 1229.0, + 1402.0, + 1263.0, + 323.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1260.0, + 722.0, + 1260.0, + 722.0, + 1293.0, + 323.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 337.0, + 1405.0, + 337.0, + 1405.0, + 371.0, + 298.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 369.0, + 1406.0, + 369.0, + 1406.0, + 403.0, + 323.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 399.0, + 989.0, + 399.0, + 989.0, + 433.0, + 324.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 278.0, + 1147.0, + 278.0, + 1147.0, + 318.0, + 294.0, + 318.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 300, + 229, + 1405, + 229, + 1405, + 385, + 300, + 385 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 299, + 1881, + 1403, + 1881, + 1403, + 2033, + 299, + 2033 + ], + "score": 0.886 + }, + { + "category_id": 1, + "poly": [ + 298, + 1013, + 1404, + 1013, + 1404, + 1168, + 298, + 1168 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 298, + 404, + 1403, + 404, + 1403, + 528, + 298, + 528 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 297, + 868, + 1405, + 868, + 1405, + 993, + 297, + 993 + ], + "score": 0.873 + }, + { + "category_id": 1, + "poly": [ + 298, + 1188, + 1404, + 1188, + 1404, + 1311, + 298, + 1311 + ], + "score": 0.869 + }, + { + "category_id": 1, + "poly": [ + 297, + 723, + 1405, + 723, + 1405, + 849, + 297, + 849 + ], + "score": 0.862 + }, + { + "category_id": 1, + "poly": [ + 298, + 1591, + 1404, + 1591, + 1404, + 1685, + 298, + 1685 + ], + "score": 0.861 + }, + { + "category_id": 1, + "poly": [ + 296, + 549, + 1405, + 549, + 1405, + 704, + 296, + 704 + ], + "score": 0.856 + }, + { + "category_id": 1, + "poly": [ + 297, + 1705, + 1403, + 1705, + 1403, + 1859, + 297, + 1859 + ], + "score": 0.853 + }, + { + "category_id": 1, + "poly": [ + 299, + 1333, + 1404, + 1333, + 1404, + 1457, + 299, + 1457 + ], + "score": 0.851 + }, + { + "category_id": 1, + "poly": [ + 298, + 1477, + 1405, + 1477, + 1405, + 1571, + 298, + 1571 + ], + "score": 0.845 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.82 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.667 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.209 + }, + { + "category_id": 13, + "poly": [ + 1127, + 821, + 1145, + 821, + 1145, + 843, + 1127, + 843 + ], + "score": 0.56, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1127, + 1428, + 1145, + 1428, + 1145, + 1451, + 1127, + 1451 + ], + "score": 0.53, + "latex": "=" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 231.0, + 1404.0, + 231.0, + 1404.0, + 264.0, + 299.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 259.0, + 1404.0, + 259.0, + 1404.0, + 298.0, + 322.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 290.0, + 1405.0, + 290.0, + 1405.0, + 327.0, + 324.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 320.0, + 1404.0, + 320.0, + 1404.0, + 358.0, + 323.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 352.0, + 1151.0, + 352.0, + 1151.0, + 386.0, + 323.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1878.0, + 1402.0, + 1878.0, + 1402.0, + 1915.0, + 295.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1910.0, + 1403.0, + 1910.0, + 1403.0, + 1947.0, + 322.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1941.0, + 1403.0, + 1941.0, + 1403.0, + 1978.0, + 319.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1969.0, + 1404.0, + 1969.0, + 1404.0, + 2008.0, + 319.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2000.0, + 538.0, + 2000.0, + 538.0, + 2037.0, + 322.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1012.0, + 1404.0, + 1012.0, + 1404.0, + 1048.0, + 296.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1037.0, + 1408.0, + 1037.0, + 1408.0, + 1084.0, + 320.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1069.0, + 1405.0, + 1069.0, + 1405.0, + 1114.0, + 318.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1102.0, + 1404.0, + 1102.0, + 1404.0, + 1141.0, + 321.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1134.0, + 1111.0, + 1134.0, + 1111.0, + 1172.0, + 323.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 405.0, + 1405.0, + 405.0, + 1405.0, + 438.0, + 296.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 435.0, + 1403.0, + 435.0, + 1403.0, + 469.0, + 321.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 466.0, + 1402.0, + 466.0, + 1402.0, + 500.0, + 321.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 498.0, + 903.0, + 498.0, + 903.0, + 529.0, + 323.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 870.0, + 1405.0, + 870.0, + 1405.0, + 902.0, + 295.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 900.0, + 1405.0, + 900.0, + 1405.0, + 934.0, + 321.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 932.0, + 1405.0, + 932.0, + 1405.0, + 964.0, + 319.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 960.0, + 1255.0, + 960.0, + 1255.0, + 995.0, + 321.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1188.0, + 1405.0, + 1188.0, + 1405.0, + 1224.0, + 293.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1220.0, + 1402.0, + 1220.0, + 1402.0, + 1253.0, + 323.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1246.0, + 1403.0, + 1246.0, + 1403.0, + 1287.0, + 321.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1282.0, + 773.0, + 1282.0, + 773.0, + 1315.0, + 325.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 722.0, + 1406.0, + 722.0, + 1406.0, + 759.0, + 294.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 752.0, + 1405.0, + 752.0, + 1405.0, + 793.0, + 321.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 787.0, + 1405.0, + 787.0, + 1405.0, + 820.0, + 324.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 817.0, + 1126.0, + 817.0, + 1126.0, + 853.0, + 323.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 817.0, + 1305.0, + 817.0, + 1305.0, + 853.0, + 1146.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1592.0, + 1405.0, + 1592.0, + 1405.0, + 1626.0, + 297.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1621.0, + 1404.0, + 1621.0, + 1404.0, + 1657.0, + 325.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1653.0, + 1345.0, + 1653.0, + 1345.0, + 1687.0, + 323.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 551.0, + 1402.0, + 551.0, + 1402.0, + 584.0, + 296.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 575.0, + 1405.0, + 575.0, + 1405.0, + 620.0, + 320.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 612.0, + 1406.0, + 612.0, + 1406.0, + 645.0, + 323.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 640.0, + 1407.0, + 640.0, + 1407.0, + 677.0, + 323.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 672.0, + 1024.0, + 672.0, + 1024.0, + 705.0, + 323.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1705.0, + 1405.0, + 1705.0, + 1405.0, + 1742.0, + 296.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1736.0, + 1403.0, + 1736.0, + 1403.0, + 1773.0, + 322.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1769.0, + 1401.0, + 1769.0, + 1401.0, + 1802.0, + 322.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1794.0, + 1402.0, + 1794.0, + 1402.0, + 1834.0, + 320.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1826.0, + 673.0, + 1826.0, + 673.0, + 1864.0, + 320.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1331.0, + 1404.0, + 1331.0, + 1404.0, + 1369.0, + 293.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1365.0, + 1404.0, + 1365.0, + 1404.0, + 1397.0, + 322.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1396.0, + 1406.0, + 1396.0, + 1406.0, + 1429.0, + 324.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1424.0, + 1126.0, + 1424.0, + 1126.0, + 1460.0, + 322.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1424.0, + 1306.0, + 1424.0, + 1306.0, + 1460.0, + 1146.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1477.0, + 1406.0, + 1477.0, + 1406.0, + 1514.0, + 293.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1508.0, + 1403.0, + 1508.0, + 1403.0, + 1540.0, + 321.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1535.0, + 1068.0, + 1535.0, + 1068.0, + 1574.0, + 322.0, + 1574.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 304, + 1177, + 1352, + 1177, + 1352, + 1507, + 304, + 1507 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 354, + 1771, + 1297, + 1771, + 1297, + 2046, + 354, + 2046 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 478, + 1632, + 1217, + 1632, + 1217, + 1673, + 478, + 1673 + ], + "score": 0.884 + }, + { + "category_id": 1, + "poly": [ + 296, + 1694, + 369, + 1694, + 369, + 1725, + 296, + 1725 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.873 + }, + { + "category_id": 2, + "poly": [ + 1369, + 1326, + 1400, + 1326, + 1400, + 1358, + 1369, + 1358 + ], + "score": 0.824 + }, + { + "category_id": 0, + "poly": [ + 300, + 878, + 509, + 878, + 509, + 912, + 300, + 912 + ], + "score": 0.811 + }, + { + "category_id": 1, + "poly": [ + 301, + 1591, + 1128, + 1591, + 1128, + 1625, + 301, + 1625 + ], + "score": 0.798 + }, + { + "category_id": 1, + "poly": [ + 301, + 943, + 729, + 943, + 729, + 978, + 301, + 978 + ], + "score": 0.791 + }, + { + "category_id": 1, + "poly": [ + 299, + 762, + 1397, + 762, + 1397, + 824, + 299, + 824 + ], + "score": 0.771 + }, + { + "category_id": 1, + "poly": [ + 299, + 1001, + 862, + 1001, + 862, + 1035, + 299, + 1035 + ], + "score": 0.763 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 863, + 2087, + 863, + 2114, + 835, + 2114 + ], + "score": 0.755 + }, + { + "category_id": 0, + "poly": [ + 298, + 1534, + 868, + 1534, + 868, + 1568, + 298, + 1568 + ], + "score": 0.75 + }, + { + "category_id": 2, + "poly": [ + 1316, + 1894, + 1342, + 1894, + 1342, + 1921, + 1316, + 1921 + ], + "score": 0.723 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 1403, + 229, + 1403, + 415, + 297, + 415 + ], + "score": 0.712 + }, + { + "category_id": 8, + "poly": [ + 355, + 1042, + 1337, + 1042, + 1337, + 1082, + 355, + 1082 + ], + "score": 0.663 + }, + { + "category_id": 1, + "poly": [ + 299, + 541, + 1397, + 541, + 1397, + 605, + 299, + 605 + ], + "score": 0.638 + }, + { + "category_id": 1, + "poly": [ + 297, + 619, + 1403, + 619, + 1403, + 746, + 297, + 746 + ], + "score": 0.601 + }, + { + "category_id": 1, + "poly": [ + 296, + 1103, + 369, + 1103, + 369, + 1135, + 296, + 1135 + ], + "score": 0.59 + }, + { + "category_id": 1, + "poly": [ + 298, + 431, + 1399, + 431, + 1399, + 524, + 298, + 524 + ], + "score": 0.585 + }, + { + "category_id": 0, + "poly": [ + 296, + 1103, + 369, + 1103, + 369, + 1135, + 296, + 1135 + ], + "score": 0.289 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 863, + 2087, + 863, + 2114, + 835, + 2114 + ], + "score": 0.213 + }, + { + "category_id": 1, + "poly": [ + 298, + 1534, + 868, + 1534, + 868, + 1568, + 298, + 1568 + ], + "score": 0.168 + }, + { + "category_id": 0, + "poly": [ + 301, + 943, + 729, + 943, + 729, + 978, + 301, + 978 + ], + "score": 0.139 + }, + { + "category_id": 14, + "poly": [ + 355, + 1775, + 1347, + 1775, + 1347, + 2052, + 355, + 2052 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\displaystyle \\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | | p ( \\mathbf { z } ) ) = \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } { p ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } } \\\\ & { \\displaystyle = \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) } { q _ { \\phi } ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } + \\int _ { \\mathbf { X } } p _ { e m p } ( \\mathbf { X } ) \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\frac { q _ { \\phi } ( \\mathbf { z } ) } { p ( \\mathbf { z } ) } d \\mathbf { z } d \\mathbf { X } } \\\\ & { \\displaystyle = \\int _ { \\mathbf { X } } \\int _ { \\mathbf { z } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { X } ) \\log \\frac { q _ { \\phi } ( \\mathbf { z } , \\mathbf { X } ) } { q _ { \\phi } ( \\mathbf { z } ) p _ { e m p } ( \\mathbf { X } ) } d \\mathbf { z } d \\mathbf { X } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } ) | | p ( \\mathbf { z } ) ) } \\\\ & { \\displaystyle = \\mathbf { I } _ { q _ { \\phi } } ( \\mathbf { X } ; \\mathbf { z } ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } ) | | p ( \\mathbf { z } ) ) } \\end{array} \\overset { \\mathrm { U L } } { \\mathop : }" + }, + { + "category_id": 14, + "poly": [ + 310, + 1182, + 1352, + 1182, + 1352, + 1513, + 310, + 1513 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\log \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } = \\log \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\log \\frac { p ( \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } + \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) } \\\\ & { = - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) | p ( \\mathbf { z } , \\mathbf { c } ) ) + \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] } \\\\ & { = \\mathrm { w i t h ~ a s s u m p t i o n } \\left( 3 . 4 \\right) \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) | p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) | p ( \\mathbf { c } ) ) } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 482, + 1634, + 1214, + 1634, + 1214, + 1673, + 482, + 1673 + ], + "score": 0.87, + "latex": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) \\big ) = \\mathbf { I } _ { q _ { \\phi } } \\big ( \\mathbf { X } ; \\mathbf { z } \\big ) + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( \\mathbf { z } ) \\| p ( \\mathbf { z } ) \\big )" + }, + { + "category_id": 14, + "poly": [ + 359, + 1042, + 1339, + 1042, + 1339, + 1082, + 359, + 1082 + ], + "score": 0.83, + "latex": "\\log p ( \\mathbf { X } ) \\geq \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } , \\mathbf { c } | \\mathbf { X } ) } [ \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { z } , \\mathbf { c } ) ] - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } ) ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } ) )" + }, + { + "category_id": 13, + "poly": [ + 841, + 718, + 859, + 718, + 859, + 740, + 841, + 740 + ], + "score": 0.28, + "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 1330.0, + 1400.0, + 1330.0, + 1400.0, + 1362.0, + 1371.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 873.0, + 513.0, + 873.0, + 513.0, + 919.0, + 295.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1534.0, + 868.0, + 1534.0, + 868.0, + 1571.0, + 296.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1102.0, + 371.0, + 1102.0, + 371.0, + 1138.0, + 290.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 943.0, + 729.0, + 943.0, + 729.0, + 980.0, + 293.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1694.0, + 372.0, + 1694.0, + 372.0, + 1730.0, + 288.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1589.0, + 1128.0, + 1589.0, + 1128.0, + 1632.0, + 294.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 943.0, + 729.0, + 943.0, + 729.0, + 980.0, + 293.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 761.0, + 1401.0, + 761.0, + 1401.0, + 797.0, + 296.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 792.0, + 1396.0, + 792.0, + 1396.0, + 827.0, + 324.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 996.0, + 868.0, + 996.0, + 868.0, + 1042.0, + 294.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 226.0, + 1123.0, + 226.0, + 1123.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 226.0, + 1404.0, + 226.0, + 1404.0, + 266.0, + 1176.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 261.0, + 1046.0, + 261.0, + 1046.0, + 296.0, + 322.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 259.0, + 1404.0, + 259.0, + 1404.0, + 295.0, + 1095.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 285.0, + 1404.0, + 285.0, + 1404.0, + 326.0, + 320.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 318.0, + 1022.0, + 318.0, + 1022.0, + 358.0, + 320.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 319.0, + 1405.0, + 319.0, + 1405.0, + 355.0, + 1067.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 348.0, + 1407.0, + 348.0, + 1407.0, + 390.0, + 320.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 383.0, + 1172.0, + 383.0, + 1172.0, + 415.0, + 325.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 541.0, + 1403.0, + 541.0, + 1403.0, + 576.0, + 294.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 572.0, + 914.0, + 572.0, + 914.0, + 604.0, + 326.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 620.0, + 1403.0, + 620.0, + 1403.0, + 656.0, + 293.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 653.0, + 1403.0, + 653.0, + 1403.0, + 686.0, + 324.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 684.0, + 1407.0, + 684.0, + 1407.0, + 717.0, + 325.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 714.0, + 840.0, + 714.0, + 840.0, + 748.0, + 324.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 714.0, + 1021.0, + 714.0, + 1021.0, + 748.0, + 860.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1102.0, + 371.0, + 1102.0, + 371.0, + 1138.0, + 290.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 432.0, + 1404.0, + 432.0, + 1404.0, + 465.0, + 297.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 462.0, + 1401.0, + 462.0, + 1401.0, + 496.0, + 323.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 492.0, + 424.0, + 492.0, + 424.0, + 525.0, + 323.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1534.0, + 868.0, + 1534.0, + 868.0, + 1571.0, + 296.0, + 1571.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1693, + 1404, + 1693, + 1404, + 1789, + 298, + 1789 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 299, + 613, + 1402, + 613, + 1402, + 812, + 299, + 812 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 716, + 1032, + 981, + 1032, + 981, + 1107, + 716, + 1107 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 295, + 1164, + 1402, + 1164, + 1402, + 1229, + 295, + 1229 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 293, + 1524, + 1400, + 1524, + 1400, + 1590, + 293, + 1590 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 698, + 1341, + 1002, + 1341, + 1002, + 1420, + 698, + 1420 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 469, + 1952, + 1228, + 1952, + 1228, + 2042, + 469, + 2042 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 297, + 955, + 1399, + 955, + 1399, + 1020, + 297, + 1020 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 500, + 1802, + 1197, + 1802, + 1197, + 1892, + 500, + 1892 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 719, + 1600, + 980, + 1600, + 980, + 1638, + 719, + 1638 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 291, + 285, + 1404, + 285, + 1404, + 352, + 291, + 352 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 299, + 1904, + 759, + 1904, + 759, + 1938, + 299, + 1938 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 297, + 1429, + 681, + 1429, + 681, + 1464, + 297, + 1464 + ], + "score": 0.927 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 859, + 73, + 859, + 106, + 297, + 106 + ], + "score": 0.922 + }, + { + "category_id": 0, + "poly": [ + 297, + 898, + 492, + 898, + 492, + 931, + 297, + 931 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 299, + 475, + 799, + 475, + 799, + 509, + 299, + 509 + ], + "score": 0.921 + }, + { + "category_id": 8, + "poly": [ + 370, + 1477, + 1278, + 1477, + 1278, + 1517, + 370, + 1517 + ], + "score": 0.912 + }, + { + "category_id": 1, + "poly": [ + 299, + 568, + 747, + 568, + 747, + 601, + 299, + 601 + ], + "score": 0.908 + }, + { + "category_id": 8, + "poly": [ + 437, + 518, + 1257, + 518, + 1257, + 559, + 437, + 559 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 297, + 1648, + 369, + 1648, + 369, + 1678, + 297, + 1678 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1054, + 1401, + 1054, + 1401, + 1085, + 1338, + 1085 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1364, + 1401, + 1364, + 1401, + 1396, + 1338, + 1396 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 296, + 429, + 369, + 429, + 369, + 461, + 296, + 461 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1338, + 1602, + 1400, + 1602, + 1400, + 1634, + 1338, + 1634 + ], + "score": 0.896 + }, + { + "category_id": 1, + "poly": [ + 301, + 1298, + 1126, + 1298, + 1126, + 1331, + 301, + 1331 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 297, + 1118, + 369, + 1118, + 369, + 1150, + 297, + 1150 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1337, + 1831, + 1401, + 1831, + 1401, + 1863, + 1337, + 1863 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.867 + }, + { + "category_id": 0, + "poly": [ + 300, + 841, + 745, + 841, + 745, + 874, + 300, + 874 + ], + "score": 0.839 + }, + { + "category_id": 2, + "poly": [ + 1422, + 696, + 1450, + 696, + 1450, + 724, + 1422, + 724 + ], + "score": 0.772 + }, + { + "category_id": 0, + "poly": [ + 299, + 228, + 753, + 228, + 753, + 262, + 299, + 262 + ], + "score": 0.697 + }, + { + "category_id": 8, + "poly": [ + 314, + 364, + 1381, + 364, + 1381, + 405, + 314, + 405 + ], + "score": 0.66 + }, + { + "category_id": 8, + "poly": [ + 457, + 1241, + 1211, + 1241, + 1211, + 1291, + 457, + 1291 + ], + "score": 0.616 + }, + { + "category_id": 2, + "poly": [ + 1300, + 1482, + 1326, + 1482, + 1326, + 1508, + 1300, + 1508 + ], + "score": 0.514 + }, + { + "category_id": 1, + "poly": [ + 299, + 228, + 753, + 228, + 753, + 262, + 299, + 262 + ], + "score": 0.267 + }, + { + "category_id": 14, + "poly": [ + 471, + 1949, + 1229, + 1949, + 1229, + 2043, + 471, + 2043 + ], + "score": 0.94, + "latex": "\\mathcal { L } ( \\tilde { \\pi } , t ) = \\lambda D _ { \\mathrm { K L } } ( \\pi _ { 1 } \\| \\tilde { \\pi } ) + ( 1 - \\lambda ) D _ { \\mathrm { K L } } ( \\pi _ { 2 } \\| \\tilde { \\pi } ) + t * ( \\sum _ { i = 1 } ^ { K } \\tilde { \\pi } _ { i } - 1 )" + }, + { + "category_id": 13, + "poly": [ + 585, + 1196, + 754, + 1196, + 754, + 1229, + 585, + 1229 + ], + "score": 0.93, + "latex": "\\mathcal { N } ( \\mathbf { z } _ { 1 } ; \\mu _ { 1 } , \\pmb { \\Sigma } _ { 1 } )" + }, + { + "category_id": 13, + "poly": [ + 763, + 1556, + 872, + 1556, + 872, + 1590, + 763, + 1590 + ], + "score": 0.93, + "latex": "\\lambda \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 495, + 1556, + 703, + 1556, + 703, + 1590, + 495, + 1590 + ], + "score": 0.93, + "latex": "{ \\bf c } _ { 2 } \\sim \\mathrm { M u l t } ( K , \\pi _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 665, + 987, + 773, + 987, + 773, + 1020, + 665, + 1020 + ], + "score": 0.93, + "latex": "\\lambda \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 298, + 1556, + 443, + 1556, + 443, + 1590, + 298, + 1590 + ], + "score": 0.92, + "latex": "\\mathrm { M u l t } ( K , \\pi _ { 1 } )" + }, + { + "category_id": 13, + "poly": [ + 344, + 986, + 603, + 986, + 603, + 1021, + 344, + 1021 + ], + "score": 0.92, + "latex": "\\mathbf z _ { 2 } \\sim \\mathcal N ( \\pmb { \\mu } _ { 2 } , \\mathrm { d i a g } ( \\pmb { \\sigma } _ { 2 } ^ { 2 } ) )" + }, + { + "category_id": 14, + "poly": [ + 310, + 612, + 1421, + 612, + 1421, + 814, + 310, + 814 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { \\log p ( \\mathbf { X } ) = \\displaystyle \\int _ { z , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log p ( \\mathbf { X } ) d \\mathbf { z } d \\mathbf { c } = \\displaystyle \\int _ { z , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } d \\mathbf { z } d \\mathbf { c } } \\\\ & { = \\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } \\log \\frac { p ( \\mathbf { X } , \\mathbf { z } , \\mathbf { c } ) } { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } + \\displaystyle \\int _ { \\mathbf { z } , \\mathbf { c } } q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\log \\frac { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } { p ( \\mathbf { z } | \\mathbf { X } ) p ( \\mathbf { c } | \\mathbf { X } ) } } \\\\ & { = \\mathrm { E L B O } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1097, + 1195, + 1265, + 1195, + 1265, + 1229, + 1097, + 1229 + ], + "score": 0.92, + "latex": "\\bar { \\mathcal { N } } ( \\mathbf { z } _ { 2 } ; \\mu _ { 2 } , \\Sigma _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1134, + 288, + 1346, + 288, + 1346, + 325, + 1134, + 325 + ], + "score": 0.92, + "latex": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } \\log p ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 573, + 1756, + 783, + 1756, + 783, + 1789, + 573, + 1789 + ], + "score": 0.92, + "latex": "{ \\bf c } _ { 1 } \\sim \\mathrm { M u l t } ( K , \\pi _ { 1 } )" + }, + { + "category_id": 14, + "poly": [ + 496, + 1799, + 1200, + 1799, + 1200, + 1895, + 496, + 1895 + ], + "score": 0.92, + "latex": "\\operatorname* { m i n } _ { \\tilde { \\pi } } \\lambda D _ { \\mathrm { K L } } ( \\pi _ { 1 } \\| \\tilde { \\pi } ) + ( 1 - \\lambda ) D _ { \\mathrm { K L } } ( \\pi _ { 2 } \\| \\tilde { \\pi } ) \\qquad s . t . \\sum _ { i = 1 } ^ { K } \\tilde { \\pi } _ { i } = 1" + }, + { + "category_id": 13, + "poly": [ + 834, + 1756, + 1043, + 1756, + 1043, + 1789, + 834, + 1789 + ], + "score": 0.92, + "latex": "{ \\bf c } _ { 2 } \\sim \\bf { M u l t } ( { \\cal K } , \\pi _ { 2 } )" + }, + { + "category_id": 14, + "poly": [ + 695, + 1339, + 1004, + 1339, + 1004, + 1422, + 695, + 1422 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\mathbf { z } _ { \\lambda } = ( 1 - \\lambda ) \\mathbf { z } _ { 1 } + \\lambda T ( \\mathbf { z } _ { 1 } ) } \\\\ { \\mathbf { T } = d i a g ( \\pmb { \\sigma } _ { 1 } / \\pmb { \\sigma } _ { 2 } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1103, + 1757, + 1211, + 1757, + 1211, + 1789, + 1103, + 1789 + ], + "score": 0.91, + "latex": "\\lambda \\in [ 0 , 1 ]" + }, + { + "category_id": 14, + "poly": [ + 715, + 1028, + 983, + 1028, + 983, + 1110, + 715, + 1110 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\tilde { \\pmb { \\mu } } = \\lambda \\pmb { \\mu } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\mu } _ { 2 } } \\\\ { \\tilde { \\pmb { \\sigma } } = \\lambda \\pmb { \\sigma } _ { 1 } + ( 1 - \\lambda ) \\pmb { \\sigma } _ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1043, + 1725, + 1249, + 1725, + 1249, + 1757, + 1043, + 1757 + ], + "score": 0.9, + "latex": "\\mathbf { c } _ { \\lambda } \\sim \\mathbf { M u l t } ( K , \\tilde { \\pi } )" + }, + { + "category_id": 13, + "poly": [ + 1134, + 955, + 1399, + 955, + 1399, + 991, + 1134, + 991 + ], + "score": 0.9, + "latex": "\\mathbf z _ { 1 } \\sim \\mathcal N ( \\pmb { \\mu } _ { 1 } , \\mathrm { d i a g } ( \\pmb { \\sigma } _ { 1 } ^ { 2 } ) )" + }, + { + "category_id": 14, + "poly": [ + 474, + 1237, + 1222, + 1237, + 1222, + 1290, + 474, + 1290 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { { \\mathbf z } \\to { \\mathcal T } ( { \\mathbf z } ) = \\mu _ { 2 } + { \\mathbf T } ( { \\mathbf z } - { \\boldsymbol \\mu _ { 1 } } ) ; \\qquad { \\mathbf T } = { \\mathbf { \\Sigma } } _ { 1 } ^ { - \\frac { 1 } { 2 } } \\bigr ( { \\mathbf { \\Sigma } } _ { 1 } ^ { \\frac { 1 } { 2 } } { \\mathbf { \\Sigma } } _ { 2 } { \\mathbf { \\Sigma } } _ { 1 } ^ { \\frac { 1 } { 2 } } \\bigr ) ^ { \\frac { 1 } { 2 } } { \\mathbf { \\Sigma } } _ { 1 } ^ { - \\frac { 1 } { 2 } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 319, + 364, + 1377, + 364, + 1377, + 406, + 319, + 406 + ], + "score": 0.89, + "latex": "\\mathbb { E } _ { p _ { c m p } ( \\mathbf { X } ) } [ \\log p ( \\mathbf { X } ) - E L B O ] = \\mathbb { E } _ { p _ { c m p } ( \\mathbf { X } ) } [ D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) ) ]" + }, + { + "category_id": 14, + "poly": [ + 719, + 1599, + 981, + 1599, + 981, + 1637, + 719, + 1637 + ], + "score": 0.89, + "latex": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 370, + 1475, + 1311, + 1475, + 1311, + 1516, + 370, + 1516 + ], + "score": 0.89, + "latex": "{ \\bf z } _ { \\lambda } \\sim \\mathcal { N } ( \\tilde { \\pmb { \\mu } } , d i a g ( \\tilde { \\pmb { \\sigma } } ^ { 2 } ) ) ; \\qquad \\tilde { \\pmb { \\mu } } = \\lambda { \\pmb { \\mu } } _ { 1 } + ( 1 - \\lambda ) { \\pmb { \\mu } } _ { 2 } ; \\qquad \\tilde { \\pmb { \\sigma } } = \\lambda { \\pmb { \\sigma } } _ { 1 } + ( 1 - \\lambda ) { \\pmb { \\sigma } } _ { 2 } \\quad [ \\lambda \\in \\mathcal { N } ^ { \\pmb { \\mu } } ] ." + }, + { + "category_id": 13, + "poly": [ + 656, + 569, + 719, + 569, + 719, + 602, + 656, + 602 + ], + "score": 0.87, + "latex": "( 3 , 4 )" + }, + { + "category_id": 14, + "poly": [ + 438, + 519, + 1259, + 519, + 1259, + 558, + 438, + 558 + ], + "score": 0.87, + "latex": "\\log p ( \\mathbf { X } ) - \\mathrm { E L B O } = D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) \\| p ( \\mathbf { z } | \\mathbf { X } ) ) + D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) \\| p ( \\mathbf { c } | \\mathbf { X } ) )" + }, + { + "category_id": 13, + "poly": [ + 298, + 319, + 475, + 319, + 475, + 355, + 298, + 355 + ], + "score": 0.86, + "latex": "\\mathbb { E } _ { p _ { e m p } ( \\mathbf { X } ) } E L B O" + }, + { + "category_id": 13, + "poly": [ + 1078, + 1300, + 1097, + 1300, + 1097, + 1325, + 1078, + 1325 + ], + "score": 0.84, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 1336, + 1531, + 1403, + 1531, + 1403, + 1557, + 1336, + 1557 + ], + "score": 0.81, + "latex": "\\mathbf { c } _ { 1 } \\sim" + }, + { + "category_id": 13, + "poly": [ + 1287, + 1696, + 1329, + 1696, + 1329, + 1723, + 1287, + 1723 + ], + "score": 0.28, + "latex": "\\mathrm { K L }" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 859.0, + 72.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 896.0, + 496.0, + 896.0, + 496.0, + 936.0, + 294.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1650.0, + 371.0, + 1650.0, + 371.0, + 1681.0, + 289.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 432.0, + 371.0, + 432.0, + 371.0, + 464.0, + 289.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.75, + 1116.5, + 374.75, + 1116.5, + 374.75, + 1156.5, + 288.75, + 1156.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 841.0, + 747.0, + 841.0, + 747.0, + 876.0, + 296.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 700.0, + 1450.0, + 700.0, + 1450.0, + 726.0, + 1425.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 753.0, + 229.0, + 753.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 1485.0, + 1327.0, + 1485.0, + 1327.0, + 1512.0, + 1312.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1691.0, + 1286.0, + 1691.0, + 1286.0, + 1730.0, + 295.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 1691.0, + 1404.0, + 1691.0, + 1404.0, + 1730.0, + 1330.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1725.0, + 1042.0, + 1725.0, + 1042.0, + 1761.0, + 292.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1725.0, + 1406.0, + 1725.0, + 1406.0, + 1761.0, + 1250.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1756.0, + 572.0, + 1756.0, + 572.0, + 1790.0, + 296.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1756.0, + 833.0, + 1756.0, + 833.0, + 1790.0, + 784.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1756.0, + 1102.0, + 1756.0, + 1102.0, + 1790.0, + 1044.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 1756.0, + 1292.0, + 1756.0, + 1292.0, + 1790.0, + 1212.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1162.0, + 1405.0, + 1162.0, + 1405.0, + 1200.0, + 295.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1194.0, + 584.0, + 1194.0, + 584.0, + 1230.0, + 295.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1194.0, + 1096.0, + 1194.0, + 1096.0, + 1230.0, + 755.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 1194.0, + 1296.0, + 1194.0, + 1296.0, + 1230.0, + 1266.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1519.0, + 1335.0, + 1519.0, + 1335.0, + 1564.0, + 292.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1556.0, + 494.0, + 1556.0, + 494.0, + 1592.0, + 444.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1556.0, + 762.0, + 1556.0, + 762.0, + 1592.0, + 704.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1556.0, + 902.0, + 1556.0, + 902.0, + 1592.0, + 873.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 951.0, + 1133.0, + 951.0, + 1133.0, + 995.0, + 295.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 951.0, + 1403.0, + 951.0, + 1403.0, + 995.0, + 1400.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 985.0, + 343.0, + 985.0, + 343.0, + 1021.0, + 296.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 985.0, + 664.0, + 985.0, + 664.0, + 1021.0, + 604.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 985.0, + 804.0, + 985.0, + 804.0, + 1021.0, + 774.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 281.0, + 1133.0, + 281.0, + 1133.0, + 329.0, + 292.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 281.0, + 1410.0, + 281.0, + 1410.0, + 329.0, + 1347.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 316.0, + 297.0, + 316.0, + 297.0, + 359.0, + 290.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 316.0, + 510.0, + 316.0, + 510.0, + 359.0, + 476.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1903.0, + 761.0, + 1903.0, + 761.0, + 1943.0, + 297.0, + 1943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1425.0, + 684.0, + 1425.0, + 684.0, + 1471.0, + 296.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 473.0, + 801.0, + 473.0, + 801.0, + 514.0, + 295.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 566.0, + 655.0, + 566.0, + 655.0, + 605.0, + 295.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 566.0, + 750.0, + 566.0, + 750.0, + 605.0, + 720.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1296.0, + 1077.0, + 1296.0, + 1077.0, + 1335.0, + 297.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1296.0, + 1125.0, + 1296.0, + 1125.0, + 1335.0, + 1098.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 753.0, + 229.0, + 753.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 318, + 276, + 1382, + 276, + 1382, + 533, + 318, + 533 + ], + "score": 0.983, + "html": "
MNISTSVHN(one-stage)SVHNCifar10Cifar100
hmaxtmaxhmaxtmaxhmaxtmaxhmaxtmaxhmaxtmax
305011751e-31501e-31501e-1150
B3050117511501e-31501e-3150
117.5505017512801502001501280150
Ic1750501752.31502.31504.6150
WMz///一1e-31501e-31501e-1150
WMc/1140012801280
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1776, + 1404, + 1776, + 1404, + 1900, + 297, + 1900 + ], + "score": 0.979 + }, + { + "category_id": 8, + "poly": [ + 626, + 646, + 1074, + 646, + 1074, + 813, + 626, + 813 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 424, + 1127, + 1209, + 1127, + 1209, + 1658, + 424, + 1658 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 296, + 1971, + 1401, + 1971, + 1401, + 2035, + 296, + 2035 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 296, + 838, + 978, + 838, + 978, + 873, + 296, + 873 + ], + "score": 0.935 + }, + { + "category_id": 8, + "poly": [ + 576, + 1910, + 1123, + 1910, + 1123, + 1953, + 576, + 1953 + ], + "score": 0.935 + }, + { + "category_id": 4, + "poly": [ + 549, + 1678, + 1150, + 1678, + 1150, + 1712, + 549, + 1712 + ], + "score": 0.93 + }, + { + "category_id": 8, + "poly": [ + 696, + 897, + 959, + 897, + 959, + 935, + 696, + 935 + ], + "score": 0.926 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 859, + 74, + 859, + 105, + 297, + 105 + ], + "score": 0.918 + }, + { + "category_id": 6, + "poly": [ + 683, + 224, + 1017, + 224, + 1017, + 256, + 683, + 256 + ], + "score": 0.917 + }, + { + "category_id": 9, + "poly": [ + 1339, + 1916, + 1400, + 1916, + 1400, + 1947, + 1339, + 1947 + ], + "score": 0.892 + }, + { + "category_id": 1, + "poly": [ + 297, + 597, + 692, + 597, + 692, + 629, + 297, + 629 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1338, + 716, + 1402, + 716, + 1402, + 747, + 1338, + 747 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.874 + }, + { + "category_id": 0, + "poly": [ + 300, + 981, + 633, + 981, + 633, + 1012, + 300, + 1012 + ], + "score": 0.698 + }, + { + "category_id": 2, + "poly": [ + 977, + 902, + 1004, + 902, + 1004, + 928, + 977, + 928 + ], + "score": 0.393 + }, + { + "category_id": 1, + "poly": [ + 300, + 981, + 633, + 981, + 633, + 1012, + 300, + 1012 + ], + "score": 0.224 + }, + { + "category_id": 14, + "poly": [ + 624, + 647, + 1075, + 647, + 1075, + 816, + 624, + 816 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\frac { \\partial \\mathcal { L } ( \\tilde { \\boldsymbol { \\pi } } , t ) } { \\partial \\tilde { \\boldsymbol { \\pi } } } = t - \\frac { \\lambda \\boldsymbol { \\pi } _ { 1 } + ( 1 - \\lambda ) \\boldsymbol { \\pi } _ { 2 } } { \\tilde { \\boldsymbol { \\pi } } } = 0 } \\\\ { t \\ast ( \\displaystyle \\sum _ { i = 1 } ^ { K } \\tilde { \\boldsymbol { \\pi } } _ { i } - 1 ) = 0 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 584, + 1841, + 643, + 1841, + 643, + 1869, + 584, + 1869 + ], + "score": 0.91, + "latex": "t _ { m a x }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1839, + 362, + 1839, + 362, + 1870, + 298, + 1870 + ], + "score": 0.9, + "latex": "h _ { m a x }" + }, + { + "category_id": 14, + "poly": [ + 576, + 1910, + 1120, + 1910, + 1120, + 1950, + 576, + 1950 + ], + "score": 0.88, + "latex": "h _ { t } = h _ { m a x } \\times \\exp { [ - 5 * ( 1 - \\operatorname* { m i n } ( 1 , t / t _ { m a x } ) ) ^ { 2 } ] }" + }, + { + "category_id": 13, + "poly": [ + 447, + 1871, + 475, + 1871, + 475, + 1900, + 447, + 1900 + ], + "score": 0.88, + "latex": "h _ { t }" + }, + { + "category_id": 14, + "poly": [ + 694, + 895, + 1008, + 895, + 1008, + 935, + 694, + 935 + ], + "score": 0.84, + "latex": "\\tilde { \\pi } = \\lambda \\pi _ { 1 } + ( 1 - \\lambda ) \\pi _ { 2 } \\quad \\bigsqcup" + }, + { + "category_id": 13, + "poly": [ + 922, + 841, + 944, + 841, + 944, + 867, + 922, + 867 + ], + "score": 0.83, + "latex": "\\tilde { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 588, + 1810, + 608, + 1810, + 608, + 1837, + 588, + 1837 + ], + "score": 0.8, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 350, + 349, + 379, + 349, + 379, + 375, + 350, + 375 + ], + "score": 0.77, + "latex": "\\beta _ { \\mathbf { z } }" + }, + { + "category_id": 13, + "poly": [ + 424, + 316, + 486, + 316, + 486, + 344, + 424, + 344 + ], + "score": 0.52, + "latex": "\\overline { { h _ { m a x } } }" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1131.0, + 532.0, + 1131.0, + 532.0, + 1178.0, + 459.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1213.0, + 533.0, + 1213.0, + 533.0, + 1263.0, + 460.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1291.0, + 533.0, + 1291.0, + 533.0, + 1347.0, + 459.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1323.0, + 473.0, + 1323.0, + 473.0, + 1395.0, + 421.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1375.0, + 531.0, + 1375.0, + 531.0, + 1427.0, + 460.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1460.0, + 519.0, + 1460.0, + 519.0, + 1506.0, + 463.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1539.0, + 530.0, + 1539.0, + 530.0, + 1583.0, + 463.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1581.0, + 576.0, + 1581.0, + 576.0, + 1621.0, + 524.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1582.0, + 699.0, + 1582.0, + 699.0, + 1620.0, + 652.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 1583.0, + 825.0, + 1583.0, + 825.0, + 1619.0, + 775.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1581.0, + 952.0, + 1581.0, + 952.0, + 1621.0, + 898.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1582.0, + 1071.0, + 1582.0, + 1071.0, + 1619.0, + 1023.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1574.0, + 1199.0, + 1574.0, + 1199.0, + 1622.0, + 1144.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 1609.0, + 989.0, + 1609.0, + 989.0, + 1668.0, + 737.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1676.0, + 1151.0, + 1676.0, + 1151.0, + 1716.0, + 549.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 220.0, + 1018.0, + 220.0, + 1018.0, + 259.0, + 681.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 976.0, + 634.0, + 976.0, + 634.0, + 1017.0, + 295.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1775.0, + 1405.0, + 1775.0, + 1405.0, + 1812.0, + 294.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1810.0, + 587.0, + 1810.0, + 587.0, + 1843.0, + 294.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1810.0, + 1404.0, + 1810.0, + 1404.0, + 1843.0, + 609.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1838.0, + 297.0, + 1838.0, + 297.0, + 1874.0, + 294.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1838.0, + 583.0, + 1838.0, + 583.0, + 1874.0, + 363.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 1838.0, + 1405.0, + 1838.0, + 1405.0, + 1874.0, + 644.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1867.0, + 446.0, + 1867.0, + 446.0, + 1903.0, + 294.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1867.0, + 511.0, + 1867.0, + 511.0, + 1903.0, + 476.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2007.0, + 296.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 2003.0, + 961.0, + 2003.0, + 961.0, + 2035.0, + 297.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 837.0, + 921.0, + 837.0, + 921.0, + 876.0, + 295.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 837.0, + 980.0, + 837.0, + 980.0, + 876.0, + 945.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 598.0, + 694.0, + 598.0, + 694.0, + 631.0, + 298.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 976.0, + 634.0, + 976.0, + 634.0, + 1017.0, + 295.0, + 1017.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 296, + 276, + 1459, + 276, + 1459, + 596, + 296, + 596 + ], + "score": 0.981, + "html": "
MNISTSVHN(one-stage)SVHNCifar10Cifar100
Latent Dim(z/c)10/1032/10128/10128/10128/100
Mutual Info(z/c)17.5/17.050/501280/2.3200/2.31280/4.6
loss of -log pe(X|c,z)BCEBCEMSEMSEBCE
a of pmixup(X)一一222
optimizerAdamSGD
learning rate5e-41e-30.1
lr scheduler(decay ratio)一一every 50 epoch after 200-th(0.5)[500,600,650](0.2)
weight decay005e-4
" + }, + { + "category_id": 1, + "poly": [ + 299, + 746, + 1405, + 746, + 1405, + 839, + 299, + 839 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 424, + 1630, + 1403, + 1630, + 1403, + 1963, + 424, + 1963 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 422, + 946, + 1406, + 946, + 1406, + 1525, + 422, + 1525 + ], + "score": 0.92 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.907 + }, + { + "category_id": 6, + "poly": [ + 649, + 224, + 1052, + 224, + 1052, + 257, + 649, + 257 + ], + "score": 0.906 + }, + { + "category_id": 1, + "poly": [ + 365, + 1562, + 854, + 1562, + 854, + 1594, + 365, + 1594 + ], + "score": 0.893 + }, + { + "category_id": 1, + "poly": [ + 295, + 2002, + 1230, + 2002, + 1230, + 2035, + 295, + 2035 + ], + "score": 0.89 + }, + { + "category_id": 1, + "poly": [ + 366, + 878, + 763, + 878, + 763, + 909, + 366, + 909 + ], + "score": 0.862 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.855 + }, + { + "category_id": 0, + "poly": [ + 301, + 677, + 757, + 677, + 757, + 709, + 301, + 709 + ], + "score": 0.792 + }, + { + "category_id": 13, + "poly": [ + 929, + 1179, + 1076, + 1179, + 1076, + 1213, + 929, + 1213 + ], + "score": 0.93, + "latex": "\\mathbf { I } _ { \\mathbf { c } } = \\log ( K )" + }, + { + "category_id": 13, + "poly": [ + 536, + 1324, + 727, + 1324, + 727, + 1357, + 536, + 1357 + ], + "score": 0.93, + "latex": "- \\log p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 504, + 1466, + 592, + 1466, + 592, + 1497, + 504, + 1497 + ], + "score": 0.92, + "latex": "\\beta ( \\alpha , \\alpha )" + }, + { + "category_id": 13, + "poly": [ + 452, + 1262, + 577, + 1262, + 577, + 1296, + 452, + 1296 + ], + "score": 0.91, + "latex": "p _ { \\pmb { \\theta } } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 1075, + 1262, + 1237, + 1262, + 1237, + 1296, + 1075, + 1296 + ], + "score": 0.91, + "latex": "\\mathcal { N } ( f _ { \\boldsymbol { \\theta } } ( \\mathbf { c } , \\mathbf { z } ) , \\mathbf { I } )" + }, + { + "category_id": 13, + "poly": [ + 618, + 1228, + 997, + 1228, + 997, + 1265, + 618, + 1265 + ], + "score": 0.9, + "latex": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 508, + 1434, + 635, + 1434, + 635, + 1466, + 508, + 1466 + ], + "score": 0.9, + "latex": "p _ { m i x u p } ( \\mathbf { X } )" + }, + { + "category_id": 13, + "poly": [ + 874, + 1937, + 906, + 1937, + 906, + 1962, + 874, + 1962 + ], + "score": 0.89, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 731, + 1293, + 1002, + 1293, + 1002, + 1327, + 731, + 1327 + ], + "score": 0.89, + "latex": "\\mathbf { M u l t } ( d i m ( \\mathbf { X } ) , f _ { \\theta } ( \\mathbf { c } , \\mathbf { z } ) )" + }, + { + "category_id": 13, + "poly": [ + 452, + 1440, + 474, + 1440, + 474, + 1461, + 452, + 1461 + ], + "score": 0.73, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 485, + 1008, + 513, + 1008, + 513, + 1035, + 485, + 1035 + ], + "score": 0.73, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 794, + 1013, + 812, + 1013, + 812, + 1035, + 794, + 1035 + ], + "score": 0.65, + "latex": "\\mathbf { z }" + }, + { + "category_id": 13, + "poly": [ + 359, + 412, + 381, + 412, + 381, + 433, + 359, + 433 + ], + "score": 0.47, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 433, + 313, + 546, + 313, + 546, + 345, + 433, + 345 + ], + "score": 0.3, + "latex": "\\overline { { \\mathrm { D i m } ( \\mathbf { z } / \\mathbf { c } ) } }" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 221.0, + 1053.0, + 221.0, + 1053.0, + 260.0, + 645.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 675.0, + 760.0, + 675.0, + 760.0, + 712.0, + 296.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 747.0, + 1405.0, + 747.0, + 1405.0, + 780.0, + 295.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 777.0, + 1405.0, + 777.0, + 1405.0, + 811.0, + 295.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 807.0, + 1299.0, + 807.0, + 1299.0, + 841.0, + 295.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1632.0, + 572.0, + 1632.0, + 572.0, + 1665.0, + 423.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1660.0, + 1401.0, + 1660.0, + 1401.0, + 1695.0, + 446.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1708.0, + 608.0, + 1708.0, + 608.0, + 1746.0, + 425.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1740.0, + 1402.0, + 1740.0, + 1402.0, + 1775.0, + 449.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1774.0, + 538.0, + 1774.0, + 538.0, + 1807.0, + 450.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 1817.0, + 900.0, + 1817.0, + 900.0, + 1858.0, + 442.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1849.0, + 1309.0, + 1849.0, + 1309.0, + 1887.0, + 449.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1895.0, + 609.0, + 1895.0, + 609.0, + 1939.0, + 423.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1930.0, + 873.0, + 1930.0, + 873.0, + 1968.0, + 449.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 1930.0, + 1078.0, + 1930.0, + 1078.0, + 1968.0, + 907.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 948.0, + 579.0, + 948.0, + 579.0, + 976.0, + 424.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 974.0, + 1406.0, + 974.0, + 1406.0, + 1011.0, + 448.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1009.0, + 484.0, + 1009.0, + 484.0, + 1040.0, + 450.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1009.0, + 793.0, + 1009.0, + 793.0, + 1040.0, + 514.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1009.0, + 1320.0, + 1009.0, + 1320.0, + 1040.0, + 813.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1058.0, + 677.0, + 1058.0, + 677.0, + 1089.0, + 425.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1086.0, + 1403.0, + 1086.0, + 1403.0, + 1122.0, + 450.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1117.0, + 1406.0, + 1117.0, + 1406.0, + 1153.0, + 450.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1150.0, + 1405.0, + 1150.0, + 1405.0, + 1181.0, + 450.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1181.0, + 928.0, + 1181.0, + 928.0, + 1212.0, + 451.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1181.0, + 1086.0, + 1181.0, + 1086.0, + 1212.0, + 1077.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1222.0, + 617.0, + 1222.0, + 617.0, + 1272.0, + 418.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1222.0, + 1001.0, + 1222.0, + 1001.0, + 1272.0, + 998.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1260.0, + 451.0, + 1260.0, + 451.0, + 1298.0, + 448.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1260.0, + 1074.0, + 1260.0, + 1074.0, + 1298.0, + 578.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1260.0, + 1406.0, + 1260.0, + 1406.0, + 1298.0, + 1238.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1294.0, + 730.0, + 1294.0, + 730.0, + 1328.0, + 449.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1294.0, + 1405.0, + 1294.0, + 1405.0, + 1328.0, + 1003.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1321.0, + 535.0, + 1321.0, + 535.0, + 1358.0, + 449.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1321.0, + 1405.0, + 1321.0, + 1405.0, + 1358.0, + 728.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1354.0, + 1405.0, + 1354.0, + 1405.0, + 1388.0, + 450.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1386.0, + 772.0, + 1386.0, + 772.0, + 1418.0, + 451.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1432.0, + 451.0, + 1432.0, + 451.0, + 1471.0, + 423.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1432.0, + 507.0, + 1432.0, + 507.0, + 1471.0, + 475.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1432.0, + 640.0, + 1432.0, + 640.0, + 1471.0, + 636.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1461.0, + 503.0, + 1461.0, + 503.0, + 1502.0, + 448.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 1461.0, + 1405.0, + 1461.0, + 1405.0, + 1502.0, + 593.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1496.0, + 774.0, + 1496.0, + 774.0, + 1530.0, + 450.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1555.0, + 857.0, + 1555.0, + 857.0, + 1603.0, + 357.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2000.0, + 1233.0, + 2000.0, + 1233.0, + 2038.0, + 296.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 878.0, + 763.0, + 878.0, + 763.0, + 911.0, + 364.0, + 911.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 352, + 1480, + 1347, + 1480, + 1347, + 1967, + 352, + 1967 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 299, + 800, + 1391, + 800, + 1391, + 1151, + 299, + 1151 + ], + "score": 0.959 + }, + { + "category_id": 3, + "poly": [ + 300, + 230, + 1393, + 230, + 1393, + 582, + 300, + 582 + ], + "score": 0.958 + }, + { + "category_id": 4, + "poly": [ + 418, + 1993, + 1280, + 1993, + 1280, + 2026, + 418, + 2026 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 298, + 1391, + 1143, + 1391, + 1143, + 1425, + 298, + 1425 + ], + "score": 0.918 + }, + { + "category_id": 4, + "poly": [ + 432, + 609, + 1262, + 609, + 1262, + 679, + 432, + 679 + ], + "score": 0.91 + }, + { + "category_id": 2, + "poly": [ + 297, + 76, + 858, + 76, + 858, + 105, + 297, + 105 + ], + "score": 0.904 + }, + { + "category_id": 0, + "poly": [ + 300, + 1324, + 718, + 1324, + 718, + 1355, + 300, + 1355 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.865 + }, + { + "category_id": 4, + "poly": [ + 418, + 1177, + 1272, + 1177, + 1272, + 1245, + 418, + 1245 + ], + "score": 0.831 + }, + { + "category_id": 13, + "poly": [ + 879, + 641, + 1259, + 641, + 1259, + 678, + 879, + 678 + ], + "score": 0.91, + "latex": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )" + }, + { + "category_id": 13, + "poly": [ + 879, + 1208, + 1259, + 1208, + 1259, + 1245, + 879, + 1245 + ], + "score": 0.9, + "latex": "\\mathbb { E } _ { q _ { \\phi } ( \\mathbf { z } | \\mathbf { X } ) , q _ { \\phi } ( \\mathbf { c } | \\mathbf { X } ) } - \\log p _ { \\theta } ( \\mathbf { X } | \\mathbf { c } , \\mathbf { z } )" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1492.0, + 391.0, + 1492.0, + 391.0, + 1530.0, + 365.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1488.0, + 440.0, + 1488.0, + 440.0, + 1573.0, + 406.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1494.0, + 477.0, + 1494.0, + 477.0, + 1524.0, + 455.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1493.0, + 522.0, + 1493.0, + 522.0, + 1519.0, + 498.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1491.0, + 569.0, + 1491.0, + 569.0, + 1539.0, + 540.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1494.0, + 609.0, + 1494.0, + 609.0, + 1525.0, + 588.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1493.0, + 654.0, + 1493.0, + 654.0, + 1531.0, + 628.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 1493.0, + 696.0, + 1493.0, + 696.0, + 1523.0, + 675.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1492.0, + 742.0, + 1492.0, + 742.0, + 1526.0, + 716.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1492.0, + 787.0, + 1492.0, + 787.0, + 1539.0, + 757.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1480.0, + 1349.0, + 1480.0, + 1349.0, + 1534.0, + 901.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1539.0, + 778.0, + 1539.0, + 778.0, + 1560.0, + 770.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1531.0, + 950.0, + 1531.0, + 950.0, + 1572.0, + 913.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1538.0, + 982.0, + 1538.0, + 982.0, + 1568.0, + 954.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 1538.0, + 1030.0, + 1538.0, + 1030.0, + 1564.0, + 1000.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1561.0, + 441.0, + 1561.0, + 441.0, + 1711.0, + 405.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1561.0, + 962.0, + 1561.0, + 962.0, + 1757.0, + 893.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1568.0, + 398.0, + 1568.0, + 398.0, + 1799.0, + 358.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1568.0, + 1213.0, + 1568.0, + 1213.0, + 1619.0, + 1173.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1569.0, + 792.0, + 1569.0, + 792.0, + 1923.0, + 753.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1571.0, + 989.0, + 1571.0, + 989.0, + 1621.0, + 951.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1571.0, + 1081.0, + 1571.0, + 1081.0, + 1621.0, + 1041.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1571.0, + 1170.0, + 1571.0, + 1170.0, + 1619.0, + 1130.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1571.0, + 1261.0, + 1571.0, + 1261.0, + 1620.0, + 1216.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1572.0, + 521.0, + 1572.0, + 521.0, + 1699.0, + 499.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1573.0, + 1123.0, + 1573.0, + 1123.0, + 1613.0, + 1086.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1574.0, + 1340.0, + 1574.0, + 1340.0, + 1615.0, + 1301.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 1576.0, + 650.0, + 1576.0, + 650.0, + 1621.0, + 631.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 1580.0, + 694.0, + 1580.0, + 694.0, + 1612.0, + 675.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1587.0, + 732.0, + 1587.0, + 732.0, + 1601.0, + 719.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1623.0, + 474.0, + 1623.0, + 474.0, + 1652.0, + 458.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1616.0, + 564.0, + 1616.0, + 564.0, + 1704.0, + 543.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1633.0, + 601.0, + 1633.0, + 601.0, + 1643.0, + 592.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 1612.0, + 650.0, + 1612.0, + 650.0, + 1664.0, + 631.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1615.0, + 740.0, + 1615.0, + 740.0, + 1707.0, + 718.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 1612.0, + 992.0, + 1612.0, + 992.0, + 1660.0, + 952.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1615.0, + 1125.0, + 1615.0, + 1125.0, + 1660.0, + 997.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1611.0, + 1172.0, + 1611.0, + 1172.0, + 1663.0, + 1128.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1617.0, + 1302.0, + 1617.0, + 1302.0, + 1660.0, + 1216.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 1618.0, + 1340.0, + 1618.0, + 1340.0, + 1659.0, + 1305.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1662.0, + 652.0, + 1662.0, + 652.0, + 1706.0, + 629.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 1658.0, + 694.0, + 1658.0, + 694.0, + 1698.0, + 675.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 1654.0, + 994.0, + 1654.0, + 994.0, + 1712.0, + 952.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1659.0, + 1039.0, + 1659.0, + 1039.0, + 1706.0, + 998.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1660.0, + 1079.0, + 1660.0, + 1079.0, + 1701.0, + 1041.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1660.0, + 1121.0, + 1660.0, + 1121.0, + 1707.0, + 1086.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 1659.0, + 1165.0, + 1659.0, + 1165.0, + 1701.0, + 1125.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1662.0, + 1249.0, + 1662.0, + 1249.0, + 1706.0, + 1210.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1659.0, + 1302.0, + 1659.0, + 1302.0, + 1707.0, + 1252.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 1662.0, + 1337.0, + 1662.0, + 1337.0, + 1699.0, + 1305.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1720.0, + 427.0, + 1720.0, + 427.0, + 1735.0, + 418.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1715.0, + 474.0, + 1715.0, + 474.0, + 1742.0, + 458.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1707.0, + 564.0, + 1707.0, + 564.0, + 1785.0, + 543.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1716.0, + 692.0, + 1716.0, + 692.0, + 1743.0, + 674.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1714.0, + 739.0, + 1714.0, + 739.0, + 1787.0, + 721.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1715.0, + 1045.0, + 1715.0, + 1045.0, + 1732.0, + 1032.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1686.0, + 1221.0, + 1686.0, + 1221.0, + 1759.0, + 1123.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1704.0, + 1299.0, + 1704.0, + 1299.0, + 1748.0, + 1257.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1756.0, + 429.0, + 1756.0, + 429.0, + 1785.0, + 417.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 1750.0, + 475.0, + 1750.0, + 475.0, + 1784.0, + 456.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1751.0, + 648.0, + 1751.0, + 648.0, + 1784.0, + 636.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1751.0, + 693.0, + 1751.0, + 693.0, + 1783.0, + 677.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1737.0, + 1035.0, + 1737.0, + 1035.0, + 1885.0, + 897.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1746.0, + 1165.0, + 1746.0, + 1165.0, + 1796.0, + 1131.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1748.0, + 1251.0, + 1748.0, + 1251.0, + 1792.0, + 1221.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1738.0, + 1341.0, + 1738.0, + 1341.0, + 1844.0, + 1252.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1795.0, + 394.0, + 1795.0, + 394.0, + 1918.0, + 361.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 1791.0, + 992.0, + 1791.0, + 992.0, + 1834.0, + 949.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 1790.0, + 1122.0, + 1790.0, + 1122.0, + 1835.0, + 1000.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1796.0, + 1156.0, + 1796.0, + 1156.0, + 1824.0, + 1136.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1797.0, + 1339.0, + 1797.0, + 1339.0, + 1836.0, + 1299.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1852.0, + 429.0, + 1852.0, + 429.0, + 1879.0, + 417.0, + 1879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1832.0, + 479.0, + 1832.0, + 479.0, + 1916.0, + 448.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1830.0, + 575.0, + 1830.0, + 575.0, + 1925.0, + 531.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1836.0, + 659.0, + 1836.0, + 659.0, + 1918.0, + 625.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1835.0, + 697.0, + 1835.0, + 697.0, + 1881.0, + 674.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1838.0, + 737.0, + 1838.0, + 737.0, + 1870.0, + 722.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 1832.0, + 993.0, + 1832.0, + 993.0, + 1878.0, + 950.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1835.0, + 1032.0, + 1835.0, + 1032.0, + 1876.0, + 1001.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1832.0, + 1080.0, + 1832.0, + 1080.0, + 1878.0, + 1039.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1832.0, + 1167.0, + 1832.0, + 1167.0, + 1878.0, + 1084.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 1831.0, + 1256.0, + 1831.0, + 1256.0, + 1881.0, + 1207.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 1837.0, + 1340.0, + 1837.0, + 1340.0, + 1877.0, + 1262.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1881.0, + 435.0, + 1881.0, + 435.0, + 1915.0, + 410.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1884.0, + 523.0, + 1884.0, + 523.0, + 1915.0, + 496.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 1881.0, + 611.0, + 1881.0, + 611.0, + 1918.0, + 579.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1883.0, + 696.0, + 1883.0, + 696.0, + 1912.0, + 674.0, + 1912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 1877.0, + 990.0, + 1877.0, + 990.0, + 1918.0, + 909.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1886.0, + 1029.0, + 1886.0, + 1029.0, + 1912.0, + 1003.0, + 1912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1879.0, + 1076.0, + 1879.0, + 1076.0, + 1920.0, + 1039.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1884.0, + 1116.0, + 1884.0, + 1116.0, + 1914.0, + 1089.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1886.0, + 1162.0, + 1886.0, + 1162.0, + 1915.0, + 1130.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 1881.0, + 1209.0, + 1881.0, + 1209.0, + 1916.0, + 1178.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 1884.0, + 1289.0, + 1884.0, + 1289.0, + 1913.0, + 1266.0, + 1913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1881.0, + 1340.0, + 1881.0, + 1340.0, + 1916.0, + 1307.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1931.0, + 641.0, + 1931.0, + 641.0, + 1973.0, + 508.0, + 1973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1062.0, + 1932.0, + 1186.0, + 1932.0, + 1186.0, + 1973.0, + 1062.0, + 1973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1577.0, + 1281.0, + 1577.0, + 1281.0, + 1609.0, + 1278.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1163.25, + 1619.0, + 1222.25, + 1619.0, + 1222.25, + 1656.0, + 1163.25, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.75, + 1663.0, + 1194.75, + 1663.0, + 1194.75, + 1697.5, + 1184.75, + 1697.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 1704.0, + 1239.0, + 1704.0, + 1239.0, + 1746.0, + 1228.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.75, + 1699.5, + 1006.75, + 1699.5, + 1006.75, + 1746.5, + 933.75, + 1746.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.25, + 1703.0, + 1050.25, + 1703.0, + 1050.25, + 1742.0, + 986.25, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.25, + 1705.0, + 1137.25, + 1705.0, + 1137.25, + 1745.5, + 1060.25, + 1745.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1297.25, + 1708.5, + 1347.25, + 1708.5, + 1347.25, + 1743.5, + 1297.25, + 1743.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1752.5, + 1090.0, + 1752.5, + 1090.0, + 1784.0, + 1033.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.75, + 1750.0, + 1131.75, + 1750.0, + 1131.75, + 1791.0, + 1071.75, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 1754.0, + 1221.0, + 1754.0, + 1221.0, + 1785.5, + 1162.0, + 1785.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.75, + 1789.5, + 1271.75, + 1789.5, + 1271.75, + 1836.5, + 1196.75, + 1836.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 1832.5, + 1232.0, + 1832.5, + 1232.0, + 1875.5, + 1152.0, + 1875.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 1880.0, + 1240.0, + 1880.0, + 1240.0, + 1919.0, + 1230.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 853.0, + 356.0, + 853.0, + 356.0, + 891.0, + 300.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 829.0, + 893.0, + 829.0, + 893.0, + 881.0, + 838.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 870.0, + 893.0, + 870.0, + 893.0, + 927.0, + 841.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 915.0, + 360.0, + 915.0, + 360.0, + 964.0, + 297.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 923.0, + 892.0, + 923.0, + 892.0, + 972.0, + 843.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 983.0, + 358.0, + 983.0, + 358.0, + 1031.0, + 296.0, + 1031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 967.0, + 892.0, + 967.0, + 892.0, + 1023.0, + 849.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1010.0, + 892.0, + 1010.0, + 892.0, + 1069.0, + 850.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1052.0, + 357.0, + 1052.0, + 357.0, + 1095.0, + 299.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 1063.0, + 888.0, + 1063.0, + 888.0, + 1098.0, + 865.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1111.0, + 805.0, + 1111.0, + 805.0, + 1151.0, + 398.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1113.0, + 1354.0, + 1113.0, + 1354.0, + 1152.0, + 936.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.25, + 846.0, + 1229.25, + 846.0, + 1229.25, + 927.0, + 1079.25, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.25, + 897.5, + 1145.25, + 897.5, + 1145.25, + 903.5, + 1133.25, + 903.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 239.0, + 894.0, + 239.0, + 894.0, + 296.0, + 843.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 271.0, + 359.0, + 271.0, + 359.0, + 314.0, + 301.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 292.0, + 894.0, + 292.0, + 894.0, + 341.0, + 849.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 325.0, + 359.0, + 325.0, + 359.0, + 369.0, + 301.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 341.0, + 894.0, + 341.0, + 894.0, + 392.0, + 849.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 378.0, + 362.0, + 378.0, + 362.0, + 425.0, + 298.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 389.0, + 892.0, + 389.0, + 892.0, + 446.0, + 850.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 433.0, + 361.0, + 433.0, + 361.0, + 479.0, + 311.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 439.0, + 891.0, + 439.0, + 891.0, + 489.0, + 851.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 491.0, + 358.0, + 491.0, + 358.0, + 529.0, + 303.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 495.0, + 888.0, + 495.0, + 888.0, + 529.0, + 865.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 552.0, + 409.0, + 552.0, + 409.0, + 576.0, + 390.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 543.0, + 824.0, + 543.0, + 824.0, + 584.0, + 423.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 546.0, + 1377.0, + 546.0, + 1377.0, + 586.0, + 917.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 292.5, + 808.0, + 292.5, + 808.0, + 324.5, + 669.0, + 324.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.75, + 348.5, + 729.75, + 348.5, + 729.75, + 496.5, + 340.75, + 496.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1990.0, + 1282.0, + 1990.0, + 1282.0, + 2030.0, + 416.0, + 2030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 607.0, + 1080.0, + 607.0, + 1080.0, + 643.0, + 616.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 631.0, + 878.0, + 631.0, + 878.0, + 688.0, + 424.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 631.0, + 1265.0, + 631.0, + 1265.0, + 688.0, + 1260.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1324.0, + 720.0, + 1324.0, + 720.0, + 1357.0, + 297.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 1174.0, + 1088.0, + 1174.0, + 1088.0, + 1210.0, + 610.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1198.0, + 878.0, + 1198.0, + 878.0, + 1254.0, + 425.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 1198.0, + 1265.0, + 1198.0, + 1265.0, + 1254.0, + 1260.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1389.0, + 1145.0, + 1389.0, + 1145.0, + 1429.0, + 294.0, + 1429.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 352, + 223, + 1347, + 223, + 1347, + 709, + 352, + 709 + ], + "score": 0.969 + }, + { + "category_id": 3, + "poly": [ + 353, + 833, + 1347, + 833, + 1347, + 1320, + 353, + 1320 + ], + "score": 0.968 + }, + { + "category_id": 4, + "poly": [ + 342, + 736, + 1346, + 736, + 1346, + 770, + 342, + 770 + ], + "score": 0.915 + }, + { + "category_id": 4, + "poly": [ + 338, + 1346, + 1354, + 1346, + 1354, + 1381, + 338, + 1381 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 858, + 76, + 858, + 105, + 298, + 105 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.843 + }, + { + "category_id": 15, + "poly": [ + 364.0, + 226.0, + 740.0, + 226.0, + 740.0, + 276.0, + 364.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 314.0, + 615.0, + 314.0, + 615.0, + 362.0, + 431.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 326.0, + 743.0, + 326.0, + 743.0, + 354.0, + 710.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 365.0, + 531.0, + 365.0, + 531.0, + 397.0, + 475.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 365.0, + 618.0, + 365.0, + 618.0, + 397.0, + 532.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 396.0, + 771.0, + 396.0, + 771.0, + 454.0, + 388.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 441.0, + 742.0, + 441.0, + 742.0, + 499.0, + 432.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 542.0, + 747.0, + 542.0, + 747.0, + 572.0, + 709.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 582.0, + 715.0, + 582.0, + 715.0, + 626.0, + 523.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 623.0, + 448.0, + 623.0, + 448.0, + 666.0, + 354.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 623.0, + 491.0, + 623.0, + 491.0, + 665.0, + 427.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 637.0, + 520.0, + 637.0, + 520.0, + 648.0, + 502.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 676.0, + 649.0, + 676.0, + 649.0, + 713.0, + 499.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 672.0, + 1217.0, + 672.0, + 1217.0, + 716.0, + 1032.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 542.0, + 447.0, + 542.0, + 447.0, + 575.5, + 407.0, + 575.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 833.0, + 795.0, + 833.0, + 795.0, + 892.0, + 358.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 834.0, + 1348.0, + 834.0, + 1348.0, + 894.0, + 905.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 930.0, + 663.0, + 930.0, + 663.0, + 966.0, + 621.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 924.0, + 1315.0, + 924.0, + 1315.0, + 975.0, + 948.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 986.0, + 434.0, + 986.0, + 434.0, + 997.0, + 412.0, + 997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 974.0, + 526.0, + 974.0, + 526.0, + 1007.0, + 493.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 979.0, + 988.0, + 979.0, + 988.0, + 1000.0, + 957.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 967.0, + 1203.0, + 967.0, + 1203.0, + 1017.0, + 1000.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 1015.0, + 491.0, + 1015.0, + 491.0, + 1058.0, + 401.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1011.0, + 582.0, + 1011.0, + 582.0, + 1063.0, + 498.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 1013.0, + 1033.0, + 1013.0, + 1033.0, + 1062.0, + 948.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1024.0, + 1073.0, + 1024.0, + 1073.0, + 1051.0, + 1044.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1034.0, + 1104.0, + 1034.0, + 1104.0, + 1042.0, + 1088.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 1032.0, + 1157.0, + 1032.0, + 1157.0, + 1048.0, + 1143.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 1060.0, + 1007.0, + 1060.0, + 1007.0, + 1102.0, + 947.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1060.0, + 1081.0, + 1060.0, + 1081.0, + 1101.0, + 1032.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 1066.0, + 1038.0, + 1066.0, + 1038.0, + 1096.0, + 1000.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1114.0, + 481.0, + 1114.0, + 481.0, + 1133.0, + 460.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1118.0, + 520.0, + 1118.0, + 520.0, + 1130.0, + 505.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1110.0, + 571.0, + 1110.0, + 571.0, + 1136.0, + 538.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 1119.0, + 737.0, + 1119.0, + 737.0, + 1136.0, + 713.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1096.0, + 1178.0, + 1096.0, + 1178.0, + 1152.0, + 897.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1145.0, + 629.0, + 1145.0, + 629.0, + 1191.0, + 407.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 1137.0, + 1260.0, + 1137.0, + 1260.0, + 1199.0, + 903.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1185.0, + 785.0, + 1185.0, + 785.0, + 1240.0, + 390.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1189.0, + 1256.0, + 1189.0, + 1256.0, + 1240.0, + 920.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1251.0, + 391.0, + 1251.0, + 391.0, + 1268.0, + 367.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1253.0, + 603.0, + 1253.0, + 603.0, + 1261.0, + 587.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 1230.0, + 797.0, + 1230.0, + 797.0, + 1281.0, + 613.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1225.0, + 1348.0, + 1225.0, + 1348.0, + 1284.0, + 898.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1287.0, + 650.0, + 1287.0, + 650.0, + 1328.0, + 499.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1287.0, + 1217.0, + 1287.0, + 1217.0, + 1329.0, + 1032.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 892.5, + 1033.0, + 892.5, + 1033.0, + 914.0, + 1005.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 734.0, + 1348.0, + 734.0, + 1348.0, + 774.0, + 352.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1346.0, + 1355.0, + 1346.0, + 1355.0, + 1382.0, + 343.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 871.0, + 2085.0, + 871.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/S1xcx3C5FX/images/1bb3f9f8dfa3ae432b6af626ea97c0a97a8a333ecd1f0b3e0b0d43c589129e90.jpg b/parse/train/S1xcx3C5FX/images/1bb3f9f8dfa3ae432b6af626ea97c0a97a8a333ecd1f0b3e0b0d43c589129e90.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d433aca4659dab48e077ded8a4e622a3a9b7847c --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/1bb3f9f8dfa3ae432b6af626ea97c0a97a8a333ecd1f0b3e0b0d43c589129e90.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03ac86859cc791944fde43920914f7628762b4236ff24d86bf2a10edcab4746d +size 7318 diff --git a/parse/train/S1xcx3C5FX/images/2379988b4abaa8ac1c75836e1fa0ccaf304f675d5568104790af6c526b43fa3b.jpg b/parse/train/S1xcx3C5FX/images/2379988b4abaa8ac1c75836e1fa0ccaf304f675d5568104790af6c526b43fa3b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5149c3628e0d0d89f55344f92512ad1d2e9dfc50 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/2379988b4abaa8ac1c75836e1fa0ccaf304f675d5568104790af6c526b43fa3b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0531d8b6c5ffdf1b7bcc868a671467e28fb6ef669d1737ce19919c2821548ad +size 7073 diff --git a/parse/train/S1xcx3C5FX/images/2b16c5f2276b7f993a087c8054a28c33b20f273cb4548f4a62e36fa107baf5eb.jpg b/parse/train/S1xcx3C5FX/images/2b16c5f2276b7f993a087c8054a28c33b20f273cb4548f4a62e36fa107baf5eb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fa67dcb488abc44b52f9656e1ee3a5a8e580eefa --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/2b16c5f2276b7f993a087c8054a28c33b20f273cb4548f4a62e36fa107baf5eb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1522055c578ff3a35be2e6acabe3b7e5996dcf1e54d33527b0d6789095c7660a +size 223154 diff --git a/parse/train/S1xcx3C5FX/images/2cdce4b2a7179d5380c4d30c7f7dc1729638e1c8c51a22c79a73d61054d3e602.jpg b/parse/train/S1xcx3C5FX/images/2cdce4b2a7179d5380c4d30c7f7dc1729638e1c8c51a22c79a73d61054d3e602.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa10dba64101dd02b6882f6e8fb1d2632ce79bbe --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/2cdce4b2a7179d5380c4d30c7f7dc1729638e1c8c51a22c79a73d61054d3e602.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df29934e3a9f804ff832bec717e45bc775203e7663437ef8ae23f8d6696582da +size 10039 diff --git a/parse/train/S1xcx3C5FX/images/333e09f2c4bfdfe7850938524282bc5f8274a38bb1d37674b16c0e999bf2a78b.jpg b/parse/train/S1xcx3C5FX/images/333e09f2c4bfdfe7850938524282bc5f8274a38bb1d37674b16c0e999bf2a78b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cca57b5660ade6566629fb6c5f97fee0285312a9 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/333e09f2c4bfdfe7850938524282bc5f8274a38bb1d37674b16c0e999bf2a78b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fcc710cd8df614666cf7151c773bf31413dad327449040ced9411afe2fe54fb3 +size 19824 diff --git a/parse/train/S1xcx3C5FX/images/4065acd2c67839128b60e9e168d0ac71169201bb3745fbe1f678de7d326185c8.jpg b/parse/train/S1xcx3C5FX/images/4065acd2c67839128b60e9e168d0ac71169201bb3745fbe1f678de7d326185c8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..33a67eb2cedf3e283af60c6394033e8759383bb2 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/4065acd2c67839128b60e9e168d0ac71169201bb3745fbe1f678de7d326185c8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a408ebf00282b87aab133d3d9fba776137b76c3a21780b366bd4a47719c4524e +size 5749 diff --git a/parse/train/S1xcx3C5FX/images/52b646502ec36ee715d1048153dc8b8976704256e2dce78be7427a03a796dd29.jpg b/parse/train/S1xcx3C5FX/images/52b646502ec36ee715d1048153dc8b8976704256e2dce78be7427a03a796dd29.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dca171967e9a6d87c307845531d6c09140bd272d --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/52b646502ec36ee715d1048153dc8b8976704256e2dce78be7427a03a796dd29.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f28db873273fd41b43b15d178767ee3ffa4bf5565c1a9be19fa2907c885f7f1 +size 8000 diff --git a/parse/train/S1xcx3C5FX/images/607a17a4c53e088c2556dbe6548132950b8e9ff0aa4e4123c886c89276891582.jpg b/parse/train/S1xcx3C5FX/images/607a17a4c53e088c2556dbe6548132950b8e9ff0aa4e4123c886c89276891582.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a7c75ad650369aa5c029ccaab26f68bf62e79076 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/607a17a4c53e088c2556dbe6548132950b8e9ff0aa4e4123c886c89276891582.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8342e330e3fe12b994b489daeaff16592464f283d0df0773ebc74fadb8e70579 +size 7678 diff --git a/parse/train/S1xcx3C5FX/images/6568e0428774ba5c2f2a3ea350da24301a3337f4f571146783a9fe5d1f54a75d.jpg b/parse/train/S1xcx3C5FX/images/6568e0428774ba5c2f2a3ea350da24301a3337f4f571146783a9fe5d1f54a75d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a483490928f034d60db9e7b574b20434eb5f7e1 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/6568e0428774ba5c2f2a3ea350da24301a3337f4f571146783a9fe5d1f54a75d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f2e0997bd2b2159a8a2dece1e63d2dc05006d5c1ddc0baf400797b33d43ffbc +size 49154 diff --git a/parse/train/S1xcx3C5FX/images/6bd9d43ef3cf2db172db1dce702caec2377b98c7c98cc1c7000f19a4874c4053.jpg b/parse/train/S1xcx3C5FX/images/6bd9d43ef3cf2db172db1dce702caec2377b98c7c98cc1c7000f19a4874c4053.jpg new file mode 100644 index 0000000000000000000000000000000000000000..621b5a211ab8ee6a56c627a1e3175571b441f25d --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/6bd9d43ef3cf2db172db1dce702caec2377b98c7c98cc1c7000f19a4874c4053.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c61c0623aeba475ffa58806abd49299cec6f2065e4dd6dece0dc77407e0652bc +size 16497 diff --git a/parse/train/S1xcx3C5FX/images/8029cb4ca15b59a78050d1f377234d8cbd12d52f8a546216004e99bf1f12bbfc.jpg b/parse/train/S1xcx3C5FX/images/8029cb4ca15b59a78050d1f377234d8cbd12d52f8a546216004e99bf1f12bbfc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0ea897f8f6f71f6b67ff4bd0f4c7614b47b23b30 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/8029cb4ca15b59a78050d1f377234d8cbd12d52f8a546216004e99bf1f12bbfc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a77db23877b8e936dd53c415733005afa544a49d67e7b28f18b52be643b45d1 +size 4216 diff --git a/parse/train/S1xcx3C5FX/images/91513f347ec2cf31b0a6f22246102991a12bbc154e6d294aa7c7e7d03c9e4118.jpg b/parse/train/S1xcx3C5FX/images/91513f347ec2cf31b0a6f22246102991a12bbc154e6d294aa7c7e7d03c9e4118.jpg new file mode 100644 index 0000000000000000000000000000000000000000..05ca3f9fa9b7576ce53f5a0812c09b1d1b82a451 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/91513f347ec2cf31b0a6f22246102991a12bbc154e6d294aa7c7e7d03c9e4118.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4e7237b6cfec084112779fa4b4cf0da7b9cef98fe29edcfab02cb9fc2b49418 +size 40240 diff --git a/parse/train/S1xcx3C5FX/images/9a2fa8d36feea61d6e520710070082e48918cd5769559018690919b1b2b952d4.jpg b/parse/train/S1xcx3C5FX/images/9a2fa8d36feea61d6e520710070082e48918cd5769559018690919b1b2b952d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b08c24ab9e19c297c99f652f9c59bf492777fd31 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/9a2fa8d36feea61d6e520710070082e48918cd5769559018690919b1b2b952d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86cf06a33ec788e2c5341c1c090aa9244d9f1cb048f69057687c1aa2ce2f5fe6 +size 13317 diff --git a/parse/train/S1xcx3C5FX/images/cd7e3a803a07b7347e19be42ce25137d5b4eba6f1eb47943a6af2e8ef5f7e5b0.jpg b/parse/train/S1xcx3C5FX/images/cd7e3a803a07b7347e19be42ce25137d5b4eba6f1eb47943a6af2e8ef5f7e5b0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..561dcdd2bb19af4d13ac40d5fe101eff0e25065a --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/cd7e3a803a07b7347e19be42ce25137d5b4eba6f1eb47943a6af2e8ef5f7e5b0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:632aef4a34696a1bca66b03cf62113b24df422faa2aaef3296f3af1a35c734ae +size 7585 diff --git a/parse/train/S1xcx3C5FX/images/d3db2c4ab334b7d4fe0de68aba529986c882c765291c39307c7bfb0509cc4825.jpg b/parse/train/S1xcx3C5FX/images/d3db2c4ab334b7d4fe0de68aba529986c882c765291c39307c7bfb0509cc4825.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eaa603e56fe7c32b90d2cf0fc8b2a38511cf5563 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/d3db2c4ab334b7d4fe0de68aba529986c882c765291c39307c7bfb0509cc4825.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e8c6fda9cae9ef24aff5da9d6a901a908a0951c46d1eda59519d45b5f8ac8d5 +size 8018 diff --git a/parse/train/S1xcx3C5FX/images/d4d13d542c86056ad5fd9404ae667d8973494a44dc95edffb4946301ca2d780e.jpg b/parse/train/S1xcx3C5FX/images/d4d13d542c86056ad5fd9404ae667d8973494a44dc95edffb4946301ca2d780e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..36e342917e1ccdf71a81e977e8c60fa8e4bbfc50 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/d4d13d542c86056ad5fd9404ae667d8973494a44dc95edffb4946301ca2d780e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2fa7aacb4d1bc07b0dbea4038b7cf9c660ca735a520035ddb443b68863734168 +size 5654 diff --git a/parse/train/S1xcx3C5FX/images/e2badc0338d6d9b794c87b92d587eb723b1a3c2e4215c6d7658d05bcd24758f3.jpg b/parse/train/S1xcx3C5FX/images/e2badc0338d6d9b794c87b92d587eb723b1a3c2e4215c6d7658d05bcd24758f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3122b3e126b7a7537b839fb62f50fb54878e19b2 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/e2badc0338d6d9b794c87b92d587eb723b1a3c2e4215c6d7658d05bcd24758f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26802376590b8a2d6400bb30c16a07347152bf00bc693e2de3196aff5b52e14d +size 1467 diff --git a/parse/train/S1xcx3C5FX/images/e6b79d5cdbf41291e6e0c755441802d3716aa158949ba795eb9e79c77ffb5bfb.jpg b/parse/train/S1xcx3C5FX/images/e6b79d5cdbf41291e6e0c755441802d3716aa158949ba795eb9e79c77ffb5bfb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..108f75910ffd3f695e0f8a8f20e79b41b35ae608 --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/e6b79d5cdbf41291e6e0c755441802d3716aa158949ba795eb9e79c77ffb5bfb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d79c0a93c9782ae36e8a795bab74f62a5d68afe0643ab2905b338cb9da37a3ef +size 3288 diff --git a/parse/train/S1xcx3C5FX/images/f1c93372546a0b2c3f4aa86b07ea649f365549350084e5a6c78717032e819855.jpg b/parse/train/S1xcx3C5FX/images/f1c93372546a0b2c3f4aa86b07ea649f365549350084e5a6c78717032e819855.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5f423c1e85caaeef9375d4833ba7e08ceb6ebc9f --- /dev/null +++ b/parse/train/S1xcx3C5FX/images/f1c93372546a0b2c3f4aa86b07ea649f365549350084e5a6c78717032e819855.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8bb9f761f1c9dec732c3a1630427ace4e9268b1aa743a69680e086327867b457 +size 78625 diff --git a/parse/train/SJNRHiAcYX/images/02c82db9088b9862f97dac6f0dc163b524770f3137635065c86abb606525c077.jpg b/parse/train/SJNRHiAcYX/images/02c82db9088b9862f97dac6f0dc163b524770f3137635065c86abb606525c077.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8c781cd12acb281b1776cce4a7a0467701879ad8 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/02c82db9088b9862f97dac6f0dc163b524770f3137635065c86abb606525c077.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b02fbbae500081b5cf5af66d0255ef861d77457731d2ea046d187be1c81fa1c +size 5920 diff --git a/parse/train/SJNRHiAcYX/images/08453d1fa6e889d21a8f6b652cce8155d40a7d38bd9794b9736eb417fc087b6f.jpg b/parse/train/SJNRHiAcYX/images/08453d1fa6e889d21a8f6b652cce8155d40a7d38bd9794b9736eb417fc087b6f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..08b17c02c52cce29d5b60a58dcca55555a17de25 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/08453d1fa6e889d21a8f6b652cce8155d40a7d38bd9794b9736eb417fc087b6f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a96df5f05c5fac542d72b12a90fde1f4d8c2c92d056be8e94cf303905fa7776 +size 8208 diff --git a/parse/train/SJNRHiAcYX/images/08c4a0f7290b44b8bc49fcb2423d38289934641c2da18548ddc32bba5408de90.jpg b/parse/train/SJNRHiAcYX/images/08c4a0f7290b44b8bc49fcb2423d38289934641c2da18548ddc32bba5408de90.jpg new file mode 100644 index 0000000000000000000000000000000000000000..911f612fc184aca4de81820c8a27a75add2bb1bb --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/08c4a0f7290b44b8bc49fcb2423d38289934641c2da18548ddc32bba5408de90.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4d78fe2af2936094970ca9448dfe32d5aeded0199526fc050c174ffaf41a017 +size 32295 diff --git a/parse/train/SJNRHiAcYX/images/25081e444af1801b4fe9ed511a868db1c99f51d051c484e35b3903d12154a81a.jpg b/parse/train/SJNRHiAcYX/images/25081e444af1801b4fe9ed511a868db1c99f51d051c484e35b3903d12154a81a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cc9a0aad55dbbf176aeb9742596cb423f4bc1a2d --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/25081e444af1801b4fe9ed511a868db1c99f51d051c484e35b3903d12154a81a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5bad7eb642eb5949d494ce11e72227d55e097a3ab888857a4e651d4cf4e97204 +size 3877 diff --git a/parse/train/SJNRHiAcYX/images/300d64b88c6ee17175c1583af6b75f5d9fc7fb2947e796c2ebdb556e4e8916bb.jpg b/parse/train/SJNRHiAcYX/images/300d64b88c6ee17175c1583af6b75f5d9fc7fb2947e796c2ebdb556e4e8916bb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c115148f89f5fa7b713f7891bfa825f266e0fb60 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/300d64b88c6ee17175c1583af6b75f5d9fc7fb2947e796c2ebdb556e4e8916bb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb6d169819cbbc77cbc2524206da34db851eacb3c2793c3bbb91e5a67769a6ec +size 5245 diff --git a/parse/train/SJNRHiAcYX/images/382ad2607c66aa1b614136605a7fcc744118685dcb65d67bf9320674470657d2.jpg b/parse/train/SJNRHiAcYX/images/382ad2607c66aa1b614136605a7fcc744118685dcb65d67bf9320674470657d2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a6da31a2fe0783eb0923ab8b925695670fdd5550 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/382ad2607c66aa1b614136605a7fcc744118685dcb65d67bf9320674470657d2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e43373a48d5bc1bd401126a2ddf54e726d615d19300bf4d5e0e5cc91339a3927 +size 3809 diff --git a/parse/train/SJNRHiAcYX/images/3bccfd8dfd24597447db5a1d292a0c3c1890916ebabcfc9e4a8f6c871ea27326.jpg b/parse/train/SJNRHiAcYX/images/3bccfd8dfd24597447db5a1d292a0c3c1890916ebabcfc9e4a8f6c871ea27326.jpg new file mode 100644 index 0000000000000000000000000000000000000000..22ac679c9087259f1425d6940b8f53d8aee40d8a --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/3bccfd8dfd24597447db5a1d292a0c3c1890916ebabcfc9e4a8f6c871ea27326.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:239aec18d2de00e7073ca024b473663b9b2e26862809b1d2cab72bc28e922018 +size 36171 diff --git a/parse/train/SJNRHiAcYX/images/5424ac5ae4db875606faa5a7c24ea47fd8b4ee7f13c0beb6a6f4440f2f674348.jpg b/parse/train/SJNRHiAcYX/images/5424ac5ae4db875606faa5a7c24ea47fd8b4ee7f13c0beb6a6f4440f2f674348.jpg new file mode 100644 index 0000000000000000000000000000000000000000..323084fbbe039c634f3963441fa9170d251067ba --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/5424ac5ae4db875606faa5a7c24ea47fd8b4ee7f13c0beb6a6f4440f2f674348.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a57a863e01643653a05fb121ea06fb8c5b6e458a6e09127cfba08c0f3775200 +size 10456 diff --git a/parse/train/SJNRHiAcYX/images/554505b4d3a997f25040636ff558c0b0c6b821f4c676082e8da541f875c86e2e.jpg b/parse/train/SJNRHiAcYX/images/554505b4d3a997f25040636ff558c0b0c6b821f4c676082e8da541f875c86e2e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b764f13467fa269f726ff0fa0a9b11262734f35a --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/554505b4d3a997f25040636ff558c0b0c6b821f4c676082e8da541f875c86e2e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5ebf161c0d3d2fdaff5c4c717d2305120cf7cdec82e1a6bec9833810b1e1906 +size 35944 diff --git a/parse/train/SJNRHiAcYX/images/74a1781bb77963bbafaff71357bc7ff59af28d530a594f0e01c8577ce550b21d.jpg b/parse/train/SJNRHiAcYX/images/74a1781bb77963bbafaff71357bc7ff59af28d530a594f0e01c8577ce550b21d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cdfe2ab3b28683897a12f5174b9ec87f1be6d34a --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/74a1781bb77963bbafaff71357bc7ff59af28d530a594f0e01c8577ce550b21d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3d4bd13cdfc966ec0a0f8daab72f25121ef22f13d5a3bb97e53649ddefb726f +size 8179 diff --git a/parse/train/SJNRHiAcYX/images/8187e5102a1a49022bca8bb86868380fba65667bd01a23600b13a5e687d45406.jpg b/parse/train/SJNRHiAcYX/images/8187e5102a1a49022bca8bb86868380fba65667bd01a23600b13a5e687d45406.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9077229628aa26efaed420935f05fa4df93204a2 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/8187e5102a1a49022bca8bb86868380fba65667bd01a23600b13a5e687d45406.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0eed34956215302656297625b7d4055395e0b49d8997202358e48832a2f394f +size 7017 diff --git a/parse/train/SJNRHiAcYX/images/8933ca49c1e077791251576d2e198024004e91e92e5bc3fd8dab951a1c44a92a.jpg b/parse/train/SJNRHiAcYX/images/8933ca49c1e077791251576d2e198024004e91e92e5bc3fd8dab951a1c44a92a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6024a904be1214c55d310657c9b81fc420836a1e --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/8933ca49c1e077791251576d2e198024004e91e92e5bc3fd8dab951a1c44a92a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8cd94ee713189a2504f7b3645ea7ae00bc550f83f884718b8d3057a53ddafd3b +size 6493 diff --git a/parse/train/SJNRHiAcYX/images/920e032fcb273f6067148c7c16f0fa5a24bad1c1014642fd46379cfd79d00369.jpg b/parse/train/SJNRHiAcYX/images/920e032fcb273f6067148c7c16f0fa5a24bad1c1014642fd46379cfd79d00369.jpg new file mode 100644 index 0000000000000000000000000000000000000000..231d72da5ea3553a5718e561ee5563ce2dfc7392 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/920e032fcb273f6067148c7c16f0fa5a24bad1c1014642fd46379cfd79d00369.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4e4ac53152cfd3c26d85e48f74f7289e8bef2a92cd64723020e0ce107f25299 +size 1778 diff --git a/parse/train/SJNRHiAcYX/images/9a729e521700f37a39f067232e5df347ff65dd0fadf0e3e5561f7292c5932643.jpg b/parse/train/SJNRHiAcYX/images/9a729e521700f37a39f067232e5df347ff65dd0fadf0e3e5561f7292c5932643.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d82abcacc3cf90289147592e1f8eca8695c27e55 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/9a729e521700f37a39f067232e5df347ff65dd0fadf0e3e5561f7292c5932643.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03d87f45a3b96274e77925be194d8843d4515e2c78effbabd450114db94ab4c4 +size 98435 diff --git a/parse/train/SJNRHiAcYX/images/a5310da199a75eada0f1a088fea858f199b1db35b7950c991aa07d1d6fcf0137.jpg b/parse/train/SJNRHiAcYX/images/a5310da199a75eada0f1a088fea858f199b1db35b7950c991aa07d1d6fcf0137.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e7f2514da5ae48f32bebbfe13cb5afb634c12ede --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/a5310da199a75eada0f1a088fea858f199b1db35b7950c991aa07d1d6fcf0137.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5f244cf40ab24b6eafa09132e0c4a71032ecdd2bbdc00e40785557f269e07e3 +size 61402 diff --git a/parse/train/SJNRHiAcYX/images/ae13d7a56a1424e54244ace29afdd7bff06a9f11b0f5c3832c664b6dc73d74d3.jpg b/parse/train/SJNRHiAcYX/images/ae13d7a56a1424e54244ace29afdd7bff06a9f11b0f5c3832c664b6dc73d74d3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a34cc92d926bcf2448942a3195ecd36d48493b60 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/ae13d7a56a1424e54244ace29afdd7bff06a9f11b0f5c3832c664b6dc73d74d3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b4d6b8cdf86fc035cffa3a5138fc9fe4cc85ce61d92230aca2fabc8055d1dee +size 3866 diff --git a/parse/train/SJNRHiAcYX/images/b80ef8bbc910deb2ec669a0df4de5bb8c0dc347ef47e0c6f390a10d119d76993.jpg b/parse/train/SJNRHiAcYX/images/b80ef8bbc910deb2ec669a0df4de5bb8c0dc347ef47e0c6f390a10d119d76993.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a86521ef45e1a7ad4b4753eee294a0797915dd91 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/b80ef8bbc910deb2ec669a0df4de5bb8c0dc347ef47e0c6f390a10d119d76993.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d929c80b644e91a65a945e12d5433dac10288e005309ab3de57759ac7057922 +size 59040 diff --git a/parse/train/SJNRHiAcYX/images/d4c881ad5852ed73b895204daad8a6c4449b633e5220c699658617809aca49f9.jpg b/parse/train/SJNRHiAcYX/images/d4c881ad5852ed73b895204daad8a6c4449b633e5220c699658617809aca49f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..877e0cb9f6daa9868be531323505adcdaa019b3f --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/d4c881ad5852ed73b895204daad8a6c4449b633e5220c699658617809aca49f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9081852bb802096bf22ebbad2db7f2b6db5cad8257ed2b197aa224e7aadf0b83 +size 52420 diff --git a/parse/train/SJNRHiAcYX/images/dc525dcabfc4ef0169089b81bfebf2c033dcb05e7c6e37d2412f7dca05e0596d.jpg b/parse/train/SJNRHiAcYX/images/dc525dcabfc4ef0169089b81bfebf2c033dcb05e7c6e37d2412f7dca05e0596d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..22ffcec015edc7fbbd87f0a7c85a978684d3705d --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/dc525dcabfc4ef0169089b81bfebf2c033dcb05e7c6e37d2412f7dca05e0596d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c525223ed4d2af6be781cd2d503d439dee1dc5f9ca18f271a25e91da338cba54 +size 24303 diff --git a/parse/train/SJNRHiAcYX/images/e94e6939a4ac9c97585fcae3155d464fc26be5033da0425d3d1c2d7a7dc5500a.jpg b/parse/train/SJNRHiAcYX/images/e94e6939a4ac9c97585fcae3155d464fc26be5033da0425d3d1c2d7a7dc5500a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6d99f7e77c30199f5ad7cfae310b47d2d330f499 --- /dev/null +++ b/parse/train/SJNRHiAcYX/images/e94e6939a4ac9c97585fcae3155d464fc26be5033da0425d3d1c2d7a7dc5500a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e66d04b4178877067ca2ea1868afdab2ec53e9d69a014eb8a58ab59d55f91394 +size 54083 diff --git a/parse/train/SJe3HiC5KX/images/06955a667ea2ca0823275ef5c34c9e6f612d7e3b425b334e8688f43b6b265c89.jpg b/parse/train/SJe3HiC5KX/images/06955a667ea2ca0823275ef5c34c9e6f612d7e3b425b334e8688f43b6b265c89.jpg new file mode 100644 index 0000000000000000000000000000000000000000..882807eca08f5c434964db66e45c77976c233b20 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/06955a667ea2ca0823275ef5c34c9e6f612d7e3b425b334e8688f43b6b265c89.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e56a795d3102687646ee16493711a4ae85023d0c79e83acfa4c597af127efe36 +size 6562 diff --git a/parse/train/SJe3HiC5KX/images/4eee80a340c3f3f33eae0d8c1930637e0924fdfa84202e223380c9d35fe39e64.jpg b/parse/train/SJe3HiC5KX/images/4eee80a340c3f3f33eae0d8c1930637e0924fdfa84202e223380c9d35fe39e64.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4352ca2643dac8a5c326b4347cdf13b3b5045cb2 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/4eee80a340c3f3f33eae0d8c1930637e0924fdfa84202e223380c9d35fe39e64.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab327f3dea6542a8a5beb6c3790c220a8e0764bc3224bf31dd9c961dfa24f14c +size 15371 diff --git a/parse/train/SJe3HiC5KX/images/54e9b6361cedeccada8f1c30741246a865437f64d6db311c4eb123598dcfe012.jpg b/parse/train/SJe3HiC5KX/images/54e9b6361cedeccada8f1c30741246a865437f64d6db311c4eb123598dcfe012.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e9b327271e1fe8e766bac13a4bef1b42d27236a9 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/54e9b6361cedeccada8f1c30741246a865437f64d6db311c4eb123598dcfe012.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:adcb6cc2e56cf21ff4ca7d8566355fcc498936fb9f7e087db77d8c4c8ca93208 +size 3475 diff --git a/parse/train/SJe3HiC5KX/images/5a9a0aa985c90f4dd588532c6b7ffad64001c89dca70529fb6eafbbff051f83d.jpg b/parse/train/SJe3HiC5KX/images/5a9a0aa985c90f4dd588532c6b7ffad64001c89dca70529fb6eafbbff051f83d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6fa346c367adbb25be4859cb87b75ab5904ebf74 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/5a9a0aa985c90f4dd588532c6b7ffad64001c89dca70529fb6eafbbff051f83d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5041b31405d4ff161869c50c3865b34a7eede93c4bfbd562dfca97277f3929d9 +size 5973 diff --git a/parse/train/SJe3HiC5KX/images/606708df43996c4c3ee16011056478d696446a214d52ff399281d4c293088779.jpg b/parse/train/SJe3HiC5KX/images/606708df43996c4c3ee16011056478d696446a214d52ff399281d4c293088779.jpg new file mode 100644 index 0000000000000000000000000000000000000000..048a82d9056c9229e2c9b2b126487f1e5d1d871d --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/606708df43996c4c3ee16011056478d696446a214d52ff399281d4c293088779.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f7206a5f4b68f89140b5a1f186752809c044acc7352a16ebbc54541eecdb36de +size 8070 diff --git a/parse/train/SJe3HiC5KX/images/6af5dcaeacd2773b50d2e1e0ede51fb7a086b5fe46272a7400190bc2ae5d4a5d.jpg b/parse/train/SJe3HiC5KX/images/6af5dcaeacd2773b50d2e1e0ede51fb7a086b5fe46272a7400190bc2ae5d4a5d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..19aa724ddb730b2bf9393c87d6164538760d1298 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/6af5dcaeacd2773b50d2e1e0ede51fb7a086b5fe46272a7400190bc2ae5d4a5d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e136fb21e545e43afdf16e49233d53ab567628a0ab388eb3e32567e1ff746cf8 +size 76846 diff --git a/parse/train/SJe3HiC5KX/images/7371dcf571c6a2370810a973499090f8b06cbfba572289f2e1864fba2a9e6a42.jpg b/parse/train/SJe3HiC5KX/images/7371dcf571c6a2370810a973499090f8b06cbfba572289f2e1864fba2a9e6a42.jpg new file mode 100644 index 0000000000000000000000000000000000000000..924468ffd2b0286ce4ce8a7d642cd5b1157d09f6 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/7371dcf571c6a2370810a973499090f8b06cbfba572289f2e1864fba2a9e6a42.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:788787952efa8762ac17f0cd427b34d5e23ec2b38dda5877fa2c0192b569e7c8 +size 72967 diff --git a/parse/train/SJe3HiC5KX/images/7d2f29cf4001d94a3f3885b5178acbf97ec3cb8dc7e43ac7475f30850d87875b.jpg b/parse/train/SJe3HiC5KX/images/7d2f29cf4001d94a3f3885b5178acbf97ec3cb8dc7e43ac7475f30850d87875b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ddda3bfb7774a7e58cad8f55e376429239486398 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/7d2f29cf4001d94a3f3885b5178acbf97ec3cb8dc7e43ac7475f30850d87875b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96e46eb9ef3eb59cdf4c730f5236f8bce4ed599a47754bc69541a51225198fae +size 25100 diff --git a/parse/train/SJe3HiC5KX/images/83aa450125b6ed8ad171484ca199a1b14a521bc1902386d2cdec09b3ef04749a.jpg b/parse/train/SJe3HiC5KX/images/83aa450125b6ed8ad171484ca199a1b14a521bc1902386d2cdec09b3ef04749a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cb8d7abda3d0e453fa1d6ec7ad6c4ab2831684f3 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/83aa450125b6ed8ad171484ca199a1b14a521bc1902386d2cdec09b3ef04749a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4223f51fb3589c3f2a7aa8a82ea733dd1176c81b36792118f03d441ef2cd3419 +size 9611 diff --git a/parse/train/SJe3HiC5KX/images/8ee7e9346a61f58f347d8877435cc5a9ffe24cc2ce8caac2212b2230efb43686.jpg b/parse/train/SJe3HiC5KX/images/8ee7e9346a61f58f347d8877435cc5a9ffe24cc2ce8caac2212b2230efb43686.jpg new file mode 100644 index 0000000000000000000000000000000000000000..04e6126c70d3f2fa3f75738cb9e0605d76dc5aaa --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/8ee7e9346a61f58f347d8877435cc5a9ffe24cc2ce8caac2212b2230efb43686.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aec85750dc8ecbc081bdfb866269af0700ffe8ca19ee27708da94306af47abb5 +size 153047 diff --git a/parse/train/SJe3HiC5KX/images/a26dca4589c4287166db7102109a80b7876b04b5045729b931a2134125559640.jpg b/parse/train/SJe3HiC5KX/images/a26dca4589c4287166db7102109a80b7876b04b5045729b931a2134125559640.jpg new file mode 100644 index 0000000000000000000000000000000000000000..04cb844fab8a58a01f41ec0e94b122901a03fb54 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/a26dca4589c4287166db7102109a80b7876b04b5045729b931a2134125559640.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5584b5f4ee126012f3913a087dbd27685b69b7f99e86993d7272abe8e06d0f91 +size 22761 diff --git a/parse/train/SJe3HiC5KX/images/a4cec1062f3cabe4da7a316f96ab8b73126525037273f676a43ae2a72aa32461.jpg b/parse/train/SJe3HiC5KX/images/a4cec1062f3cabe4da7a316f96ab8b73126525037273f676a43ae2a72aa32461.jpg new file mode 100644 index 0000000000000000000000000000000000000000..90968d40c09129f6a3c3cc5f819db518baba144d --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/a4cec1062f3cabe4da7a316f96ab8b73126525037273f676a43ae2a72aa32461.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67ae0ba6bc58a8e427f2f0175a5b2c0f863f259241a119479f629ec00176fa8e +size 18547 diff --git a/parse/train/SJe3HiC5KX/images/a518c73202ee86aca28c5911a774c9336b156fd29f7e95cc8643bf18110f8ee9.jpg b/parse/train/SJe3HiC5KX/images/a518c73202ee86aca28c5911a774c9336b156fd29f7e95cc8643bf18110f8ee9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5142974d6c56f729f8f313ebd34e415a7ea5dcfc --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/a518c73202ee86aca28c5911a774c9336b156fd29f7e95cc8643bf18110f8ee9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:787051e2bce119ed85c59e92e5155c9226a02230853cbc8d0645e6c2c16503f0 +size 162306 diff --git a/parse/train/SJe3HiC5KX/images/ccf432ea8a26c6aba8fb50180e084bef065608157fc15cbad90443853971b9f0.jpg b/parse/train/SJe3HiC5KX/images/ccf432ea8a26c6aba8fb50180e084bef065608157fc15cbad90443853971b9f0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..beebf9302e881eb08d9698f98519db2a2bcc6895 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/ccf432ea8a26c6aba8fb50180e084bef065608157fc15cbad90443853971b9f0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5310fb927c145b801a1c7f158d38add1006ae494c07b313f6116fc285b899d9 +size 5535 diff --git a/parse/train/SJe3HiC5KX/images/e064d3974e5e9c0fb0c605063d71cb08f06bd13e992196deb75d0d3a0109b414.jpg b/parse/train/SJe3HiC5KX/images/e064d3974e5e9c0fb0c605063d71cb08f06bd13e992196deb75d0d3a0109b414.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b97c5445cb9b4da2ffc7ab08da1ac05e98707d07 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/e064d3974e5e9c0fb0c605063d71cb08f06bd13e992196deb75d0d3a0109b414.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8210bc8168257d9b64c29e62ab42880f9f29badd2f5fd69b9050137ea8dcfa9f +size 6247 diff --git a/parse/train/SJe3HiC5KX/images/e11cb5d6fcdf5370381a318c998557aef3776053c73c21d1ac6ad328e9b12fe8.jpg b/parse/train/SJe3HiC5KX/images/e11cb5d6fcdf5370381a318c998557aef3776053c73c21d1ac6ad328e9b12fe8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..598f743fe0edfa63a391262b9f6ad18649d3fdd0 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/e11cb5d6fcdf5370381a318c998557aef3776053c73c21d1ac6ad328e9b12fe8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95840928b6f2a7b17c5b3ab83b2189f40c99190fbc4169911af28a919e35f8b2 +size 5845 diff --git a/parse/train/SJe3HiC5KX/images/f6e8307a91cba11a9779b2312f183596f19aa74ded692c38ca1f932f398ec5fb.jpg b/parse/train/SJe3HiC5KX/images/f6e8307a91cba11a9779b2312f183596f19aa74ded692c38ca1f932f398ec5fb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9623c941d6063120c0ddd3f5753438ae3ebdc6d0 --- /dev/null +++ b/parse/train/SJe3HiC5KX/images/f6e8307a91cba11a9779b2312f183596f19aa74ded692c38ca1f932f398ec5fb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a25ffcc1c135fdc53d6494a1c2cc68ffdceced0a14494bd7e500726972048f0 +size 17115 diff --git a/parse/train/SJlJSaEFwS/SJlJSaEFwS.md b/parse/train/SJlJSaEFwS/SJlJSaEFwS.md new file mode 100644 index 0000000000000000000000000000000000000000..49db8cca8fe85cddd695b8d6034ec3bfbebe7b23 --- /dev/null +++ b/parse/train/SJlJSaEFwS/SJlJSaEFwS.md @@ -0,0 +1,273 @@ +# ROBUST CROSS-LINGUAL EMBEDDINGS FROM PARALLEL SENTENCES + +Anonymous authors Paper under double-blind review + +# ABSTRACT + +Recent advances in cross-lingual word embeddings have primarily relied on mapping-based methods, which project pre-trained word embeddings from different languages into a shared space through a linear transformation. However, these approaches assume word embedding spaces are isomorphic between different languages, which has been shown not to hold in practice (Søgaard et al., 2018), and fundamentally limits their performance. This motivates investigating joint learning methods which can overcome this impediment, by simultaneously learning embeddings across languages via a cross-lingual term in the training objective. Given the abundance of parallel data available (Tiedemann, 2012), we propose a bilingual extension of the CBOW method which leverages sentencealigned corpora to obtain robust cross-lingual word and sentence representations. Our approach significantly improves cross-lingual sentence retrieval performance over all other approaches, as well as convincingly outscores mapping methods while maintaining parity with jointly trained methods on word-translation. It also achieves parity with a deep RNN method on a zero-shot cross-lingual document classification task, requiring far fewer computational resources for training and inference. As an additional advantage, our bilingual method also improves the quality of monolingual word vectors despite training on much smaller datasets. We make our code and models publicly available. + +# 1 INTRODUCTION + +Cross-lingual representations—such as embeddings of words and phrases into a single comparable feature space—have become a key technique in multilingual natural language processing. They offer strong promise towards the goal of a joint understanding of concepts across languages, as well as for enabling the transfer of knowledge and machine learning models between different languages. Therefore, cross-lingual embeddings can serve a variety of downstream tasks such as bilingual lexicon induction, cross-lingual information retrieval, machine translation and many applications of zero-shot transfer learning, which is particularly impactful from resource-rich to low-resource languages. + +Existing methods can be broadly classified into two groups (Ruder et al., 2017): mapping methods leverage existing monolingual embeddings which are treated as independent, and apply a postprocess step to map the embeddings of each language into a shared space, through a linear transformation (Mikolov et al., 2013b; Conneau et al., 2017; Joulin et al., 2018). On the other hand, joint methods learn representations concurrently for multiple languages, by combining monolingual and cross-lingual training tasks (Luong et al., 2015; Coulmance et al., 2015; Gouws et al., 2015; Vulic & Moens, 2015; Chandar et al., 2014; Hermann & Blunsom, 2013). + +While recent work on word embeddings has focused almost exclusively on mapping methods, which require little to no cross-lingual supervision, (Søgaard et al., 2018) establish that their performance is hindered by linguistic and domain divergences in general, and for distant language pairs in particular. Principally, their analysis shows that cross-lingual hubness, where a few words (hubs) in the source language are nearest cross-lingual neighbours of many words in the target language, and structural non-isometry between embeddings do impose a fundamental barrier to the performance of linear mapping methods. + +(Ormazabal et al., 2019) propose using joint learning as a means of mitigating these issues. Given parallel data, such as sentences, a joint model learns to predict either the word or context in both source and target languages. As we will demonstrate with results from our algorithm, joint methods yield compatible embeddings which are closer to isomorphic, less sensitive to hubness, and perform better on cross-lingual benchmarks. + +Contributions. We propose the BI-SENT2VEC algorithm, which extends the SENT2VEC algorithm (Pagliardini et al., 2018; Gupta et al., 2019) to the cross-lingual setting. We also revisit TRANSGRAM Coulmance et al. (2015), another joint learning method, to assess the effectiveness of joint learning over mapping-based methods. Our contributions are + +• On cross-lingual sentence-retrieval and monolingual word representation quality evaluations, BI-SENT2VEC significantly outperforms competing methods, both jointly trained as well as mapping-based ones while preserving state-of-the-art performance on cross-lingual word retrieval tasks. For dis-similar language pairs, BI-SENT2VEC outperform their competitors by an even larger margin on all the tasks hinting towards the robustness of our method. BI-SENT2VEC performs on par with a multilingual RNN based sentence encoder, LASER (Artetxe & Schwenk, 2018), on MLDoc (Schwenk & Li, 2018), a zero-shot crosslingual transfer task on documents in multiple languages. Compared to LASER, our method improves computational efficiency by an order of magnitude for both training and inference, making it suitable for resource or latency-constrained on-device cross-lingual NLP applications. +We verify that joint learning methods consistently dominate state-of-the-art mapping methods on standard benchmarks, i.e., cross-lingual word and sentence retrieval. +• Training on parallel data additionally enriches monolingual representation quality, evident by the superior performance of BI-SENT2VEC over FASTTEXT embeddings trained on a $1 0 0 \times$ larger corpus. + +We make our models and code publicly available. + +# 2 RELATED WORK + +The literature on cross-lingual representation learning is extensive. Most recent advances in the field pursue unsupervised (Artetxe et al., 2017; Conneau et al., 2017; Chen & Cardie, 2018; Hoshen & Wolf, 2018; Grave et al., 2018b) or supervised (Joulin et al., 2018; Conneau et al., 2017) mapping or alignment-based algorithms. All these methods use existing monolingual word embeddings, followed by a cross-lingual alignment procedure as a post-processing step— that is to learn a simple (typically linear) mapping from the source language embedding space to the target language embedding space. + +Supervised learning of a linear map from a source embedding space to another target embedding space (Mikolov et al., 2013b) based on a bilingual dictionary was one of the first approaches towards cross-lingual word embeddings. Additionally enforcing orthogonality constraints on the linear map results in rotations, and can be formulated as an orthogonal Procrustes problem (Smith et al., 2017). However, the authors found the translated embeddings to suffer from hubness, which they mitigate by introducing the inverted softmax as a corrective search metric at inference time. (Artetxe et al., 2017) align embedding spaces starting from a parallel seed lexicon such as digits and iteratively build a larger bilingual dictionary during training. + +In their seminal work, (Conneau et al., 2017) propose an adversarial training method to learn a linear orthogonal map, avoiding bilingual supervision altogether. They further refine the learnt mapping by applying the Procrustes procedure iteratively with a synthetic dictionary generated through adversarial training. They also introduce the ‘Cross-Domain Similarity Local Scaling’ (CSLS) retrieval criterion for translating between spaces, which further improves on the word translation accuracy over nearest-neighbour and inverted softmax metrics. They refer to their work as Multilingual Unsupervised and Supervised Embeddings (MUSE). In this paper, we will use MUSE to denote the unsupervised embeddings introduced by them, and “Procrustes $^ +$ refine” to denote the supervised embeddings obtained by them. (Chen & Cardie, 2018) similarly use “multilingual adversarial training” followed by “pseudo-supervised refinement” to obtain unsupervised multilingual word embeddings (UMWE), as opposed to bilingual word embeddings by (Conneau et al., 2017). Hoshen & Wolf (2018) describe an unsupervised approach where they align the second moment of the two word embedding distributions followed by a further refinement. Building on the success of CSLS in reducing retrieval sensitivity to hubness, (Joulin et al., 2018) directly optimize a convex relaxation of the CSLS function (RCSLS) to align existing mono-lingual embeddings using a bilingual dictionary. + +While none of the methods described above require parallel corpora, all assume structural isomorphism between existing embeddings for each language (Mikolov et al., 2013b), i.e. there exists a simple (typically linear) mapping function which aligns all existing embeddings. However, this is not always a realistic assumption (Søgaard et al., 2018)—even in small toy-examples it is clear that many geometric configurations of points can not be linearly mapped to their targets. + +Joint learning algorithms such as TRANSGRAM (Coulmance et al., 2015) and Cr5 (Josifoski et al., 2019) , circumvent this restriction by simultaneously learning embeddings as well as their alignment. TRANSGRAM, for example, extends the Skipgram (Mikolov et al., 2013a) method to jointly train bilingual embeddings in the same space, on a corpus composed of parallel sentences. In addition to the monolingual Skipgram loss for both languages, they introduce a similar cross-lingual loss where a word from a sentence in one language is trained to predict the word-contents of the sentence in the other. Cr5, on the other hand, uses document-aligned corpora to achieve state-of-the-art results for cross-lingual document retrieval while staying competitive at cross-lingual sentence and word retrieval. TRANSGRAM embeddings have been absent from discussion in most of the recent work. However, the growing abundance of sentence-aligned parallel data (Tiedemann, 2012) merits a reappraisal of their performance. + +(Ormazabal et al., 2019) use BIVEC (Luong et al., 2015), another bilingual extension of Skipgram, which uses a bilingual dictionary in addition to parallel sentences to obtain word-alignments and compare it with the unsupervised version of VECMAP (Artetxe et al., 2018b), another mappingbased method. Our experiments show this extra level of supervision in the case of BIVEC is redundant in obtaining state-of-the-art performance. + +# 3 MODEL + +Proposed Model. Our BI-SENT2VEC model is a cross-lingual extension of SENT2VEC proposed by (Pagliardini et al., 2018), which in turn is an extension of the $C$ -BOW embedding method (Mikolov et al., 2013a). SENT2VEC is trained on sentence contexts, with the word and higher-order word n-gram embeddings specifically optimized toward obtaining robust sentence embeddings using additive composition. Formally, SENT2VEC obtains representation ${ \pmb v } _ { s }$ of a sentence $S$ by averaging the word-ngram embeddings (including unigrams) as $\begin{array} { r } { \mathbf { \dot { \boldsymbol { v } } } _ { s } : = \frac { 1 } { R ( S ) } \sum _ { w \in R ( S ) } \pmb { v } _ { w } } \end{array}$ where $R ( S )$ is the set of word n-grams in the sentence $S$ . + +The SENT2VEC training objective aims to predict a masked word token $w _ { t }$ in the sentence $S$ using the rest of the sentence representation ${ \pmb v } _ { S \backslash \{ { u v } _ { t } \} }$ . To formulate the training objective, we use logistic loss $\ell : x \mapsto \log { ( 1 + e ^ { - x } ) }$ in conjunction with negative sampling. More precisely, for a raw text corpus $C$ , the monolingual training objective for SENT2VEC is given by + +$$ +\operatorname* { m i n } _ { U , V } \sum _ { S \in C } \sum _ { w _ { t } \in S } \left( \ell \big ( \boldsymbol { u } _ { w _ { t } } ^ { \top } \boldsymbol { v } _ { S \setminus \{ w _ { t } \} } \big ) + \sum _ { w ^ { \prime } \in N _ { w _ { t } } } \ell \big ( - \boldsymbol { u } _ { w ^ { \prime } } ^ { \top } \boldsymbol { v } _ { S \setminus \{ w _ { t } \} } \big ) \right) +$$ + +where $w _ { t }$ is the target word and, $V$ and $U$ are the source n-gram and target word embedding matrices respectively. Here, the set of negative words $N _ { w _ { t } }$ is sampled from a multinomial distribution where the probability of picking a word is directly proportional to the square root of its frequency in the corpus. Each target word $w _ { t }$ is sampled with probability $m i n \{ 1 , \sqrt { t / f _ { w _ { t } } } + t / f _ { w _ { t } } \}$ where $f _ { w _ { t } }$ is the frequency of the word in the corpus. + +We adapt the SENT2VEC model to bilingual corpora by introducing a cross-lingual loss in addition to the monolingual loss in equation (1). Given a sentence pair $S = ( S _ { l _ { 1 } } , S _ { l _ { 2 } } )$ where $S _ { l _ { 1 } }$ and $S _ { l _ { 2 } }$ are translations of each other in languages $l _ { 1 }$ and $l _ { 2 }$ , the cross-lingual loss for a target word $w _ { t }$ in $l _ { 1 }$ is given by + +$$ +\ell \big ( { \pmb u } _ { w _ { t } } ^ { \top } { \pmb v } _ { S _ { l _ { 2 } } } \big ) + \sum _ { w ^ { \prime } \in N _ { w _ { t } } } \ell \big ( - { \pmb u } _ { w _ { t } ^ { \prime } } ^ { \top } { \pmb v } _ { S _ { l _ { 2 } } } \big ) +$$ + +Thus, we use the sentence $S _ { l _ { 1 } }$ to predict the constituent words of $S _ { l _ { 2 } }$ and vice-versa in a similar fashion to the monolingual SENT2VEC, shown in Figure 1. This ensures that the word and $\mathbf { n }$ -gram embeddings of both languages lie in the same space. + +![](images/447f2c4e9599c41e36d0eb4b008a68d7f6bba88815213ca6273055255c6f9dd4.jpg) +Figure 1: An illustration of the BI-SENT2VEC training process. A word from a sentence pair is chosen as a target and the algorithm learns to predict it using the rest of the sentence(monolingual training component) and the translation of the sentence(cross-lingual component). + +Assuming $C$ to be a sentence aligned bilingual corpus and combining equations (1) and (2), our BI-SENT2VEC model objective function is formulated as + +$$ +\operatorname* { m i n } _ { U , V } \sum _ { \underbrace { l , l ^ { \prime } \in [ l _ { 1 } , l _ { 2 } ] } _ { l \neq l ^ { \prime } } } \sum _ { w _ { t } \in S _ { l } \atop l \neq l ^ { \prime } } \left( \underbrace { \ell ( u _ { w _ { t } } ^ { \top } v _ { S _ { l } \backslash \{ w _ { t } \} } ) + \sum _ { w ^ { \prime } \in N _ { w _ { t } } } \ell ( - u _ { w _ { t } ^ { \prime } } v _ { S _ { l } \backslash \{ w _ { t } \} } ) } _ { \mathrm { m o n o i n g u a l ~ l o s s } } + \underbrace { \ell ( u _ { w _ { t } } ^ { \top } v _ { S _ { l ^ { \prime } } } ) + \sum _ { w ^ { \prime } \in N _ { w _ { t } } } \ell ( - u _ { w _ { t } ^ { \prime } } v _ { S _ { l ^ { \prime } } } ) } _ { \mathrm { c r o s s i n g u a l ~ l o s s } } \right) +$$ + +Implementation Details. We build our $\mathrm { C } { + } { + }$ implementation on the top of the FASTTEXT library (Bojanowski et al., 2016; Joulin et al., 2016). Model parameters are updated by asynchronous SGD with a linearly decaying learning rate. + +Our model is trained on the ParaCrawl (Espla-Gomis, 2019) v4.0 datasets for the English-Italian, \` English-German, English-French, English-Spanish, English-Hungarian and English-Finnish language pairs. For the English-Russian language pair, we concatenate the OpenSubtitle corpus1(Lison & Tiedemann, 2016) and the Tanzil project2 (Quran translations) corpus. The number of parallel sentence pairs in the corpora except for those of English-Finnish and English-Hungarian used by us range from 17-32 Million. Number of parallel sentence pairs for the dis-similar language pairs(English-Hungarian and English-Finnish) is approximately 2 million. Evaluation results for these two language pairs can be found in Subsection 4.4. Exact statistics regarding the different corpora can be found in the Table 7 in the Appendix. All the sentences were tokenized using Spacy tokenizers3 for their respective languages. + +For each dataset, we trained two different models: one with unigram embeddings only, and the other additionally augmented with bigrams. The earlier TRANSGRAM models (Coulmance et al., 2015) were trained on a small amount of data (Europarl Corpus (Koehn, 2005)). To facilitate a fair comparison, we train new TRANSGRAM embeddings on the same data used for BI-SENT2VEC. Given that TRANSGRAM and BI-SENT2VEC are a cross-lingual extension of Skipgram and SENT2VEC respectively, we use the same parameters as (Bojanowski et al., 2016) and (Gupta et al., 2019), except increasing the number of epochs for TRANSGRAM to 8, and decreasing the same for BI-SENT2VEC to 5. Additionally, a preliminary hyperparameter search (except changing the number of epochs) on BI-SENT2VEC and TRANSGRAM did not improve the results. All parameters for training the TRANSGRAM and BI-SENT2VEC models can be found in the Table 6 in the Appendix. + +In order to make the comparison more extensive, we also train VECMAP (mapping-based) (Artetxe et al., 2018b;a) and BIVEC (joint-training)(Luong et al., 2015) methods on the same corpora using the exact pipeline as (Ormazabal et al., 2019). + +# 4 EVALUATION + +To assess the quality of the word and sentence embeddings obtained as well as their cross-lingual alignment quality, we compare our results using the following four benchmarks + +• Cross-lingual word retrieval • Monolingual word representation quality + +• Cross-lingual sentence retrieval • Zero-shot cross-lingual transfer of document classifiers + +where benchmarks are presented in order of increasing linguistic granularity, i.e. word, sentence, and document level. We also analyze the effect of training data by studying the relationship between representation quality and corpus size. + +We use the code available in the MUSE library4 (Conneau et al., 2017) for all evaluations except the zero-shot classifier transfer, which is tested on the MLDoc task (Schwenk & Li, 2018)5. + +# 4.1 WORD TRANSLATION + +The task involves retrieving correct translation(s) of a word in a source language from a target language. To evaluate translation accuracy, we use the bilingual dictionaries constructed by (Conneau et al., 2017). We consider 1500 source-test queries and $2 0 0 \mathrm { k }$ target words for each language pair and report $\mathrm { P @ 1 }$ scores for the supervised and unsupervised baselines as well as our models in Table 1. + +Table 1: Word translation retrieval $\mathbf { P } @ \mathbf { 1 }$ for various language pairs of MUSE evaluation dictionary (Conneau et al., 2017). NN: nearest neighbours. CSLS: Cross-Domain Similarity Local Scaling. (‘en’ is English, ‘fr’ is French, ‘de’ is German, ‘ru’ is Russian, ‘it’ is Italian) (‘uni.’ and ‘bi.’ denote unigrams and bigrams respectively) denotes translation from the first language to the second and $\gets$ the other way around.) + +
Methoden-esen-fren-deen-ruen-itavg.
→↑→↑→↑→↑→↑
MUSE (Conneau et al.,2017)81.7 83.382.3 82.174.0 72.244.0 59.178.6 77.973.5
UMWE(Chen & Cardie,2018)82.5 83.182.5 82.174.6 72.549.5 61.778.3 77.074.4
Procrustes + refine (Conneau et al., 2017)82.4 83.982.3 83.275.3 73.250.1 63.577.5 77.674.9
RCSLS (Joulin et al., 2018)83.7 87.184.1 84.779.2 77.560.9 70.281.1 82.779.1
TRANSGRAM (Coulmance et al., 2015)91.6 88.689.1 90.187.5 87.265.6 73.788.6 89.585.2
VECMAP (unsupervised) (Artetxe et al.,2018b)87.4 87.888.3 88.584.3 87.248.6 50.587.4 86.579.6
VECMAP (supervised) (Artetxe et al.,2018a)87.2 90.287.6 90.487.3 86.849.7 65.687.2 89.282.1
BIVEC NN (Luong et al., 2015)87.4 88.686.8 89.187.5 87.264.0 59.186.8 84.081.7
BIVEC CSLS (Luong et al., 2015)87.6 89.188.8 90.386.4 87.266.1 70.687.6 87.884.3
BI-SENT2VEC uni. NN86.9 91.686.9 91.086.0 88.758.0 72.888.3 92.484.3
BI-SENT2VEC uni. + bi. NN89.4 92.989.3 92.886.7 89.359.0 70.289.5 91.885.1
BI-SENT2VEC uni.CSLS86.0 91.786.4 91.484.6 88.860.5 73.088.2 91.884.2
BI-SENT2VEC uni. + bi. CSLS89.0 92.188.9 92.486.5 89.061.0 73.589.6 91.485.3
+ +# 4.2 MONOLINGUAL WORD REPRESENTATION QUALITY + +We assess the monolingual quality improvement of our proposed cross-lingual training by evaluating performance on monolingual word similarity tasks. To disentangle the specific contribution of the cross-lingual loss, we train the monolingual counterpart of BI-SENT2VEC, SENT2VEC on the same corpora as our method. + +Performance on monolingual word-similarity tasks is evaluated using the English SimLex-999 (Hill et al., 2014) and its Italian and German translations, English WS-353 (Finkelstein et al., 2001) and its German, Italian and Spanish translations. For French, we use a translation of the RG-65 (Joubarne & Inkpen, 2011) dataset. Pearson scores are used to measure the correlation between human-annotated word similarities and predicted cosine similarities. We also include FASTTEXT monolingual vectors trained on CommonCrawl data (Grave et al., 2018a) which is comprised of 600 billion, 68 billion, 66 billion, 72 billion and 36 billion words of English, French, German, Spanish and Italian respectively and is at least $1 0 0 \times$ larger than the corpora on which we trained BI-SENT2VEC. We report Pearson correlation scores on different word-similarity datasets for En-It pair in Table 2. Evaluation results on other language pairs are similar and can be found in the appendix in Tables 8, 9, and 10. + +Table 2: Monolingual word similarity task performance of our methods when trained on en-it ParaCrawl data. We report Pearson correlation scores. + +
Method\DatasetSimLex-999WS-353
en iten it
MUSE0.380.300.74 0.64
RCSLS0.38 0.300.740.64
FASTTEXT- Common Crawl0.49 0.320.750.57
BIVEC0.40 0.360.700.60
TRANSGRAM0.43 0.370.730.63
SENT2VEC uni.0.49 0.380.730.60
BI-SENT2VEC uni.0.570.47 0.790.65
BI-SENT2VEC uni. + bi.0.580.50 0.800.69
+ +# 4.3 CROSS-LINGUAL SENTENCE RETRIEVAL + +The primary contribution of our work is to deliver improved cross-lingual sentence representations. We test sentence embeddings for each method obtained by bag-of-words composition for sentence retrieval across different languages on the Europarl corpus. In particular, the tf-idf weighted average is used to construct sentence embeddings from word embeddings. We consider 2000 sentences in the source language dataset and retrieve their translation among 200K sentences in the target language dataset. The other 300K sentences in the Europarl corpus are used to calculate tf-idf weights. Results for $\mathrm { P @ 1 }$ of unsupervised and supervised benchmarks vs our models are included in Table 3. + +
Methoden-esen-fren-deen-itavg.
→↑→↑→↑→↑
MUSE72.771.569.268.853.353.466.164.364.9
RCSLS26.926.719.321.28.811.315.117.618.4
TRANSGRAM83.581.480.481.664.869.977.277.977.1
VECMAP (unsupervised)81.782.179.880.462.864.669.071.174.0
VECMAP (supervised)81.38180.480.762.664.367.87173.6
BIVEC NN69.877.154.775.556.144.158.245.160.1
BIVEC CSLS81.683.478.181.671.668.174.272.476.4
BI-SENT2VEC uni. NN87.886.485.283.482.380.285.985.884.6
BI-SENT2VEC uni. + bi. NN87.987.886.183.979.579.785.185.384.4
BI-SENT2VEC uni. CSLS89.588.587.186.484.483.088.287.586.8
BI-SENT2VEC uni. + bi. CSLS89.789.687.887.484.284.087.987.687.3
Reduction in error37.5% 37.3%37.8% 31.5%44.4% 46.8%46.9% 43.9%1
+ +Table 3: Cross-lingual Sentence retrieval. We report $\mathrm { P @ 1 }$ scores for 2000 source queries searching over 200 000 target sentences. Reduction in error is calculated with respect to BI-SENT2VEC uni. $^ +$ bi. CSLS and the best non-BI-SENT2VEC method. + +# 4.4 PERFORMANCE ON DIS-SIMILAR LANGUAGE PAIRS + +We report a substantial improvement on the performance of previous models on cross-lingual word and sentence retrieval tasks for the dis-similar language pairs(English-Finnish and EnglishHungarian). We use the same evaluation scheme as in Subsections 4.1 and 4.3 Results for these pairs are included in Table 4. + +# 4.5 ZERO-SHOT CROSS-LINGUAL TRANSFER OF DOCUMENT CLASSIFIERS + +The MLDoc multilingual document classification task (Schwenk & Li, 2018) consists of news documents given in 8 different languages, which need to be classified into 4 different categories. To demonstrate the ability to transfer trained classifiers in a robust fashion between languages, we use a zero-shot setting, i.e., we train a classifier on embeddings in the source language, and report the accuracy of the same classifier applied to the target language. As the classifier, we use a simple feed-forward neural network with two hidden layers of size 10 and 8 respectively, optimized using the Adam optimizer. Each document is represented using the sum of its sentence embeddings. + +Table 4: Cross-lingual Word and Sentence retrieval for dis-similar language pairs $( \mathbf { P } @ \mathbf { 1 }$ scores). ‘en’ is English, ‘fi’ is Finnish, ‘hu’ is Hungarian + +
Method word retrieval sentence retrieval
en-fi →↑en-hu → ↑→en-fi ↑en-hu →↑
MUSE48.159.553.9 64.921.729.539.146.7
RCSLS61.869.967.0 73.03.24.83.65.1
VECMAP (unsupervised)62.566.861.6 68.713.214.720.519.3
VECMAP (supervised)62.678.363.7 76.615.016.920.921.7
BIVEC NN62.155.362.1 53.714.29.726.213.7
BIVEC CSLS69.678.072.4 78.433.332.046.741.3
TRANSGRAM69.781.173.1 80.835.440.552.155
B1-SENT2VEC uni. NN71.285.475.683.963.5 64.275.2 76.2
BI-SENT2VEC uni. + bi. NN68.581.771.4 79.457.5 55.965.865.2
B1-SENT2VEC uni. CSLS72.0 86.576.3 85.170.2 69.081.480.8
BI-SENT2VEC uni. + bi. CSLS70.184.473.781.766 64.173.874.5
+ +
Methoden-esen-fren-deavg.
→↑→冏↑→↑→en-it ↑
LASER79.3 69.678.0 80.186.3 80.870.2 74.277.3
BI-SENT2VEC74.0 71.581.6 82.286.5 79.275.0 72.677.8
+ +Table 5: MLDoc Benchmark results (Schwenk & Li, 2018). A document classifier was trained on one language and tested on another without additional training/fine-tuning. We report $\%$ accuracy. + +We compare the performance of BI-SENT2VEC with the LASER sentence embeddings (Artetxe & Schwenk, 2018) in Table 5. LASER sentence embedding model is a multi-lingual sentence embedding model which is composed of a biLSTM encoder and an LSTM decoder. It uses a shared byte pair encoding based vocabulary of 50k words. The LASER model which we compare to was trained on 223M sentences for 93 languages and requires 5 days to train on 16 V100 GPUs compared to our model which takes 1-2.5 hours for each language pair on 30 CPU threads. + +# 4.6 EFFECT OF CORPUS SIZE ON REPRESENTATION QUALITY + +We conduct an ablation study on how BI-SENT2VEC embeddings’ performance depends on the size of the training corpus. We uniformly sample smaller subsets of the En-Fr ParaCrawl dataset and train a BI-SENT2VEC model on them. We test word/sentence translation performance with the CSLS retrieval criterion, and monolingual embedding quality for En-Fr with increasing ParaCrawl corpus size. The results are illustrated in Figures 2 and 3. + +# 5 DISCUSSION + +In the following section, we discuss the results on monolingual and cross-lingual benchmarks, presented in Tables 1 - 5, and a data ablation study for how the model behaves with increasing parallel corpus size in Figure $2 \cdot 3$ . The most impressive outcome of our experiments is improved crosslingual sentence retrieval performance, which we elaborate on along with word translation in the next subsection. + +![](images/bb3e3da4961fcc13096a65b06ddba37fd74cb2a40ffb7fe4be5b7a1bd56621c1.jpg) +Figure 2: Effect of corpus size on cross-lingual word/sentence retrieval performance. + +![](images/1145c4987fe2df1428c48252033511b35e1e13156a25dc7e8955255c49fc87e7.jpg) +Figure 3: Effect of corpus size on monolingual word quality. We use SimLex-999, WS-353, and FR-RG datasets for measuring monolingual word embedding quality. + +Cross-lingual evaluations For cross-lingual tasks, we observe in Table 1 that jointly trained embeddings produce much better results on cross-lingual word and sentence retrieval tasks. BISENT2VEC’s performance on word-retrieval tasks is uniformly superior to mapping methods, achieving up to $1 1 . 5 \%$ more in $\mathrm { P @ 1 }$ than RCSLS for the English to German language pair, consistent with the results from (Ormazabal et al., 2019). It is also on-par with, or better than competing joint methods except on translation from Russian to English, where TRANSGRAM receives a significantly better score. For word retrieval tasks, there is no discernible difference between CSLS/NN criteria for BI-SENT2VEC, suggesting the relative absence of the hubness phenomenon which significantly hinders the performance of cross-lingual word embedding methods. + +Our principal contribution is in improving cross-lingual sentence retrieval. Table 3 shows BISENT2VEC decisively outperforms all other methods by a wide margin, reducing the relative $\mathrm { P @ 1 }$ error anywhere from $3 1 . 5 \%$ to $5 5 . 1 \%$ . Our model displays considerably less variance than others in quality across language pairs, with at most a $\approx 5 \%$ deficit between best and worst, and nearly symmetric accuracy within a language pair. + +TRANSGRAM also outperforms the mapping-based methods, but still falls significantly short of BISENT2VEC’s. These results can be attributed to the fact that BI-SENT2VEC directly optimizes for obtaining robust sentence embeddings using additive composition of its word embeddings. Since BI-SENT2VEC’s learning objective is closest to a sentence retrieval task amongst current state-ofthe-art methods, it can surpass them without sacrificing performance on other tasks. + +Cross-lingual evaluations on dis-similar language pairs Unlike other language pairs in the evaluation, English-Finnish and English-Hungarian pairs are composed of languages from two different language families(English being an Indo-European language and the other language being a Finno-Ugric language). In Table 4, we see that the performance boost achieved by BI-SENT2VEC on competing methods methods is more pronounced in the case of dis-similar language pairs as compared to paris of languages close to each other. This observation affirms the suitaibility of BISENT2VEC for learning joint representations on languages from different families. + +Monolingual word quality For the monolingual word similarity tasks, we observe large gains over existing methods. SENT2VEC is trained on the same corpora as us, and FASTTEXT vectors are trained on the CommonCrawl corpora which are more than 100 times larger than ParaCrawl v4.0. In Table 2, we see that BI-SENT2VEC outperforms them by a significant margin on SimLex-999 and WS-353, two important monolingual word quality benchmarks. This observation is in accordance with the fact (Faruqui & Dyer, 2014) that bilingual contexts can be surprisingly effective for learning monolingual word representations. However, amongst the joint-training methods, BI-SENT2VEC also outperforms TRANSGRAM and BIVEC trained on the same corpora by a significant margin, again hinting at the superiority of the sentence level loss function over a fixed context window loss. + +Effect of n-grams (Gupta et al., 2019) report improved results on monolingual word representation evaluation tasks for SENT2VEC and FASTTEXT word vectors by training them alongside word n-grams. Our method incorporates their results based on the observation that unigram vectors trained alongside with bigrams significantly outperform unigrams alone on the majority of the evaluation tasks. We can see from Tables 1 - 3 that this holds for the bilingual case as well. However, in case of dis-similar language pairs(Table 4), we observe that using n-grams degrades the cross-lingual performance of the embeddings. This observation suggests that use of higher order n-grams may not be helpful for language pairs where the grammatical structures are contrasting. + +Effect of corpus size Considering the cross-lingual performance curve exhibited by BISENT2VEC in Figure 2, increasing corpus size for the English-French datasets up to 1-3.1M lines appears to saturate the performance of the model on cross-lingual word/sentence retrieval, after which it either plateaus or degrades slightly. This is an encouraging result, indicating that joint methods can use significantly less data to obtain promising performance. This implies that joint methods may not necessarily be constrained to high-resource language pairs as previously assumed, though further experimentation is needed to verify this claim. + +It should be noted from Figure 3 that the monolingual quality does keep improving with an increase in the size of the corpus. A potential way to overcome this issue of plateauing cross-lingual performance is to give different weights to the monolingual and cross-lingual component of the loss with the weights possibly being dependent on other factors such as training progress. + +Comparison with a cross-lingual sentence embedding model and performance on document level task On the MLDoc classifier transfer task (Schwenk & Li, 2018) where we evaluate a classifier learned on documents in one language on documents in another, Table 5 shows we achieve parity with the performance of the LASER model for language pairs involving English, where BISENT2VEC’s average accuracy of $7 7 . 8 \%$ is slightly higher than LASER’s $7 7 . 3 \%$ . While the comparison is not completely justified as LASER is multilingual in nature and is trained on a different dataset, one must emphasize that BI-SENT2VEC is a bag-of-words method as compared to LASER which uses a multi-layered biLSTM sentence encoder. Our method only requires to average a set of vectors to encode sentences reducing its computational footprint significantly. This makes BI-SENT2VEC an ideal candidate for on-device computationally efficient cross-lingual NLP, unlike LASER which has a huge computational overhead and specialized hardware requirement for encoding sentences. + +# 6 CONCLUSION AND FUTURE WORK + +We introduce a cross-lingual extension of an existing monolingual word and sentence embedding method. The proposed model is tested at three levels of linguistic granularity: words, sentences and documents. The model outperforms all other methods by a wide margin on the cross-lingual sentence retrieval task while maintaining parity with the best-performing methods on word translation tasks. Our method achieves parity with LASER on zero-shot document classification, despite being a much simpler model. We also demonstrate that training on parallel data yields a significant improvement in the monolingual word representation quality. + +The success of our model on the bilingual level calls for its extension to the multilingual level especially for pairs which have little or no parallel corpora. While the amount of bilingual/multilingual parallel data has grown in abundance, the amount of monolingual data available is practically limitless. Consequently, we would like to explore training cross-lingual embeddings with a large amount of raw text combined with a smaller amount of parallel data. + +# REFERENCES + +Mikel Artetxe and Holger Schwenk. Massively multilingual sentence embeddings for zero-shot cross-lingual transfer and beyond. Transactions of the Association for Computational Linguistics, 7:597–610, 2018. + +Mikel Artetxe, Gorka Labaka, and Eneko Agirre. Learning bilingual word embeddings with (almost) no bilingual data. In ACL, 2017. + +Mikel Artetxe, Gorka Labaka, and Eneko Agirre. Generalizing and improving bilingual word embedding mappings with a multi-step framework of linear transformations. In Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, pp. 5012–5019, 2018a. + +Mikel Artetxe, Gorka Labaka, and Eneko Agirre. A robust self-learning method for fully unsupervised cross-lingual mappings of word embeddings. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 789–798, 2018b. + +Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. Enriching word vectors with subword information. Transactions of the Association for Computational Linguistics, 5:135–146, 2016. + +A. P. Sarath Chandar, Stanislas Lauly, Hugo Larochelle, Mitesh M. Khapra, Balaraman Ravindran, Vikas C. Raykar, and Amrita Saha. An autoencoder approach to learning bilingual word representations. In NIPS, 2014. + +Xilun Chen and Claire Cardie. Unsupervised multilingual word embeddings. In EMNLP, 2018. + +Alexis Conneau, Guillaume Lample, Marc’Aurelio Ranzato, Ludovic Denoyer, and Herve J ´ egou.´ Word translation without parallel data. ArXiv, abs/1710.04087, 2017. + +Jocelyn Coulmance, Jean-Marc Marty, Guillaume Wenzek, and Amine Benhalloum. Trans-gram, Fast Cross-lingual Word-embeddings. In EMNLP - Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pp. 1109–1113, 2015. + +M. Espla-Gomis. ParaCrawl: Web-scale parallel corpora for the languages of the EU. 2019. \` + +Manaal Faruqui and Chris Dyer. Improving vector space word representations using multilingual correlation. In EACL, 2014. + +Lev Finkelstein, Evgeniy Gabrilovich, Yossi Matias, Ehud Rivlin, Zach Solan, Gadi Wolfman, and Eytan Ruppin. Placing search in context: the concept revisited. In WWW, 2001. + +Stephan Gouws, Yoshua Bengio, and Greg Corrado. Bilbowa: Fast bilingual distributed representations without word alignments. 2015. + +Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Armand Joulin, and Tomas Mikolov. Learning word vectors for 157 languages. ArXiv, abs/1802.06893, 2018a. + +Edouard Grave, Armand Joulin, and Quentin Berthet. Unsupervised alignment of embeddings with wasserstein procrustes. In AISTATS, 2018b. + +Prakhar Gupta, Matteo Pagliardini, and Martin Jaggi. Better word embeddings by disentangling contextual n-gram information. In NAACL-HLT, 2019. + +Karl Moritz Hermann and Phil Blunsom. Multilingual distributed representations without word alignment. In ICLR 2014, 2013. + +Felix Hill, Roi Reichart, and Anna Korhonen. Simlex-999: Evaluating semantic models with (genuine) similarity estimation. Computational Linguistics, 41:665–695, 2014. + +Yedid Hoshen and Lior Wolf. Non-adversarial unsupervised word translation. In EMNLP, 2018. + +Martin Josifoski, Ivan S Paskov, Hristo S Paskov, Martin Jaggi, and Robert West. Crosslingual document embedding as reduced-rank ridge regression. In Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining, pp. 744–752. ACM, 2019. + +Colette Joubarne and Diana Inkpen. Comparison of semantic similarity for different languages using the google n-gram corpus and second-order co-occurrence measures. In Canadian Conference on AI, 2011. + +Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. Bag of tricks for efficient text classification. In EACL, 2016. + +Armand Joulin, Piotr Bojanowski, Tomas Mikolov, Herve J ´ egou, and Edouard Grave. Loss in trans-´ lation: Learning bilingual word mapping with a retrieval criterion. In EMNLP, 2018. + +Philipp Koehn. Europarl: A parallel corpus for statistical machine translation. 2005. + +Pierre Lison and Jorg Tiedemann. Opensubtitles2016: Extracting large parallel corpora from movie ¨ and tv subtitles. In LREC, 2016. + +Thang Luong, Hieu Pham, and Christopher D Manning. Bilingual word representations with monolingual quality in mind. In Proceedings of the 1st Workshop on Vector Space Modeling for Natural Language Processing, pp. 151–159, 2015. + +Tomas Mikolov, Kai Chen, Gregory S. Corrado, and Jeffrey Dean. Efficient estimation of word representations in vector space. CoRR, abs/1301.3781, 2013a. + +Tomas Mikolov, Quoc V. Le, and Ilya Sutskever. Exploiting similarities among languages for machine translation. ArXiv, abs/1309.4168, 2013b. + +Aitor Ormazabal, Mikel Artetxe, Gorka Labaka, Aitor Soroa, and Eneko Agirre. Analyzing the limitations of cross-lingual word embedding mappings. In ACL, 2019. + +Matteo Pagliardini, Prakhar Gupta, and Martin Jaggi. Unsupervised learning of sentence embeddings using compositional n-gram features. In NAACL-HLT, 2018. + +Sebastian Ruder, Ivan Vulic, and Anders Søgaard. A survey of cross-lingual word embedding mod-´ els. arXiv preprint arXiv:1706.04902, 2017. + +Holger Schwenk and Xian Li. A corpus for multilingual document classification in eight languages. arXiv preprint arXiv:1805.09821, 2018. + +Samuel L. Smith, David H. P. Turban, Steven Hamblin, and Nils Y. Hammerla. Offline bilingual word vectors, orthogonal transformations and the inverted softmax. ArXiv, abs/1702.03859, 2017. + +Anders Søgaard, Sebastian Ruder, and Ivan Vulic. On the limitations of unsupervised bilingual dictionary induction. In ACL, 2018. + +Jorg Tiedemann. Parallel data, tools and interfaces in opus. In ¨ Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC-2012), pp. 2214–2218, 2012. + +Ivan Vulic and Marie-Francine Moens. Bilingual word embeddings from non-parallel documentaligned data applied to bilingual lexicon induction. In ACL, 2015. + +# A APPENDIX + +A.1 DATASET STATISTICS + +
DatasetNumber of sentencesNumber of tokens(English tokens if bilingual)
En-De ParaCrawl v4.017 Million308 Million
En-Es ParaCrawl v4.022 Million477 Million
En-FiParaCrawl v4.02.16 Million42 Million
En-FrParaCrawl v4.032 Million665 Million
En-Hu ParaCrawl v4.01.91Million31Million
En-It ParaCrawl v4.013 Million261 Million
En-Ru OpenSubtitles+ Tanzil27 Million363Million
Wikipedia - En70 Million1792 Million
Wikipedia - De二1384 Million
Wikipedia - Fr二1108Million
Wikipedia - Es二797 Million
Wikipedia - It二702 Million
Wikipedia - Ru二824 Million
Common Crawl - En二600 Billion
Common Crawl - De二66 Billion
Common Crawl -Fr二68 Billion
Common Crawl - It二36 Billion
Common Crawl -Es172 Billion
+ +Table 6: Dataset Sizes. ‘En’,‘De’,‘Fi’,‘Fr’,‘Hu’,‘It’,‘Es’ and ‘Ru’ stand for English, German, Finnish, French, Hungarian, Italian, Spanish and Russian respectively. + +We used ParaCrawl $\mathrm { v } 4 . 0$ corpora for training BI-SENT2VEC, SENT2VEC,BIVEC,VECMAP and TRANSGRAM embeddings except for En-Ru pair for which we used OpenSubtitles and Tanzil corpora combined. MUSE and RCSLS vectors were trained from FASTTEXT vectors obtained from Wikipedia dumps(Grave et al., 2018a). + +# A.2 TRAINING PARAMETERS FOR TRAINED MODELS + +Table 7: Hyperparameters for the trained models + +
ModelBI-SENT2VECuni.BI-SENT2VECuni. + bi.SENT2VECuni.TRANSGRAM
Embedding dimension300300300300
Maxvocabulary size750k750k750k750k
Minimum word count5855
Initial Learning Rate0.20.20.20.025
Epochs5558
Subsampling hyper-parameter1·10-55:10-61·10-51·10-4
Word-Ngrams Bucket Size二2M二二
Word-Ngrams dropped per context141二
Window size5
Number of negatives sampled1010105
+ +A.3 ADDITIONAL MONOLINGUAL QUALITY TABLES + +
SimLex-999 Method\DatasetWS-353
enenes
MUSE RCSLS0.380.740.61
FASTTEXT- Common Crawl0.38 0.490.74 0.750.62 0.54
BIVEC0.400.720.57
TRANSGRAM0.420.740.59
SENT2VEC uni.0.490.580.51
BI-SENT2VEC uni.0.570.780.60
BI-SENT2VEC uni. + bi.0.600.820.66
+ +Table 8: Monolingual word similarity task performance of our methods when trained on en-es ParaCrawl data. We report Pearson correlation scores. + +
Method\DatasetSimLex-999 enWS-353 enRG-65 fr
MUSE0.380.740.72
RCSLS0.380.740.70
FASTTEXT- Common Crawl0.490.750.76
BIVEC0.400.700.74
TRANSGRAM0.390.74
SENT2VEC uni.0.72
0.460.750.71
BI-SENT2VEC uni. BI-SENT2VEC uni. + bi.0.55 0.590.78 0.790.74 0.78
+ +Table 9: Monolingual word similarity task performance of our methods when trained on en-fr ParaCrawl data. We report Pearson correlation scores. +Table 10: Monolingual word similarity task performance of our methods when trained on ende ParaCrawl data. We report Pearson correlation scores. + +
Method\DatasetSimLex-999WS-353
en deende
MUSE0.38 0.410.740.68
RCSLS0.38 0.430.740.70
FASTTEXT- Common Crawl0.49 0.390.750.64
BIVEC0.40 0.410.710.62
TRANSGRAM0.42 0.420.740.66
SENT2VEC uni.0.48 0.380.700.63
BI-SENT2VEC uni.0.56 0.470.760.68
BI-SENT2VEC uni. + bi.0.59 0.530.750.70
\ No newline at end of file diff --git a/parse/train/SJlJSaEFwS/SJlJSaEFwS_content_list.json b/parse/train/SJlJSaEFwS/SJlJSaEFwS_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..45b9e90940033f16b61f7fefd60d7a1950e2e2d6 --- /dev/null +++ b/parse/train/SJlJSaEFwS/SJlJSaEFwS_content_list.json @@ -0,0 +1,1461 @@ +[ + { + "type": "text", + "text": "ROBUST CROSS-LINGUAL EMBEDDINGS FROM PARALLEL SENTENCES ", + "text_level": 1, + "bbox": [ + 176, + 98, + 730, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Anonymous authors Paper under double-blind review ", + "bbox": [ + 176, + 171, + 392, + 198 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 236, + 544, + 251 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Recent advances in cross-lingual word embeddings have primarily relied on mapping-based methods, which project pre-trained word embeddings from different languages into a shared space through a linear transformation. However, these approaches assume word embedding spaces are isomorphic between different languages, which has been shown not to hold in practice (Søgaard et al., 2018), and fundamentally limits their performance. This motivates investigating joint learning methods which can overcome this impediment, by simultaneously learning embeddings across languages via a cross-lingual term in the training objective. Given the abundance of parallel data available (Tiedemann, 2012), we propose a bilingual extension of the CBOW method which leverages sentencealigned corpora to obtain robust cross-lingual word and sentence representations. Our approach significantly improves cross-lingual sentence retrieval performance over all other approaches, as well as convincingly outscores mapping methods while maintaining parity with jointly trained methods on word-translation. It also achieves parity with a deep RNN method on a zero-shot cross-lingual document classification task, requiring far fewer computational resources for training and inference. As an additional advantage, our bilingual method also improves the quality of monolingual word vectors despite training on much smaller datasets. We make our code and models publicly available. ", + "bbox": [ + 233, + 262, + 764, + 526 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 544, + 336, + 559 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Cross-lingual representations—such as embeddings of words and phrases into a single comparable feature space—have become a key technique in multilingual natural language processing. They offer strong promise towards the goal of a joint understanding of concepts across languages, as well as for enabling the transfer of knowledge and machine learning models between different languages. Therefore, cross-lingual embeddings can serve a variety of downstream tasks such as bilingual lexicon induction, cross-lingual information retrieval, machine translation and many applications of zero-shot transfer learning, which is particularly impactful from resource-rich to low-resource languages. ", + "bbox": [ + 174, + 568, + 825, + 680 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Existing methods can be broadly classified into two groups (Ruder et al., 2017): mapping methods leverage existing monolingual embeddings which are treated as independent, and apply a postprocess step to map the embeddings of each language into a shared space, through a linear transformation (Mikolov et al., 2013b; Conneau et al., 2017; Joulin et al., 2018). On the other hand, joint methods learn representations concurrently for multiple languages, by combining monolingual and cross-lingual training tasks (Luong et al., 2015; Coulmance et al., 2015; Gouws et al., 2015; Vulic & Moens, 2015; Chandar et al., 2014; Hermann & Blunsom, 2013). ", + "bbox": [ + 174, + 686, + 823, + 784 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "While recent work on word embeddings has focused almost exclusively on mapping methods, which require little to no cross-lingual supervision, (Søgaard et al., 2018) establish that their performance is hindered by linguistic and domain divergences in general, and for distant language pairs in particular. Principally, their analysis shows that cross-lingual hubness, where a few words (hubs) in the source language are nearest cross-lingual neighbours of many words in the target language, and structural non-isometry between embeddings do impose a fundamental barrier to the performance of linear mapping methods. ", + "bbox": [ + 174, + 791, + 823, + 888 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "(Ormazabal et al., 2019) propose using joint learning as a means of mitigating these issues. Given parallel data, such as sentences, a joint model learns to predict either the word or context in both source and target languages. As we will demonstrate with results from our algorithm, joint methods yield compatible embeddings which are closer to isomorphic, less sensitive to hubness, and perform better on cross-lingual benchmarks. ", + "bbox": [ + 176, + 895, + 821, + 922 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 103, + 821, + 145 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Contributions. We propose the BI-SENT2VEC algorithm, which extends the SENT2VEC algorithm (Pagliardini et al., 2018; Gupta et al., 2019) to the cross-lingual setting. We also revisit TRANSGRAM Coulmance et al. (2015), another joint learning method, to assess the effectiveness of joint learning over mapping-based methods. Our contributions are ", + "bbox": [ + 176, + 152, + 823, + 208 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• On cross-lingual sentence-retrieval and monolingual word representation quality evaluations, BI-SENT2VEC significantly outperforms competing methods, both jointly trained as well as mapping-based ones while preserving state-of-the-art performance on cross-lingual word retrieval tasks. For dis-similar language pairs, BI-SENT2VEC outperform their competitors by an even larger margin on all the tasks hinting towards the robustness of our method. BI-SENT2VEC performs on par with a multilingual RNN based sentence encoder, LASER (Artetxe & Schwenk, 2018), on MLDoc (Schwenk & Li, 2018), a zero-shot crosslingual transfer task on documents in multiple languages. Compared to LASER, our method improves computational efficiency by an order of magnitude for both training and inference, making it suitable for resource or latency-constrained on-device cross-lingual NLP applications. \nWe verify that joint learning methods consistently dominate state-of-the-art mapping methods on standard benchmarks, i.e., cross-lingual word and sentence retrieval. \n• Training on parallel data additionally enriches monolingual representation quality, evident by the superior performance of BI-SENT2VEC over FASTTEXT embeddings trained on a $1 0 0 \\times$ larger corpus. ", + "bbox": [ + 215, + 215, + 825, + 468 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "We make our models and code publicly available. ", + "bbox": [ + 176, + 477, + 498, + 492 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 RELATED WORK ", + "text_level": 1, + "bbox": [ + 176, + 507, + 344, + 523 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The literature on cross-lingual representation learning is extensive. Most recent advances in the field pursue unsupervised (Artetxe et al., 2017; Conneau et al., 2017; Chen & Cardie, 2018; Hoshen & Wolf, 2018; Grave et al., 2018b) or supervised (Joulin et al., 2018; Conneau et al., 2017) mapping or alignment-based algorithms. All these methods use existing monolingual word embeddings, followed by a cross-lingual alignment procedure as a post-processing step— that is to learn a simple (typically linear) mapping from the source language embedding space to the target language embedding space. ", + "bbox": [ + 174, + 535, + 825, + 632 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Supervised learning of a linear map from a source embedding space to another target embedding space (Mikolov et al., 2013b) based on a bilingual dictionary was one of the first approaches towards cross-lingual word embeddings. Additionally enforcing orthogonality constraints on the linear map results in rotations, and can be formulated as an orthogonal Procrustes problem (Smith et al., 2017). However, the authors found the translated embeddings to suffer from hubness, which they mitigate by introducing the inverted softmax as a corrective search metric at inference time. (Artetxe et al., 2017) align embedding spaces starting from a parallel seed lexicon such as digits and iteratively build a larger bilingual dictionary during training. ", + "bbox": [ + 174, + 638, + 825, + 751 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In their seminal work, (Conneau et al., 2017) propose an adversarial training method to learn a linear orthogonal map, avoiding bilingual supervision altogether. They further refine the learnt mapping by applying the Procrustes procedure iteratively with a synthetic dictionary generated through adversarial training. They also introduce the ‘Cross-Domain Similarity Local Scaling’ (CSLS) retrieval criterion for translating between spaces, which further improves on the word translation accuracy over nearest-neighbour and inverted softmax metrics. They refer to their work as Multilingual Unsupervised and Supervised Embeddings (MUSE). In this paper, we will use MUSE to denote the unsupervised embeddings introduced by them, and “Procrustes $^ +$ refine” to denote the supervised embeddings obtained by them. (Chen & Cardie, 2018) similarly use “multilingual adversarial training” followed by “pseudo-supervised refinement” to obtain unsupervised multilingual word embeddings (UMWE), as opposed to bilingual word embeddings by (Conneau et al., 2017). Hoshen & Wolf (2018) describe an unsupervised approach where they align the second moment of the two word embedding distributions followed by a further refinement. Building on the success of CSLS in reducing retrieval sensitivity to hubness, (Joulin et al., 2018) directly optimize a convex relaxation of the CSLS function (RCSLS) to align existing mono-lingual embeddings using a bilingual dictionary. ", + "bbox": [ + 174, + 757, + 825, + 924 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 103, + 823, + 146 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "While none of the methods described above require parallel corpora, all assume structural isomorphism between existing embeddings for each language (Mikolov et al., 2013b), i.e. there exists a simple (typically linear) mapping function which aligns all existing embeddings. However, this is not always a realistic assumption (Søgaard et al., 2018)—even in small toy-examples it is clear that many geometric configurations of points can not be linearly mapped to their targets. ", + "bbox": [ + 174, + 152, + 825, + 223 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Joint learning algorithms such as TRANSGRAM (Coulmance et al., 2015) and Cr5 (Josifoski et al., 2019) , circumvent this restriction by simultaneously learning embeddings as well as their alignment. TRANSGRAM, for example, extends the Skipgram (Mikolov et al., 2013a) method to jointly train bilingual embeddings in the same space, on a corpus composed of parallel sentences. In addition to the monolingual Skipgram loss for both languages, they introduce a similar cross-lingual loss where a word from a sentence in one language is trained to predict the word-contents of the sentence in the other. Cr5, on the other hand, uses document-aligned corpora to achieve state-of-the-art results for cross-lingual document retrieval while staying competitive at cross-lingual sentence and word retrieval. TRANSGRAM embeddings have been absent from discussion in most of the recent work. However, the growing abundance of sentence-aligned parallel data (Tiedemann, 2012) merits a reappraisal of their performance. ", + "bbox": [ + 174, + 229, + 825, + 382 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "(Ormazabal et al., 2019) use BIVEC (Luong et al., 2015), another bilingual extension of Skipgram, which uses a bilingual dictionary in addition to parallel sentences to obtain word-alignments and compare it with the unsupervised version of VECMAP (Artetxe et al., 2018b), another mappingbased method. Our experiments show this extra level of supervision in the case of BIVEC is redundant in obtaining state-of-the-art performance. ", + "bbox": [ + 174, + 388, + 825, + 459 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 MODEL ", + "text_level": 1, + "bbox": [ + 174, + 473, + 267, + 489 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Proposed Model. Our BI-SENT2VEC model is a cross-lingual extension of SENT2VEC proposed by (Pagliardini et al., 2018), which in turn is an extension of the $C$ -BOW embedding method (Mikolov et al., 2013a). SENT2VEC is trained on sentence contexts, with the word and higher-order word n-gram embeddings specifically optimized toward obtaining robust sentence embeddings using additive composition. Formally, SENT2VEC obtains representation ${ \\pmb v } _ { s }$ of a sentence $S$ by averaging the word-ngram embeddings (including unigrams) as $\\begin{array} { r } { \\mathbf { \\dot { \\boldsymbol { v } } } _ { s } : = \\frac { 1 } { R ( S ) } \\sum _ { w \\in R ( S ) } \\pmb { v } _ { w } } \\end{array}$ where $R ( S )$ is the set of word n-grams in the sentence $S$ . ", + "bbox": [ + 173, + 497, + 825, + 599 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The SENT2VEC training objective aims to predict a masked word token $w _ { t }$ in the sentence $S$ using the rest of the sentence representation ${ \\pmb v } _ { S \\backslash \\{ { u v } _ { t } \\} }$ . To formulate the training objective, we use logistic loss $\\ell : x \\mapsto \\log { ( 1 + e ^ { - x } ) }$ in conjunction with negative sampling. More precisely, for a raw text corpus $C$ , the monolingual training objective for SENT2VEC is given by ", + "bbox": [ + 174, + 606, + 825, + 664 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/d8aa3b9ad8140d3cea01015e60fceeebdd5dd9e761c326f1689084ce0bd64d63.jpg", + "text": "$$\n\\operatorname* { m i n } _ { U , V } \\sum _ { S \\in C } \\sum _ { w _ { t } \\in S } \\left( \\ell \\big ( \\boldsymbol { u } _ { w _ { t } } ^ { \\top } \\boldsymbol { v } _ { S \\setminus \\{ w _ { t } \\} } \\big ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell \\big ( - \\boldsymbol { u } _ { w ^ { \\prime } } ^ { \\top } \\boldsymbol { v } _ { S \\setminus \\{ w _ { t } \\} } \\big ) \\right)\n$$", + "text_format": "latex", + "bbox": [ + 297, + 667, + 700, + 708 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $w _ { t }$ is the target word and, $V$ and $U$ are the source n-gram and target word embedding matrices respectively. Here, the set of negative words $N _ { w _ { t } }$ is sampled from a multinomial distribution where the probability of picking a word is directly proportional to the square root of its frequency in the corpus. Each target word $w _ { t }$ is sampled with probability $m i n \\{ 1 , \\sqrt { t / f _ { w _ { t } } } + t / f _ { w _ { t } } \\}$ where $f _ { w _ { t } }$ is the frequency of the word in the corpus. ", + "bbox": [ + 173, + 712, + 825, + 785 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We adapt the SENT2VEC model to bilingual corpora by introducing a cross-lingual loss in addition to the monolingual loss in equation (1). Given a sentence pair $S = ( S _ { l _ { 1 } } , S _ { l _ { 2 } } )$ where $S _ { l _ { 1 } }$ and $S _ { l _ { 2 } }$ are translations of each other in languages $l _ { 1 }$ and $l _ { 2 }$ , the cross-lingual loss for a target word $w _ { t }$ in $l _ { 1 }$ is given by ", + "bbox": [ + 173, + 790, + 825, + 845 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/c243177fd7c6b452de3f5e756d7406bfd6c15aa46bf4c87fa832e1c1d9acf376.jpg", + "text": "$$\n\\ell \\big ( { \\pmb u } _ { w _ { t } } ^ { \\top } { \\pmb v } _ { S _ { l _ { 2 } } } \\big ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell \\big ( - { \\pmb u } _ { w _ { t } ^ { \\prime } } ^ { \\top } { \\pmb v } _ { S _ { l _ { 2 } } } \\big )\n$$", + "text_format": "latex", + "bbox": [ + 383, + 844, + 614, + 881 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Thus, we use the sentence $S _ { l _ { 1 } }$ to predict the constituent words of $S _ { l _ { 2 } }$ and vice-versa in a similar fashion to the monolingual SENT2VEC, shown in Figure 1. This ensures that the word and $\\mathbf { n }$ -gram embeddings of both languages lie in the same space. ", + "bbox": [ + 174, + 882, + 823, + 924 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/447f2c4e9599c41e36d0eb4b008a68d7f6bba88815213ca6273055255c6f9dd4.jpg", + "image_caption": [ + "Figure 1: An illustration of the BI-SENT2VEC training process. A word from a sentence pair is chosen as a target and the algorithm learns to predict it using the rest of the sentence(monolingual training component) and the translation of the sentence(cross-lingual component). " + ], + "image_footnote": [], + "bbox": [ + 179, + 103, + 820, + 186 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Assuming $C$ to be a sentence aligned bilingual corpus and combining equations (1) and (2), our BI-SENT2VEC model objective function is formulated as ", + "bbox": [ + 173, + 268, + 821, + 297 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/5bb3669b20c8bc86b17d6b82f9a667ac7f8d629adf9e63c00475c2b0e2226775.jpg", + "text": "$$\n\\operatorname* { m i n } _ { U , V } \\sum _ { \\underbrace { l , l ^ { \\prime } \\in [ l _ { 1 } , l _ { 2 } ] } _ { l \\neq l ^ { \\prime } } } \\sum _ { w _ { t } \\in S _ { l } \\atop l \\neq l ^ { \\prime } } \\left( \\underbrace { \\ell ( u _ { w _ { t } } ^ { \\top } v _ { S _ { l } \\backslash \\{ w _ { t } \\} } ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell ( - u _ { w _ { t } ^ { \\prime } } v _ { S _ { l } \\backslash \\{ w _ { t } \\} } ) } _ { \\mathrm { m o n o i n g u a l ~ l o s s } } + \\underbrace { \\ell ( u _ { w _ { t } } ^ { \\top } v _ { S _ { l ^ { \\prime } } } ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell ( - u _ { w _ { t } ^ { \\prime } } v _ { S _ { l ^ { \\prime } } } ) } _ { \\mathrm { c r o s s i n g u a l ~ l o s s } } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 174, + 304, + 816, + 361 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Implementation Details. We build our $\\mathrm { C } { + } { + }$ implementation on the top of the FASTTEXT library (Bojanowski et al., 2016; Joulin et al., 2016). Model parameters are updated by asynchronous SGD with a linearly decaying learning rate. ", + "bbox": [ + 174, + 377, + 823, + 420 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Our model is trained on the ParaCrawl (Espla-Gomis, 2019) v4.0 datasets for the English-Italian, \\` English-German, English-French, English-Spanish, English-Hungarian and English-Finnish language pairs. For the English-Russian language pair, we concatenate the OpenSubtitle corpus1(Lison & Tiedemann, 2016) and the Tanzil project2 (Quran translations) corpus. The number of parallel sentence pairs in the corpora except for those of English-Finnish and English-Hungarian used by us range from 17-32 Million. Number of parallel sentence pairs for the dis-similar language pairs(English-Hungarian and English-Finnish) is approximately 2 million. Evaluation results for these two language pairs can be found in Subsection 4.4. Exact statistics regarding the different corpora can be found in the Table 7 in the Appendix. All the sentences were tokenized using Spacy tokenizers3 for their respective languages. ", + "bbox": [ + 173, + 426, + 825, + 566 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "For each dataset, we trained two different models: one with unigram embeddings only, and the other additionally augmented with bigrams. The earlier TRANSGRAM models (Coulmance et al., 2015) were trained on a small amount of data (Europarl Corpus (Koehn, 2005)). To facilitate a fair comparison, we train new TRANSGRAM embeddings on the same data used for BI-SENT2VEC. Given that TRANSGRAM and BI-SENT2VEC are a cross-lingual extension of Skipgram and SENT2VEC respectively, we use the same parameters as (Bojanowski et al., 2016) and (Gupta et al., 2019), except increasing the number of epochs for TRANSGRAM to 8, and decreasing the same for BI-SENT2VEC to 5. Additionally, a preliminary hyperparameter search (except changing the number of epochs) on BI-SENT2VEC and TRANSGRAM did not improve the results. All parameters for training the TRANSGRAM and BI-SENT2VEC models can be found in the Table 6 in the Appendix. ", + "bbox": [ + 174, + 571, + 825, + 712 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In order to make the comparison more extensive, we also train VECMAP (mapping-based) (Artetxe et al., 2018b;a) and BIVEC (joint-training)(Luong et al., 2015) methods on the same corpora using the exact pipeline as (Ormazabal et al., 2019). ", + "bbox": [ + 174, + 719, + 825, + 761 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4 EVALUATION ", + "text_level": 1, + "bbox": [ + 176, + 776, + 315, + 791 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "To assess the quality of the word and sentence embeddings obtained as well as their cross-lingual alignment quality, we compare our results using the following four benchmarks ", + "bbox": [ + 173, + 803, + 823, + 832 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "• Cross-lingual word retrieval • Monolingual word representation quality ", + "bbox": [ + 215, + 844, + 501, + 873 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "• Cross-lingual sentence retrieval • Zero-shot cross-lingual transfer of document classifiers ", + "bbox": [ + 217, + 103, + 594, + 132 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where benchmarks are presented in order of increasing linguistic granularity, i.e. word, sentence, and document level. We also analyze the effect of training data by studying the relationship between representation quality and corpus size. ", + "bbox": [ + 174, + 147, + 825, + 189 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We use the code available in the MUSE library4 (Conneau et al., 2017) for all evaluations except the zero-shot classifier transfer, which is tested on the MLDoc task (Schwenk & Li, 2018)5. ", + "bbox": [ + 176, + 195, + 823, + 224 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.1 WORD TRANSLATION ", + "text_level": 1, + "bbox": [ + 176, + 238, + 366, + 252 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The task involves retrieving correct translation(s) of a word in a source language from a target language. To evaluate translation accuracy, we use the bilingual dictionaries constructed by (Conneau et al., 2017). We consider 1500 source-test queries and $2 0 0 \\mathrm { k }$ target words for each language pair and report $\\mathrm { P @ 1 }$ scores for the supervised and unsupervised baselines as well as our models in Table 1. ", + "bbox": [ + 174, + 262, + 825, + 319 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/3dcea9e7aa1406f38ea34781b8850b294bca917fba2f98685efd6ef2b933ea6d.jpg", + "table_caption": [ + "Table 1: Word translation retrieval $\\mathbf { P } @ \\mathbf { 1 }$ for various language pairs of MUSE evaluation dictionary (Conneau et al., 2017). NN: nearest neighbours. CSLS: Cross-Domain Similarity Local Scaling. (‘en’ is English, ‘fr’ is French, ‘de’ is German, ‘ru’ is Russian, ‘it’ is Italian) (‘uni.’ and ‘bi.’ denote unigrams and bigrams respectively) denotes translation from the first language to the second and $\\gets$ the other way around.) " + ], + "table_footnote": [], + "table_body": "
Methoden-esen-fren-deen-ruen-itavg.
→↑→↑→↑→↑→↑
MUSE (Conneau et al.,2017)81.7 83.382.3 82.174.0 72.244.0 59.178.6 77.973.5
UMWE(Chen & Cardie,2018)82.5 83.182.5 82.174.6 72.549.5 61.778.3 77.074.4
Procrustes + refine (Conneau et al., 2017)82.4 83.982.3 83.275.3 73.250.1 63.577.5 77.674.9
RCSLS (Joulin et al., 2018)83.7 87.184.1 84.779.2 77.560.9 70.281.1 82.779.1
TRANSGRAM (Coulmance et al., 2015)91.6 88.689.1 90.187.5 87.265.6 73.788.6 89.585.2
VECMAP (unsupervised) (Artetxe et al.,2018b)87.4 87.888.3 88.584.3 87.248.6 50.587.4 86.579.6
VECMAP (supervised) (Artetxe et al.,2018a)87.2 90.287.6 90.487.3 86.849.7 65.687.2 89.282.1
BIVEC NN (Luong et al., 2015)87.4 88.686.8 89.187.5 87.264.0 59.186.8 84.081.7
BIVEC CSLS (Luong et al., 2015)87.6 89.188.8 90.386.4 87.266.1 70.687.6 87.884.3
BI-SENT2VEC uni. NN86.9 91.686.9 91.086.0 88.758.0 72.888.3 92.484.3
BI-SENT2VEC uni. + bi. NN89.4 92.989.3 92.886.7 89.359.0 70.289.5 91.885.1
BI-SENT2VEC uni.CSLS86.0 91.786.4 91.484.6 88.860.5 73.088.2 91.884.2
BI-SENT2VEC uni. + bi. CSLS89.0 92.188.9 92.486.5 89.061.0 73.589.6 91.485.3
", + "bbox": [ + 173, + 333, + 823, + 550 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.2 MONOLINGUAL WORD REPRESENTATION QUALITY ", + "text_level": 1, + "bbox": [ + 174, + 657, + 571, + 671 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We assess the monolingual quality improvement of our proposed cross-lingual training by evaluating performance on monolingual word similarity tasks. To disentangle the specific contribution of the cross-lingual loss, we train the monolingual counterpart of BI-SENT2VEC, SENT2VEC on the same corpora as our method. ", + "bbox": [ + 174, + 681, + 825, + 736 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Performance on monolingual word-similarity tasks is evaluated using the English SimLex-999 (Hill et al., 2014) and its Italian and German translations, English WS-353 (Finkelstein et al., 2001) and its German, Italian and Spanish translations. For French, we use a translation of the RG-65 (Joubarne & Inkpen, 2011) dataset. Pearson scores are used to measure the correlation between human-annotated word similarities and predicted cosine similarities. We also include FASTTEXT monolingual vectors trained on CommonCrawl data (Grave et al., 2018a) which is comprised of 600 billion, 68 billion, 66 billion, 72 billion and 36 billion words of English, French, German, Spanish and Italian respectively and is at least $1 0 0 \\times$ larger than the corpora on which we trained BI-SENT2VEC. We report Pearson correlation scores on different word-similarity datasets for En-It pair in Table 2. Evaluation results on other language pairs are similar and can be found in the appendix in Tables 8, 9, and 10. ", + "bbox": [ + 174, + 743, + 825, + 883 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/283fd29dd3c6466d32f63f2aa830962d367d5512f1bdda7a83e27aa8a345bd28.jpg", + "table_caption": [ + "Table 2: Monolingual word similarity task performance of our methods when trained on en-it ParaCrawl data. We report Pearson correlation scores. " + ], + "table_footnote": [], + "table_body": "
Method\\DatasetSimLex-999WS-353
en iten it
MUSE0.380.300.74 0.64
RCSLS0.38 0.300.740.64
FASTTEXT- Common Crawl0.49 0.320.750.57
BIVEC0.40 0.360.700.60
TRANSGRAM0.43 0.370.730.63
SENT2VEC uni.0.49 0.380.730.60
BI-SENT2VEC uni.0.570.47 0.790.65
BI-SENT2VEC uni. + bi.0.580.50 0.800.69
", + "bbox": [ + 294, + 95, + 704, + 275 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.3 CROSS-LINGUAL SENTENCE RETRIEVAL ", + "text_level": 1, + "bbox": [ + 174, + 334, + 495, + 348 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The primary contribution of our work is to deliver improved cross-lingual sentence representations. We test sentence embeddings for each method obtained by bag-of-words composition for sentence retrieval across different languages on the Europarl corpus. In particular, the tf-idf weighted average is used to construct sentence embeddings from word embeddings. We consider 2000 sentences in the source language dataset and retrieve their translation among 200K sentences in the target language dataset. The other 300K sentences in the Europarl corpus are used to calculate tf-idf weights. Results for $\\mathrm { P @ 1 }$ of unsupervised and supervised benchmarks vs our models are included in Table 3. ", + "bbox": [ + 173, + 356, + 825, + 455 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/a66fd06c0021769a89fe836ddbc7a35cea2224df45f811fb2a378d697d07753b.jpg", + "table_caption": [], + "table_footnote": [ + "Table 3: Cross-lingual Sentence retrieval. We report $\\mathrm { P @ 1 }$ scores for 2000 source queries searching over 200 000 target sentences. Reduction in error is calculated with respect to BI-SENT2VEC uni. $^ +$ bi. CSLS and the best non-BI-SENT2VEC method. " + ], + "table_body": "
Methoden-esen-fren-deen-itavg.
→↑→↑→↑→↑
MUSE72.771.569.268.853.353.466.164.364.9
RCSLS26.926.719.321.28.811.315.117.618.4
TRANSGRAM83.581.480.481.664.869.977.277.977.1
VECMAP (unsupervised)81.782.179.880.462.864.669.071.174.0
VECMAP (supervised)81.38180.480.762.664.367.87173.6
BIVEC NN69.877.154.775.556.144.158.245.160.1
BIVEC CSLS81.683.478.181.671.668.174.272.476.4
BI-SENT2VEC uni. NN87.886.485.283.482.380.285.985.884.6
BI-SENT2VEC uni. + bi. NN87.987.886.183.979.579.785.185.384.4
BI-SENT2VEC uni. CSLS89.588.587.186.484.483.088.287.586.8
BI-SENT2VEC uni. + bi. CSLS89.789.687.887.484.284.087.987.687.3
Reduction in error37.5% 37.3%37.8% 31.5%44.4% 46.8%46.9% 43.9%1
", + "bbox": [ + 173, + 467, + 821, + 702 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.4 PERFORMANCE ON DIS-SIMILAR LANGUAGE PAIRS ", + "text_level": 1, + "bbox": [ + 174, + 775, + 565, + 789 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We report a substantial improvement on the performance of previous models on cross-lingual word and sentence retrieval tasks for the dis-similar language pairs(English-Finnish and EnglishHungarian). We use the same evaluation scheme as in Subsections 4.1 and 4.3 Results for these pairs are included in Table 4. ", + "bbox": [ + 174, + 797, + 825, + 853 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.5 ZERO-SHOT CROSS-LINGUAL TRANSFER OF DOCUMENT CLASSIFIERS ", + "text_level": 1, + "bbox": [ + 176, + 858, + 702, + 873 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The MLDoc multilingual document classification task (Schwenk & Li, 2018) consists of news documents given in 8 different languages, which need to be classified into 4 different categories. To demonstrate the ability to transfer trained classifiers in a robust fashion between languages, we use a zero-shot setting, i.e., we train a classifier on embeddings in the source language, and report the accuracy of the same classifier applied to the target language. As the classifier, we use a simple feed-forward neural network with two hidden layers of size 10 and 8 respectively, optimized using the Adam optimizer. Each document is represented using the sum of its sentence embeddings. ", + "bbox": [ + 176, + 881, + 825, + 924 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/df08d014e7400a856ca7b23caeb7e4d2e39a6c924d2082bfd199f5231b4e5868.jpg", + "table_caption": [ + "Table 4: Cross-lingual Word and Sentence retrieval for dis-similar language pairs $( \\mathbf { P } @ \\mathbf { 1 }$ scores). ‘en’ is English, ‘fi’ is Finnish, ‘hu’ is Hungarian " + ], + "table_footnote": [], + "table_body": "
Method word retrieval sentence retrieval
en-fi →↑en-hu → ↑→en-fi ↑en-hu →↑
MUSE48.159.553.9 64.921.729.539.146.7
RCSLS61.869.967.0 73.03.24.83.65.1
VECMAP (unsupervised)62.566.861.6 68.713.214.720.519.3
VECMAP (supervised)62.678.363.7 76.615.016.920.921.7
BIVEC NN62.155.362.1 53.714.29.726.213.7
BIVEC CSLS69.678.072.4 78.433.332.046.741.3
TRANSGRAM69.781.173.1 80.835.440.552.155
B1-SENT2VEC uni. NN71.285.475.683.963.5 64.275.2 76.2
BI-SENT2VEC uni. + bi. NN68.581.771.4 79.457.5 55.965.865.2
B1-SENT2VEC uni. CSLS72.0 86.576.3 85.170.2 69.081.480.8
BI-SENT2VEC uni. + bi. CSLS70.184.473.781.766 64.173.874.5
", + "bbox": [ + 173, + 101, + 823, + 377 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 440, + 825, + 497 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/938340f7ee36091f27e2f02534b875eeec8920f5d4897f23d52629e79e5776f2.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
Methoden-esen-fren-deavg.
→↑→冏↑→↑→en-it ↑
LASER79.3 69.678.0 80.186.3 80.870.2 74.277.3
BI-SENT2VEC74.0 71.581.6 82.286.5 79.275.0 72.677.8
", + "bbox": [ + 261, + 507, + 736, + 580 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Table 5: MLDoc Benchmark results (Schwenk & Li, 2018). A document classifier was trained on one language and tested on another without additional training/fine-tuning. We report $\\%$ accuracy. ", + "bbox": [ + 174, + 593, + 823, + 622 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We compare the performance of BI-SENT2VEC with the LASER sentence embeddings (Artetxe & Schwenk, 2018) in Table 5. LASER sentence embedding model is a multi-lingual sentence embedding model which is composed of a biLSTM encoder and an LSTM decoder. It uses a shared byte pair encoding based vocabulary of 50k words. The LASER model which we compare to was trained on 223M sentences for 93 languages and requires 5 days to train on 16 V100 GPUs compared to our model which takes 1-2.5 hours for each language pair on 30 CPU threads. ", + "bbox": [ + 174, + 636, + 825, + 719 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.6 EFFECT OF CORPUS SIZE ON REPRESENTATION QUALITY ", + "text_level": 1, + "bbox": [ + 174, + 729, + 611, + 744 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We conduct an ablation study on how BI-SENT2VEC embeddings’ performance depends on the size of the training corpus. We uniformly sample smaller subsets of the En-Fr ParaCrawl dataset and train a BI-SENT2VEC model on them. We test word/sentence translation performance with the CSLS retrieval criterion, and monolingual embedding quality for En-Fr with increasing ParaCrawl corpus size. The results are illustrated in Figures 2 and 3. ", + "bbox": [ + 174, + 751, + 825, + 821 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5 DISCUSSION ", + "text_level": 1, + "bbox": [ + 174, + 828, + 312, + 844 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In the following section, we discuss the results on monolingual and cross-lingual benchmarks, presented in Tables 1 - 5, and a data ablation study for how the model behaves with increasing parallel corpus size in Figure $2 \\cdot 3$ . The most impressive outcome of our experiments is improved crosslingual sentence retrieval performance, which we elaborate on along with word translation in the next subsection. ", + "bbox": [ + 174, + 853, + 823, + 924 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/bb3e3da4961fcc13096a65b06ddba37fd74cb2a40ffb7fe4be5b7a1bd56621c1.jpg", + "image_caption": [ + "Figure 2: Effect of corpus size on cross-lingual word/sentence retrieval performance. " + ], + "image_footnote": [], + "bbox": [ + 181, + 109, + 794, + 266 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/1145c4987fe2df1428c48252033511b35e1e13156a25dc7e8955255c49fc87e7.jpg", + "image_caption": [ + "Figure 3: Effect of corpus size on monolingual word quality. We use SimLex-999, WS-353, and FR-RG datasets for measuring monolingual word embedding quality. " + ], + "image_footnote": [], + "bbox": [ + 349, + 314, + 629, + 470 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Cross-lingual evaluations For cross-lingual tasks, we observe in Table 1 that jointly trained embeddings produce much better results on cross-lingual word and sentence retrieval tasks. BISENT2VEC’s performance on word-retrieval tasks is uniformly superior to mapping methods, achieving up to $1 1 . 5 \\%$ more in $\\mathrm { P @ 1 }$ than RCSLS for the English to German language pair, consistent with the results from (Ormazabal et al., 2019). It is also on-par with, or better than competing joint methods except on translation from Russian to English, where TRANSGRAM receives a significantly better score. For word retrieval tasks, there is no discernible difference between CSLS/NN criteria for BI-SENT2VEC, suggesting the relative absence of the hubness phenomenon which significantly hinders the performance of cross-lingual word embedding methods. ", + "bbox": [ + 174, + 540, + 825, + 666 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Our principal contribution is in improving cross-lingual sentence retrieval. Table 3 shows BISENT2VEC decisively outperforms all other methods by a wide margin, reducing the relative $\\mathrm { P @ 1 }$ error anywhere from $3 1 . 5 \\%$ to $5 5 . 1 \\%$ . Our model displays considerably less variance than others in quality across language pairs, with at most a $\\approx 5 \\%$ deficit between best and worst, and nearly symmetric accuracy within a language pair. ", + "bbox": [ + 174, + 672, + 823, + 742 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "TRANSGRAM also outperforms the mapping-based methods, but still falls significantly short of BISENT2VEC’s. These results can be attributed to the fact that BI-SENT2VEC directly optimizes for obtaining robust sentence embeddings using additive composition of its word embeddings. Since BI-SENT2VEC’s learning objective is closest to a sentence retrieval task amongst current state-ofthe-art methods, it can surpass them without sacrificing performance on other tasks. ", + "bbox": [ + 174, + 750, + 825, + 819 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Cross-lingual evaluations on dis-similar language pairs Unlike other language pairs in the evaluation, English-Finnish and English-Hungarian pairs are composed of languages from two different language families(English being an Indo-European language and the other language being a Finno-Ugric language). In Table 4, we see that the performance boost achieved by BI-SENT2VEC on competing methods methods is more pronounced in the case of dis-similar language pairs as compared to paris of languages close to each other. This observation affirms the suitaibility of BISENT2VEC for learning joint representations on languages from different families. ", + "bbox": [ + 174, + 827, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Monolingual word quality For the monolingual word similarity tasks, we observe large gains over existing methods. SENT2VEC is trained on the same corpora as us, and FASTTEXT vectors are trained on the CommonCrawl corpora which are more than 100 times larger than ParaCrawl v4.0. In Table 2, we see that BI-SENT2VEC outperforms them by a significant margin on SimLex-999 and WS-353, two important monolingual word quality benchmarks. This observation is in accordance with the fact (Faruqui & Dyer, 2014) that bilingual contexts can be surprisingly effective for learning monolingual word representations. However, amongst the joint-training methods, BI-SENT2VEC also outperforms TRANSGRAM and BIVEC trained on the same corpora by a significant margin, again hinting at the superiority of the sentence level loss function over a fixed context window loss. ", + "bbox": [ + 174, + 103, + 825, + 229 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Effect of n-grams (Gupta et al., 2019) report improved results on monolingual word representation evaluation tasks for SENT2VEC and FASTTEXT word vectors by training them alongside word n-grams. Our method incorporates their results based on the observation that unigram vectors trained alongside with bigrams significantly outperform unigrams alone on the majority of the evaluation tasks. We can see from Tables 1 - 3 that this holds for the bilingual case as well. However, in case of dis-similar language pairs(Table 4), we observe that using n-grams degrades the cross-lingual performance of the embeddings. This observation suggests that use of higher order n-grams may not be helpful for language pairs where the grammatical structures are contrasting. ", + "bbox": [ + 174, + 236, + 825, + 347 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Effect of corpus size Considering the cross-lingual performance curve exhibited by BISENT2VEC in Figure 2, increasing corpus size for the English-French datasets up to 1-3.1M lines appears to saturate the performance of the model on cross-lingual word/sentence retrieval, after which it either plateaus or degrades slightly. This is an encouraging result, indicating that joint methods can use significantly less data to obtain promising performance. This implies that joint methods may not necessarily be constrained to high-resource language pairs as previously assumed, though further experimentation is needed to verify this claim. ", + "bbox": [ + 174, + 354, + 825, + 452 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "It should be noted from Figure 3 that the monolingual quality does keep improving with an increase in the size of the corpus. A potential way to overcome this issue of plateauing cross-lingual performance is to give different weights to the monolingual and cross-lingual component of the loss with the weights possibly being dependent on other factors such as training progress. ", + "bbox": [ + 174, + 458, + 825, + 515 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Comparison with a cross-lingual sentence embedding model and performance on document level task On the MLDoc classifier transfer task (Schwenk & Li, 2018) where we evaluate a classifier learned on documents in one language on documents in another, Table 5 shows we achieve parity with the performance of the LASER model for language pairs involving English, where BISENT2VEC’s average accuracy of $7 7 . 8 \\%$ is slightly higher than LASER’s $7 7 . 3 \\%$ . While the comparison is not completely justified as LASER is multilingual in nature and is trained on a different dataset, one must emphasize that BI-SENT2VEC is a bag-of-words method as compared to LASER which uses a multi-layered biLSTM sentence encoder. Our method only requires to average a set of vectors to encode sentences reducing its computational footprint significantly. This makes BI-SENT2VEC an ideal candidate for on-device computationally efficient cross-lingual NLP, unlike LASER which has a huge computational overhead and specialized hardware requirement for encoding sentences. ", + "bbox": [ + 173, + 522, + 825, + 674 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "6 CONCLUSION AND FUTURE WORK ", + "text_level": 1, + "bbox": [ + 174, + 709, + 495, + 726 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We introduce a cross-lingual extension of an existing monolingual word and sentence embedding method. The proposed model is tested at three levels of linguistic granularity: words, sentences and documents. The model outperforms all other methods by a wide margin on the cross-lingual sentence retrieval task while maintaining parity with the best-performing methods on word translation tasks. Our method achieves parity with LASER on zero-shot document classification, despite being a much simpler model. We also demonstrate that training on parallel data yields a significant improvement in the monolingual word representation quality. ", + "bbox": [ + 174, + 750, + 825, + 847 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "The success of our model on the bilingual level calls for its extension to the multilingual level especially for pairs which have little or no parallel corpora. While the amount of bilingual/multilingual parallel data has grown in abundance, the amount of monolingual data available is practically limitless. Consequently, we would like to explore training cross-lingual embeddings with a large amount of raw text combined with a smaller amount of parallel data. ", + "bbox": [ + 174, + 854, + 823, + 924 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 176, + 102, + 287, + 118 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Mikel Artetxe and Holger Schwenk. Massively multilingual sentence embeddings for zero-shot cross-lingual transfer and beyond. Transactions of the Association for Computational Linguistics, 7:597–610, 2018. ", + "bbox": [ + 173, + 128, + 823, + 171 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. Learning bilingual word embeddings with (almost) no bilingual data. In ACL, 2017. ", + "bbox": [ + 171, + 179, + 823, + 208 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. Generalizing and improving bilingual word embedding mappings with a multi-step framework of linear transformations. In Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, pp. 5012–5019, 2018a. ", + "bbox": [ + 173, + 217, + 821, + 261 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. A robust self-learning method for fully unsupervised cross-lingual mappings of word embeddings. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 789–798, 2018b. ", + "bbox": [ + 174, + 267, + 823, + 311 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. Enriching word vectors with subword information. Transactions of the Association for Computational Linguistics, 5:135–146, 2016. ", + "bbox": [ + 173, + 319, + 826, + 362 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "A. P. Sarath Chandar, Stanislas Lauly, Hugo Larochelle, Mitesh M. Khapra, Balaraman Ravindran, Vikas C. Raykar, and Amrita Saha. An autoencoder approach to learning bilingual word representations. In NIPS, 2014. ", + "bbox": [ + 174, + 371, + 825, + 412 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Xilun Chen and Claire Cardie. Unsupervised multilingual word embeddings. In EMNLP, 2018. ", + "bbox": [ + 171, + 421, + 799, + 438 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Alexis Conneau, Guillaume Lample, Marc’Aurelio Ranzato, Ludovic Denoyer, and Herve J ´ egou.´ Word translation without parallel data. ArXiv, abs/1710.04087, 2017. ", + "bbox": [ + 173, + 445, + 820, + 474 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Jocelyn Coulmance, Jean-Marc Marty, Guillaume Wenzek, and Amine Benhalloum. Trans-gram, Fast Cross-lingual Word-embeddings. In EMNLP - Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pp. 1109–1113, 2015. ", + "bbox": [ + 173, + 482, + 825, + 526 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "M. Espla-Gomis. ParaCrawl: Web-scale parallel corpora for the languages of the EU. 2019. \\` ", + "bbox": [ + 171, + 534, + 776, + 550 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Manaal Faruqui and Chris Dyer. Improving vector space word representations using multilingual correlation. In EACL, 2014. ", + "bbox": [ + 171, + 558, + 820, + 587 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Lev Finkelstein, Evgeniy Gabrilovich, Yossi Matias, Ehud Rivlin, Zach Solan, Gadi Wolfman, and Eytan Ruppin. Placing search in context: the concept revisited. In WWW, 2001. ", + "bbox": [ + 174, + 594, + 821, + 625 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Stephan Gouws, Yoshua Bengio, and Greg Corrado. Bilbowa: Fast bilingual distributed representations without word alignments. 2015. ", + "bbox": [ + 173, + 632, + 820, + 662 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Armand Joulin, and Tomas Mikolov. Learning word vectors for 157 languages. ArXiv, abs/1802.06893, 2018a. ", + "bbox": [ + 171, + 670, + 823, + 699 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Edouard Grave, Armand Joulin, and Quentin Berthet. Unsupervised alignment of embeddings with wasserstein procrustes. In AISTATS, 2018b. ", + "bbox": [ + 173, + 708, + 821, + 737 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Prakhar Gupta, Matteo Pagliardini, and Martin Jaggi. Better word embeddings by disentangling contextual n-gram information. In NAACL-HLT, 2019. ", + "bbox": [ + 173, + 744, + 821, + 775 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Karl Moritz Hermann and Phil Blunsom. Multilingual distributed representations without word alignment. In ICLR 2014, 2013. ", + "bbox": [ + 173, + 782, + 821, + 811 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Felix Hill, Roi Reichart, and Anna Korhonen. Simlex-999: Evaluating semantic models with (genuine) similarity estimation. Computational Linguistics, 41:665–695, 2014. ", + "bbox": [ + 173, + 819, + 820, + 849 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Yedid Hoshen and Lior Wolf. Non-adversarial unsupervised word translation. In EMNLP, 2018. ", + "bbox": [ + 173, + 858, + 805, + 873 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Martin Josifoski, Ivan S Paskov, Hristo S Paskov, Martin Jaggi, and Robert West. Crosslingual document embedding as reduced-rank ridge regression. In Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining, pp. 744–752. ACM, 2019. ", + "bbox": [ + 174, + 882, + 823, + 924 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Colette Joubarne and Diana Inkpen. Comparison of semantic similarity for different languages using the google n-gram corpus and second-order co-occurrence measures. In Canadian Conference on AI, 2011. ", + "bbox": [ + 176, + 103, + 823, + 146 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. Bag of tricks for efficient text classification. In EACL, 2016. ", + "bbox": [ + 173, + 155, + 823, + 184 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Armand Joulin, Piotr Bojanowski, Tomas Mikolov, Herve J ´ egou, and Edouard Grave. Loss in trans-´ lation: Learning bilingual word mapping with a retrieval criterion. In EMNLP, 2018. ", + "bbox": [ + 173, + 193, + 823, + 222 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Philipp Koehn. Europarl: A parallel corpus for statistical machine translation. 2005. ", + "bbox": [ + 174, + 231, + 727, + 246 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Pierre Lison and Jorg Tiedemann. Opensubtitles2016: Extracting large parallel corpora from movie ¨ and tv subtitles. In LREC, 2016. ", + "bbox": [ + 171, + 253, + 823, + 284 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Thang Luong, Hieu Pham, and Christopher D Manning. Bilingual word representations with monolingual quality in mind. In Proceedings of the 1st Workshop on Vector Space Modeling for Natural Language Processing, pp. 151–159, 2015. ", + "bbox": [ + 176, + 291, + 823, + 335 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Tomas Mikolov, Kai Chen, Gregory S. Corrado, and Jeffrey Dean. Efficient estimation of word representations in vector space. CoRR, abs/1301.3781, 2013a. ", + "bbox": [ + 171, + 343, + 823, + 372 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Tomas Mikolov, Quoc V. Le, and Ilya Sutskever. Exploiting similarities among languages for machine translation. ArXiv, abs/1309.4168, 2013b. ", + "bbox": [ + 171, + 381, + 823, + 410 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Aitor Ormazabal, Mikel Artetxe, Gorka Labaka, Aitor Soroa, and Eneko Agirre. Analyzing the limitations of cross-lingual word embedding mappings. In ACL, 2019. ", + "bbox": [ + 173, + 419, + 823, + 449 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Matteo Pagliardini, Prakhar Gupta, and Martin Jaggi. Unsupervised learning of sentence embeddings using compositional n-gram features. In NAACL-HLT, 2018. ", + "bbox": [ + 173, + 457, + 820, + 486 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Sebastian Ruder, Ivan Vulic, and Anders Søgaard. A survey of cross-lingual word embedding mod-´ els. arXiv preprint arXiv:1706.04902, 2017. ", + "bbox": [ + 174, + 494, + 820, + 523 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Holger Schwenk and Xian Li. A corpus for multilingual document classification in eight languages. arXiv preprint arXiv:1805.09821, 2018. ", + "bbox": [ + 174, + 532, + 821, + 560 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Samuel L. Smith, David H. P. Turban, Steven Hamblin, and Nils Y. Hammerla. Offline bilingual word vectors, orthogonal transformations and the inverted softmax. ArXiv, abs/1702.03859, 2017. ", + "bbox": [ + 173, + 570, + 821, + 599 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Anders Søgaard, Sebastian Ruder, and Ivan Vulic. On the limitations of unsupervised bilingual dictionary induction. In ACL, 2018. ", + "bbox": [ + 173, + 607, + 821, + 637 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jorg Tiedemann. Parallel data, tools and interfaces in opus. In ¨ Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC-2012), pp. 2214–2218, 2012. ", + "bbox": [ + 173, + 645, + 821, + 675 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ivan Vulic and Marie-Francine Moens. Bilingual word embeddings from non-parallel documentaligned data applied to bilingual lexicon induction. In ACL, 2015. ", + "bbox": [ + 173, + 683, + 821, + 712 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "A APPENDIX ", + "text_level": 1, + "bbox": [ + 176, + 102, + 299, + 117 + ], + "page_idx": 11 + }, + { + "type": "table", + "img_path": "images/34da396cee228ad7f3a75f949b12e1420d31d2c471b6948a734534e6f4f80a95.jpg", + "table_caption": [ + "A.1 DATASET STATISTICS " + ], + "table_footnote": [], + "table_body": "
DatasetNumber of sentencesNumber of tokens(English tokens if bilingual)
En-De ParaCrawl v4.017 Million308 Million
En-Es ParaCrawl v4.022 Million477 Million
En-FiParaCrawl v4.02.16 Million42 Million
En-FrParaCrawl v4.032 Million665 Million
En-Hu ParaCrawl v4.01.91Million31Million
En-It ParaCrawl v4.013 Million261 Million
En-Ru OpenSubtitles+ Tanzil27 Million363Million
Wikipedia - En70 Million1792 Million
Wikipedia - De二1384 Million
Wikipedia - Fr二1108Million
Wikipedia - Es二797 Million
Wikipedia - It二702 Million
Wikipedia - Ru二824 Million
Common Crawl - En二600 Billion
Common Crawl - De二66 Billion
Common Crawl -Fr二68 Billion
Common Crawl - It二36 Billion
Common Crawl -Es172 Billion
", + "bbox": [ + 173, + 160, + 715, + 450 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Table 6: Dataset Sizes. ‘En’,‘De’,‘Fi’,‘Fr’,‘Hu’,‘It’,‘Es’ and ‘Ru’ stand for English, German, Finnish, French, Hungarian, Italian, Spanish and Russian respectively. ", + "bbox": [ + 173, + 459, + 821, + 488 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We used ParaCrawl $\\mathrm { v } 4 . 0$ corpora for training BI-SENT2VEC, SENT2VEC,BIVEC,VECMAP and TRANSGRAM embeddings except for En-Ru pair for which we used OpenSubtitles and Tanzil corpora combined. MUSE and RCSLS vectors were trained from FASTTEXT vectors obtained from Wikipedia dumps(Grave et al., 2018a). ", + "bbox": [ + 174, + 502, + 825, + 558 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "A.2 TRAINING PARAMETERS FOR TRAINED MODELS ", + "text_level": 1, + "bbox": [ + 176, + 104, + 545, + 118 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/5d36ad4d3a232e416a5e0819224e5e023f5c7a3f99da04945ab280cf5ab981d7.jpg", + "table_caption": [ + "Table 7: Hyperparameters for the trained models " + ], + "table_footnote": [], + "table_body": "
ModelBI-SENT2VECuni.BI-SENT2VECuni. + bi.SENT2VECuni.TRANSGRAM
Embedding dimension300300300300
Maxvocabulary size750k750k750k750k
Minimum word count5855
Initial Learning Rate0.20.20.20.025
Epochs5558
Subsampling hyper-parameter1·10-55:10-61·10-51·10-4
Word-Ngrams Bucket Size二2M二二
Word-Ngrams dropped per context141二
Window size5
Number of negatives sampled1010105
", + "bbox": [ + 173, + 133, + 803, + 309 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/6dd016b50cfe64778b72f82487b56858d38fb6a0aec44d5ed0fc98072d36ebf1.jpg", + "table_caption": [ + "A.3 ADDITIONAL MONOLINGUAL QUALITY TABLES " + ], + "table_footnote": [], + "table_body": "
SimLex-999 Method\\DatasetWS-353
enenes
MUSE RCSLS0.380.740.61
FASTTEXT- Common Crawl0.38 0.490.74 0.750.62 0.54
BIVEC0.400.720.57
TRANSGRAM0.420.740.59
SENT2VEC uni.0.490.580.51
BI-SENT2VEC uni.0.570.780.60
BI-SENT2VEC uni. + bi.0.600.820.66
", + "bbox": [ + 302, + 135, + 694, + 315 + ], + "page_idx": 13 + }, + { + "type": "table", + "img_path": "images/ab74ae0a4ec9a8376403443db7f1f5b502fc937b4df7852c62d53cae7d9000ec.jpg", + "table_caption": [ + "Table 8: Monolingual word similarity task performance of our methods when trained on en-es ParaCrawl data. We report Pearson correlation scores. " + ], + "table_footnote": [], + "table_body": "
Method\\DatasetSimLex-999 enWS-353 enRG-65 fr
MUSE0.380.740.72
RCSLS0.380.740.70
FASTTEXT- Common Crawl0.490.750.76
BIVEC0.400.700.74
TRANSGRAM0.390.74
SENT2VEC uni.0.72
0.460.750.71
BI-SENT2VEC uni. BI-SENT2VEC uni. + bi.0.55 0.590.78 0.790.74 0.78
", + "bbox": [ + 276, + 381, + 722, + 560 + ], + "page_idx": 13 + }, + { + "type": "table", + "img_path": "images/1008cbd31964a96ebb37aa7d023437a153bbcaa3186ee3ec0be8366bf85f1dcc.jpg", + "table_caption": [ + "Table 9: Monolingual word similarity task performance of our methods when trained on en-fr ParaCrawl data. We report Pearson correlation scores. ", + "Table 10: Monolingual word similarity task performance of our methods when trained on ende ParaCrawl data. We report Pearson correlation scores. " + ], + "table_footnote": [], + "table_body": "
Method\\DatasetSimLex-999WS-353
en deende
MUSE0.38 0.410.740.68
RCSLS0.38 0.430.740.70
FASTTEXT- Common Crawl0.49 0.390.750.64
BIVEC0.40 0.410.710.62
TRANSGRAM0.42 0.420.740.66
SENT2VEC uni.0.48 0.380.700.63
BI-SENT2VEC uni.0.56 0.470.760.68
BI-SENT2VEC uni. + bi.0.59 0.530.750.70
", + "bbox": [ + 294, + 627, + 705, + 806 + ], + "page_idx": 13 + } +] \ No newline at end of file diff --git a/parse/train/SJlJSaEFwS/SJlJSaEFwS_middle.json b/parse/train/SJlJSaEFwS/SJlJSaEFwS_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..30c1ec67dd882bf7c485ebaf7df29b1b9e4ccb28 --- /dev/null +++ b/parse/train/SJlJSaEFwS/SJlJSaEFwS_middle.json @@ -0,0 +1,29132 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 447, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 449, + 98 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 449, + 98 + ], + "score": 1.0, + "content": "ROBUST CROSS-LINGUAL EMBEDDINGS FROM", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 279, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 279, + 118 + ], + "score": 1.0, + "content": "PARALLEL SENTENCES", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 108, + 136, + 240, + 157 + ], + "lines": [ + { + "bbox": [ + 108, + 135, + 198, + 148 + ], + "spans": [ + { + "bbox": [ + 108, + 135, + 198, + 148 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 108, + 146, + 241, + 159 + ], + "spans": [ + { + "bbox": [ + 108, + 146, + 241, + 159 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 278, + 187, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 276, + 186, + 335, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 186, + 335, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 208, + 468, + 417 + ], + "lines": [ + { + "bbox": [ + 141, + 208, + 470, + 222 + ], + "spans": [ + { + "bbox": [ + 141, + 208, + 470, + 222 + ], + "score": 1.0, + "content": "Recent advances in cross-lingual word embeddings have primarily relied on", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 221, + 469, + 232 + ], + "spans": [ + { + "bbox": [ + 141, + 221, + 469, + 232 + ], + "score": 1.0, + "content": "mapping-based methods, which project pre-trained word embeddings from dif-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 231, + 470, + 245 + ], + "spans": [ + { + "bbox": [ + 141, + 231, + 470, + 245 + ], + "score": 1.0, + "content": "ferent languages into a shared space through a linear transformation. However,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 242, + 470, + 255 + ], + "spans": [ + { + "bbox": [ + 141, + 242, + 470, + 255 + ], + "score": 1.0, + "content": "these approaches assume word embedding spaces are isomorphic between dif-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 253, + 469, + 265 + ], + "spans": [ + { + "bbox": [ + 141, + 253, + 469, + 265 + ], + "score": 1.0, + "content": "ferent languages, which has been shown not to hold in practice (Søgaard et al.,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 264, + 469, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 264, + 469, + 277 + ], + "score": 1.0, + "content": "2018), and fundamentally limits their performance. This motivates investigating", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "score": 1.0, + "content": "joint learning methods which can overcome this impediment, by simultaneously", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 287, + 469, + 299 + ], + "spans": [ + { + "bbox": [ + 141, + 287, + 469, + 299 + ], + "score": 1.0, + "content": "learning embeddings across languages via a cross-lingual term in the training ob-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 297, + 470, + 310 + ], + "spans": [ + { + "bbox": [ + 141, + 297, + 470, + 310 + ], + "score": 1.0, + "content": "jective. Given the abundance of parallel data available (Tiedemann, 2012), we", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 140, + 307, + 470, + 322 + ], + "spans": [ + { + "bbox": [ + 140, + 307, + 470, + 322 + ], + "score": 1.0, + "content": "propose a bilingual extension of the CBOW method which leverages sentence-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 320, + 469, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 320, + 469, + 331 + ], + "score": 1.0, + "content": "aligned corpora to obtain robust cross-lingual word and sentence representations.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 330, + 470, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 330, + 470, + 343 + ], + "score": 1.0, + "content": "Our approach significantly improves cross-lingual sentence retrieval performance", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 340, + 470, + 354 + ], + "spans": [ + { + "bbox": [ + 141, + 340, + 470, + 354 + ], + "score": 1.0, + "content": "over all other approaches, as well as convincingly outscores mapping methods", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 352, + 470, + 364 + ], + "spans": [ + { + "bbox": [ + 141, + 352, + 470, + 364 + ], + "score": 1.0, + "content": "while maintaining parity with jointly trained methods on word-translation. It also", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 142, + 363, + 470, + 375 + ], + "spans": [ + { + "bbox": [ + 142, + 363, + 470, + 375 + ], + "score": 1.0, + "content": "achieves parity with a deep RNN method on a zero-shot cross-lingual document", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 374, + 470, + 387 + ], + "spans": [ + { + "bbox": [ + 141, + 374, + 470, + 387 + ], + "score": 1.0, + "content": "classification task, requiring far fewer computational resources for training and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 384, + 469, + 397 + ], + "spans": [ + { + "bbox": [ + 141, + 384, + 469, + 397 + ], + "score": 1.0, + "content": "inference. As an additional advantage, our bilingual method also improves the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 396, + 469, + 408 + ], + "spans": [ + { + "bbox": [ + 141, + 396, + 469, + 408 + ], + "score": 1.0, + "content": "quality of monolingual word vectors despite training on much smaller datasets.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 406, + 343, + 419 + ], + "spans": [ + { + "bbox": [ + 142, + 406, + 343, + 419 + ], + "score": 1.0, + "content": "We make our code and models publicly available.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 431, + 206, + 443 + ], + "lines": [ + { + "bbox": [ + 105, + 430, + 208, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 208, + 446 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 450, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "score": 1.0, + "content": "Cross-lingual representations—such as embeddings of words and phrases into a single comparable", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "score": 1.0, + "content": "feature space—have become a key technique in multilingual natural language processing. They of-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "fer strong promise towards the goal of a joint understanding of concepts across languages, as well as", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 482, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 506, + 497 + ], + "score": 1.0, + "content": "for enabling the transfer of knowledge and machine learning models between different languages.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "Therefore, cross-lingual embeddings can serve a variety of downstream tasks such as bilingual lex-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "icon induction, cross-lingual information retrieval, machine translation and many applications of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "zero-shot transfer learning, which is particularly impactful from resource-rich to low-resource lan-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 529, + 140, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 140, + 542 + ], + "score": 1.0, + "content": "guages.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 544, + 504, + 621 + ], + "lines": [ + { + "bbox": [ + 105, + 544, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 505, + 558 + ], + "score": 1.0, + "content": "Existing methods can be broadly classified into two groups (Ruder et al., 2017): mapping meth-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "score": 1.0, + "content": "ods leverage existing monolingual embeddings which are treated as independent, and apply a post-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "process step to map the embeddings of each language into a shared space, through a linear transfor-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "score": 1.0, + "content": "mation (Mikolov et al., 2013b; Conneau et al., 2017; Joulin et al., 2018). On the other hand, joint", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 588, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 600 + ], + "score": 1.0, + "content": "methods learn representations concurrently for multiple languages, by combining monolingual and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "score": 1.0, + "content": "cross-lingual training tasks (Luong et al., 2015; Coulmance et al., 2015; Gouws et al., 2015; Vulic", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 610, + 379, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 379, + 622 + ], + "score": 1.0, + "content": "& Moens, 2015; Chandar et al., 2014; Hermann & Blunsom, 2013).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 627, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "While recent work on word embeddings has focused almost exclusively on mapping methods, which", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "require little to no cross-lingual supervision, (Søgaard et al., 2018) establish that their performance is", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "hindered by linguistic and domain divergences in general, and for distant language pairs in particular.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 660, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 506, + 672 + ], + "score": 1.0, + "content": "Principally, their analysis shows that cross-lingual hubness, where a few words (hubs) in the source", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "language are nearest cross-lingual neighbours of many words in the target language, and structural", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "score": 1.0, + "content": "non-isometry between embeddings do impose a fundamental barrier to the performance of linear", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 693, + 183, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 183, + 705 + ], + "score": 1.0, + "content": "mapping methods.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 503, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "(Ormazabal et al., 2019) propose using joint learning as a means of mitigating these issues. Given", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "parallel data, such as sentences, a joint model learns to predict either the word or context in both", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 447, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 449, + 98 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 449, + 98 + ], + "score": 1.0, + "content": "ROBUST CROSS-LINGUAL EMBEDDINGS FROM", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 279, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 279, + 118 + ], + "score": 1.0, + "content": "PARALLEL SENTENCES", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 108, + 136, + 240, + 157 + ], + "lines": [ + { + "bbox": [ + 108, + 135, + 198, + 148 + ], + "spans": [ + { + "bbox": [ + 108, + 135, + 198, + 148 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 108, + 146, + 241, + 159 + ], + "spans": [ + { + "bbox": [ + 108, + 146, + 241, + 159 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 108, + 135, + 241, + 159 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 187, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 276, + 186, + 335, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 186, + 335, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 208, + 468, + 417 + ], + "lines": [ + { + "bbox": [ + 141, + 208, + 470, + 222 + ], + "spans": [ + { + "bbox": [ + 141, + 208, + 470, + 222 + ], + "score": 1.0, + "content": "Recent advances in cross-lingual word embeddings have primarily relied on", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 221, + 469, + 232 + ], + "spans": [ + { + "bbox": [ + 141, + 221, + 469, + 232 + ], + "score": 1.0, + "content": "mapping-based methods, which project pre-trained word embeddings from dif-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 231, + 470, + 245 + ], + "spans": [ + { + "bbox": [ + 141, + 231, + 470, + 245 + ], + "score": 1.0, + "content": "ferent languages into a shared space through a linear transformation. However,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 242, + 470, + 255 + ], + "spans": [ + { + "bbox": [ + 141, + 242, + 470, + 255 + ], + "score": 1.0, + "content": "these approaches assume word embedding spaces are isomorphic between dif-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 253, + 469, + 265 + ], + "spans": [ + { + "bbox": [ + 141, + 253, + 469, + 265 + ], + "score": 1.0, + "content": "ferent languages, which has been shown not to hold in practice (Søgaard et al.,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 264, + 469, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 264, + 469, + 277 + ], + "score": 1.0, + "content": "2018), and fundamentally limits their performance. This motivates investigating", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "score": 1.0, + "content": "joint learning methods which can overcome this impediment, by simultaneously", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 287, + 469, + 299 + ], + "spans": [ + { + "bbox": [ + 141, + 287, + 469, + 299 + ], + "score": 1.0, + "content": "learning embeddings across languages via a cross-lingual term in the training ob-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 297, + 470, + 310 + ], + "spans": [ + { + "bbox": [ + 141, + 297, + 470, + 310 + ], + "score": 1.0, + "content": "jective. Given the abundance of parallel data available (Tiedemann, 2012), we", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 140, + 307, + 470, + 322 + ], + "spans": [ + { + "bbox": [ + 140, + 307, + 470, + 322 + ], + "score": 1.0, + "content": "propose a bilingual extension of the CBOW method which leverages sentence-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 320, + 469, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 320, + 469, + 331 + ], + "score": 1.0, + "content": "aligned corpora to obtain robust cross-lingual word and sentence representations.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 330, + 470, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 330, + 470, + 343 + ], + "score": 1.0, + "content": "Our approach significantly improves cross-lingual sentence retrieval performance", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 340, + 470, + 354 + ], + "spans": [ + { + "bbox": [ + 141, + 340, + 470, + 354 + ], + "score": 1.0, + "content": "over all other approaches, as well as convincingly outscores mapping methods", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 352, + 470, + 364 + ], + "spans": [ + { + "bbox": [ + 141, + 352, + 470, + 364 + ], + "score": 1.0, + "content": "while maintaining parity with jointly trained methods on word-translation. It also", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 142, + 363, + 470, + 375 + ], + "spans": [ + { + "bbox": [ + 142, + 363, + 470, + 375 + ], + "score": 1.0, + "content": "achieves parity with a deep RNN method on a zero-shot cross-lingual document", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 374, + 470, + 387 + ], + "spans": [ + { + "bbox": [ + 141, + 374, + 470, + 387 + ], + "score": 1.0, + "content": "classification task, requiring far fewer computational resources for training and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 384, + 469, + 397 + ], + "spans": [ + { + "bbox": [ + 141, + 384, + 469, + 397 + ], + "score": 1.0, + "content": "inference. As an additional advantage, our bilingual method also improves the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 396, + 469, + 408 + ], + "spans": [ + { + "bbox": [ + 141, + 396, + 469, + 408 + ], + "score": 1.0, + "content": "quality of monolingual word vectors despite training on much smaller datasets.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 406, + 343, + 419 + ], + "spans": [ + { + "bbox": [ + 142, + 406, + 343, + 419 + ], + "score": 1.0, + "content": "We make our code and models publicly available.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 14, + "bbox_fs": [ + 140, + 208, + 470, + 419 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 431, + 206, + 443 + ], + "lines": [ + { + "bbox": [ + 105, + 430, + 208, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 208, + 446 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 450, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 463 + ], + "score": 1.0, + "content": "Cross-lingual representations—such as embeddings of words and phrases into a single comparable", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 475 + ], + "score": 1.0, + "content": "feature space—have become a key technique in multilingual natural language processing. They of-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "fer strong promise towards the goal of a joint understanding of concepts across languages, as well as", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 482, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 506, + 497 + ], + "score": 1.0, + "content": "for enabling the transfer of knowledge and machine learning models between different languages.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "Therefore, cross-lingual embeddings can serve a variety of downstream tasks such as bilingual lex-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "icon induction, cross-lingual information retrieval, machine translation and many applications of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "zero-shot transfer learning, which is particularly impactful from resource-rich to low-resource lan-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 529, + 140, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 140, + 542 + ], + "score": 1.0, + "content": "guages.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 450, + 506, + 542 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 544, + 504, + 621 + ], + "lines": [ + { + "bbox": [ + 105, + 544, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 505, + 558 + ], + "score": 1.0, + "content": "Existing methods can be broadly classified into two groups (Ruder et al., 2017): mapping meth-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "score": 1.0, + "content": "ods leverage existing monolingual embeddings which are treated as independent, and apply a post-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "process step to map the embeddings of each language into a shared space, through a linear transfor-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "score": 1.0, + "content": "mation (Mikolov et al., 2013b; Conneau et al., 2017; Joulin et al., 2018). On the other hand, joint", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 588, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 600 + ], + "score": 1.0, + "content": "methods learn representations concurrently for multiple languages, by combining monolingual and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "score": 1.0, + "content": "cross-lingual training tasks (Luong et al., 2015; Coulmance et al., 2015; Gouws et al., 2015; Vulic", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 610, + 379, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 379, + 622 + ], + "score": 1.0, + "content": "& Moens, 2015; Chandar et al., 2014; Hermann & Blunsom, 2013).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 544, + 506, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 627, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "While recent work on word embeddings has focused almost exclusively on mapping methods, which", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "require little to no cross-lingual supervision, (Søgaard et al., 2018) establish that their performance is", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "hindered by linguistic and domain divergences in general, and for distant language pairs in particular.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 660, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 506, + 672 + ], + "score": 1.0, + "content": "Principally, their analysis shows that cross-lingual hubness, where a few words (hubs) in the source", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "language are nearest cross-lingual neighbours of many words in the target language, and structural", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 695 + ], + "score": 1.0, + "content": "non-isometry between embeddings do impose a fundamental barrier to the performance of linear", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 693, + 183, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 183, + 705 + ], + "score": 1.0, + "content": "mapping methods.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 626, + 506, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 503, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "(Ormazabal et al., 2019) propose using joint learning as a means of mitigating these issues. Given", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "parallel data, such as sentences, a joint model learns to predict either the word or context in both", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "source and target languages. As we will demonstrate with results from our algorithm, joint methods", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "yield compatible embeddings which are closer to isomorphic, less sensitive to hubness, and perform", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 251, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 251, + 117 + ], + "score": 1.0, + "content": "better on cross-lingual benchmarks.", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 47.5, + "bbox_fs": [ + 105, + 709, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 503, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "source and target languages. As we will demonstrate with results from our algorithm, joint methods", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "yield compatible embeddings which are closer to isomorphic, less sensitive to hubness, and perform", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 251, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 251, + 117 + ], + "score": 1.0, + "content": "better on cross-lingual benchmarks.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 108, + 121, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 506, + 134 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 506, + 134 + ], + "score": 1.0, + "content": "Contributions. We propose the BI-SENT2VEC algorithm, which extends the SENT2VEC algorithm", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "score": 1.0, + "content": "(Pagliardini et al., 2018; Gupta et al., 2019) to the cross-lingual setting. We also revisit TRANS-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 143, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 155 + ], + "score": 1.0, + "content": "GRAM Coulmance et al. (2015), another joint learning method, to assess the effectiveness of joint", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 155, + 352, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 352, + 167 + ], + "score": 1.0, + "content": "learning over mapping-based methods. Our contributions are", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 132, + 171, + 505, + 371 + ], + "lines": [ + { + "bbox": [ + 134, + 173, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 134, + 173, + 505, + 185 + ], + "score": 1.0, + "content": "• On cross-lingual sentence-retrieval and monolingual word representation quality evalua-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 183, + 506, + 197 + ], + "spans": [ + { + "bbox": [ + 141, + 183, + 506, + 197 + ], + "score": 1.0, + "content": "tions, BI-SENT2VEC significantly outperforms competing methods, both jointly trained as", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 142, + 195, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 142, + 195, + 505, + 207 + ], + "score": 1.0, + "content": "well as mapping-based ones while preserving state-of-the-art performance on cross-lingual", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 142, + 206, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 142, + 206, + 505, + 218 + ], + "score": 1.0, + "content": "word retrieval tasks. For dis-similar language pairs, BI-SENT2VEC outperform their com-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 217, + 506, + 229 + ], + "spans": [ + { + "bbox": [ + 141, + 217, + 506, + 229 + ], + "score": 1.0, + "content": "petitors by an even larger margin on all the tasks hinting towards the robustness of our", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 228, + 176, + 238 + ], + "spans": [ + { + "bbox": [ + 141, + 228, + 176, + 238 + ], + "score": 1.0, + "content": "method.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 140, + 242, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 140, + 242, + 505, + 255 + ], + "score": 1.0, + "content": "BI-SENT2VEC performs on par with a multilingual RNN based sentence encoder,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 254, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 142, + 254, + 505, + 265 + ], + "score": 1.0, + "content": "LASER (Artetxe & Schwenk, 2018), on MLDoc (Schwenk & Li, 2018), a zero-shot cross-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 263, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 263, + 505, + 277 + ], + "score": 1.0, + "content": "lingual transfer task on documents in multiple languages. Compared to LASER, our method", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "improves computational efficiency by an order of magnitude for both training and infer-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 286, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 141, + 286, + 505, + 299 + ], + "score": 1.0, + "content": "ence, making it suitable for resource or latency-constrained on-device cross-lingual NLP", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 298, + 195, + 309 + ], + "spans": [ + { + "bbox": [ + 142, + 298, + 195, + 309 + ], + "score": 1.0, + "content": "applications.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 138, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 138, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "We verify that joint learning methods consistently dominate state-of-the-art mapping meth-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 324, + 446, + 336 + ], + "spans": [ + { + "bbox": [ + 142, + 324, + 446, + 336 + ], + "score": 1.0, + "content": "ods on standard benchmarks, i.e., cross-lingual word and sentence retrieval.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 134, + 337, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 134, + 337, + 506, + 351 + ], + "score": 1.0, + "content": "• Training on parallel data additionally enriches monolingual representation quality, evident", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 349, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 141, + 349, + 506, + 362 + ], + "score": 1.0, + "content": "by the superior performance of BI-SENT2VEC over FASTTEXT embeddings trained on a", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 360, + 225, + 374 + ], + "spans": [ + { + "bbox": [ + 142, + 361, + 166, + 371 + ], + "score": 0.86, + "content": "1 0 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 360, + 225, + 374 + ], + "score": 1.0, + "content": "larger corpus.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 108, + 378, + 305, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 377, + 307, + 391 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 307, + 391 + ], + "score": 1.0, + "content": "We make our models and code publicly available.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 108, + 402, + 211, + 415 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 213, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 213, + 417 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 505, + 501 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 505, + 437 + ], + "score": 1.0, + "content": "The literature on cross-lingual representation learning is extensive. Most recent advances in the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "field pursue unsupervised (Artetxe et al., 2017; Conneau et al., 2017; Chen & Cardie, 2018; Hoshen", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 444, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 460 + ], + "score": 1.0, + "content": "& Wolf, 2018; Grave et al., 2018b) or supervised (Joulin et al., 2018; Conneau et al., 2017) map-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "score": 1.0, + "content": "ping or alignment-based algorithms. All these methods use existing monolingual word embeddings,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 466, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 505, + 481 + ], + "score": 1.0, + "content": "followed by a cross-lingual alignment procedure as a post-processing step— that is to learn a sim-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "score": 1.0, + "content": "ple (typically linear) mapping from the source language embedding space to the target language", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 488, + 181, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 181, + 503 + ], + "score": 1.0, + "content": "embedding space.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 505, + 595 + ], + "lines": [ + { + "bbox": [ + 106, + 505, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 505, + 520 + ], + "score": 1.0, + "content": "Supervised learning of a linear map from a source embedding space to another target embedding", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "score": 1.0, + "content": "space (Mikolov et al., 2013b) based on a bilingual dictionary was one of the first approaches towards", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 527, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 542 + ], + "score": 1.0, + "content": "cross-lingual word embeddings. Additionally enforcing orthogonality constraints on the linear map", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "results in rotations, and can be formulated as an orthogonal Procrustes problem (Smith et al., 2017).", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "score": 1.0, + "content": "However, the authors found the translated embeddings to suffer from hubness, which they mitigate", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "by introducing the inverted softmax as a corrective search metric at inference time. (Artetxe et al.,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 571, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 505, + 585 + ], + "score": 1.0, + "content": "2017) align embedding spaces starting from a parallel seed lexicon such as digits and iteratively", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 583, + 307, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 307, + 596 + ], + "score": 1.0, + "content": "build a larger bilingual dictionary during training.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "In their seminal work, (Conneau et al., 2017) propose an adversarial training method to learn a linear", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 610, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 624 + ], + "score": 1.0, + "content": "orthogonal map, avoiding bilingual supervision altogether. They further refine the learnt mapping by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "applying the Procrustes procedure iteratively with a synthetic dictionary generated through adver-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "score": 1.0, + "content": "sarial training. They also introduce the ‘Cross-Domain Similarity Local Scaling’ (CSLS) retrieval", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 643, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 658 + ], + "score": 1.0, + "content": "criterion for translating between spaces, which further improves on the word translation accuracy", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "over nearest-neighbour and inverted softmax metrics. They refer to their work as Multilingual Un-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "supervised and Supervised Embeddings (MUSE). In this paper, we will use MUSE to denote the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 364, + 690 + ], + "score": 1.0, + "content": "unsupervised embeddings introduced by them, and “Procrustes", + "type": "text" + }, + { + "bbox": [ + 365, + 678, + 374, + 687 + ], + "score": 0.65, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "refine” to denote the supervised", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "embeddings obtained by them. (Chen & Cardie, 2018) similarly use “multilingual adversarial train-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "ing” followed by “pseudo-supervised refinement” to obtain unsupervised multilingual word embed-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "dings (UMWE), as opposed to bilingual word embeddings by (Conneau et al., 2017). Hoshen &", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "Wolf (2018) describe an unsupervised approach where they align the second moment of the two", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 46.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 503, + 115 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 105, + 82, + 505, + 117 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 108, + 121, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 506, + 134 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 506, + 134 + ], + "score": 1.0, + "content": "Contributions. We propose the BI-SENT2VEC algorithm, which extends the SENT2VEC algorithm", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "score": 1.0, + "content": "(Pagliardini et al., 2018; Gupta et al., 2019) to the cross-lingual setting. We also revisit TRANS-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 143, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 155 + ], + "score": 1.0, + "content": "GRAM Coulmance et al. (2015), another joint learning method, to assess the effectiveness of joint", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 155, + 352, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 352, + 167 + ], + "score": 1.0, + "content": "learning over mapping-based methods. Our contributions are", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 121, + 506, + 167 + ] + }, + { + "type": "list", + "bbox": [ + 132, + 171, + 505, + 371 + ], + "lines": [ + { + "bbox": [ + 134, + 173, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 134, + 173, + 505, + 185 + ], + "score": 1.0, + "content": "• On cross-lingual sentence-retrieval and monolingual word representation quality evalua-", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 183, + 506, + 197 + ], + "spans": [ + { + "bbox": [ + 141, + 183, + 506, + 197 + ], + "score": 1.0, + "content": "tions, BI-SENT2VEC significantly outperforms competing methods, both jointly trained as", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 142, + 195, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 142, + 195, + 505, + 207 + ], + "score": 1.0, + "content": "well as mapping-based ones while preserving state-of-the-art performance on cross-lingual", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 142, + 206, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 142, + 206, + 505, + 218 + ], + "score": 1.0, + "content": "word retrieval tasks. For dis-similar language pairs, BI-SENT2VEC outperform their com-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 217, + 506, + 229 + ], + "spans": [ + { + "bbox": [ + 141, + 217, + 506, + 229 + ], + "score": 1.0, + "content": "petitors by an even larger margin on all the tasks hinting towards the robustness of our", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 228, + 176, + 238 + ], + "spans": [ + { + "bbox": [ + 141, + 228, + 176, + 238 + ], + "score": 1.0, + "content": "method.", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 140, + 242, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 140, + 242, + 505, + 255 + ], + "score": 1.0, + "content": "BI-SENT2VEC performs on par with a multilingual RNN based sentence encoder,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 254, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 142, + 254, + 505, + 265 + ], + "score": 1.0, + "content": "LASER (Artetxe & Schwenk, 2018), on MLDoc (Schwenk & Li, 2018), a zero-shot cross-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 263, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 263, + 505, + 277 + ], + "score": 1.0, + "content": "lingual transfer task on documents in multiple languages. Compared to LASER, our method", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "improves computational efficiency by an order of magnitude for both training and infer-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 286, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 141, + 286, + 505, + 299 + ], + "score": 1.0, + "content": "ence, making it suitable for resource or latency-constrained on-device cross-lingual NLP", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 298, + 195, + 309 + ], + "spans": [ + { + "bbox": [ + 142, + 298, + 195, + 309 + ], + "score": 1.0, + "content": "applications.", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 138, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 138, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "We verify that joint learning methods consistently dominate state-of-the-art mapping meth-", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 324, + 446, + 336 + ], + "spans": [ + { + "bbox": [ + 142, + 324, + 446, + 336 + ], + "score": 1.0, + "content": "ods on standard benchmarks, i.e., cross-lingual word and sentence retrieval.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 134, + 337, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 134, + 337, + 506, + 351 + ], + "score": 1.0, + "content": "• Training on parallel data additionally enriches monolingual representation quality, evident", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 349, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 141, + 349, + 506, + 362 + ], + "score": 1.0, + "content": "by the superior performance of BI-SENT2VEC over FASTTEXT embeddings trained on a", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 360, + 225, + 374 + ], + "spans": [ + { + "bbox": [ + 142, + 361, + 166, + 371 + ], + "score": 0.86, + "content": "1 0 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 360, + 225, + 374 + ], + "score": 1.0, + "content": "larger corpus.", + "type": "text" + } + ], + "index": 23, + "is_list_end_line": true + } + ], + "index": 15, + "bbox_fs": [ + 134, + 173, + 506, + 374 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 378, + 305, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 377, + 307, + 391 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 307, + 391 + ], + "score": 1.0, + "content": "We make our models and code publicly available.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 377, + 307, + 391 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 402, + 211, + 415 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 213, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 213, + 417 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 505, + 501 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 505, + 437 + ], + "score": 1.0, + "content": "The literature on cross-lingual representation learning is extensive. Most recent advances in the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "field pursue unsupervised (Artetxe et al., 2017; Conneau et al., 2017; Chen & Cardie, 2018; Hoshen", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 444, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 460 + ], + "score": 1.0, + "content": "& Wolf, 2018; Grave et al., 2018b) or supervised (Joulin et al., 2018; Conneau et al., 2017) map-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "score": 1.0, + "content": "ping or alignment-based algorithms. All these methods use existing monolingual word embeddings,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 466, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 505, + 481 + ], + "score": 1.0, + "content": "followed by a cross-lingual alignment procedure as a post-processing step— that is to learn a sim-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "score": 1.0, + "content": "ple (typically linear) mapping from the source language embedding space to the target language", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 488, + 181, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 181, + 503 + ], + "score": 1.0, + "content": "embedding space.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 423, + 505, + 503 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 505, + 595 + ], + "lines": [ + { + "bbox": [ + 106, + 505, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 505, + 520 + ], + "score": 1.0, + "content": "Supervised learning of a linear map from a source embedding space to another target embedding", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "score": 1.0, + "content": "space (Mikolov et al., 2013b) based on a bilingual dictionary was one of the first approaches towards", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 527, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 542 + ], + "score": 1.0, + "content": "cross-lingual word embeddings. Additionally enforcing orthogonality constraints on the linear map", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "results in rotations, and can be formulated as an orthogonal Procrustes problem (Smith et al., 2017).", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "score": 1.0, + "content": "However, the authors found the translated embeddings to suffer from hubness, which they mitigate", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "by introducing the inverted softmax as a corrective search metric at inference time. (Artetxe et al.,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 571, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 505, + 585 + ], + "score": 1.0, + "content": "2017) align embedding spaces starting from a parallel seed lexicon such as digits and iteratively", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 583, + 307, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 307, + 596 + ], + "score": 1.0, + "content": "build a larger bilingual dictionary during training.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 505, + 505, + 596 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "In their seminal work, (Conneau et al., 2017) propose an adversarial training method to learn a linear", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 610, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 624 + ], + "score": 1.0, + "content": "orthogonal map, avoiding bilingual supervision altogether. They further refine the learnt mapping by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "applying the Procrustes procedure iteratively with a synthetic dictionary generated through adver-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "score": 1.0, + "content": "sarial training. They also introduce the ‘Cross-Domain Similarity Local Scaling’ (CSLS) retrieval", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 643, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 658 + ], + "score": 1.0, + "content": "criterion for translating between spaces, which further improves on the word translation accuracy", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "over nearest-neighbour and inverted softmax metrics. They refer to their work as Multilingual Un-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "supervised and Supervised Embeddings (MUSE). In this paper, we will use MUSE to denote the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 364, + 690 + ], + "score": 1.0, + "content": "unsupervised embeddings introduced by them, and “Procrustes", + "type": "text" + }, + { + "bbox": [ + 365, + 678, + 374, + 687 + ], + "score": 0.65, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "refine” to denote the supervised", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "embeddings obtained by them. (Chen & Cardie, 2018) similarly use “multilingual adversarial train-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "ing” followed by “pseudo-supervised refinement” to obtain unsupervised multilingual word embed-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "dings (UMWE), as opposed to bilingual word embeddings by (Conneau et al., 2017). Hoshen &", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "Wolf (2018) describe an unsupervised approach where they align the second moment of the two", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "word embedding distributions followed by a further refinement. Building on the success of CSLS in", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "reducing retrieval sensitivity to hubness, (Joulin et al., 2018) directly optimize a convex relaxation of", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 118 + ], + "score": 1.0, + "content": "the CSLS function (RCSLS) to align existing mono-lingual embeddings using a bilingual dictionary.", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 46.5, + "bbox_fs": [ + 105, + 600, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "word embedding distributions followed by a further refinement. Building on the success of CSLS in", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "reducing retrieval sensitivity to hubness, (Joulin et al., 2018) directly optimize a convex relaxation of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 118 + ], + "score": 1.0, + "content": "the CSLS function (RCSLS) to align existing mono-lingual embeddings using a bilingual dictionary.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 121, + 505, + 177 + ], + "lines": [ + { + "bbox": [ + 106, + 122, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 505, + 133 + ], + "score": 1.0, + "content": "While none of the methods described above require parallel corpora, all assume structural isomor-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "phism between existing embeddings for each language (Mikolov et al., 2013b), i.e. there exists a", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "simple (typically linear) mapping function which aligns all existing embeddings. However, this is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "not always a realistic assumption (Søgaard et al., 2018)—even in small toy-examples it is clear that", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 443, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 443, + 179 + ], + "score": 1.0, + "content": "many geometric configurations of points can not be linearly mapped to their targets.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 182, + 505, + 303 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 505, + 195 + ], + "score": 1.0, + "content": "Joint learning algorithms such as TRANSGRAM (Coulmance et al., 2015) and Cr5 (Josifoski et al.,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 192, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 505, + 206 + ], + "score": 1.0, + "content": "2019) , circumvent this restriction by simultaneously learning embeddings as well as their alignment.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 203, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 217 + ], + "score": 1.0, + "content": "TRANSGRAM, for example, extends the Skipgram (Mikolov et al., 2013a) method to jointly train", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "score": 1.0, + "content": "bilingual embeddings in the same space, on a corpus composed of parallel sentences. In addition to", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 226, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 238 + ], + "score": 1.0, + "content": "the monolingual Skipgram loss for both languages, they introduce a similar cross-lingual loss where", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 506, + 250 + ], + "score": 1.0, + "content": "a word from a sentence in one language is trained to predict the word-contents of the sentence", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 247, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 506, + 260 + ], + "score": 1.0, + "content": "in the other. Cr5, on the other hand, uses document-aligned corpora to achieve state-of-the-art", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 259, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 271 + ], + "score": 1.0, + "content": "results for cross-lingual document retrieval while staying competitive at cross-lingual sentence and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "score": 1.0, + "content": "word retrieval. TRANSGRAM embeddings have been absent from discussion in most of the recent", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 282, + 504, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 504, + 293 + ], + "score": 1.0, + "content": "work. However, the growing abundance of sentence-aligned parallel data (Tiedemann, 2012) merits", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 292, + 246, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 246, + 304 + ], + "score": 1.0, + "content": "a reappraisal of their performance.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 308, + 505, + 364 + ], + "lines": [ + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "score": 1.0, + "content": "(Ormazabal et al., 2019) use BIVEC (Luong et al., 2015), another bilingual extension of Skipgram,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "score": 1.0, + "content": "which uses a bilingual dictionary in addition to parallel sentences to obtain word-alignments and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 329, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 345 + ], + "score": 1.0, + "content": "compare it with the unsupervised version of VECMAP (Artetxe et al., 2018b), another mapping-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 341, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 505, + 354 + ], + "score": 1.0, + "content": "based method. Our experiments show this extra level of supervision in the case of BIVEC is redun-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 352, + 293, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 293, + 365 + ], + "score": 1.0, + "content": "dant in obtaining state-of-the-art performance.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "title", + "bbox": [ + 107, + 375, + 164, + 388 + ], + "lines": [ + { + "bbox": [ + 104, + 373, + 167, + 390 + ], + "spans": [ + { + "bbox": [ + 104, + 373, + 167, + 390 + ], + "score": 1.0, + "content": "3 MODEL", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 394, + 505, + 475 + ], + "lines": [ + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "Proposed Model. Our BI-SENT2VEC model is a cross-lingual extension of SENT2VEC proposed", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 388, + 419 + ], + "score": 1.0, + "content": "by (Pagliardini et al., 2018), which in turn is an extension of the", + "type": "text" + }, + { + "bbox": [ + 388, + 407, + 396, + 416 + ], + "score": 0.28, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "-BOW embedding method", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 417, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 505, + 430 + ], + "score": 1.0, + "content": "(Mikolov et al., 2013a). SENT2VEC is trained on sentence contexts, with the word and higher-order", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "word n-gram embeddings specifically optimized toward obtaining robust sentence embeddings using", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 437, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 104, + 437, + 375, + 454 + ], + "score": 1.0, + "content": "additive composition. Formally, SENT2VEC obtains representation", + "type": "text" + }, + { + "bbox": [ + 375, + 441, + 386, + 450 + ], + "score": 0.84, + "content": "{ \\pmb v } _ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 437, + 442, + 454 + ], + "score": 1.0, + "content": "of a sentence", + "type": "text" + }, + { + "bbox": [ + 442, + 440, + 450, + 449 + ], + "score": 0.81, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 437, + 506, + 454 + ], + "score": 1.0, + "content": "by averaging", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 103, + 447, + 507, + 469 + ], + "spans": [ + { + "bbox": [ + 103, + 447, + 324, + 469 + ], + "score": 1.0, + "content": "the word-ngram embeddings (including unigrams) as", + "type": "text" + }, + { + "bbox": [ + 325, + 450, + 430, + 465 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathbf { \\dot { \\boldsymbol { v } } } _ { s } : = \\frac { 1 } { R ( S ) } \\sum _ { w \\in R ( S ) } \\pmb { v } _ { w } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 447, + 455, + 469 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 456, + 450, + 478, + 462 + ], + "score": 0.91, + "content": "R ( S )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 447, + 507, + 469 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 463, + 262, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 250, + 475 + ], + "score": 1.0, + "content": "set of word n-grams in the sentence", + "type": "text" + }, + { + "bbox": [ + 251, + 463, + 258, + 473 + ], + "score": 0.8, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 463, + 262, + 475 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 480, + 505, + 526 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 396, + 493 + ], + "score": 1.0, + "content": "The SENT2VEC training objective aims to predict a masked word token", + "type": "text" + }, + { + "bbox": [ + 396, + 482, + 408, + 491 + ], + "score": 0.85, + "content": "w _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 479, + 471, + 493 + ], + "score": 1.0, + "content": "in the sentence", + "type": "text" + }, + { + "bbox": [ + 472, + 481, + 479, + 490 + ], + "score": 0.82, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 479, + 505, + 493 + ], + "score": 1.0, + "content": "using", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 490, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 104, + 490, + 260, + 506 + ], + "score": 1.0, + "content": "the rest of the sentence representation", + "type": "text" + }, + { + "bbox": [ + 261, + 493, + 294, + 504 + ], + "score": 0.91, + "content": "{ \\pmb v } _ { S \\backslash \\{ { u v } _ { t } \\} }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 490, + 506, + 506 + ], + "score": 1.0, + "content": ". To formulate the training objective, we use logistic", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 503, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 125, + 517 + ], + "score": 1.0, + "content": "loss", + "type": "text" + }, + { + "bbox": [ + 125, + 503, + 219, + 516 + ], + "score": 0.9, + "content": "\\ell : x \\mapsto \\log { ( 1 + e ^ { - x } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 503, + 505, + 517 + ], + "score": 1.0, + "content": "in conjunction with negative sampling. More precisely, for a raw text", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 514, + 397, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 136, + 528 + ], + "score": 1.0, + "content": "corpus", + "type": "text" + }, + { + "bbox": [ + 136, + 515, + 145, + 524 + ], + "score": 0.81, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 514, + 397, + 528 + ], + "score": 1.0, + "content": ", the monolingual training objective for SENT2VEC is given by", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "interline_equation", + "bbox": [ + 182, + 529, + 429, + 561 + ], + "lines": [ + { + "bbox": [ + 182, + 529, + 429, + 561 + ], + "spans": [ + { + "bbox": [ + 182, + 529, + 429, + 561 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { U , V } \\sum _ { S \\in C } \\sum _ { w _ { t } \\in S } \\left( \\ell \\big ( \\boldsymbol { u } _ { w _ { t } } ^ { \\top } \\boldsymbol { v } _ { S \\setminus \\{ w _ { t } \\} } \\big ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell \\big ( - \\boldsymbol { u } _ { w ^ { \\prime } } ^ { \\top } \\boldsymbol { v } _ { S \\setminus \\{ w _ { t } \\} } \\big ) \\right)", + "type": "interline_equation", + "image_path": "d8aa3b9ad8140d3cea01015e60fceeebdd5dd9e761c326f1689084ce0bd64d63.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 182, + 529, + 429, + 539.6666666666666 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 182, + 539.6666666666666, + 429, + 550.3333333333333 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 182, + 550.3333333333333, + 429, + 560.9999999999999 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 505, + 622 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 132, + 577 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 566, + 145, + 575 + ], + "score": 0.85, + "content": "w _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 564, + 234, + 577 + ], + "score": 1.0, + "content": "is the target word and,", + "type": "text" + }, + { + "bbox": [ + 234, + 564, + 244, + 574 + ], + "score": 0.83, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 564, + 261, + 577 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 262, + 564, + 272, + 574 + ], + "score": 0.82, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 564, + 505, + 577 + ], + "score": 1.0, + "content": "are the source n-gram and target word embedding matrices", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 575, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 285, + 589 + ], + "score": 1.0, + "content": "respectively. Here, the set of negative words", + "type": "text" + }, + { + "bbox": [ + 285, + 576, + 303, + 587 + ], + "score": 0.91, + "content": "N _ { w _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 575, + 505, + 589 + ], + "score": 1.0, + "content": "is sampled from a multinomial distribution where", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 586, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 586, + 505, + 600 + ], + "score": 1.0, + "content": "the probability of picking a word is directly proportional to the square root of its frequency in the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 597, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 208, + 613 + ], + "score": 1.0, + "content": "corpus. Each target word", + "type": "text" + }, + { + "bbox": [ + 208, + 600, + 220, + 610 + ], + "score": 0.86, + "content": "w _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 597, + 331, + 613 + ], + "score": 1.0, + "content": "is sampled with probability", + "type": "text" + }, + { + "bbox": [ + 332, + 597, + 437, + 611 + ], + "score": 0.9, + "content": "m i n \\{ 1 , \\sqrt { t / f _ { w _ { t } } } + t / f _ { w _ { t } } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 597, + 465, + 613 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 465, + 599, + 480, + 611 + ], + "score": 0.9, + "content": "f _ { w _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 597, + 505, + 613 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 609, + 253, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 253, + 622 + ], + "score": 1.0, + "content": "frequency of the word in the corpus.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 670 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "We adapt the SENT2VEC model to bilingual corpora by introducing a cross-lingual loss in addition", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 355, + 651 + ], + "score": 1.0, + "content": "to the monolingual loss in equation (1). Given a sentence pair", + "type": "text" + }, + { + "bbox": [ + 355, + 638, + 414, + 650 + ], + "score": 0.92, + "content": "S = ( S _ { l _ { 1 } } , S _ { l _ { 2 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 637, + 442, + 651 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 442, + 638, + 456, + 649 + ], + "score": 0.92, + "content": "S _ { l _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 637, + 475, + 651 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 475, + 638, + 488, + 649 + ], + "score": 0.89, + "content": "S _ { l _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "are", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 263, + 661 + ], + "score": 1.0, + "content": "translations of each other in languages", + "type": "text" + }, + { + "bbox": [ + 264, + 649, + 272, + 659 + ], + "score": 0.87, + "content": "l _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 648, + 291, + 661 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 291, + 649, + 300, + 659 + ], + "score": 0.86, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 648, + 461, + 661 + ], + "score": 1.0, + "content": ", the cross-lingual loss for a target word", + "type": "text" + }, + { + "bbox": [ + 462, + 650, + 473, + 659 + ], + "score": 0.85, + "content": "w _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 648, + 486, + 661 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 486, + 649, + 494, + 659 + ], + "score": 0.87, + "content": "l _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 659, + 144, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 144, + 673 + ], + "score": 1.0, + "content": "given by", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5 + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 669, + 376, + 698 + ], + "lines": [ + { + "bbox": [ + 235, + 669, + 376, + 698 + ], + "spans": [ + { + "bbox": [ + 235, + 669, + 376, + 698 + ], + "score": 0.94, + "content": "\\ell \\big ( { \\pmb u } _ { w _ { t } } ^ { \\top } { \\pmb v } _ { S _ { l _ { 2 } } } \\big ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell \\big ( - { \\pmb u } _ { w _ { t } ^ { \\prime } } ^ { \\top } { \\pmb v } _ { S _ { l _ { 2 } } } \\big )", + "type": "interline_equation", + "image_path": "c243177fd7c6b452de3f5e756d7406bfd6c15aa46bf4c87fa832e1c1d9acf376.jpg" + } + ] + } + ], + "index": 48.5, + "virtual_lines": [ + { + "bbox": [ + 235, + 669, + 376, + 683.5 + ], + "spans": [], + "index": 48 + }, + { + "bbox": [ + 235, + 683.5, + 376, + 698.0 + ], + "spans": [], + "index": 49 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 217, + 712 + ], + "score": 1.0, + "content": "Thus, we use the sentence", + "type": "text" + }, + { + "bbox": [ + 218, + 700, + 231, + 710 + ], + "score": 0.89, + "content": "S _ { l _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 698, + 378, + 712 + ], + "score": 1.0, + "content": "to predict the constituent words of", + "type": "text" + }, + { + "bbox": [ + 378, + 700, + 391, + 711 + ], + "score": 0.9, + "content": "S _ { l _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "and vice-versa in a similar", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 473, + 722 + ], + "score": 1.0, + "content": "fashion to the monolingual SENT2VEC, shown in Figure 1. This ensures that the word and", + "type": "text" + }, + { + "bbox": [ + 474, + 712, + 480, + 720 + ], + "score": 0.29, + "content": "\\mathbf { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "-gram", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 720, + 318, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 318, + 733 + ], + "score": 1.0, + "content": "embeddings of both languages lie in the same space.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 51 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 105, + 82, + 506, + 118 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 121, + 505, + 177 + ], + "lines": [ + { + "bbox": [ + 106, + 122, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 505, + 133 + ], + "score": 1.0, + "content": "While none of the methods described above require parallel corpora, all assume structural isomor-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "phism between existing embeddings for each language (Mikolov et al., 2013b), i.e. there exists a", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "simple (typically linear) mapping function which aligns all existing embeddings. However, this is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "not always a realistic assumption (Søgaard et al., 2018)—even in small toy-examples it is clear that", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 443, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 443, + 179 + ], + "score": 1.0, + "content": "many geometric configurations of points can not be linearly mapped to their targets.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 122, + 505, + 179 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 182, + 505, + 303 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 505, + 195 + ], + "score": 1.0, + "content": "Joint learning algorithms such as TRANSGRAM (Coulmance et al., 2015) and Cr5 (Josifoski et al.,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 192, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 505, + 206 + ], + "score": 1.0, + "content": "2019) , circumvent this restriction by simultaneously learning embeddings as well as their alignment.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 203, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 217 + ], + "score": 1.0, + "content": "TRANSGRAM, for example, extends the Skipgram (Mikolov et al., 2013a) method to jointly train", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "score": 1.0, + "content": "bilingual embeddings in the same space, on a corpus composed of parallel sentences. In addition to", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 226, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 238 + ], + "score": 1.0, + "content": "the monolingual Skipgram loss for both languages, they introduce a similar cross-lingual loss where", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 506, + 250 + ], + "score": 1.0, + "content": "a word from a sentence in one language is trained to predict the word-contents of the sentence", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 247, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 506, + 260 + ], + "score": 1.0, + "content": "in the other. Cr5, on the other hand, uses document-aligned corpora to achieve state-of-the-art", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 259, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 271 + ], + "score": 1.0, + "content": "results for cross-lingual document retrieval while staying competitive at cross-lingual sentence and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "score": 1.0, + "content": "word retrieval. TRANSGRAM embeddings have been absent from discussion in most of the recent", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 282, + 504, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 504, + 293 + ], + "score": 1.0, + "content": "work. However, the growing abundance of sentence-aligned parallel data (Tiedemann, 2012) merits", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 292, + 246, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 246, + 304 + ], + "score": 1.0, + "content": "a reappraisal of their performance.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 182, + 506, + 304 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 308, + 505, + 364 + ], + "lines": [ + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "score": 1.0, + "content": "(Ormazabal et al., 2019) use BIVEC (Luong et al., 2015), another bilingual extension of Skipgram,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "score": 1.0, + "content": "which uses a bilingual dictionary in addition to parallel sentences to obtain word-alignments and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 329, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 345 + ], + "score": 1.0, + "content": "compare it with the unsupervised version of VECMAP (Artetxe et al., 2018b), another mapping-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 341, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 505, + 354 + ], + "score": 1.0, + "content": "based method. Our experiments show this extra level of supervision in the case of BIVEC is redun-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 352, + 293, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 293, + 365 + ], + "score": 1.0, + "content": "dant in obtaining state-of-the-art performance.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 308, + 505, + 365 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 375, + 164, + 388 + ], + "lines": [ + { + "bbox": [ + 104, + 373, + 167, + 390 + ], + "spans": [ + { + "bbox": [ + 104, + 373, + 167, + 390 + ], + "score": 1.0, + "content": "3 MODEL", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 394, + 505, + 475 + ], + "lines": [ + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "Proposed Model. Our BI-SENT2VEC model is a cross-lingual extension of SENT2VEC proposed", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 388, + 419 + ], + "score": 1.0, + "content": "by (Pagliardini et al., 2018), which in turn is an extension of the", + "type": "text" + }, + { + "bbox": [ + 388, + 407, + 396, + 416 + ], + "score": 0.28, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "-BOW embedding method", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 417, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 505, + 430 + ], + "score": 1.0, + "content": "(Mikolov et al., 2013a). SENT2VEC is trained on sentence contexts, with the word and higher-order", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "word n-gram embeddings specifically optimized toward obtaining robust sentence embeddings using", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 437, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 104, + 437, + 375, + 454 + ], + "score": 1.0, + "content": "additive composition. Formally, SENT2VEC obtains representation", + "type": "text" + }, + { + "bbox": [ + 375, + 441, + 386, + 450 + ], + "score": 0.84, + "content": "{ \\pmb v } _ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 437, + 442, + 454 + ], + "score": 1.0, + "content": "of a sentence", + "type": "text" + }, + { + "bbox": [ + 442, + 440, + 450, + 449 + ], + "score": 0.81, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 437, + 506, + 454 + ], + "score": 1.0, + "content": "by averaging", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 103, + 447, + 507, + 469 + ], + "spans": [ + { + "bbox": [ + 103, + 447, + 324, + 469 + ], + "score": 1.0, + "content": "the word-ngram embeddings (including unigrams) as", + "type": "text" + }, + { + "bbox": [ + 325, + 450, + 430, + 465 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathbf { \\dot { \\boldsymbol { v } } } _ { s } : = \\frac { 1 } { R ( S ) } \\sum _ { w \\in R ( S ) } \\pmb { v } _ { w } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 447, + 455, + 469 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 456, + 450, + 478, + 462 + ], + "score": 0.91, + "content": "R ( S )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 447, + 507, + 469 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 463, + 262, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 250, + 475 + ], + "score": 1.0, + "content": "set of word n-grams in the sentence", + "type": "text" + }, + { + "bbox": [ + 251, + 463, + 258, + 473 + ], + "score": 0.8, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 463, + 262, + 475 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28, + "bbox_fs": [ + 103, + 395, + 507, + 475 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 480, + 505, + 526 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 396, + 493 + ], + "score": 1.0, + "content": "The SENT2VEC training objective aims to predict a masked word token", + "type": "text" + }, + { + "bbox": [ + 396, + 482, + 408, + 491 + ], + "score": 0.85, + "content": "w _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 479, + 471, + 493 + ], + "score": 1.0, + "content": "in the sentence", + "type": "text" + }, + { + "bbox": [ + 472, + 481, + 479, + 490 + ], + "score": 0.82, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 479, + 505, + 493 + ], + "score": 1.0, + "content": "using", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 490, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 104, + 490, + 260, + 506 + ], + "score": 1.0, + "content": "the rest of the sentence representation", + "type": "text" + }, + { + "bbox": [ + 261, + 493, + 294, + 504 + ], + "score": 0.91, + "content": "{ \\pmb v } _ { S \\backslash \\{ { u v } _ { t } \\} }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 490, + 506, + 506 + ], + "score": 1.0, + "content": ". To formulate the training objective, we use logistic", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 503, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 125, + 517 + ], + "score": 1.0, + "content": "loss", + "type": "text" + }, + { + "bbox": [ + 125, + 503, + 219, + 516 + ], + "score": 0.9, + "content": "\\ell : x \\mapsto \\log { ( 1 + e ^ { - x } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 503, + 505, + 517 + ], + "score": 1.0, + "content": "in conjunction with negative sampling. More precisely, for a raw text", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 514, + 397, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 136, + 528 + ], + "score": 1.0, + "content": "corpus", + "type": "text" + }, + { + "bbox": [ + 136, + 515, + 145, + 524 + ], + "score": 0.81, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 514, + 397, + 528 + ], + "score": 1.0, + "content": ", the monolingual training objective for SENT2VEC is given by", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 479, + 506, + 528 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 182, + 529, + 429, + 561 + ], + "lines": [ + { + "bbox": [ + 182, + 529, + 429, + 561 + ], + "spans": [ + { + "bbox": [ + 182, + 529, + 429, + 561 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { U , V } \\sum _ { S \\in C } \\sum _ { w _ { t } \\in S } \\left( \\ell \\big ( \\boldsymbol { u } _ { w _ { t } } ^ { \\top } \\boldsymbol { v } _ { S \\setminus \\{ w _ { t } \\} } \\big ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell \\big ( - \\boldsymbol { u } _ { w ^ { \\prime } } ^ { \\top } \\boldsymbol { v } _ { S \\setminus \\{ w _ { t } \\} } \\big ) \\right)", + "type": "interline_equation", + "image_path": "d8aa3b9ad8140d3cea01015e60fceeebdd5dd9e761c326f1689084ce0bd64d63.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 182, + 529, + 429, + 539.6666666666666 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 182, + 539.6666666666666, + 429, + 550.3333333333333 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 182, + 550.3333333333333, + 429, + 560.9999999999999 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 505, + 622 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 132, + 577 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 566, + 145, + 575 + ], + "score": 0.85, + "content": "w _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 564, + 234, + 577 + ], + "score": 1.0, + "content": "is the target word and,", + "type": "text" + }, + { + "bbox": [ + 234, + 564, + 244, + 574 + ], + "score": 0.83, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 564, + 261, + 577 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 262, + 564, + 272, + 574 + ], + "score": 0.82, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 564, + 505, + 577 + ], + "score": 1.0, + "content": "are the source n-gram and target word embedding matrices", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 575, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 285, + 589 + ], + "score": 1.0, + "content": "respectively. Here, the set of negative words", + "type": "text" + }, + { + "bbox": [ + 285, + 576, + 303, + 587 + ], + "score": 0.91, + "content": "N _ { w _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 575, + 505, + 589 + ], + "score": 1.0, + "content": "is sampled from a multinomial distribution where", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 586, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 586, + 505, + 600 + ], + "score": 1.0, + "content": "the probability of picking a word is directly proportional to the square root of its frequency in the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 597, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 208, + 613 + ], + "score": 1.0, + "content": "corpus. Each target word", + "type": "text" + }, + { + "bbox": [ + 208, + 600, + 220, + 610 + ], + "score": 0.86, + "content": "w _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 597, + 331, + 613 + ], + "score": 1.0, + "content": "is sampled with probability", + "type": "text" + }, + { + "bbox": [ + 332, + 597, + 437, + 611 + ], + "score": 0.9, + "content": "m i n \\{ 1 , \\sqrt { t / f _ { w _ { t } } } + t / f _ { w _ { t } } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 597, + 465, + 613 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 465, + 599, + 480, + 611 + ], + "score": 0.9, + "content": "f _ { w _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 597, + 505, + 613 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 609, + 253, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 253, + 622 + ], + "score": 1.0, + "content": "frequency of the word in the corpus.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 564, + 505, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 670 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "We adapt the SENT2VEC model to bilingual corpora by introducing a cross-lingual loss in addition", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 355, + 651 + ], + "score": 1.0, + "content": "to the monolingual loss in equation (1). Given a sentence pair", + "type": "text" + }, + { + "bbox": [ + 355, + 638, + 414, + 650 + ], + "score": 0.92, + "content": "S = ( S _ { l _ { 1 } } , S _ { l _ { 2 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 637, + 442, + 651 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 442, + 638, + 456, + 649 + ], + "score": 0.92, + "content": "S _ { l _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 637, + 475, + 651 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 475, + 638, + 488, + 649 + ], + "score": 0.89, + "content": "S _ { l _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "are", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 263, + 661 + ], + "score": 1.0, + "content": "translations of each other in languages", + "type": "text" + }, + { + "bbox": [ + 264, + 649, + 272, + 659 + ], + "score": 0.87, + "content": "l _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 648, + 291, + 661 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 291, + 649, + 300, + 659 + ], + "score": 0.86, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 648, + 461, + 661 + ], + "score": 1.0, + "content": ", the cross-lingual loss for a target word", + "type": "text" + }, + { + "bbox": [ + 462, + 650, + 473, + 659 + ], + "score": 0.85, + "content": "w _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 648, + 486, + 661 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 486, + 649, + 494, + 659 + ], + "score": 0.87, + "content": "l _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 659, + 144, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 144, + 673 + ], + "score": 1.0, + "content": "given by", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 626, + 506, + 673 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 669, + 376, + 698 + ], + "lines": [ + { + "bbox": [ + 235, + 669, + 376, + 698 + ], + "spans": [ + { + "bbox": [ + 235, + 669, + 376, + 698 + ], + "score": 0.94, + "content": "\\ell \\big ( { \\pmb u } _ { w _ { t } } ^ { \\top } { \\pmb v } _ { S _ { l _ { 2 } } } \\big ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell \\big ( - { \\pmb u } _ { w _ { t } ^ { \\prime } } ^ { \\top } { \\pmb v } _ { S _ { l _ { 2 } } } \\big )", + "type": "interline_equation", + "image_path": "c243177fd7c6b452de3f5e756d7406bfd6c15aa46bf4c87fa832e1c1d9acf376.jpg" + } + ] + } + ], + "index": 48.5, + "virtual_lines": [ + { + "bbox": [ + 235, + 669, + 376, + 683.5 + ], + "spans": [], + "index": 48 + }, + { + "bbox": [ + 235, + 683.5, + 376, + 698.0 + ], + "spans": [], + "index": 49 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 217, + 712 + ], + "score": 1.0, + "content": "Thus, we use the sentence", + "type": "text" + }, + { + "bbox": [ + 218, + 700, + 231, + 710 + ], + "score": 0.89, + "content": "S _ { l _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 698, + 378, + 712 + ], + "score": 1.0, + "content": "to predict the constituent words of", + "type": "text" + }, + { + "bbox": [ + 378, + 700, + 391, + 711 + ], + "score": 0.9, + "content": "S _ { l _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "and vice-versa in a similar", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 473, + 722 + ], + "score": 1.0, + "content": "fashion to the monolingual SENT2VEC, shown in Figure 1. This ensures that the word and", + "type": "text" + }, + { + "bbox": [ + 474, + 712, + 480, + 720 + ], + "score": 0.29, + "content": "\\mathbf { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "-gram", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 720, + 318, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 318, + 733 + ], + "score": 1.0, + "content": "embeddings of both languages lie in the same space.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 51, + "bbox_fs": [ + 105, + 698, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 82, + 502, + 148 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 82, + 502, + 148 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 82, + 502, + 148 + ], + "spans": [ + { + "bbox": [ + 110, + 82, + 502, + 148 + ], + "score": 0.954, + "type": "image", + "image_path": "447f2c4e9599c41e36d0eb4b008a68d7f6bba88815213ca6273055255c6f9dd4.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 82, + 502, + 104.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 104.0, + 502, + 126.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 126.0, + 502, + 148.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 159, + 506, + 193 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "Figure 1: An illustration of the BI-SENT2VEC training process. A word from a sentence pair is", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 170, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 170, + 506, + 182 + ], + "score": 1.0, + "content": "chosen as a target and the algorithm learns to predict it using the rest of the sentence(monolingual", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 181, + 435, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 435, + 194 + ], + "score": 1.0, + "content": "training component) and the translation of the sentence(cross-lingual component).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 503, + 236 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 504, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 150, + 227 + ], + "score": 1.0, + "content": "Assuming", + "type": "text" + }, + { + "bbox": [ + 150, + 214, + 160, + 224 + ], + "score": 0.78, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 213, + 504, + 227 + ], + "score": 1.0, + "content": "to be a sentence aligned bilingual corpus and combining equations (1) and (2), our", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 225, + 336, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 336, + 237 + ], + "score": 1.0, + "content": "BI-SENT2VEC model objective function is formulated as", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "interline_equation", + "bbox": [ + 107, + 241, + 500, + 286 + ], + "lines": [ + { + "bbox": [ + 107, + 241, + 500, + 286 + ], + "spans": [ + { + "bbox": [ + 107, + 241, + 500, + 286 + ], + "score": 0.8, + "content": "\\operatorname* { m i n } _ { U , V } \\sum _ { \\underbrace { l , l ^ { \\prime } \\in [ l _ { 1 } , l _ { 2 } ] } _ { l \\neq l ^ { \\prime } } } \\sum _ { w _ { t } \\in S _ { l } \\atop l \\neq l ^ { \\prime } } \\left( \\underbrace { \\ell ( u _ { w _ { t } } ^ { \\top } v _ { S _ { l } \\backslash \\{ w _ { t } \\} } ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell ( - u _ { w _ { t } ^ { \\prime } } v _ { S _ { l } \\backslash \\{ w _ { t } \\} } ) } _ { \\mathrm { m o n o i n g u a l ~ l o s s } } + \\underbrace { \\ell ( u _ { w _ { t } } ^ { \\top } v _ { S _ { l ^ { \\prime } } } ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell ( - u _ { w _ { t } ^ { \\prime } } v _ { S _ { l ^ { \\prime } } } ) } _ { \\mathrm { c r o s s i n g u a l ~ l o s s } } \\right)", + "type": "interline_equation", + "image_path": "5bb3669b20c8bc86b17d6b82f9a667ac7f8d629adf9e63c00475c2b0e2226775.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 107, + 241, + 500, + 256.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 107, + 256.0, + 500, + 271.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 107, + 271.0, + 500, + 286.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 299, + 504, + 333 + ], + "lines": [ + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 279, + 312 + ], + "score": 1.0, + "content": "Implementation Details. We build our", + "type": "text" + }, + { + "bbox": [ + 279, + 300, + 299, + 310 + ], + "score": 0.86, + "content": "\\mathrm { C } { + } { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 300, + 504, + 312 + ], + "score": 1.0, + "content": "implementation on the top of the FASTTEXT li-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "score": 1.0, + "content": "brary (Bojanowski et al., 2016; Joulin et al., 2016). Model parameters are updated by asynchronous", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 321, + 282, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 282, + 334 + ], + "score": 1.0, + "content": "SGD with a linearly decaying learning rate.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 337, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 351 + ], + "score": 1.0, + "content": "Our model is trained on the ParaCrawl (Espla-Gomis, 2019) v4.0 datasets for the English-Italian, `", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "English-German, English-French, English-Spanish, English-Hungarian and English-Finnish lan-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "score": 1.0, + "content": "guage pairs. For the English-Russian language pair, we concatenate the OpenSubtitle corpus1(Lison", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 505, + 384 + ], + "score": 1.0, + "content": "& Tiedemann, 2016) and the Tanzil project2 (Quran translations) corpus. The number of paral-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 381, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 104, + 381, + 506, + 396 + ], + "score": 1.0, + "content": "lel sentence pairs in the corpora except for those of English-Finnish and English-Hungarian used", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 393, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 407 + ], + "score": 1.0, + "content": "by us range from 17-32 Million. Number of parallel sentence pairs for the dis-similar language", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 404, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 104, + 404, + 506, + 417 + ], + "score": 1.0, + "content": "pairs(English-Hungarian and English-Finnish) is approximately 2 million. Evaluation results for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "these two language pairs can be found in Subsection 4.4. Exact statistics regarding the different", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 425, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 505, + 440 + ], + "score": 1.0, + "content": "corpora can be found in the Table 7 in the Appendix. All the sentences were tokenized using Spacy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 436, + 276, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 276, + 450 + ], + "score": 1.0, + "content": "tokenizers3 for their respective languages.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 453, + 505, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 506, + 466 + ], + "score": 1.0, + "content": "For each dataset, we trained two different models: one with unigram embeddings only, and the other", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "score": 1.0, + "content": "additionally augmented with bigrams. The earlier TRANSGRAM models (Coulmance et al., 2015)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "score": 1.0, + "content": "were trained on a small amount of data (Europarl Corpus (Koehn, 2005)). To facilitate a fair com-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "parison, we train new TRANSGRAM embeddings on the same data used for BI-SENT2VEC. Given", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 499, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 505, + 510 + ], + "score": 1.0, + "content": "that TRANSGRAM and BI-SENT2VEC are a cross-lingual extension of Skipgram and SENT2VEC re-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "spectively, we use the same parameters as (Bojanowski et al., 2016) and (Gupta et al., 2019), except", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 520, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 533 + ], + "score": 1.0, + "content": "increasing the number of epochs for TRANSGRAM to 8, and decreasing the same for BI-SENT2VEC", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 531, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 544 + ], + "score": 1.0, + "content": "to 5. Additionally, a preliminary hyperparameter search (except changing the number of epochs)", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "score": 1.0, + "content": "on BI-SENT2VEC and TRANSGRAM did not improve the results. All parameters for training the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 552, + 456, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 456, + 565 + ], + "score": 1.0, + "content": "TRANSGRAM and BI-SENT2VEC models can be found in the Table 6 in the Appendix.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 570, + 505, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 569, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 505, + 583 + ], + "score": 1.0, + "content": "In order to make the comparison more extensive, we also train VECMAP (mapping-based) (Artetxe", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 579, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 505, + 595 + ], + "score": 1.0, + "content": "et al., 2018b;a) and BIVEC (joint-training)(Luong et al., 2015) methods on the same corpora using", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 592, + 291, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 291, + 603 + ], + "score": 1.0, + "content": "the exact pipeline as (Ormazabal et al., 2019).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 108, + 615, + 193, + 627 + ], + "lines": [ + { + "bbox": [ + 105, + 613, + 195, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 195, + 630 + ], + "score": 1.0, + "content": "4 EVALUATION", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 636, + 504, + 659 + ], + "lines": [ + { + "bbox": [ + 106, + 636, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 649 + ], + "score": 1.0, + "content": "To assess the quality of the word and sentence embeddings obtained as well as their cross-lingual", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 647, + 426, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 426, + 660 + ], + "score": 1.0, + "content": "alignment quality, we compare our results using the following four benchmarks", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 132, + 669, + 307, + 692 + ], + "lines": [ + { + "bbox": [ + 132, + 669, + 256, + 681 + ], + "spans": [ + { + "bbox": [ + 132, + 669, + 256, + 681 + ], + "score": 1.0, + "content": "• Cross-lingual word retrieval", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 132, + 679, + 308, + 693 + ], + "spans": [ + { + "bbox": [ + 132, + 679, + 308, + 693 + ], + "score": 1.0, + "content": "• Monolingual word representation quality", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 699, + 280, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 698, + 281, + 712 + ], + "spans": [ + { + "bbox": [ + 119, + 698, + 281, + 712 + ], + "score": 1.0, + "content": "1http://www.opensubtitles.org/", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 708, + 222, + 723 + ], + "spans": [ + { + "bbox": [ + 118, + 708, + 222, + 723 + ], + "score": 1.0, + "content": "2http://tanzil.net/", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 719, + 217, + 734 + ], + "spans": [ + { + "bbox": [ + 118, + 719, + 217, + 734 + ], + "score": 1.0, + "content": "3https://spacy.io/", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 82, + 502, + 148 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 82, + 502, + 148 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 82, + 502, + 148 + ], + "spans": [ + { + "bbox": [ + 110, + 82, + 502, + 148 + ], + "score": 0.954, + "type": "image", + "image_path": "447f2c4e9599c41e36d0eb4b008a68d7f6bba88815213ca6273055255c6f9dd4.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 82, + 502, + 104.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 104.0, + 502, + 126.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 126.0, + 502, + 148.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 159, + 506, + 193 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "Figure 1: An illustration of the BI-SENT2VEC training process. A word from a sentence pair is", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 170, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 170, + 506, + 182 + ], + "score": 1.0, + "content": "chosen as a target and the algorithm learns to predict it using the rest of the sentence(monolingual", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 181, + 435, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 435, + 194 + ], + "score": 1.0, + "content": "training component) and the translation of the sentence(cross-lingual component).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 503, + 236 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 504, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 150, + 227 + ], + "score": 1.0, + "content": "Assuming", + "type": "text" + }, + { + "bbox": [ + 150, + 214, + 160, + 224 + ], + "score": 0.78, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 213, + 504, + 227 + ], + "score": 1.0, + "content": "to be a sentence aligned bilingual corpus and combining equations (1) and (2), our", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 225, + 336, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 336, + 237 + ], + "score": 1.0, + "content": "BI-SENT2VEC model objective function is formulated as", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 213, + 504, + 237 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 107, + 241, + 500, + 286 + ], + "lines": [ + { + "bbox": [ + 107, + 241, + 500, + 286 + ], + "spans": [ + { + "bbox": [ + 107, + 241, + 500, + 286 + ], + "score": 0.8, + "content": "\\operatorname* { m i n } _ { U , V } \\sum _ { \\underbrace { l , l ^ { \\prime } \\in [ l _ { 1 } , l _ { 2 } ] } _ { l \\neq l ^ { \\prime } } } \\sum _ { w _ { t } \\in S _ { l } \\atop l \\neq l ^ { \\prime } } \\left( \\underbrace { \\ell ( u _ { w _ { t } } ^ { \\top } v _ { S _ { l } \\backslash \\{ w _ { t } \\} } ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell ( - u _ { w _ { t } ^ { \\prime } } v _ { S _ { l } \\backslash \\{ w _ { t } \\} } ) } _ { \\mathrm { m o n o i n g u a l ~ l o s s } } + \\underbrace { \\ell ( u _ { w _ { t } } ^ { \\top } v _ { S _ { l ^ { \\prime } } } ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell ( - u _ { w _ { t } ^ { \\prime } } v _ { S _ { l ^ { \\prime } } } ) } _ { \\mathrm { c r o s s i n g u a l ~ l o s s } } \\right)", + "type": "interline_equation", + "image_path": "5bb3669b20c8bc86b17d6b82f9a667ac7f8d629adf9e63c00475c2b0e2226775.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 107, + 241, + 500, + 256.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 107, + 256.0, + 500, + 271.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 107, + 271.0, + 500, + 286.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 299, + 504, + 333 + ], + "lines": [ + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 279, + 312 + ], + "score": 1.0, + "content": "Implementation Details. We build our", + "type": "text" + }, + { + "bbox": [ + 279, + 300, + 299, + 310 + ], + "score": 0.86, + "content": "\\mathrm { C } { + } { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 300, + 504, + 312 + ], + "score": 1.0, + "content": "implementation on the top of the FASTTEXT li-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "score": 1.0, + "content": "brary (Bojanowski et al., 2016; Joulin et al., 2016). Model parameters are updated by asynchronous", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 321, + 282, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 282, + 334 + ], + "score": 1.0, + "content": "SGD with a linearly decaying learning rate.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 106, + 300, + 505, + 334 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 337, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 351 + ], + "score": 1.0, + "content": "Our model is trained on the ParaCrawl (Espla-Gomis, 2019) v4.0 datasets for the English-Italian, `", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "English-German, English-French, English-Spanish, English-Hungarian and English-Finnish lan-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "score": 1.0, + "content": "guage pairs. For the English-Russian language pair, we concatenate the OpenSubtitle corpus1(Lison", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 505, + 384 + ], + "score": 1.0, + "content": "& Tiedemann, 2016) and the Tanzil project2 (Quran translations) corpus. The number of paral-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 381, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 104, + 381, + 506, + 396 + ], + "score": 1.0, + "content": "lel sentence pairs in the corpora except for those of English-Finnish and English-Hungarian used", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 393, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 407 + ], + "score": 1.0, + "content": "by us range from 17-32 Million. Number of parallel sentence pairs for the dis-similar language", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 404, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 104, + 404, + 506, + 417 + ], + "score": 1.0, + "content": "pairs(English-Hungarian and English-Finnish) is approximately 2 million. Evaluation results for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "these two language pairs can be found in Subsection 4.4. Exact statistics regarding the different", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 425, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 505, + 440 + ], + "score": 1.0, + "content": "corpora can be found in the Table 7 in the Appendix. All the sentences were tokenized using Spacy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 436, + 276, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 276, + 450 + ], + "score": 1.0, + "content": "tokenizers3 for their respective languages.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 18.5, + "bbox_fs": [ + 104, + 337, + 506, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 453, + 505, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 506, + 466 + ], + "score": 1.0, + "content": "For each dataset, we trained two different models: one with unigram embeddings only, and the other", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "score": 1.0, + "content": "additionally augmented with bigrams. The earlier TRANSGRAM models (Coulmance et al., 2015)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "score": 1.0, + "content": "were trained on a small amount of data (Europarl Corpus (Koehn, 2005)). To facilitate a fair com-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "parison, we train new TRANSGRAM embeddings on the same data used for BI-SENT2VEC. Given", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 499, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 505, + 510 + ], + "score": 1.0, + "content": "that TRANSGRAM and BI-SENT2VEC are a cross-lingual extension of Skipgram and SENT2VEC re-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "spectively, we use the same parameters as (Bojanowski et al., 2016) and (Gupta et al., 2019), except", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 520, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 533 + ], + "score": 1.0, + "content": "increasing the number of epochs for TRANSGRAM to 8, and decreasing the same for BI-SENT2VEC", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 531, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 544 + ], + "score": 1.0, + "content": "to 5. Additionally, a preliminary hyperparameter search (except changing the number of epochs)", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "score": 1.0, + "content": "on BI-SENT2VEC and TRANSGRAM did not improve the results. All parameters for training the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 552, + 456, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 456, + 565 + ], + "score": 1.0, + "content": "TRANSGRAM and BI-SENT2VEC models can be found in the Table 6 in the Appendix.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 454, + 506, + 565 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 570, + 505, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 569, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 505, + 583 + ], + "score": 1.0, + "content": "In order to make the comparison more extensive, we also train VECMAP (mapping-based) (Artetxe", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 579, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 505, + 595 + ], + "score": 1.0, + "content": "et al., 2018b;a) and BIVEC (joint-training)(Luong et al., 2015) methods on the same corpora using", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 592, + 291, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 291, + 603 + ], + "score": 1.0, + "content": "the exact pipeline as (Ormazabal et al., 2019).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 569, + 505, + 603 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 615, + 193, + 627 + ], + "lines": [ + { + "bbox": [ + 105, + 613, + 195, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 195, + 630 + ], + "score": 1.0, + "content": "4 EVALUATION", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 636, + 504, + 659 + ], + "lines": [ + { + "bbox": [ + 106, + 636, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 649 + ], + "score": 1.0, + "content": "To assess the quality of the word and sentence embeddings obtained as well as their cross-lingual", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 647, + 426, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 426, + 660 + ], + "score": 1.0, + "content": "alignment quality, we compare our results using the following four benchmarks", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 106, + 636, + 505, + 660 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 669, + 307, + 692 + ], + "lines": [ + { + "bbox": [ + 132, + 669, + 256, + 681 + ], + "spans": [ + { + "bbox": [ + 132, + 669, + 256, + 681 + ], + "score": 1.0, + "content": "• Cross-lingual word retrieval", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 132, + 679, + 308, + 693 + ], + "spans": [ + { + "bbox": [ + 132, + 679, + 308, + 693 + ], + "score": 1.0, + "content": "• Monolingual word representation quality", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 132, + 669, + 308, + 693 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 133, + 82, + 364, + 105 + ], + "lines": [ + { + "bbox": [ + 132, + 82, + 270, + 94 + ], + "spans": [ + { + "bbox": [ + 132, + 82, + 270, + 94 + ], + "score": 1.0, + "content": "• Cross-lingual sentence retrieval", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 132, + 93, + 365, + 106 + ], + "spans": [ + { + "bbox": [ + 132, + 93, + 365, + 106 + ], + "score": 1.0, + "content": "• Zero-shot cross-lingual transfer of document classifiers", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 117, + 505, + 150 + ], + "lines": [ + { + "bbox": [ + 105, + 115, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 130 + ], + "score": 1.0, + "content": "where benchmarks are presented in order of increasing linguistic granularity, i.e. word, sentence,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 505, + 141 + ], + "score": 1.0, + "content": "and document level. We also analyze the effect of training data by studying the relationship between", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 139, + 262, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 262, + 151 + ], + "score": 1.0, + "content": "representation quality and corpus size.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 108, + 155, + 504, + 178 + ], + "lines": [ + { + "bbox": [ + 106, + 155, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 505, + 168 + ], + "score": 1.0, + "content": "We use the code available in the MUSE library4 (Conneau et al., 2017) for all evaluations except the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 166, + 459, + 179 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 459, + 179 + ], + "score": 1.0, + "content": "zero-shot classifier transfer, which is tested on the MLDoc task (Schwenk & Li, 2018)5.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "title", + "bbox": [ + 108, + 189, + 224, + 200 + ], + "lines": [ + { + "bbox": [ + 105, + 188, + 225, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 188, + 225, + 201 + ], + "score": 1.0, + "content": "4.1 WORD TRANSLATION", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 208, + 505, + 253 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 221 + ], + "score": 1.0, + "content": "The task involves retrieving correct translation(s) of a word in a source language from a target lan-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "score": 1.0, + "content": "guage. To evaluate translation accuracy, we use the bilingual dictionaries constructed by (Conneau", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 230, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 325, + 243 + ], + "score": 1.0, + "content": "et al., 2017). We consider 1500 source-test queries and", + "type": "text" + }, + { + "bbox": [ + 325, + 230, + 347, + 241 + ], + "score": 0.57, + "content": "2 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 230, + 505, + 243 + ], + "score": 1.0, + "content": "target words for each language pair and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 241, + 501, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 132, + 253 + ], + "score": 1.0, + "content": "report", + "type": "text" + }, + { + "bbox": [ + 132, + 241, + 154, + 252 + ], + "score": 0.75, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 242, + 501, + 253 + ], + "score": 1.0, + "content": "scores for the supervised and unsupervised baselines as well as our models in Table 1.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "table", + "bbox": [ + 106, + 264, + 504, + 436 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 264, + 504, + 436 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 264, + 504, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 264, + 504, + 436 + ], + "score": 0.984, + "html": "
Methoden-esen-fren-deen-ruen-itavg.
→↑→↑→↑→↑→↑
MUSE (Conneau et al.,2017)81.7 83.382.3 82.174.0 72.244.0 59.178.6 77.973.5
UMWE(Chen & Cardie,2018)82.5 83.182.5 82.174.6 72.549.5 61.778.3 77.074.4
Procrustes + refine (Conneau et al., 2017)82.4 83.982.3 83.275.3 73.250.1 63.577.5 77.674.9
RCSLS (Joulin et al., 2018)83.7 87.184.1 84.779.2 77.560.9 70.281.1 82.779.1
TRANSGRAM (Coulmance et al., 2015)91.6 88.689.1 90.187.5 87.265.6 73.788.6 89.585.2
VECMAP (unsupervised) (Artetxe et al.,2018b)87.4 87.888.3 88.584.3 87.248.6 50.587.4 86.579.6
VECMAP (supervised) (Artetxe et al.,2018a)87.2 90.287.6 90.487.3 86.849.7 65.687.2 89.282.1
BIVEC NN (Luong et al., 2015)87.4 88.686.8 89.187.5 87.264.0 59.186.8 84.081.7
BIVEC CSLS (Luong et al., 2015)87.6 89.188.8 90.386.4 87.266.1 70.687.6 87.884.3
BI-SENT2VEC uni. NN86.9 91.686.9 91.086.0 88.758.0 72.888.3 92.484.3
BI-SENT2VEC uni. + bi. NN89.4 92.989.3 92.886.7 89.359.0 70.289.5 91.885.1
BI-SENT2VEC uni.CSLS86.0 91.786.4 91.484.6 88.860.5 73.088.2 91.884.2
BI-SENT2VEC uni. + bi. CSLS89.0 92.188.9 92.486.5 89.061.0 73.589.6 91.485.3
", + "type": "table", + "image_path": "3dcea9e7aa1406f38ea34781b8850b294bca917fba2f98685efd6ef2b933ea6d.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 106, + 264, + 504, + 321.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 106, + 321.3333333333333, + 504, + 378.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 378.66666666666663, + 504, + 435.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 107, + 446, + 505, + 502 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 445, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 258, + 459 + ], + "score": 1.0, + "content": "Table 1: Word translation retrieval", + "type": "text" + }, + { + "bbox": [ + 258, + 446, + 280, + 457 + ], + "score": 0.67, + "content": "\\mathbf { P } @ \\mathbf { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 445, + 505, + 459 + ], + "score": 1.0, + "content": "for various language pairs of MUSE evaluation dic-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 457, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 505, + 470 + ], + "score": 1.0, + "content": "tionary (Conneau et al., 2017). NN: nearest neighbours. CSLS: Cross-Domain Similarity Local", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 468, + 504, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 504, + 480 + ], + "score": 1.0, + "content": "Scaling. (‘en’ is English, ‘fr’ is French, ‘de’ is German, ‘ru’ is Russian, ‘it’ is Italian) (‘uni.’ and", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 307, + 492 + ], + "score": 1.0, + "content": "‘bi.’ denote unigrams and bigrams respectively)", + "type": "text" + }, + { + "bbox": [ + 307, + 480, + 320, + 489 + ], + "score": 0.81, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "denotes translation from the first language to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 489, + 273, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 167, + 502 + ], + "score": 1.0, + "content": "the second and", + "type": "text" + }, + { + "bbox": [ + 167, + 491, + 180, + 500 + ], + "score": 0.83, + "content": "\\gets", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 489, + 273, + 502 + ], + "score": 1.0, + "content": "the other way around.)", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17 + } + ], + "index": 15.0 + }, + { + "type": "title", + "bbox": [ + 107, + 521, + 350, + 532 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 351, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 351, + 534 + ], + "score": 1.0, + "content": "4.2 MONOLINGUAL WORD REPRESENTATION QUALITY", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 553 + ], + "score": 1.0, + "content": "We assess the monolingual quality improvement of our proposed cross-lingual training by evaluating", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 551, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 563 + ], + "score": 1.0, + "content": "performance on monolingual word similarity tasks. To disentangle the specific contribution of the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 562, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 575 + ], + "score": 1.0, + "content": "cross-lingual loss, we train the monolingual counterpart of BI-SENT2VEC, SENT2VEC on the same", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 574, + 201, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 201, + 584 + ], + "score": 1.0, + "content": "corpora as our method.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 589, + 505, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "Performance on monolingual word-similarity tasks is evaluated using the English SimLex-999 (Hill", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "et al., 2014) and its Italian and German translations, English WS-353 (Finkelstein et al., 2001) and its", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 612, + 504, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 504, + 623 + ], + "score": 1.0, + "content": "German, Italian and Spanish translations. For French, we use a translation of the RG-65 (Joubarne &", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "score": 1.0, + "content": "Inkpen, 2011) dataset. Pearson scores are used to measure the correlation between human-annotated", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 634, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 506, + 645 + ], + "score": 1.0, + "content": "word similarities and predicted cosine similarities. We also include FASTTEXT monolingual vectors", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 645, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 505, + 657 + ], + "score": 1.0, + "content": "trained on CommonCrawl data (Grave et al., 2018a) which is comprised of 600 billion, 68 billion, 66", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 654, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 669 + ], + "score": 1.0, + "content": "billion, 72 billion and 36 billion words of English, French, German, Spanish and Italian respectively", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 666, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 163, + 679 + ], + "score": 1.0, + "content": "and is at least", + "type": "text" + }, + { + "bbox": [ + 163, + 667, + 187, + 677 + ], + "score": 0.88, + "content": "1 0 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 666, + 505, + 679 + ], + "score": 1.0, + "content": "larger than the corpora on which we trained BI-SENT2VEC. We report Pearson", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "correlation scores on different word-similarity datasets for En-It pair in Table 2. Evaluation results", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 689, + 472, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 472, + 700 + ], + "score": 1.0, + "content": "on other language pairs are similar and can be found in the appendix in Tables 8, 9, and 10.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 711, + 345, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 708, + 341, + 723 + ], + "spans": [ + { + "bbox": [ + 118, + 708, + 341, + 723 + ], + "score": 1.0, + "content": "4https://github.com/facebookresearch/MUSE", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 719, + 346, + 733 + ], + "spans": [ + { + "bbox": [ + 118, + 719, + 346, + 733 + ], + "score": 1.0, + "content": "5https://github.com/facebookresearch/MLDoc", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 133, + 82, + 364, + 105 + ], + "lines": [ + { + "bbox": [ + 132, + 82, + 270, + 94 + ], + "spans": [ + { + "bbox": [ + 132, + 82, + 270, + 94 + ], + "score": 1.0, + "content": "• Cross-lingual sentence retrieval", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 132, + 93, + 365, + 106 + ], + "spans": [ + { + "bbox": [ + 132, + 93, + 365, + 106 + ], + "score": 1.0, + "content": "• Zero-shot cross-lingual transfer of document classifiers", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 132, + 82, + 365, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 117, + 505, + 150 + ], + "lines": [ + { + "bbox": [ + 105, + 115, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 130 + ], + "score": 1.0, + "content": "where benchmarks are presented in order of increasing linguistic granularity, i.e. word, sentence,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 505, + 141 + ], + "score": 1.0, + "content": "and document level. We also analyze the effect of training data by studying the relationship between", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 139, + 262, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 262, + 151 + ], + "score": 1.0, + "content": "representation quality and corpus size.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 115, + 505, + 151 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 155, + 504, + 178 + ], + "lines": [ + { + "bbox": [ + 106, + 155, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 505, + 168 + ], + "score": 1.0, + "content": "We use the code available in the MUSE library4 (Conneau et al., 2017) for all evaluations except the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 166, + 459, + 179 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 459, + 179 + ], + "score": 1.0, + "content": "zero-shot classifier transfer, which is tested on the MLDoc task (Schwenk & Li, 2018)5.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 106, + 155, + 505, + 179 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 189, + 224, + 200 + ], + "lines": [ + { + "bbox": [ + 105, + 188, + 225, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 188, + 225, + 201 + ], + "score": 1.0, + "content": "4.1 WORD TRANSLATION", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 208, + 505, + 253 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 221 + ], + "score": 1.0, + "content": "The task involves retrieving correct translation(s) of a word in a source language from a target lan-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "score": 1.0, + "content": "guage. To evaluate translation accuracy, we use the bilingual dictionaries constructed by (Conneau", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 230, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 325, + 243 + ], + "score": 1.0, + "content": "et al., 2017). We consider 1500 source-test queries and", + "type": "text" + }, + { + "bbox": [ + 325, + 230, + 347, + 241 + ], + "score": 0.57, + "content": "2 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 230, + 505, + 243 + ], + "score": 1.0, + "content": "target words for each language pair and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 241, + 501, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 132, + 253 + ], + "score": 1.0, + "content": "report", + "type": "text" + }, + { + "bbox": [ + 132, + 241, + 154, + 252 + ], + "score": 0.75, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 242, + 501, + 253 + ], + "score": 1.0, + "content": "scores for the supervised and unsupervised baselines as well as our models in Table 1.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 207, + 505, + 253 + ] + }, + { + "type": "table", + "bbox": [ + 106, + 264, + 504, + 436 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 264, + 504, + 436 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 264, + 504, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 264, + 504, + 436 + ], + "score": 0.984, + "html": "
Methoden-esen-fren-deen-ruen-itavg.
→↑→↑→↑→↑→↑
MUSE (Conneau et al.,2017)81.7 83.382.3 82.174.0 72.244.0 59.178.6 77.973.5
UMWE(Chen & Cardie,2018)82.5 83.182.5 82.174.6 72.549.5 61.778.3 77.074.4
Procrustes + refine (Conneau et al., 2017)82.4 83.982.3 83.275.3 73.250.1 63.577.5 77.674.9
RCSLS (Joulin et al., 2018)83.7 87.184.1 84.779.2 77.560.9 70.281.1 82.779.1
TRANSGRAM (Coulmance et al., 2015)91.6 88.689.1 90.187.5 87.265.6 73.788.6 89.585.2
VECMAP (unsupervised) (Artetxe et al.,2018b)87.4 87.888.3 88.584.3 87.248.6 50.587.4 86.579.6
VECMAP (supervised) (Artetxe et al.,2018a)87.2 90.287.6 90.487.3 86.849.7 65.687.2 89.282.1
BIVEC NN (Luong et al., 2015)87.4 88.686.8 89.187.5 87.264.0 59.186.8 84.081.7
BIVEC CSLS (Luong et al., 2015)87.6 89.188.8 90.386.4 87.266.1 70.687.6 87.884.3
BI-SENT2VEC uni. NN86.9 91.686.9 91.086.0 88.758.0 72.888.3 92.484.3
BI-SENT2VEC uni. + bi. NN89.4 92.989.3 92.886.7 89.359.0 70.289.5 91.885.1
BI-SENT2VEC uni.CSLS86.0 91.786.4 91.484.6 88.860.5 73.088.2 91.884.2
BI-SENT2VEC uni. + bi. CSLS89.0 92.188.9 92.486.5 89.061.0 73.589.6 91.485.3
", + "type": "table", + "image_path": "3dcea9e7aa1406f38ea34781b8850b294bca917fba2f98685efd6ef2b933ea6d.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 106, + 264, + 504, + 321.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 106, + 321.3333333333333, + 504, + 378.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 378.66666666666663, + 504, + 435.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 107, + 446, + 505, + 502 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 445, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 258, + 459 + ], + "score": 1.0, + "content": "Table 1: Word translation retrieval", + "type": "text" + }, + { + "bbox": [ + 258, + 446, + 280, + 457 + ], + "score": 0.67, + "content": "\\mathbf { P } @ \\mathbf { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 445, + 505, + 459 + ], + "score": 1.0, + "content": "for various language pairs of MUSE evaluation dic-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 457, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 505, + 470 + ], + "score": 1.0, + "content": "tionary (Conneau et al., 2017). NN: nearest neighbours. CSLS: Cross-Domain Similarity Local", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 468, + 504, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 504, + 480 + ], + "score": 1.0, + "content": "Scaling. (‘en’ is English, ‘fr’ is French, ‘de’ is German, ‘ru’ is Russian, ‘it’ is Italian) (‘uni.’ and", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 307, + 492 + ], + "score": 1.0, + "content": "‘bi.’ denote unigrams and bigrams respectively)", + "type": "text" + }, + { + "bbox": [ + 307, + 480, + 320, + 489 + ], + "score": 0.81, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "denotes translation from the first language to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 489, + 273, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 167, + 502 + ], + "score": 1.0, + "content": "the second and", + "type": "text" + }, + { + "bbox": [ + 167, + 491, + 180, + 500 + ], + "score": 0.83, + "content": "\\gets", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 489, + 273, + 502 + ], + "score": 1.0, + "content": "the other way around.)", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17 + } + ], + "index": 15.0 + }, + { + "type": "title", + "bbox": [ + 107, + 521, + 350, + 532 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 351, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 351, + 534 + ], + "score": 1.0, + "content": "4.2 MONOLINGUAL WORD REPRESENTATION QUALITY", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 553 + ], + "score": 1.0, + "content": "We assess the monolingual quality improvement of our proposed cross-lingual training by evaluating", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 551, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 563 + ], + "score": 1.0, + "content": "performance on monolingual word similarity tasks. To disentangle the specific contribution of the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 562, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 575 + ], + "score": 1.0, + "content": "cross-lingual loss, we train the monolingual counterpart of BI-SENT2VEC, SENT2VEC on the same", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 574, + 201, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 201, + 584 + ], + "score": 1.0, + "content": "corpora as our method.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 539, + 505, + 584 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 589, + 505, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "Performance on monolingual word-similarity tasks is evaluated using the English SimLex-999 (Hill", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "et al., 2014) and its Italian and German translations, English WS-353 (Finkelstein et al., 2001) and its", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 612, + 504, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 504, + 623 + ], + "score": 1.0, + "content": "German, Italian and Spanish translations. For French, we use a translation of the RG-65 (Joubarne &", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "score": 1.0, + "content": "Inkpen, 2011) dataset. Pearson scores are used to measure the correlation between human-annotated", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 634, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 506, + 645 + ], + "score": 1.0, + "content": "word similarities and predicted cosine similarities. We also include FASTTEXT monolingual vectors", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 645, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 505, + 657 + ], + "score": 1.0, + "content": "trained on CommonCrawl data (Grave et al., 2018a) which is comprised of 600 billion, 68 billion, 66", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 654, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 669 + ], + "score": 1.0, + "content": "billion, 72 billion and 36 billion words of English, French, German, Spanish and Italian respectively", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 666, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 163, + 679 + ], + "score": 1.0, + "content": "and is at least", + "type": "text" + }, + { + "bbox": [ + 163, + 667, + 187, + 677 + ], + "score": 0.88, + "content": "1 0 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 666, + 505, + 679 + ], + "score": 1.0, + "content": "larger than the corpora on which we trained BI-SENT2VEC. We report Pearson", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "correlation scores on different word-similarity datasets for En-It pair in Table 2. Evaluation results", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 689, + 472, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 472, + 700 + ], + "score": 1.0, + "content": "on other language pairs are similar and can be found in the appendix in Tables 8, 9, and 10.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 590, + 506, + 700 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 180, + 76, + 431, + 218 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 180, + 76, + 431, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 76, + 431, + 218 + ], + "spans": [ + { + "bbox": [ + 180, + 76, + 431, + 218 + ], + "score": 0.98, + "html": "
Method\\DatasetSimLex-999WS-353
en iten it
MUSE0.380.300.74 0.64
RCSLS0.38 0.300.740.64
FASTTEXT- Common Crawl0.49 0.320.750.57
BIVEC0.40 0.360.700.60
TRANSGRAM0.43 0.370.730.63
SENT2VEC uni.0.49 0.380.730.60
BI-SENT2VEC uni.0.570.47 0.790.65
BI-SENT2VEC uni. + bi.0.580.50 0.800.69
", + "type": "table", + "image_path": "283fd29dd3c6466d32f63f2aa830962d367d5512f1bdda7a83e27aa8a345bd28.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 180, + 76, + 431, + 123.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 180, + 123.33333333333334, + 431, + 170.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 180, + 170.66666666666669, + 431, + 218.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 229, + 502, + 252 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 229, + 504, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 229, + 504, + 241 + ], + "score": 1.0, + "content": "Table 2: Monolingual word similarity task performance of our methods when trained on en-it", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 239, + 332, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 332, + 253 + ], + "score": 1.0, + "content": "ParaCrawl data. We report Pearson correlation scores.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 107, + 265, + 303, + 276 + ], + "lines": [ + { + "bbox": [ + 105, + 263, + 304, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 304, + 278 + ], + "score": 1.0, + "content": "4.3 CROSS-LINGUAL SENTENCE RETRIEVAL", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 282, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 106, + 283, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 505, + 296 + ], + "score": 1.0, + "content": "The primary contribution of our work is to deliver improved cross-lingual sentence representations.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 294, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 294, + 506, + 307 + ], + "score": 1.0, + "content": "We test sentence embeddings for each method obtained by bag-of-words composition for sentence", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 304, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 104, + 304, + 506, + 319 + ], + "score": 1.0, + "content": "retrieval across different languages on the Europarl corpus. In particular, the tf-idf weighted average", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 317, + 504, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 504, + 328 + ], + "score": 1.0, + "content": "is used to construct sentence embeddings from word embeddings. We consider 2000 sentences in the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 326, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 505, + 341 + ], + "score": 1.0, + "content": "source language dataset and retrieve their translation among 200K sentences in the target language", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "dataset. The other 300K sentences in the Europarl corpus are used to calculate tf-idf weights. Results", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 349, + 475, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 120, + 360 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 120, + 349, + 142, + 359 + ], + "score": 0.85, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 349, + 475, + 360 + ], + "score": 1.0, + "content": "of unsupervised and supervised benchmarks vs our models are included in Table 3.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9 + }, + { + "type": "table", + "bbox": [ + 106, + 370, + 503, + 556 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 370, + 503, + 556 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 370, + 503, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 503, + 556 + ], + "score": 0.92, + "html": "
Methoden-esen-fren-deen-itavg.
→↑→↑→↑→↑
MUSE72.771.569.268.853.353.466.164.364.9
RCSLS26.926.719.321.28.811.315.117.618.4
TRANSGRAM83.581.480.481.664.869.977.277.977.1
VECMAP (unsupervised)81.782.179.880.462.864.669.071.174.0
VECMAP (supervised)81.38180.480.762.664.367.87173.6
BIVEC NN69.877.154.775.556.144.158.245.160.1
BIVEC CSLS81.683.478.181.671.668.174.272.476.4
BI-SENT2VEC uni. NN87.886.485.283.482.380.285.985.884.6
BI-SENT2VEC uni. + bi. NN87.987.886.183.979.579.785.185.384.4
BI-SENT2VEC uni. CSLS89.588.587.186.484.483.088.287.586.8
BI-SENT2VEC uni. + bi. CSLS89.789.687.887.484.284.087.987.687.3
Reduction in error37.5% 37.3%37.8% 31.5%44.4% 46.8%46.9% 43.9%1
", + "type": "table", + "image_path": "a66fd06c0021769a89fe836ddbc7a35cea2224df45f811fb2a378d697d07753b.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 106, + 370, + 503, + 432.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 432.0, + 503, + 494.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 106, + 494.0, + 503, + 556.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 107, + 563, + 505, + 596 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 321, + 577 + ], + "score": 1.0, + "content": "Table 3: Cross-lingual Sentence retrieval. We report", + "type": "text" + }, + { + "bbox": [ + 321, + 564, + 343, + 574 + ], + "score": 0.79, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 562, + 505, + 577 + ], + "score": 1.0, + "content": "scores for 2000 source queries searching", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 574, + 504, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 504, + 586 + ], + "score": 1.0, + "content": "over 200 000 target sentences. Reduction in error is calculated with respect to BI-SENT2VEC uni.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 586, + 320, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 586, + 114, + 596 + ], + "score": 0.41, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 586, + 320, + 596 + ], + "score": 1.0, + "content": "bi. CSLS and the best non-BI-SENT2VEC method.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 107, + 614, + 346, + 625 + ], + "lines": [ + { + "bbox": [ + 105, + 613, + 347, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 347, + 626 + ], + "score": 1.0, + "content": "4.4 PERFORMANCE ON DIS-SIMILAR LANGUAGE PAIRS", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 632, + 505, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 631, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 645 + ], + "score": 1.0, + "content": "We report a substantial improvement on the performance of previous models on cross-lingual", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 643, + 504, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 504, + 655 + ], + "score": 1.0, + "content": "word and sentence retrieval tasks for the dis-similar language pairs(English-Finnish and English-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 654, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 666 + ], + "score": 1.0, + "content": "Hungarian). We use the same evaluation scheme as in Subsections 4.1 and 4.3 Results for these", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 666, + 225, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 225, + 676 + ], + "score": 1.0, + "content": "pairs are included in Table 4.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 108, + 680, + 430, + 692 + ], + "lines": [ + { + "bbox": [ + 106, + 680, + 431, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 431, + 693 + ], + "score": 1.0, + "content": "4.5 ZERO-SHOT CROSS-LINGUAL TRANSFER OF DOCUMENT CLASSIFIERS", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "The MLDoc multilingual document classification task (Schwenk & Li, 2018) consists of news", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "documents given in 8 different languages, which need to be classified into 4 different categories.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "To demonstrate the ability to transfer trained classifiers in a robust fashion between languages, we", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 180, + 76, + 431, + 218 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 180, + 76, + 431, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 76, + 431, + 218 + ], + "spans": [ + { + "bbox": [ + 180, + 76, + 431, + 218 + ], + "score": 0.98, + "html": "
Method\\DatasetSimLex-999WS-353
en iten it
MUSE0.380.300.74 0.64
RCSLS0.38 0.300.740.64
FASTTEXT- Common Crawl0.49 0.320.750.57
BIVEC0.40 0.360.700.60
TRANSGRAM0.43 0.370.730.63
SENT2VEC uni.0.49 0.380.730.60
BI-SENT2VEC uni.0.570.47 0.790.65
BI-SENT2VEC uni. + bi.0.580.50 0.800.69
", + "type": "table", + "image_path": "283fd29dd3c6466d32f63f2aa830962d367d5512f1bdda7a83e27aa8a345bd28.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 180, + 76, + 431, + 123.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 180, + 123.33333333333334, + 431, + 170.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 180, + 170.66666666666669, + 431, + 218.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 229, + 502, + 252 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 229, + 504, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 229, + 504, + 241 + ], + "score": 1.0, + "content": "Table 2: Monolingual word similarity task performance of our methods when trained on en-it", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 239, + 332, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 332, + 253 + ], + "score": 1.0, + "content": "ParaCrawl data. We report Pearson correlation scores.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 107, + 265, + 303, + 276 + ], + "lines": [ + { + "bbox": [ + 105, + 263, + 304, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 304, + 278 + ], + "score": 1.0, + "content": "4.3 CROSS-LINGUAL SENTENCE RETRIEVAL", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 282, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 106, + 283, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 505, + 296 + ], + "score": 1.0, + "content": "The primary contribution of our work is to deliver improved cross-lingual sentence representations.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 294, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 294, + 506, + 307 + ], + "score": 1.0, + "content": "We test sentence embeddings for each method obtained by bag-of-words composition for sentence", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 304, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 104, + 304, + 506, + 319 + ], + "score": 1.0, + "content": "retrieval across different languages on the Europarl corpus. In particular, the tf-idf weighted average", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 317, + 504, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 504, + 328 + ], + "score": 1.0, + "content": "is used to construct sentence embeddings from word embeddings. We consider 2000 sentences in the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 326, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 505, + 341 + ], + "score": 1.0, + "content": "source language dataset and retrieve their translation among 200K sentences in the target language", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "dataset. The other 300K sentences in the Europarl corpus are used to calculate tf-idf weights. Results", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 349, + 475, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 120, + 360 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 120, + 349, + 142, + 359 + ], + "score": 0.85, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 349, + 475, + 360 + ], + "score": 1.0, + "content": "of unsupervised and supervised benchmarks vs our models are included in Table 3.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9, + "bbox_fs": [ + 104, + 283, + 506, + 360 + ] + }, + { + "type": "table", + "bbox": [ + 106, + 370, + 503, + 556 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 370, + 503, + 556 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 370, + 503, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 503, + 556 + ], + "score": 0.92, + "html": "
Methoden-esen-fren-deen-itavg.
→↑→↑→↑→↑
MUSE72.771.569.268.853.353.466.164.364.9
RCSLS26.926.719.321.28.811.315.117.618.4
TRANSGRAM83.581.480.481.664.869.977.277.977.1
VECMAP (unsupervised)81.782.179.880.462.864.669.071.174.0
VECMAP (supervised)81.38180.480.762.664.367.87173.6
BIVEC NN69.877.154.775.556.144.158.245.160.1
BIVEC CSLS81.683.478.181.671.668.174.272.476.4
BI-SENT2VEC uni. NN87.886.485.283.482.380.285.985.884.6
BI-SENT2VEC uni. + bi. NN87.987.886.183.979.579.785.185.384.4
BI-SENT2VEC uni. CSLS89.588.587.186.484.483.088.287.586.8
BI-SENT2VEC uni. + bi. CSLS89.789.687.887.484.284.087.987.687.3
Reduction in error37.5% 37.3%37.8% 31.5%44.4% 46.8%46.9% 43.9%1
", + "type": "table", + "image_path": "a66fd06c0021769a89fe836ddbc7a35cea2224df45f811fb2a378d697d07753b.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 106, + 370, + 503, + 432.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 432.0, + 503, + 494.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 106, + 494.0, + 503, + 556.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 107, + 563, + 505, + 596 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 321, + 577 + ], + "score": 1.0, + "content": "Table 3: Cross-lingual Sentence retrieval. We report", + "type": "text" + }, + { + "bbox": [ + 321, + 564, + 343, + 574 + ], + "score": 0.79, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 562, + 505, + 577 + ], + "score": 1.0, + "content": "scores for 2000 source queries searching", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 574, + 504, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 504, + 586 + ], + "score": 1.0, + "content": "over 200 000 target sentences. Reduction in error is calculated with respect to BI-SENT2VEC uni.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 586, + 320, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 586, + 114, + 596 + ], + "score": 0.41, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 586, + 320, + 596 + ], + "score": 1.0, + "content": "bi. CSLS and the best non-BI-SENT2VEC method.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 107, + 614, + 346, + 625 + ], + "lines": [ + { + "bbox": [ + 105, + 613, + 347, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 347, + 626 + ], + "score": 1.0, + "content": "4.4 PERFORMANCE ON DIS-SIMILAR LANGUAGE PAIRS", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 632, + 505, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 631, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 645 + ], + "score": 1.0, + "content": "We report a substantial improvement on the performance of previous models on cross-lingual", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 643, + 504, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 504, + 655 + ], + "score": 1.0, + "content": "word and sentence retrieval tasks for the dis-similar language pairs(English-Finnish and English-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 654, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 666 + ], + "score": 1.0, + "content": "Hungarian). We use the same evaluation scheme as in Subsections 4.1 and 4.3 Results for these", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 666, + 225, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 225, + 676 + ], + "score": 1.0, + "content": "pairs are included in Table 4.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 631, + 505, + 676 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 680, + 430, + 692 + ], + "lines": [ + { + "bbox": [ + 106, + 680, + 431, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 431, + 693 + ], + "score": 1.0, + "content": "4.5 ZERO-SHOT CROSS-LINGUAL TRANSFER OF DOCUMENT CLASSIFIERS", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "The MLDoc multilingual document classification task (Schwenk & Li, 2018) consists of news", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "documents given in 8 different languages, which need to be classified into 4 different categories.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "To demonstrate the ability to transfer trained classifiers in a robust fashion between languages, we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "use a zero-shot setting, i.e., we train a classifier on embeddings in the source language, and report", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 360, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 505, + 374 + ], + "score": 1.0, + "content": "the accuracy of the same classifier applied to the target language. As the classifier, we use a simple", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "score": 1.0, + "content": "feed-forward neural network with two hidden layers of size 10 and 8 respectively, optimized using", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 383, + 483, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 483, + 396 + ], + "score": 1.0, + "content": "the Adam optimizer. Each document is represented using the sum of its sentence embeddings.", + "type": "text", + "cross_page": true + } + ], + "index": 8 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 698, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 80, + 504, + 299 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 80, + 504, + 299 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 504, + 299 + ], + "score": 0.983, + "html": "
Method word retrieval sentence retrieval
en-fi →↑en-hu → ↑→en-fi ↑en-hu →↑
MUSE48.159.553.9 64.921.729.539.146.7
RCSLS61.869.967.0 73.03.24.83.65.1
VECMAP (unsupervised)62.566.861.6 68.713.214.720.519.3
VECMAP (supervised)62.678.363.7 76.615.016.920.921.7
BIVEC NN62.155.362.1 53.714.29.726.213.7
BIVEC CSLS69.678.072.4 78.433.332.046.741.3
TRANSGRAM69.781.173.1 80.835.440.552.155
B1-SENT2VEC uni. NN71.285.475.683.963.5 64.275.2 76.2
BI-SENT2VEC uni. + bi. NN68.581.771.4 79.457.5 55.965.865.2
B1-SENT2VEC uni. CSLS72.0 86.576.3 85.170.2 69.081.480.8
BI-SENT2VEC uni. + bi. CSLS70.184.473.781.766 64.173.874.5
", + "type": "table", + "image_path": "df08d014e7400a856ca7b23caeb7e4d2e39a6c924d2082bfd199f5231b4e5868.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 80, + 504, + 153.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 153.0, + 504, + 226.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 226.0, + 504, + 299.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 307, + 501, + 330 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 307, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 479, + 321 + ], + "score": 1.0, + "content": "Table 4: Cross-lingual Word and Sentence retrieval for dis-similar language pairs", + "type": "text" + }, + { + "bbox": [ + 480, + 307, + 505, + 319 + ], + "score": 0.26, + "content": "( \\mathbf { P } @ \\mathbf { 1 }", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 318, + 336, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 336, + 331 + ], + "score": 1.0, + "content": "scores). ‘en’ is English, ‘fi’ is Finnish, ‘hu’ is Hungarian", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 108, + 349, + 505, + 394 + ], + "lines": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "use a zero-shot setting, i.e., we train a classifier on embeddings in the source language, and report", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 360, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 505, + 374 + ], + "score": 1.0, + "content": "the accuracy of the same classifier applied to the target language. As the classifier, we use a simple", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "score": 1.0, + "content": "feed-forward neural network with two hidden layers of size 10 and 8 respectively, optimized using", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 383, + 483, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 483, + 396 + ], + "score": 1.0, + "content": "the Adam optimizer. Each document is represented using the sum of its sentence embeddings.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "table", + "bbox": [ + 160, + 402, + 451, + 460 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 160, + 402, + 451, + 460 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 160, + 402, + 451, + 460 + ], + "spans": [ + { + "bbox": [ + 160, + 402, + 451, + 460 + ], + "score": 0.969, + "html": "
Methoden-esen-fren-deavg.
→↑→冏↑→↑→en-it ↑
LASER79.3 69.678.0 80.186.3 80.870.2 74.277.3
BI-SENT2VEC74.0 71.581.6 82.286.5 79.275.0 72.677.8
", + "type": "table", + "image_path": "938340f7ee36091f27e2f02534b875eeec8920f5d4897f23d52629e79e5776f2.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 160, + 402, + 451, + 421.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 160, + 421.3333333333333, + 451, + 440.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 160, + 440.66666666666663, + 451, + 459.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 470, + 504, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "score": 1.0, + "content": "Table 5: MLDoc Benchmark results (Schwenk & Li, 2018). A document classifier was trained on", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 480, + 500, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 449, + 495 + ], + "score": 1.0, + "content": "one language and tested on another without additional training/fine-tuning. We report", + "type": "text" + }, + { + "bbox": [ + 449, + 481, + 459, + 492 + ], + "score": 0.84, + "content": "\\%", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 480, + 500, + 495 + ], + "score": 1.0, + "content": "accuracy.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 504, + 505, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 504, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 505, + 517 + ], + "score": 1.0, + "content": "We compare the performance of BI-SENT2VEC with the LASER sentence embeddings (Artetxe &", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 515, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 505, + 528 + ], + "score": 1.0, + "content": "Schwenk, 2018) in Table 5. LASER sentence embedding model is a multi-lingual sentence embed-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 526, + 506, + 539 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 506, + 539 + ], + "score": 1.0, + "content": "ding model which is composed of a biLSTM encoder and an LSTM decoder. It uses a shared byte", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 537, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 506, + 550 + ], + "score": 1.0, + "content": "pair encoding based vocabulary of 50k words. The LASER model which we compare to was trained", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 547, + 506, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 506, + 561 + ], + "score": 1.0, + "content": "on 223M sentences for 93 languages and requires 5 days to train on 16 V100 GPUs compared to our", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 560, + 403, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 403, + 571 + ], + "score": 1.0, + "content": "model which takes 1-2.5 hours for each language pair on 30 CPU threads.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 107, + 578, + 374, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 577, + 375, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 375, + 592 + ], + "score": 1.0, + "content": "4.6 EFFECT OF CORPUS SIZE ON REPRESENTATION QUALITY", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 107, + 595, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 107, + 595, + 505, + 608 + ], + "score": 1.0, + "content": "We conduct an ablation study on how BI-SENT2VEC embeddings’ performance depends on the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 605, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 619 + ], + "score": 1.0, + "content": "size of the training corpus. We uniformly sample smaller subsets of the En-Fr ParaCrawl dataset", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "and train a BI-SENT2VEC model on them. We test word/sentence translation performance with the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "CSLS retrieval criterion, and monolingual embedding quality for En-Fr with increasing ParaCrawl", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 640, + 337, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 337, + 652 + ], + "score": 1.0, + "content": "corpus size. The results are illustrated in Figures 2 and 3.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 107, + 656, + 191, + 669 + ], + "lines": [ + { + "bbox": [ + 104, + 654, + 192, + 672 + ], + "spans": [ + { + "bbox": [ + 104, + 654, + 192, + 672 + ], + "score": 1.0, + "content": "5 DISCUSSION", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "In the following section, we discuss the results on monolingual and cross-lingual benchmarks, pre-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "sented in Tables 1 - 5, and a data ablation study for how the model behaves with increasing parallel", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 699, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 195, + 711 + ], + "score": 1.0, + "content": "corpus size in Figure", + "type": "text" + }, + { + "bbox": [ + 195, + 699, + 216, + 709 + ], + "score": 0.25, + "content": "2 \\cdot 3", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 699, + 504, + 711 + ], + "score": 1.0, + "content": ". The most impressive outcome of our experiments is improved cross-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "lingual sentence retrieval performance, which we elaborate on along with word translation in the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 721, + 173, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 173, + 732 + ], + "score": 1.0, + "content": "next subsection.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 80, + 504, + 299 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 80, + 504, + 299 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 504, + 299 + ], + "score": 0.983, + "html": "
Method word retrieval sentence retrieval
en-fi →↑en-hu → ↑→en-fi ↑en-hu →↑
MUSE48.159.553.9 64.921.729.539.146.7
RCSLS61.869.967.0 73.03.24.83.65.1
VECMAP (unsupervised)62.566.861.6 68.713.214.720.519.3
VECMAP (supervised)62.678.363.7 76.615.016.920.921.7
BIVEC NN62.155.362.1 53.714.29.726.213.7
BIVEC CSLS69.678.072.4 78.433.332.046.741.3
TRANSGRAM69.781.173.1 80.835.440.552.155
B1-SENT2VEC uni. NN71.285.475.683.963.5 64.275.2 76.2
BI-SENT2VEC uni. + bi. NN68.581.771.4 79.457.5 55.965.865.2
B1-SENT2VEC uni. CSLS72.0 86.576.3 85.170.2 69.081.480.8
BI-SENT2VEC uni. + bi. CSLS70.184.473.781.766 64.173.874.5
", + "type": "table", + "image_path": "df08d014e7400a856ca7b23caeb7e4d2e39a6c924d2082bfd199f5231b4e5868.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 80, + 504, + 153.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 153.0, + 504, + 226.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 226.0, + 504, + 299.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 307, + 501, + 330 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 307, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 479, + 321 + ], + "score": 1.0, + "content": "Table 4: Cross-lingual Word and Sentence retrieval for dis-similar language pairs", + "type": "text" + }, + { + "bbox": [ + 480, + 307, + 505, + 319 + ], + "score": 0.26, + "content": "( \\mathbf { P } @ \\mathbf { 1 }", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 318, + 336, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 336, + 331 + ], + "score": 1.0, + "content": "scores). ‘en’ is English, ‘fi’ is Finnish, ‘hu’ is Hungarian", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 108, + 349, + 505, + 394 + ], + "lines": [], + "index": 6.5, + "bbox_fs": [ + 105, + 349, + 505, + 396 + ], + "lines_deleted": true + }, + { + "type": "table", + "bbox": [ + 160, + 402, + 451, + 460 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 160, + 402, + 451, + 460 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 160, + 402, + 451, + 460 + ], + "spans": [ + { + "bbox": [ + 160, + 402, + 451, + 460 + ], + "score": 0.969, + "html": "
Methoden-esen-fren-deavg.
→↑→冏↑→↑→en-it ↑
LASER79.3 69.678.0 80.186.3 80.870.2 74.277.3
BI-SENT2VEC74.0 71.581.6 82.286.5 79.275.0 72.677.8
", + "type": "table", + "image_path": "938340f7ee36091f27e2f02534b875eeec8920f5d4897f23d52629e79e5776f2.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 160, + 402, + 451, + 421.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 160, + 421.3333333333333, + 451, + 440.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 160, + 440.66666666666663, + 451, + 459.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 470, + 504, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "score": 1.0, + "content": "Table 5: MLDoc Benchmark results (Schwenk & Li, 2018). A document classifier was trained on", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 480, + 500, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 449, + 495 + ], + "score": 1.0, + "content": "one language and tested on another without additional training/fine-tuning. We report", + "type": "text" + }, + { + "bbox": [ + 449, + 481, + 459, + 492 + ], + "score": 0.84, + "content": "\\%", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 480, + 500, + 495 + ], + "score": 1.0, + "content": "accuracy.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 470, + 505, + 495 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 504, + 505, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 504, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 505, + 517 + ], + "score": 1.0, + "content": "We compare the performance of BI-SENT2VEC with the LASER sentence embeddings (Artetxe &", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 515, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 505, + 528 + ], + "score": 1.0, + "content": "Schwenk, 2018) in Table 5. LASER sentence embedding model is a multi-lingual sentence embed-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 526, + 506, + 539 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 506, + 539 + ], + "score": 1.0, + "content": "ding model which is composed of a biLSTM encoder and an LSTM decoder. It uses a shared byte", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 537, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 506, + 550 + ], + "score": 1.0, + "content": "pair encoding based vocabulary of 50k words. The LASER model which we compare to was trained", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 547, + 506, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 506, + 561 + ], + "score": 1.0, + "content": "on 223M sentences for 93 languages and requires 5 days to train on 16 V100 GPUs compared to our", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 560, + 403, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 403, + 571 + ], + "score": 1.0, + "content": "model which takes 1-2.5 hours for each language pair on 30 CPU threads.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 504, + 506, + 571 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 578, + 374, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 577, + 375, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 375, + 592 + ], + "score": 1.0, + "content": "4.6 EFFECT OF CORPUS SIZE ON REPRESENTATION QUALITY", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 107, + 595, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 107, + 595, + 505, + 608 + ], + "score": 1.0, + "content": "We conduct an ablation study on how BI-SENT2VEC embeddings’ performance depends on the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 605, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 619 + ], + "score": 1.0, + "content": "size of the training corpus. We uniformly sample smaller subsets of the En-Fr ParaCrawl dataset", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "and train a BI-SENT2VEC model on them. We test word/sentence translation performance with the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "CSLS retrieval criterion, and monolingual embedding quality for En-Fr with increasing ParaCrawl", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 640, + 337, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 337, + 652 + ], + "score": 1.0, + "content": "corpus size. The results are illustrated in Figures 2 and 3.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 595, + 506, + 652 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 656, + 191, + 669 + ], + "lines": [ + { + "bbox": [ + 104, + 654, + 192, + 672 + ], + "spans": [ + { + "bbox": [ + 104, + 654, + 192, + 672 + ], + "score": 1.0, + "content": "5 DISCUSSION", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "In the following section, we discuss the results on monolingual and cross-lingual benchmarks, pre-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "sented in Tables 1 - 5, and a data ablation study for how the model behaves with increasing parallel", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 699, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 195, + 711 + ], + "score": 1.0, + "content": "corpus size in Figure", + "type": "text" + }, + { + "bbox": [ + 195, + 699, + 216, + 709 + ], + "score": 0.25, + "content": "2 \\cdot 3", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 699, + 504, + 711 + ], + "score": 1.0, + "content": ". The most impressive outcome of our experiments is improved cross-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "lingual sentence retrieval performance, which we elaborate on along with word translation in the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 721, + 173, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 173, + 732 + ], + "score": 1.0, + "content": "next subsection.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 676, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 87, + 486, + 211 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 87, + 486, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 87, + 486, + 211 + ], + "spans": [ + { + "bbox": [ + 111, + 87, + 486, + 211 + ], + "score": 0.961, + "type": "image", + "image_path": "bb3e3da4961fcc13096a65b06ddba37fd74cb2a40ffb7fe4be5b7a1bd56621c1.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 87, + 486, + 128.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 128.33333333333334, + 486, + 169.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 169.66666666666669, + 486, + 211.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 124, + 214, + 484, + 226 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 212, + 486, + 228 + ], + "spans": [ + { + "bbox": [ + 124, + 212, + 486, + 228 + ], + "score": 1.0, + "content": "Figure 2: Effect of corpus size on cross-lingual word/sentence retrieval performance.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "image", + "bbox": [ + 214, + 249, + 385, + 373 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 214, + 249, + 385, + 373 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 214, + 249, + 385, + 373 + ], + "spans": [ + { + "bbox": [ + 214, + 249, + 385, + 373 + ], + "score": 0.957, + "type": "image", + "image_path": "1145c4987fe2df1428c48252033511b35e1e13156a25dc7e8955255c49fc87e7.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 214, + 249, + 385, + 262.77777777777777 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 214, + 262.77777777777777, + 385, + 276.55555555555554 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 214, + 276.55555555555554, + 385, + 290.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 214, + 290.3333333333333, + 385, + 304.1111111111111 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 214, + 304.1111111111111, + 385, + 317.88888888888886 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 214, + 317.88888888888886, + 385, + 331.66666666666663 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 214, + 331.66666666666663, + 385, + 345.4444444444444 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 214, + 345.4444444444444, + 385, + 359.2222222222222 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 214, + 359.2222222222222, + 385, + 372.99999999999994 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 378, + 505, + 402 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "Figure 3: Effect of corpus size on monolingual word quality. We use SimLex-999, WS-353, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 389, + 384, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 384, + 402 + ], + "score": 1.0, + "content": "FR-RG datasets for measuring monolingual word embedding quality.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + } + ], + "index": 10.75 + }, + { + "type": "text", + "bbox": [ + 107, + 428, + 505, + 528 + ], + "lines": [ + { + "bbox": [ + 107, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 107, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "Cross-lingual evaluations For cross-lingual tasks, we observe in Table 1 that jointly trained", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 440, + 504, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 504, + 452 + ], + "score": 1.0, + "content": "embeddings produce much better results on cross-lingual word and sentence retrieval tasks. BI-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 450, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 464 + ], + "score": 1.0, + "content": "SENT2VEC’s performance on word-retrieval tasks is uniformly superior to mapping methods,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 461, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 169, + 475 + ], + "score": 1.0, + "content": "achieving up to", + "type": "text" + }, + { + "bbox": [ + 169, + 462, + 195, + 472 + ], + "score": 0.86, + "content": "1 1 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 461, + 228, + 475 + ], + "score": 1.0, + "content": "more in", + "type": "text" + }, + { + "bbox": [ + 228, + 462, + 249, + 472 + ], + "score": 0.84, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 461, + 506, + 475 + ], + "score": 1.0, + "content": "than RCSLS for the English to German language pair, consistent", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 472, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 506, + 486 + ], + "score": 1.0, + "content": "with the results from (Ormazabal et al., 2019). It is also on-par with, or better than competing joint", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 484, + 504, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 504, + 496 + ], + "score": 1.0, + "content": "methods except on translation from Russian to English, where TRANSGRAM receives a significantly", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 493, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 505, + 507 + ], + "score": 1.0, + "content": "better score. For word retrieval tasks, there is no discernible difference between CSLS/NN criteria", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 505, + 504, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 504, + 518 + ], + "score": 1.0, + "content": "for BI-SENT2VEC, suggesting the relative absence of the hubness phenomenon which significantly", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 517, + 379, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 379, + 529 + ], + "score": 1.0, + "content": "hinders the performance of cross-lingual word embedding methods.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 504, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "Our principal contribution is in improving cross-lingual sentence retrieval. Table 3 shows BI-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 545, + 504, + 555 + ], + "spans": [ + { + "bbox": [ + 107, + 545, + 482, + 555 + ], + "score": 1.0, + "content": "SENT2VEC decisively outperforms all other methods by a wide margin, reducing the relative", + "type": "text" + }, + { + "bbox": [ + 482, + 545, + 504, + 555 + ], + "score": 0.76, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 555, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 193, + 568 + ], + "score": 1.0, + "content": "error anywhere from", + "type": "text" + }, + { + "bbox": [ + 193, + 555, + 221, + 566 + ], + "score": 0.89, + "content": "3 1 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 555, + 233, + 568 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 233, + 555, + 260, + 566 + ], + "score": 0.89, + "content": "5 5 . 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 555, + 506, + 568 + ], + "score": 1.0, + "content": ". Our model displays considerably less variance than others", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 302, + 579 + ], + "score": 1.0, + "content": "in quality across language pairs, with at most a", + "type": "text" + }, + { + "bbox": [ + 303, + 566, + 329, + 576 + ], + "score": 0.86, + "content": "\\approx 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 565, + 505, + 579 + ], + "score": 1.0, + "content": "deficit between best and worst, and nearly", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 578, + 281, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 281, + 590 + ], + "score": 1.0, + "content": "symmetric accuracy within a language pair.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 594, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 593, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 505, + 607 + ], + "score": 1.0, + "content": "TRANSGRAM also outperforms the mapping-based methods, but still falls significantly short of BI-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "SENT2VEC’s. These results can be attributed to the fact that BI-SENT2VEC directly optimizes for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "obtaining robust sentence embeddings using additive composition of its word embeddings. Since", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "BI-SENT2VEC’s learning objective is closest to a sentence retrieval task amongst current state-of-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 637, + 442, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 442, + 651 + ], + "score": 1.0, + "content": "the-art methods, it can surpass them without sacrificing performance on other tasks.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "Cross-lingual evaluations on dis-similar language pairs Unlike other language pairs in the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 666, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 679 + ], + "score": 1.0, + "content": "evaluation, English-Finnish and English-Hungarian pairs are composed of languages from two dif-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "ferent language families(English being an Indo-European language and the other language being a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "Finno-Ugric language). In Table 4, we see that the performance boost achieved by BI-SENT2VEC", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "on competing methods methods is more pronounced in the case of dis-similar language pairs as", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "compared to paris of languages close to each other. This observation affirms the suitaibility of BI-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 438, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 438, + 734 + ], + "score": 1.0, + "content": "SENT2VEC for learning joint representations on languages from different families.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 87, + 486, + 211 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 87, + 486, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 87, + 486, + 211 + ], + "spans": [ + { + "bbox": [ + 111, + 87, + 486, + 211 + ], + "score": 0.961, + "type": "image", + "image_path": "bb3e3da4961fcc13096a65b06ddba37fd74cb2a40ffb7fe4be5b7a1bd56621c1.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 87, + 486, + 128.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 128.33333333333334, + 486, + 169.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 169.66666666666669, + 486, + 211.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 124, + 214, + 484, + 226 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 212, + 486, + 228 + ], + "spans": [ + { + "bbox": [ + 124, + 212, + 486, + 228 + ], + "score": 1.0, + "content": "Figure 2: Effect of corpus size on cross-lingual word/sentence retrieval performance.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "image", + "bbox": [ + 214, + 249, + 385, + 373 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 214, + 249, + 385, + 373 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 214, + 249, + 385, + 373 + ], + "spans": [ + { + "bbox": [ + 214, + 249, + 385, + 373 + ], + "score": 0.957, + "type": "image", + "image_path": "1145c4987fe2df1428c48252033511b35e1e13156a25dc7e8955255c49fc87e7.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 214, + 249, + 385, + 262.77777777777777 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 214, + 262.77777777777777, + 385, + 276.55555555555554 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 214, + 276.55555555555554, + 385, + 290.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 214, + 290.3333333333333, + 385, + 304.1111111111111 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 214, + 304.1111111111111, + 385, + 317.88888888888886 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 214, + 317.88888888888886, + 385, + 331.66666666666663 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 214, + 331.66666666666663, + 385, + 345.4444444444444 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 214, + 345.4444444444444, + 385, + 359.2222222222222 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 214, + 359.2222222222222, + 385, + 372.99999999999994 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 378, + 505, + 402 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "Figure 3: Effect of corpus size on monolingual word quality. We use SimLex-999, WS-353, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 389, + 384, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 384, + 402 + ], + "score": 1.0, + "content": "FR-RG datasets for measuring monolingual word embedding quality.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + } + ], + "index": 10.75 + }, + { + "type": "text", + "bbox": [ + 107, + 428, + 505, + 528 + ], + "lines": [ + { + "bbox": [ + 107, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 107, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "Cross-lingual evaluations For cross-lingual tasks, we observe in Table 1 that jointly trained", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 440, + 504, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 504, + 452 + ], + "score": 1.0, + "content": "embeddings produce much better results on cross-lingual word and sentence retrieval tasks. BI-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 450, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 464 + ], + "score": 1.0, + "content": "SENT2VEC’s performance on word-retrieval tasks is uniformly superior to mapping methods,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 461, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 169, + 475 + ], + "score": 1.0, + "content": "achieving up to", + "type": "text" + }, + { + "bbox": [ + 169, + 462, + 195, + 472 + ], + "score": 0.86, + "content": "1 1 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 461, + 228, + 475 + ], + "score": 1.0, + "content": "more in", + "type": "text" + }, + { + "bbox": [ + 228, + 462, + 249, + 472 + ], + "score": 0.84, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 461, + 506, + 475 + ], + "score": 1.0, + "content": "than RCSLS for the English to German language pair, consistent", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 472, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 506, + 486 + ], + "score": 1.0, + "content": "with the results from (Ormazabal et al., 2019). It is also on-par with, or better than competing joint", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 484, + 504, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 504, + 496 + ], + "score": 1.0, + "content": "methods except on translation from Russian to English, where TRANSGRAM receives a significantly", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 493, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 505, + 507 + ], + "score": 1.0, + "content": "better score. For word retrieval tasks, there is no discernible difference between CSLS/NN criteria", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 505, + 504, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 504, + 518 + ], + "score": 1.0, + "content": "for BI-SENT2VEC, suggesting the relative absence of the hubness phenomenon which significantly", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 517, + 379, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 379, + 529 + ], + "score": 1.0, + "content": "hinders the performance of cross-lingual word embedding methods.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 429, + 506, + 529 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 504, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "Our principal contribution is in improving cross-lingual sentence retrieval. Table 3 shows BI-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 545, + 504, + 555 + ], + "spans": [ + { + "bbox": [ + 107, + 545, + 482, + 555 + ], + "score": 1.0, + "content": "SENT2VEC decisively outperforms all other methods by a wide margin, reducing the relative", + "type": "text" + }, + { + "bbox": [ + 482, + 545, + 504, + 555 + ], + "score": 0.76, + "content": "\\mathrm { P @ 1 }", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 555, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 193, + 568 + ], + "score": 1.0, + "content": "error anywhere from", + "type": "text" + }, + { + "bbox": [ + 193, + 555, + 221, + 566 + ], + "score": 0.89, + "content": "3 1 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 555, + 233, + 568 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 233, + 555, + 260, + 566 + ], + "score": 0.89, + "content": "5 5 . 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 555, + 506, + 568 + ], + "score": 1.0, + "content": ". Our model displays considerably less variance than others", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 302, + 579 + ], + "score": 1.0, + "content": "in quality across language pairs, with at most a", + "type": "text" + }, + { + "bbox": [ + 303, + 566, + 329, + 576 + ], + "score": 0.86, + "content": "\\approx 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 565, + 505, + 579 + ], + "score": 1.0, + "content": "deficit between best and worst, and nearly", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 578, + 281, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 281, + 590 + ], + "score": 1.0, + "content": "symmetric accuracy within a language pair.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 533, + 506, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 594, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 593, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 505, + 607 + ], + "score": 1.0, + "content": "TRANSGRAM also outperforms the mapping-based methods, but still falls significantly short of BI-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "SENT2VEC’s. These results can be attributed to the fact that BI-SENT2VEC directly optimizes for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "obtaining robust sentence embeddings using additive composition of its word embeddings. Since", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "BI-SENT2VEC’s learning objective is closest to a sentence retrieval task amongst current state-of-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 637, + 442, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 442, + 651 + ], + "score": 1.0, + "content": "the-art methods, it can surpass them without sacrificing performance on other tasks.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 593, + 505, + 651 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "Cross-lingual evaluations on dis-similar language pairs Unlike other language pairs in the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 666, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 679 + ], + "score": 1.0, + "content": "evaluation, English-Finnish and English-Hungarian pairs are composed of languages from two dif-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "ferent language families(English being an Indo-European language and the other language being a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "Finno-Ugric language). In Table 4, we see that the performance boost achieved by BI-SENT2VEC", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "on competing methods methods is more pronounced in the case of dis-similar language pairs as", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "compared to paris of languages close to each other. This observation affirms the suitaibility of BI-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 438, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 438, + 734 + ], + "score": 1.0, + "content": "SENT2VEC for learning joint representations on languages from different families.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 655, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "Monolingual word quality For the monolingual word similarity tasks, we observe large gains", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "over existing methods. SENT2VEC is trained on the same corpora as us, and FASTTEXT vectors are", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "trained on the CommonCrawl corpora which are more than 100 times larger than ParaCrawl v4.0. In", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "Table 2, we see that BI-SENT2VEC outperforms them by a significant margin on SimLex-999 and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "WS-353, two important monolingual word quality benchmarks. This observation is in accordance", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 506, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 506, + 151 + ], + "score": 1.0, + "content": "with the fact (Faruqui & Dyer, 2014) that bilingual contexts can be surprisingly effective for learning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "score": 1.0, + "content": "monolingual word representations. However, amongst the joint-training methods, BI-SENT2VEC", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "score": 1.0, + "content": "also outperforms TRANSGRAM and BIVEC trained on the same corpora by a significant margin,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "score": 1.0, + "content": "again hinting at the superiority of the sentence level loss function over a fixed context window loss.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 275 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 200 + ], + "score": 1.0, + "content": "Effect of n-grams (Gupta et al., 2019) report improved results on monolingual word representa-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "tion evaluation tasks for SENT2VEC and FASTTEXT word vectors by training them alongside word", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 210, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 505, + 221 + ], + "score": 1.0, + "content": "n-grams. Our method incorporates their results based on the observation that unigram vectors trained", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "alongside with bigrams significantly outperform unigrams alone on the majority of the evaluation", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "tasks. We can see from Tables 1 - 3 that this holds for the bilingual case as well. However, in case", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "score": 1.0, + "content": "of dis-similar language pairs(Table 4), we observe that using n-grams degrades the cross-lingual", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 505, + 267 + ], + "score": 1.0, + "content": "performance of the embeddings. This observation suggests that use of higher order n-grams may", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 437, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 437, + 277 + ], + "score": 1.0, + "content": "not be helpful for language pairs where the grammatical structures are contrasting.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 281, + 505, + 358 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 203, + 294 + ], + "score": 1.0, + "content": "Effect of corpus size", + "type": "text" + }, + { + "bbox": [ + 218, + 281, + 505, + 293 + ], + "score": 1.0, + "content": "Considering the cross-lingual performance curve exhibited by BI-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 305 + ], + "score": 1.0, + "content": "SENT2VEC in Figure 2, increasing corpus size for the English-French datasets up to 1-3.1M lines", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "score": 1.0, + "content": "appears to saturate the performance of the model on cross-lingual word/sentence retrieval, after", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 314, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 505, + 326 + ], + "score": 1.0, + "content": "which it either plateaus or degrades slightly. This is an encouraging result, indicating that joint", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "methods can use significantly less data to obtain promising performance. This implies that joint", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 336, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 505, + 348 + ], + "score": 1.0, + "content": "methods may not necessarily be constrained to high-resource language pairs as previously assumed,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 347, + 354, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 354, + 359 + ], + "score": 1.0, + "content": "though further experimentation is needed to verify this claim.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 363, + 505, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 363, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 377 + ], + "score": 1.0, + "content": "It should be noted from Figure 3 that the monolingual quality does keep improving with an increase", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 375, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 387 + ], + "score": 1.0, + "content": "in the size of the corpus. A potential way to overcome this issue of plateauing cross-lingual perfor-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 385, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 398 + ], + "score": 1.0, + "content": "mance is to give different weights to the monolingual and cross-lingual component of the loss with", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 396, + 427, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 427, + 409 + ], + "score": 1.0, + "content": "the weights possibly being dependent on other factors such as training progress.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 414, + 505, + 534 + ], + "lines": [ + { + "bbox": [ + 106, + 413, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 426 + ], + "score": 1.0, + "content": "Comparison with a cross-lingual sentence embedding model and performance on document", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "level task On the MLDoc classifier transfer task (Schwenk & Li, 2018) where we evaluate a clas-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "sifier learned on documents in one language on documents in another, Table 5 shows we achieve", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "parity with the performance of the LASER model for language pairs involving English, where BI-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 241, + 470 + ], + "score": 1.0, + "content": "SENT2VEC’s average accuracy of", + "type": "text" + }, + { + "bbox": [ + 241, + 457, + 269, + 468 + ], + "score": 0.87, + "content": "7 7 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 456, + 394, + 470 + ], + "score": 1.0, + "content": "is slightly higher than LASER’s", + "type": "text" + }, + { + "bbox": [ + 394, + 457, + 421, + 468 + ], + "score": 0.87, + "content": "7 7 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 456, + 505, + 470 + ], + "score": 1.0, + "content": ". While the compari-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 469, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 481 + ], + "score": 1.0, + "content": "son is not completely justified as LASER is multilingual in nature and is trained on a different dataset,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "one must emphasize that BI-SENT2VEC is a bag-of-words method as compared to LASER which", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 490, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 506, + 502 + ], + "score": 1.0, + "content": "uses a multi-layered biLSTM sentence encoder. Our method only requires to average a set of vectors", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "to encode sentences reducing its computational footprint significantly. This makes BI-SENT2VEC", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 512, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 524 + ], + "score": 1.0, + "content": "an ideal candidate for on-device computationally efficient cross-lingual NLP, unlike LASER which", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 523, + 500, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 500, + 536 + ], + "score": 1.0, + "content": "has a huge computational overhead and specialized hardware requirement for encoding sentences.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33 + }, + { + "type": "title", + "bbox": [ + 107, + 562, + 303, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 560, + 304, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 304, + 577 + ], + "score": 1.0, + "content": "6 CONCLUSION AND FUTURE WORK", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 594, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 593, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 505, + 607 + ], + "score": 1.0, + "content": "We introduce a cross-lingual extension of an existing monolingual word and sentence embedding", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "method. The proposed model is tested at three levels of linguistic granularity: words, sentences", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "and documents. The model outperforms all other methods by a wide margin on the cross-lingual", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "sentence retrieval task while maintaining parity with the best-performing methods on word transla-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "score": 1.0, + "content": "tion tasks. Our method achieves parity with LASER on zero-shot document classification, despite", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "being a much simpler model. We also demonstrate that training on parallel data yields a significant", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 660, + 353, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 353, + 673 + ], + "score": 1.0, + "content": "improvement in the monolingual word representation quality.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "The success of our model on the bilingual level calls for its extension to the multilingual level espe-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 687, + 504, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 504, + 700 + ], + "score": 1.0, + "content": "cially for pairs which have little or no parallel corpora. While the amount of bilingual/multilingual", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "parallel data has grown in abundance, the amount of monolingual data available is practically limit-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "less. Consequently, we would like to explore training cross-lingual embeddings with a large amount", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 721, + 349, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 349, + 733 + ], + "score": 1.0, + "content": "of raw text combined with a smaller amount of parallel data.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 49 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "Monolingual word quality For the monolingual word similarity tasks, we observe large gains", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "over existing methods. SENT2VEC is trained on the same corpora as us, and FASTTEXT vectors are", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "trained on the CommonCrawl corpora which are more than 100 times larger than ParaCrawl v4.0. In", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "Table 2, we see that BI-SENT2VEC outperforms them by a significant margin on SimLex-999 and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "WS-353, two important monolingual word quality benchmarks. This observation is in accordance", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 506, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 506, + 151 + ], + "score": 1.0, + "content": "with the fact (Faruqui & Dyer, 2014) that bilingual contexts can be surprisingly effective for learning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "score": 1.0, + "content": "monolingual word representations. However, amongst the joint-training methods, BI-SENT2VEC", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "score": 1.0, + "content": "also outperforms TRANSGRAM and BIVEC trained on the same corpora by a significant margin,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "score": 1.0, + "content": "again hinting at the superiority of the sentence level loss function over a fixed context window loss.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 83, + 506, + 183 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 275 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 200 + ], + "score": 1.0, + "content": "Effect of n-grams (Gupta et al., 2019) report improved results on monolingual word representa-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "tion evaluation tasks for SENT2VEC and FASTTEXT word vectors by training them alongside word", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 210, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 505, + 221 + ], + "score": 1.0, + "content": "n-grams. Our method incorporates their results based on the observation that unigram vectors trained", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "alongside with bigrams significantly outperform unigrams alone on the majority of the evaluation", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "tasks. We can see from Tables 1 - 3 that this holds for the bilingual case as well. However, in case", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "score": 1.0, + "content": "of dis-similar language pairs(Table 4), we observe that using n-grams degrades the cross-lingual", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 505, + 267 + ], + "score": 1.0, + "content": "performance of the embeddings. This observation suggests that use of higher order n-grams may", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 437, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 437, + 277 + ], + "score": 1.0, + "content": "not be helpful for language pairs where the grammatical structures are contrasting.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 186, + 505, + 277 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 281, + 505, + 358 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 203, + 294 + ], + "score": 1.0, + "content": "Effect of corpus size", + "type": "text" + }, + { + "bbox": [ + 218, + 281, + 505, + 293 + ], + "score": 1.0, + "content": "Considering the cross-lingual performance curve exhibited by BI-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 305 + ], + "score": 1.0, + "content": "SENT2VEC in Figure 2, increasing corpus size for the English-French datasets up to 1-3.1M lines", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "score": 1.0, + "content": "appears to saturate the performance of the model on cross-lingual word/sentence retrieval, after", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 314, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 505, + 326 + ], + "score": 1.0, + "content": "which it either plateaus or degrades slightly. This is an encouraging result, indicating that joint", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "methods can use significantly less data to obtain promising performance. This implies that joint", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 336, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 505, + 348 + ], + "score": 1.0, + "content": "methods may not necessarily be constrained to high-resource language pairs as previously assumed,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 347, + 354, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 354, + 359 + ], + "score": 1.0, + "content": "though further experimentation is needed to verify this claim.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 280, + 506, + 359 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 363, + 505, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 363, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 377 + ], + "score": 1.0, + "content": "It should be noted from Figure 3 that the monolingual quality does keep improving with an increase", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 375, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 387 + ], + "score": 1.0, + "content": "in the size of the corpus. A potential way to overcome this issue of plateauing cross-lingual perfor-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 385, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 398 + ], + "score": 1.0, + "content": "mance is to give different weights to the monolingual and cross-lingual component of the loss with", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 396, + 427, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 427, + 409 + ], + "score": 1.0, + "content": "the weights possibly being dependent on other factors such as training progress.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 363, + 505, + 409 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 414, + 505, + 534 + ], + "lines": [ + { + "bbox": [ + 106, + 413, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 426 + ], + "score": 1.0, + "content": "Comparison with a cross-lingual sentence embedding model and performance on document", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "level task On the MLDoc classifier transfer task (Schwenk & Li, 2018) where we evaluate a clas-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "sifier learned on documents in one language on documents in another, Table 5 shows we achieve", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "parity with the performance of the LASER model for language pairs involving English, where BI-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 241, + 470 + ], + "score": 1.0, + "content": "SENT2VEC’s average accuracy of", + "type": "text" + }, + { + "bbox": [ + 241, + 457, + 269, + 468 + ], + "score": 0.87, + "content": "7 7 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 456, + 394, + 470 + ], + "score": 1.0, + "content": "is slightly higher than LASER’s", + "type": "text" + }, + { + "bbox": [ + 394, + 457, + 421, + 468 + ], + "score": 0.87, + "content": "7 7 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 456, + 505, + 470 + ], + "score": 1.0, + "content": ". While the compari-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 469, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 481 + ], + "score": 1.0, + "content": "son is not completely justified as LASER is multilingual in nature and is trained on a different dataset,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "one must emphasize that BI-SENT2VEC is a bag-of-words method as compared to LASER which", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 490, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 506, + 502 + ], + "score": 1.0, + "content": "uses a multi-layered biLSTM sentence encoder. Our method only requires to average a set of vectors", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "to encode sentences reducing its computational footprint significantly. This makes BI-SENT2VEC", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 512, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 524 + ], + "score": 1.0, + "content": "an ideal candidate for on-device computationally efficient cross-lingual NLP, unlike LASER which", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 523, + 500, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 500, + 536 + ], + "score": 1.0, + "content": "has a huge computational overhead and specialized hardware requirement for encoding sentences.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 413, + 506, + 536 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 562, + 303, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 560, + 304, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 304, + 577 + ], + "score": 1.0, + "content": "6 CONCLUSION AND FUTURE WORK", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 594, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 593, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 505, + 607 + ], + "score": 1.0, + "content": "We introduce a cross-lingual extension of an existing monolingual word and sentence embedding", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "method. The proposed model is tested at three levels of linguistic granularity: words, sentences", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "and documents. The model outperforms all other methods by a wide margin on the cross-lingual", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "sentence retrieval task while maintaining parity with the best-performing methods on word transla-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "score": 1.0, + "content": "tion tasks. Our method achieves parity with LASER on zero-shot document classification, despite", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "being a much simpler model. We also demonstrate that training on parallel data yields a significant", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 660, + 353, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 353, + 673 + ], + "score": 1.0, + "content": "improvement in the monolingual word representation quality.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 593, + 506, + 673 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "The success of our model on the bilingual level calls for its extension to the multilingual level espe-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 687, + 504, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 504, + 700 + ], + "score": 1.0, + "content": "cially for pairs which have little or no parallel corpora. While the amount of bilingual/multilingual", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "parallel data has grown in abundance, the amount of monolingual data available is practically limit-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "less. Consequently, we would like to explore training cross-lingual embeddings with a large amount", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 721, + 349, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 349, + 733 + ], + "score": 1.0, + "content": "of raw text combined with a smaller amount of parallel data.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 49, + "bbox_fs": [ + 105, + 676, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 176, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 504, + 136 + ], + "lines": [ + { + "bbox": [ + 106, + 102, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 114 + ], + "score": 1.0, + "content": "Mikel Artetxe and Holger Schwenk. Massively multilingual sentence embeddings for zero-shot", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 112, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 115, + 112, + 506, + 127 + ], + "score": 1.0, + "content": "cross-lingual transfer and beyond. Transactions of the Association for Computational Linguistics,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 124, + 189, + 136 + ], + "spans": [ + { + "bbox": [ + 116, + 124, + 189, + 136 + ], + "score": 1.0, + "content": "7:597–610, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 105, + 142, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 141, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 505, + 156 + ], + "score": 1.0, + "content": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. Learning bilingual word embeddings with (almost)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 248, + 166 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 248, + 166 + ], + "score": 1.0, + "content": "no bilingual data. In ACL, 2017.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 503, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "score": 1.0, + "content": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. Generalizing and improving bilingual word em-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 505, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 505, + 196 + ], + "score": 1.0, + "content": "bedding mappings with a multi-step framework of linear transformations. In Proceedings of the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 117, + 194, + 444, + 207 + ], + "spans": [ + { + "bbox": [ + 117, + 194, + 444, + 207 + ], + "score": 1.0, + "content": "Thirty-Second AAAI Conference on Artificial Intelligence, pp. 5012–5019, 2018a.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 212, + 504, + 247 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 505, + 226 + ], + "score": 1.0, + "content": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. A robust self-learning method for fully unsuper-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 116, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "vised cross-lingual mappings of word embeddings. In Proceedings of the 56th Annual Meeting of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 235, + 492, + 248 + ], + "spans": [ + { + "bbox": [ + 116, + 235, + 492, + 248 + ], + "score": 1.0, + "content": "the Association for Computational Linguistics (Volume 1: Long Papers), pp. 789–798, 2018b.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 253, + 506, + 287 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 505, + 266 + ], + "score": 1.0, + "content": "Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. Enriching word vectors with", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 264, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 115, + 264, + 505, + 277 + ], + "score": 1.0, + "content": "subword information. Transactions of the Association for Computational Linguistics, 5:135–146,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 275, + 142, + 288 + ], + "spans": [ + { + "bbox": [ + 115, + 275, + 142, + 288 + ], + "score": 1.0, + "content": "2016.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 294, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 307 + ], + "score": 1.0, + "content": "A. P. Sarath Chandar, Stanislas Lauly, Hugo Larochelle, Mitesh M. Khapra, Balaraman Ravindran,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 305, + 504, + 317 + ], + "spans": [ + { + "bbox": [ + 116, + 305, + 504, + 317 + ], + "score": 1.0, + "content": "Vikas C. Raykar, and Amrita Saha. An autoencoder approach to learning bilingual word repre-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 316, + 224, + 327 + ], + "spans": [ + { + "bbox": [ + 116, + 316, + 224, + 327 + ], + "score": 1.0, + "content": "sentations. In NIPS, 2014.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 105, + 334, + 489, + 347 + ], + "lines": [ + { + "bbox": [ + 106, + 334, + 491, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 491, + 348 + ], + "score": 1.0, + "content": "Xilun Chen and Claire Cardie. Unsupervised multilingual word embeddings. In EMNLP, 2018.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 353, + 502, + 376 + ], + "lines": [ + { + "bbox": [ + 106, + 353, + 503, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 503, + 366 + ], + "score": 1.0, + "content": "Alexis Conneau, Guillaume Lample, Marc’Aurelio Ranzato, Ludovic Denoyer, and Herve J ´ egou.´", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 364, + 395, + 377 + ], + "spans": [ + { + "bbox": [ + 116, + 364, + 395, + 377 + ], + "score": 1.0, + "content": "Word translation without parallel data. ArXiv, abs/1710.04087, 2017.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 382, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 104, + 383, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 104, + 383, + 505, + 397 + ], + "score": 1.0, + "content": "Jocelyn Coulmance, Jean-Marc Marty, Guillaume Wenzek, and Amine Benhalloum. Trans-gram,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 394, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 116, + 394, + 505, + 407 + ], + "score": 1.0, + "content": "Fast Cross-lingual Word-embeddings. In EMNLP - Proceedings of the 2015 Conference on Em-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 405, + 407, + 417 + ], + "spans": [ + { + "bbox": [ + 115, + 405, + 407, + 417 + ], + "score": 1.0, + "content": "pirical Methods in Natural Language Processing, pp. 1109–1113, 2015.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 105, + 423, + 475, + 436 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 476, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 476, + 437 + ], + "score": 1.0, + "content": "M. Espla-Gomis. ParaCrawl: Web-scale parallel corpora for the languages of the EU. 2019. `", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 105, + 442, + 502, + 465 + ], + "lines": [ + { + "bbox": [ + 105, + 442, + 504, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 504, + 456 + ], + "score": 1.0, + "content": "Manaal Faruqui and Chris Dyer. Improving vector space word representations using multilingual", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 454, + 231, + 465 + ], + "spans": [ + { + "bbox": [ + 116, + 454, + 231, + 465 + ], + "score": 1.0, + "content": "correlation. In EACL, 2014.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 471, + 503, + 495 + ], + "lines": [ + { + "bbox": [ + 106, + 473, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 505, + 484 + ], + "score": 1.0, + "content": "Lev Finkelstein, Evgeniy Gabrilovich, Yossi Matias, Ehud Rivlin, Zach Solan, Gadi Wolfman, and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 483, + 437, + 495 + ], + "spans": [ + { + "bbox": [ + 116, + 483, + 437, + 495 + ], + "score": 1.0, + "content": "Eytan Ruppin. Placing search in context: the concept revisited. In WWW, 2001.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 501, + 502, + 525 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 504, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 504, + 515 + ], + "score": 1.0, + "content": "Stephan Gouws, Yoshua Bengio, and Greg Corrado. Bilbowa: Fast bilingual distributed representa-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 513, + 269, + 524 + ], + "spans": [ + { + "bbox": [ + 115, + 513, + 269, + 524 + ], + "score": 1.0, + "content": "tions without word alignments. 2015.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 105, + 531, + 504, + 554 + ], + "lines": [ + { + "bbox": [ + 105, + 529, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 545 + ], + "score": 1.0, + "content": "Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Armand Joulin, and Tomas Mikolov. Learning", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 542, + 374, + 555 + ], + "spans": [ + { + "bbox": [ + 116, + 542, + 374, + 555 + ], + "score": 1.0, + "content": "word vectors for 157 languages. ArXiv, abs/1802.06893, 2018a.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 503, + 584 + ], + "lines": [ + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "Edouard Grave, Armand Joulin, and Quentin Berthet. Unsupervised alignment of embeddings with", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 572, + 293, + 584 + ], + "spans": [ + { + "bbox": [ + 115, + 572, + 293, + 584 + ], + "score": 1.0, + "content": "wasserstein procrustes. In AISTATS, 2018b.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 106, + 590, + 503, + 614 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 505, + 605 + ], + "score": 1.0, + "content": "Prakhar Gupta, Matteo Pagliardini, and Martin Jaggi. Better word embeddings by disentangling", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 602, + 338, + 614 + ], + "spans": [ + { + "bbox": [ + 116, + 602, + 338, + 614 + ], + "score": 1.0, + "content": "contextual n-gram information. In NAACL-HLT, 2019.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 620, + 503, + 643 + ], + "lines": [ + { + "bbox": [ + 106, + 620, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 505, + 633 + ], + "score": 1.0, + "content": "Karl Moritz Hermann and Phil Blunsom. Multilingual distributed representations without word", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 632, + 248, + 642 + ], + "spans": [ + { + "bbox": [ + 116, + 632, + 248, + 642 + ], + "score": 1.0, + "content": "alignment. In ICLR 2014, 2013.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 649, + 502, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 649, + 504, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 504, + 663 + ], + "score": 1.0, + "content": "Felix Hill, Roi Reichart, and Anna Korhonen. Simlex-999: Evaluating semantic models with (gen-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 661, + 417, + 674 + ], + "spans": [ + { + "bbox": [ + 115, + 661, + 417, + 674 + ], + "score": 1.0, + "content": "uine) similarity estimation. Computational Linguistics, 41:665–695, 2014.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 106, + 680, + 493, + 692 + ], + "lines": [ + { + "bbox": [ + 107, + 679, + 493, + 692 + ], + "spans": [ + { + "bbox": [ + 107, + 679, + 493, + 692 + ], + "score": 1.0, + "content": "Yedid Hoshen and Lior Wolf. Non-adversarial unsupervised word translation. In EMNLP, 2018.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Martin Josifoski, Ivan S Paskov, Hristo S Paskov, Martin Jaggi, and Robert West. Crosslingual", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 116, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "document embedding as reduced-rank ridge regression. In Proceedings of the Twelfth ACM In-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 720, + 454, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 454, + 733 + ], + "score": 1.0, + "content": "ternational Conference on Web Search and Data Mining, pp. 744–752. ACM, 2019.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 176, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 504, + 136 + ], + "lines": [ + { + "bbox": [ + 106, + 102, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 114 + ], + "score": 1.0, + "content": "Mikel Artetxe and Holger Schwenk. Massively multilingual sentence embeddings for zero-shot", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 112, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 115, + 112, + 506, + 127 + ], + "score": 1.0, + "content": "cross-lingual transfer and beyond. Transactions of the Association for Computational Linguistics,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 124, + 189, + 136 + ], + "spans": [ + { + "bbox": [ + 116, + 124, + 189, + 136 + ], + "score": 1.0, + "content": "7:597–610, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 102, + 506, + 136 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 142, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 141, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 505, + 156 + ], + "score": 1.0, + "content": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. Learning bilingual word embeddings with (almost)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 248, + 166 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 248, + 166 + ], + "score": 1.0, + "content": "no bilingual data. In ACL, 2017.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 141, + 505, + 166 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 503, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "score": 1.0, + "content": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. Generalizing and improving bilingual word em-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 505, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 505, + 196 + ], + "score": 1.0, + "content": "bedding mappings with a multi-step framework of linear transformations. In Proceedings of the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 117, + 194, + 444, + 207 + ], + "spans": [ + { + "bbox": [ + 117, + 194, + 444, + 207 + ], + "score": 1.0, + "content": "Thirty-Second AAAI Conference on Artificial Intelligence, pp. 5012–5019, 2018a.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 172, + 505, + 207 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 212, + 504, + 247 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 505, + 226 + ], + "score": 1.0, + "content": "Mikel Artetxe, Gorka Labaka, and Eneko Agirre. A robust self-learning method for fully unsuper-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 116, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "vised cross-lingual mappings of word embeddings. In Proceedings of the 56th Annual Meeting of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 235, + 492, + 248 + ], + "spans": [ + { + "bbox": [ + 116, + 235, + 492, + 248 + ], + "score": 1.0, + "content": "the Association for Computational Linguistics (Volume 1: Long Papers), pp. 789–798, 2018b.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 212, + 506, + 248 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 253, + 506, + 287 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 505, + 266 + ], + "score": 1.0, + "content": "Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. Enriching word vectors with", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 264, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 115, + 264, + 505, + 277 + ], + "score": 1.0, + "content": "subword information. Transactions of the Association for Computational Linguistics, 5:135–146,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 275, + 142, + 288 + ], + "spans": [ + { + "bbox": [ + 115, + 275, + 142, + 288 + ], + "score": 1.0, + "content": "2016.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 106, + 254, + 505, + 288 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 294, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 307 + ], + "score": 1.0, + "content": "A. P. Sarath Chandar, Stanislas Lauly, Hugo Larochelle, Mitesh M. Khapra, Balaraman Ravindran,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 305, + 504, + 317 + ], + "spans": [ + { + "bbox": [ + 116, + 305, + 504, + 317 + ], + "score": 1.0, + "content": "Vikas C. Raykar, and Amrita Saha. An autoencoder approach to learning bilingual word repre-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 316, + 224, + 327 + ], + "spans": [ + { + "bbox": [ + 116, + 316, + 224, + 327 + ], + "score": 1.0, + "content": "sentations. In NIPS, 2014.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 294, + 505, + 327 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 334, + 489, + 347 + ], + "lines": [ + { + "bbox": [ + 106, + 334, + 491, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 491, + 348 + ], + "score": 1.0, + "content": "Xilun Chen and Claire Cardie. Unsupervised multilingual word embeddings. In EMNLP, 2018.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 334, + 491, + 348 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 353, + 502, + 376 + ], + "lines": [ + { + "bbox": [ + 106, + 353, + 503, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 503, + 366 + ], + "score": 1.0, + "content": "Alexis Conneau, Guillaume Lample, Marc’Aurelio Ranzato, Ludovic Denoyer, and Herve J ´ egou.´", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 364, + 395, + 377 + ], + "spans": [ + { + "bbox": [ + 116, + 364, + 395, + 377 + ], + "score": 1.0, + "content": "Word translation without parallel data. ArXiv, abs/1710.04087, 2017.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 353, + 503, + 377 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 382, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 104, + 383, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 104, + 383, + 505, + 397 + ], + "score": 1.0, + "content": "Jocelyn Coulmance, Jean-Marc Marty, Guillaume Wenzek, and Amine Benhalloum. Trans-gram,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 394, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 116, + 394, + 505, + 407 + ], + "score": 1.0, + "content": "Fast Cross-lingual Word-embeddings. In EMNLP - Proceedings of the 2015 Conference on Em-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 405, + 407, + 417 + ], + "spans": [ + { + "bbox": [ + 115, + 405, + 407, + 417 + ], + "score": 1.0, + "content": "pirical Methods in Natural Language Processing, pp. 1109–1113, 2015.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 104, + 383, + 505, + 417 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 423, + 475, + 436 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 476, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 476, + 437 + ], + "score": 1.0, + "content": "M. Espla-Gomis. ParaCrawl: Web-scale parallel corpora for the languages of the EU. 2019. `", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 423, + 476, + 437 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 442, + 502, + 465 + ], + "lines": [ + { + "bbox": [ + 105, + 442, + 504, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 504, + 456 + ], + "score": 1.0, + "content": "Manaal Faruqui and Chris Dyer. Improving vector space word representations using multilingual", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 454, + 231, + 465 + ], + "spans": [ + { + "bbox": [ + 116, + 454, + 231, + 465 + ], + "score": 1.0, + "content": "correlation. In EACL, 2014.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 442, + 504, + 465 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 471, + 503, + 495 + ], + "lines": [ + { + "bbox": [ + 106, + 473, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 505, + 484 + ], + "score": 1.0, + "content": "Lev Finkelstein, Evgeniy Gabrilovich, Yossi Matias, Ehud Rivlin, Zach Solan, Gadi Wolfman, and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 483, + 437, + 495 + ], + "spans": [ + { + "bbox": [ + 116, + 483, + 437, + 495 + ], + "score": 1.0, + "content": "Eytan Ruppin. Placing search in context: the concept revisited. In WWW, 2001.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 473, + 505, + 495 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 501, + 502, + 525 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 504, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 504, + 515 + ], + "score": 1.0, + "content": "Stephan Gouws, Yoshua Bengio, and Greg Corrado. Bilbowa: Fast bilingual distributed representa-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 513, + 269, + 524 + ], + "spans": [ + { + "bbox": [ + 115, + 513, + 269, + 524 + ], + "score": 1.0, + "content": "tions without word alignments. 2015.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 106, + 500, + 504, + 524 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 531, + 504, + 554 + ], + "lines": [ + { + "bbox": [ + 105, + 529, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 545 + ], + "score": 1.0, + "content": "Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Armand Joulin, and Tomas Mikolov. Learning", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 542, + 374, + 555 + ], + "spans": [ + { + "bbox": [ + 116, + 542, + 374, + 555 + ], + "score": 1.0, + "content": "word vectors for 157 languages. ArXiv, abs/1802.06893, 2018a.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 529, + 505, + 555 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 503, + 584 + ], + "lines": [ + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "Edouard Grave, Armand Joulin, and Quentin Berthet. Unsupervised alignment of embeddings with", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 572, + 293, + 584 + ], + "spans": [ + { + "bbox": [ + 115, + 572, + 293, + 584 + ], + "score": 1.0, + "content": "wasserstein procrustes. In AISTATS, 2018b.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 106, + 561, + 505, + 584 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 590, + 503, + 614 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 505, + 605 + ], + "score": 1.0, + "content": "Prakhar Gupta, Matteo Pagliardini, and Martin Jaggi. Better word embeddings by disentangling", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 602, + 338, + 614 + ], + "spans": [ + { + "bbox": [ + 116, + 602, + 338, + 614 + ], + "score": 1.0, + "content": "contextual n-gram information. In NAACL-HLT, 2019.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 588, + 505, + 614 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 620, + 503, + 643 + ], + "lines": [ + { + "bbox": [ + 106, + 620, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 505, + 633 + ], + "score": 1.0, + "content": "Karl Moritz Hermann and Phil Blunsom. Multilingual distributed representations without word", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 632, + 248, + 642 + ], + "spans": [ + { + "bbox": [ + 116, + 632, + 248, + 642 + ], + "score": 1.0, + "content": "alignment. In ICLR 2014, 2013.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 106, + 620, + 505, + 642 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 649, + 502, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 649, + 504, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 504, + 663 + ], + "score": 1.0, + "content": "Felix Hill, Roi Reichart, and Anna Korhonen. Simlex-999: Evaluating semantic models with (gen-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 661, + 417, + 674 + ], + "spans": [ + { + "bbox": [ + 115, + 661, + 417, + 674 + ], + "score": 1.0, + "content": "uine) similarity estimation. Computational Linguistics, 41:665–695, 2014.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 649, + 504, + 674 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 680, + 493, + 692 + ], + "lines": [ + { + "bbox": [ + 107, + 679, + 493, + 692 + ], + "spans": [ + { + "bbox": [ + 107, + 679, + 493, + 692 + ], + "score": 1.0, + "content": "Yedid Hoshen and Lior Wolf. Non-adversarial unsupervised word translation. In EMNLP, 2018.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41, + "bbox_fs": [ + 107, + 679, + 493, + 692 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Martin Josifoski, Ivan S Paskov, Hristo S Paskov, Martin Jaggi, and Robert West. Crosslingual", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 116, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "document embedding as reduced-rank ridge regression. In Proceedings of the Twelfth ACM In-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 720, + 454, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 454, + 733 + ], + "score": 1.0, + "content": "ternational Conference on Web Search and Data Mining, pp. 744–752. ACM, 2019.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 698, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Colette Joubarne and Diana Inkpen. Comparison of semantic similarity for different languages using", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "the google n-gram corpus and second-order co-occurrence measures. In Canadian Conference on", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 104, + 156, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 104, + 156, + 117 + ], + "score": 1.0, + "content": "AI, 2011.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 123, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 106, + 123, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 505, + 136 + ], + "score": 1.0, + "content": "Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. Bag of tricks for efficient", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 135, + 257, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 135, + 257, + 146 + ], + "score": 1.0, + "content": "text classification. In EACL, 2016.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 153, + 504, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 154, + 504, + 165 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 504, + 165 + ], + "score": 1.0, + "content": "Armand Joulin, Piotr Bojanowski, Tomas Mikolov, Herve J ´ egou, and Edouard Grave. Loss in trans-´", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 164, + 458, + 176 + ], + "spans": [ + { + "bbox": [ + 115, + 164, + 458, + 176 + ], + "score": 1.0, + "content": "lation: Learning bilingual word mapping with a retrieval criterion. In EMNLP, 2018.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 107, + 183, + 445, + 195 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 446, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 446, + 196 + ], + "score": 1.0, + "content": "Philipp Koehn. Europarl: A parallel corpus for statistical machine translation. 2005.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 105, + 201, + 504, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 506, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 506, + 215 + ], + "score": 1.0, + "content": "Pierre Lison and Jorg Tiedemann. Opensubtitles2016: Extracting large parallel corpora from movie ¨", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 213, + 248, + 225 + ], + "spans": [ + { + "bbox": [ + 116, + 213, + 248, + 225 + ], + "score": 1.0, + "content": "and tv subtitles. In LREC, 2016.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 108, + 231, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 231, + 504, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 504, + 245 + ], + "score": 1.0, + "content": "Thang Luong, Hieu Pham, and Christopher D Manning. Bilingual word representations with mono-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 242, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 115, + 242, + 505, + 256 + ], + "score": 1.0, + "content": "lingual quality in mind. In Proceedings of the 1st Workshop on Vector Space Modeling for Natural", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 254, + 287, + 266 + ], + "spans": [ + { + "bbox": [ + 116, + 254, + 287, + 266 + ], + "score": 1.0, + "content": "Language Processing, pp. 151–159, 2015.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 105, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 106, + 272, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 505, + 285 + ], + "score": 1.0, + "content": "Tomas Mikolov, Kai Chen, Gregory S. Corrado, and Jeffrey Dean. Efficient estimation of word", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 284, + 367, + 296 + ], + "spans": [ + { + "bbox": [ + 116, + 284, + 367, + 296 + ], + "score": 1.0, + "content": "representations in vector space. CoRR, abs/1301.3781, 2013a.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 105, + 302, + 504, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 504, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 504, + 316 + ], + "score": 1.0, + "content": "Tomas Mikolov, Quoc V. Le, and Ilya Sutskever. Exploiting similarities among languages for ma-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 117, + 313, + 310, + 325 + ], + "spans": [ + { + "bbox": [ + 117, + 313, + 310, + 325 + ], + "score": 1.0, + "content": "chine translation. ArXiv, abs/1309.4168, 2013b.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 332, + 504, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 345 + ], + "score": 1.0, + "content": "Aitor Ormazabal, Mikel Artetxe, Gorka Labaka, Aitor Soroa, and Eneko Agirre. Analyzing the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 344, + 399, + 355 + ], + "spans": [ + { + "bbox": [ + 116, + 344, + 399, + 355 + ], + "score": 1.0, + "content": "limitations of cross-lingual word embedding mappings. In ACL, 2019.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 362, + 502, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 504, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 504, + 375 + ], + "score": 1.0, + "content": "Matteo Pagliardini, Prakhar Gupta, and Martin Jaggi. Unsupervised learning of sentence embed-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 374, + 385, + 385 + ], + "spans": [ + { + "bbox": [ + 115, + 374, + 385, + 385 + ], + "score": 1.0, + "content": "dings using compositional n-gram features. In NAACL-HLT, 2018.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 502, + 415 + ], + "lines": [ + { + "bbox": [ + 106, + 392, + 504, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 504, + 405 + ], + "score": 1.0, + "content": "Sebastian Ruder, Ivan Vulic, and Anders Søgaard. A survey of cross-lingual word embedding mod-´", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 403, + 295, + 415 + ], + "spans": [ + { + "bbox": [ + 116, + 403, + 295, + 415 + ], + "score": 1.0, + "content": "els. arXiv preprint arXiv:1706.04902, 2017.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 422, + 503, + 444 + ], + "lines": [ + { + "bbox": [ + 105, + 420, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 436 + ], + "score": 1.0, + "content": "Holger Schwenk and Xian Li. A corpus for multilingual document classification in eight languages.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 433, + 279, + 445 + ], + "spans": [ + { + "bbox": [ + 115, + 433, + 279, + 445 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1805.09821, 2018.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 452, + 503, + 475 + ], + "lines": [ + { + "bbox": [ + 106, + 451, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 465 + ], + "score": 1.0, + "content": "Samuel L. Smith, David H. P. Turban, Steven Hamblin, and Nils Y. Hammerla. Offline bilingual", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 115, + 463, + 505, + 475 + ], + "score": 1.0, + "content": "word vectors, orthogonal transformations and the inverted softmax. ArXiv, abs/1702.03859, 2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 481, + 503, + 505 + ], + "lines": [ + { + "bbox": [ + 106, + 480, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 495 + ], + "score": 1.0, + "content": "Anders Søgaard, Sebastian Ruder, and Ivan Vulic. On the limitations of unsupervised bilingual", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 492, + 263, + 505 + ], + "spans": [ + { + "bbox": [ + 116, + 492, + 263, + 505 + ], + "score": 1.0, + "content": "dictionary induction. In ACL, 2018.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 511, + 503, + 535 + ], + "lines": [ + { + "bbox": [ + 106, + 511, + 504, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 504, + 524 + ], + "score": 1.0, + "content": "Jorg Tiedemann. Parallel data, tools and interfaces in opus. In ¨ Proceedings of the Eighth Interna-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 522, + 501, + 535 + ], + "spans": [ + { + "bbox": [ + 115, + 522, + 501, + 535 + ], + "score": 1.0, + "content": "tional Conference on Language Resources and Evaluation (LREC-2012), pp. 2214–2218, 2012.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 503, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 540, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 505, + 555 + ], + "score": 1.0, + "content": "Ivan Vulic and Marie-Francine Moens. Bilingual word embeddings from non-parallel document-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 552, + 382, + 564 + ], + "spans": [ + { + "bbox": [ + 115, + 552, + 382, + 564 + ], + "score": 1.0, + "content": "aligned data applied to bilingual lexicon induction. In ACL, 2015.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Colette Joubarne and Diana Inkpen. Comparison of semantic similarity for different languages using", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "the google n-gram corpus and second-order co-occurrence measures. In Canadian Conference on", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 104, + 156, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 104, + 156, + 117 + ], + "score": 1.0, + "content": "AI, 2011.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 506, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 123, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 106, + 123, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 505, + 136 + ], + "score": 1.0, + "content": "Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. Bag of tricks for efficient", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 135, + 257, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 135, + 257, + 146 + ], + "score": 1.0, + "content": "text classification. In EACL, 2016.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 106, + 123, + 505, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 153, + 504, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 154, + 504, + 165 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 504, + 165 + ], + "score": 1.0, + "content": "Armand Joulin, Piotr Bojanowski, Tomas Mikolov, Herve J ´ egou, and Edouard Grave. Loss in trans-´", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 164, + 458, + 176 + ], + "spans": [ + { + "bbox": [ + 115, + 164, + 458, + 176 + ], + "score": 1.0, + "content": "lation: Learning bilingual word mapping with a retrieval criterion. In EMNLP, 2018.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 106, + 154, + 504, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 183, + 445, + 195 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 446, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 446, + 196 + ], + "score": 1.0, + "content": "Philipp Koehn. Europarl: A parallel corpus for statistical machine translation. 2005.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 182, + 446, + 196 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 201, + 504, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 506, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 506, + 215 + ], + "score": 1.0, + "content": "Pierre Lison and Jorg Tiedemann. Opensubtitles2016: Extracting large parallel corpora from movie ¨", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 213, + 248, + 225 + ], + "spans": [ + { + "bbox": [ + 116, + 213, + 248, + 225 + ], + "score": 1.0, + "content": "and tv subtitles. In LREC, 2016.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 200, + 506, + 225 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 231, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 231, + 504, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 504, + 245 + ], + "score": 1.0, + "content": "Thang Luong, Hieu Pham, and Christopher D Manning. Bilingual word representations with mono-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 242, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 115, + 242, + 505, + 256 + ], + "score": 1.0, + "content": "lingual quality in mind. In Proceedings of the 1st Workshop on Vector Space Modeling for Natural", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 254, + 287, + 266 + ], + "spans": [ + { + "bbox": [ + 116, + 254, + 287, + 266 + ], + "score": 1.0, + "content": "Language Processing, pp. 151–159, 2015.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 231, + 505, + 266 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 106, + 272, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 505, + 285 + ], + "score": 1.0, + "content": "Tomas Mikolov, Kai Chen, Gregory S. Corrado, and Jeffrey Dean. Efficient estimation of word", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 284, + 367, + 296 + ], + "spans": [ + { + "bbox": [ + 116, + 284, + 367, + 296 + ], + "score": 1.0, + "content": "representations in vector space. CoRR, abs/1301.3781, 2013a.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 272, + 505, + 296 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 302, + 504, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 504, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 504, + 316 + ], + "score": 1.0, + "content": "Tomas Mikolov, Quoc V. Le, and Ilya Sutskever. Exploiting similarities among languages for ma-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 117, + 313, + 310, + 325 + ], + "spans": [ + { + "bbox": [ + 117, + 313, + 310, + 325 + ], + "score": 1.0, + "content": "chine translation. ArXiv, abs/1309.4168, 2013b.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 302, + 504, + 325 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 332, + 504, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 345 + ], + "score": 1.0, + "content": "Aitor Ormazabal, Mikel Artetxe, Gorka Labaka, Aitor Soroa, and Eneko Agirre. Analyzing the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 344, + 399, + 355 + ], + "spans": [ + { + "bbox": [ + 116, + 344, + 399, + 355 + ], + "score": 1.0, + "content": "limitations of cross-lingual word embedding mappings. In ACL, 2019.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 106, + 331, + 505, + 355 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 362, + 502, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 504, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 504, + 375 + ], + "score": 1.0, + "content": "Matteo Pagliardini, Prakhar Gupta, and Martin Jaggi. Unsupervised learning of sentence embed-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 374, + 385, + 385 + ], + "spans": [ + { + "bbox": [ + 115, + 374, + 385, + 385 + ], + "score": 1.0, + "content": "dings using compositional n-gram features. In NAACL-HLT, 2018.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 362, + 504, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 502, + 415 + ], + "lines": [ + { + "bbox": [ + 106, + 392, + 504, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 504, + 405 + ], + "score": 1.0, + "content": "Sebastian Ruder, Ivan Vulic, and Anders Søgaard. A survey of cross-lingual word embedding mod-´", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 403, + 295, + 415 + ], + "spans": [ + { + "bbox": [ + 116, + 403, + 295, + 415 + ], + "score": 1.0, + "content": "els. arXiv preprint arXiv:1706.04902, 2017.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 106, + 392, + 504, + 415 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 422, + 503, + 444 + ], + "lines": [ + { + "bbox": [ + 105, + 420, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 436 + ], + "score": 1.0, + "content": "Holger Schwenk and Xian Li. A corpus for multilingual document classification in eight languages.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 433, + 279, + 445 + ], + "spans": [ + { + "bbox": [ + 115, + 433, + 279, + 445 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1805.09821, 2018.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 420, + 505, + 445 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 452, + 503, + 475 + ], + "lines": [ + { + "bbox": [ + 106, + 451, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 465 + ], + "score": 1.0, + "content": "Samuel L. Smith, David H. P. Turban, Steven Hamblin, and Nils Y. Hammerla. Offline bilingual", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 115, + 463, + 505, + 475 + ], + "score": 1.0, + "content": "word vectors, orthogonal transformations and the inverted softmax. ArXiv, abs/1702.03859, 2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 451, + 505, + 475 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 481, + 503, + 505 + ], + "lines": [ + { + "bbox": [ + 106, + 480, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 495 + ], + "score": 1.0, + "content": "Anders Søgaard, Sebastian Ruder, and Ivan Vulic. On the limitations of unsupervised bilingual", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 492, + 263, + 505 + ], + "spans": [ + { + "bbox": [ + 116, + 492, + 263, + 505 + ], + "score": 1.0, + "content": "dictionary induction. In ACL, 2018.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 480, + 505, + 505 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 511, + 503, + 535 + ], + "lines": [ + { + "bbox": [ + 106, + 511, + 504, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 504, + 524 + ], + "score": 1.0, + "content": "Jorg Tiedemann. Parallel data, tools and interfaces in opus. In ¨ Proceedings of the Eighth Interna-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 522, + 501, + 535 + ], + "spans": [ + { + "bbox": [ + 115, + 522, + 501, + 535 + ], + "score": 1.0, + "content": "tional Conference on Language Resources and Evaluation (LREC-2012), pp. 2214–2218, 2012.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 106, + 511, + 504, + 535 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 503, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 540, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 505, + 555 + ], + "score": 1.0, + "content": "Ivan Vulic and Marie-Francine Moens. Bilingual word embeddings from non-parallel document-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 552, + 382, + 564 + ], + "spans": [ + { + "bbox": [ + 115, + 552, + 382, + 564 + ], + "score": 1.0, + "content": "aligned data applied to bilingual lexicon induction. In ACL, 2015.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 540, + 505, + 564 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 183, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 79, + 184, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 79, + 184, + 96 + ], + "score": 1.0, + "content": "A APPENDIX", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 106, + 127, + 438, + 357 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 103, + 225, + 115 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 102, + 225, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 225, + 116 + ], + "score": 1.0, + "content": "A.1 DATASET STATISTICS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 106, + 127, + 438, + 357 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 127, + 438, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 438, + 357 + ], + "score": 0.985, + "html": "
DatasetNumber of sentencesNumber of tokens(English tokens if bilingual)
En-De ParaCrawl v4.017 Million308 Million
En-Es ParaCrawl v4.022 Million477 Million
En-FiParaCrawl v4.02.16 Million42 Million
En-FrParaCrawl v4.032 Million665 Million
En-Hu ParaCrawl v4.01.91Million31Million
En-It ParaCrawl v4.013 Million261 Million
En-Ru OpenSubtitles+ Tanzil27 Million363Million
Wikipedia - En70 Million1792 Million
Wikipedia - De二1384 Million
Wikipedia - Fr二1108Million
Wikipedia - Es二797 Million
Wikipedia - It二702 Million
Wikipedia - Ru二824 Million
Common Crawl - En二600 Billion
Common Crawl - De二66 Billion
Common Crawl -Fr二68 Billion
Common Crawl - It二36 Billion
Common Crawl -Es172 Billion
", + "type": "table", + "image_path": "34da396cee228ad7f3a75f949b12e1420d31d2c471b6948a734534e6f4f80a95.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 106, + 127, + 438, + 203.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 203.66666666666669, + 438, + 280.33333333333337 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 280.33333333333337, + 438, + 357.00000000000006 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 364, + 503, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 378 + ], + "score": 1.0, + "content": "Table 6: Dataset Sizes. ‘En’,‘De’,‘Fi’,‘Fr’,‘Hu’,‘It’,‘Es’ and ‘Ru’ stand for English, German,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 374, + 388, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 388, + 388 + ], + "score": 1.0, + "content": "Finnish, French, Hungarian, Italian, Spanish and Russian respectively.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 107, + 398, + 505, + 442 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 191, + 411 + ], + "score": 1.0, + "content": "We used ParaCrawl", + "type": "text" + }, + { + "bbox": [ + 191, + 398, + 211, + 409 + ], + "score": 0.33, + "content": "\\mathrm { v } 4 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 398, + 506, + 411 + ], + "score": 1.0, + "content": "corpora for training BI-SENT2VEC, SENT2VEC,BIVEC,VECMAP and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 408, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 505, + 422 + ], + "score": 1.0, + "content": "TRANSGRAM embeddings except for En-Ru pair for which we used OpenSubtitles and Tanzil cor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "pora combined. MUSE and RCSLS vectors were trained from FASTTEXT vectors obtained from", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 432, + 263, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 263, + 443 + ], + "score": 1.0, + "content": "Wikipedia dumps(Grave et al., 2018a).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 183, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 79, + 184, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 79, + 184, + 96 + ], + "score": 1.0, + "content": "A APPENDIX", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 106, + 127, + 438, + 357 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 103, + 225, + 115 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 102, + 225, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 225, + 116 + ], + "score": 1.0, + "content": "A.1 DATASET STATISTICS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 106, + 127, + 438, + 357 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 127, + 438, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 438, + 357 + ], + "score": 0.985, + "html": "
DatasetNumber of sentencesNumber of tokens(English tokens if bilingual)
En-De ParaCrawl v4.017 Million308 Million
En-Es ParaCrawl v4.022 Million477 Million
En-FiParaCrawl v4.02.16 Million42 Million
En-FrParaCrawl v4.032 Million665 Million
En-Hu ParaCrawl v4.01.91Million31Million
En-It ParaCrawl v4.013 Million261 Million
En-Ru OpenSubtitles+ Tanzil27 Million363Million
Wikipedia - En70 Million1792 Million
Wikipedia - De二1384 Million
Wikipedia - Fr二1108Million
Wikipedia - Es二797 Million
Wikipedia - It二702 Million
Wikipedia - Ru二824 Million
Common Crawl - En二600 Billion
Common Crawl - De二66 Billion
Common Crawl -Fr二68 Billion
Common Crawl - It二36 Billion
Common Crawl -Es172 Billion
", + "type": "table", + "image_path": "34da396cee228ad7f3a75f949b12e1420d31d2c471b6948a734534e6f4f80a95.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 106, + 127, + 438, + 203.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 203.66666666666669, + 438, + 280.33333333333337 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 280.33333333333337, + 438, + 357.00000000000006 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 364, + 503, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 378 + ], + "score": 1.0, + "content": "Table 6: Dataset Sizes. ‘En’,‘De’,‘Fi’,‘Fr’,‘Hu’,‘It’,‘Es’ and ‘Ru’ stand for English, German,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 374, + 388, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 388, + 388 + ], + "score": 1.0, + "content": "Finnish, French, Hungarian, Italian, Spanish and Russian respectively.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 105, + 362, + 505, + 388 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 398, + 505, + 442 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 191, + 411 + ], + "score": 1.0, + "content": "We used ParaCrawl", + "type": "text" + }, + { + "bbox": [ + 191, + 398, + 211, + 409 + ], + "score": 0.33, + "content": "\\mathrm { v } 4 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 398, + 506, + 411 + ], + "score": 1.0, + "content": "corpora for training BI-SENT2VEC, SENT2VEC,BIVEC,VECMAP and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 408, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 505, + 422 + ], + "score": 1.0, + "content": "TRANSGRAM embeddings except for En-Ru pair for which we used OpenSubtitles and Tanzil cor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "pora combined. MUSE and RCSLS vectors were trained from FASTTEXT vectors obtained from", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 432, + 263, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 263, + 443 + ], + "score": 1.0, + "content": "Wikipedia dumps(Grave et al., 2018a).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 398, + 506, + 443 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 83, + 334, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 336, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 336, + 95 + ], + "score": 1.0, + "content": "A.2 TRAINING PARAMETERS FOR TRAINED MODELS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 106, + 106, + 492, + 245 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 106, + 492, + 245 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 106, + 492, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 492, + 245 + ], + "score": 0.978, + "html": "
ModelBI-SENT2VECuni.BI-SENT2VECuni. + bi.SENT2VECuni.TRANSGRAM
Embedding dimension300300300300
Maxvocabulary size750k750k750k750k
Minimum word count5855
Initial Learning Rate0.20.20.20.025
Epochs5558
Subsampling hyper-parameter1·10-55:10-61·10-51·10-4
Word-Ngrams Bucket Size二2M二二
Word-Ngrams dropped per context141二
Window size5
Number of negatives sampled1010105
", + "type": "table", + "image_path": "5d36ad4d3a232e416a5e0819224e5e023f5c7a3f99da04945ab280cf5ab981d7.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 106, + 492, + 152.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 152.33333333333334, + 492, + 198.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 198.66666666666669, + 492, + 245.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 207, + 253, + 404, + 265 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 206, + 253, + 405, + 265 + ], + "spans": [ + { + "bbox": [ + 206, + 253, + 405, + 265 + ], + "score": 1.0, + "content": "Table 7: Hyperparameters for the trained models", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + } + ], + "index": 3.0 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 83, + 334, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 336, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 336, + 95 + ], + "score": 1.0, + "content": "A.2 TRAINING PARAMETERS FOR TRAINED MODELS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 106, + 106, + 492, + 245 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 106, + 492, + 245 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 106, + 492, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 492, + 245 + ], + "score": 0.978, + "html": "
ModelBI-SENT2VECuni.BI-SENT2VECuni. + bi.SENT2VECuni.TRANSGRAM
Embedding dimension300300300300
Maxvocabulary size750k750k750k750k
Minimum word count5855
Initial Learning Rate0.20.20.20.025
Epochs5558
Subsampling hyper-parameter1·10-55:10-61·10-51·10-4
Word-Ngrams Bucket Size二2M二二
Word-Ngrams dropped per context141二
Window size5
Number of negatives sampled1010105
", + "type": "table", + "image_path": "5d36ad4d3a232e416a5e0819224e5e023f5c7a3f99da04945ab280cf5ab981d7.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 106, + 492, + 152.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 152.33333333333334, + 492, + 198.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 198.66666666666669, + 492, + 245.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 207, + 253, + 404, + 265 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 206, + 253, + 405, + 265 + ], + "spans": [ + { + "bbox": [ + 206, + 253, + 405, + 265 + ], + "score": 1.0, + "content": "Table 7: Hyperparameters for the trained models", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + } + ], + "index": 3.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 185, + 107, + 425, + 250 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 83, + 334, + 94 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 82, + 336, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 336, + 95 + ], + "score": 1.0, + "content": "A.3 ADDITIONAL MONOLINGUAL QUALITY TABLES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 185, + 107, + 425, + 250 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 185, + 107, + 425, + 250 + ], + "spans": [ + { + "bbox": [ + 185, + 107, + 425, + 250 + ], + "score": 0.979, + "html": "
SimLex-999 Method\\DatasetWS-353
enenes
MUSE RCSLS0.380.740.61
FASTTEXT- Common Crawl0.38 0.490.74 0.750.62 0.54
BIVEC0.400.720.57
TRANSGRAM0.420.740.59
SENT2VEC uni.0.490.580.51
BI-SENT2VEC uni.0.570.780.60
BI-SENT2VEC uni. + bi.0.600.820.66
", + "type": "table", + "image_path": "6dd016b50cfe64778b72f82487b56858d38fb6a0aec44d5ed0fc98072d36ebf1.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 185, + 107, + 425, + 120.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 185, + 120.0, + 425, + 133.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 185, + 133.0, + 425, + 146.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 185, + 146.0, + 425, + 159.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 185, + 159.0, + 425, + 172.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 185, + 172.0, + 425, + 185.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 185, + 185.0, + 425, + 198.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 185, + 198.0, + 425, + 211.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 185, + 211.0, + 425, + 224.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 185, + 224.0, + 425, + 237.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 185, + 237.0, + 425, + 250.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 3.0 + }, + { + "type": "table", + "bbox": [ + 169, + 302, + 442, + 444 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 103, + 261, + 504, + 284 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 259, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 275 + ], + "score": 1.0, + "content": "Table 8: Monolingual word similarity task performance of our methods when trained on en-es", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 271, + 332, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 332, + 285 + ], + "score": 1.0, + "content": "ParaCrawl data. We report Pearson correlation scores.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "table_body", + "bbox": [ + 169, + 302, + 442, + 444 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 169, + 302, + 442, + 444 + ], + "spans": [ + { + "bbox": [ + 169, + 302, + 442, + 444 + ], + "score": 0.981, + "html": "
Method\\DatasetSimLex-999 enWS-353 enRG-65 fr
MUSE0.380.740.72
RCSLS0.380.740.70
FASTTEXT- Common Crawl0.490.750.76
BIVEC0.400.700.74
TRANSGRAM0.390.74
SENT2VEC uni.0.72
0.460.750.71
BI-SENT2VEC uni. BI-SENT2VEC uni. + bi.0.55 0.590.78 0.790.74 0.78
", + "type": "table", + "image_path": "ab74ae0a4ec9a8376403443db7f1f5b502fc937b4df7852c62d53cae7d9000ec.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 169, + 302, + 442, + 349.3333333333333 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 169, + 349.3333333333333, + 442, + 396.66666666666663 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 169, + 396.66666666666663, + 442, + 443.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + } + ], + "index": 13.75 + }, + { + "type": "table", + "bbox": [ + 180, + 497, + 432, + 639 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 455, + 502, + 478 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 455, + 504, + 467 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 504, + 467 + ], + "score": 1.0, + "content": "Table 9: Monolingual word similarity task performance of our methods when trained on en-fr", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 466, + 332, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 332, + 478 + ], + "score": 1.0, + "content": "ParaCrawl data. We report Pearson correlation scores.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "table_body", + "bbox": [ + 180, + 497, + 432, + 639 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 180, + 497, + 432, + 639 + ], + "spans": [ + { + "bbox": [ + 180, + 497, + 432, + 639 + ], + "score": 0.98, + "html": "
Method\\DatasetSimLex-999WS-353
en deende
MUSE0.38 0.410.740.68
RCSLS0.38 0.430.740.70
FASTTEXT- Common Crawl0.49 0.390.750.64
BIVEC0.40 0.410.710.62
TRANSGRAM0.42 0.420.740.66
SENT2VEC uni.0.48 0.380.700.63
BI-SENT2VEC uni.0.56 0.470.760.68
BI-SENT2VEC uni. + bi.0.59 0.530.750.70
", + "type": "table", + "image_path": "1008cbd31964a96ebb37aa7d023437a153bbcaa3186ee3ec0be8366bf85f1dcc.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 180, + 497, + 432, + 544.3333333333334 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 180, + 544.3333333333334, + 432, + 591.6666666666667 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 180, + 591.6666666666667, + 432, + 639.0000000000001 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 105, + 650, + 501, + 672 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 649, + 503, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 503, + 662 + ], + "score": 1.0, + "content": "Table 10: Monolingual word similarity task performance of our methods when trained on en-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 660, + 344, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 344, + 673 + ], + "score": 1.0, + "content": "de ParaCrawl data. We report Pearson correlation scores.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + } + ], + "index": 20 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2020", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 185, + 107, + 425, + 250 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 83, + 334, + 94 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 82, + 336, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 336, + 95 + ], + "score": 1.0, + "content": "A.3 ADDITIONAL MONOLINGUAL QUALITY TABLES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 185, + 107, + 425, + 250 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 185, + 107, + 425, + 250 + ], + "spans": [ + { + "bbox": [ + 185, + 107, + 425, + 250 + ], + "score": 0.979, + "html": "
SimLex-999 Method\\DatasetWS-353
enenes
MUSE RCSLS0.380.740.61
FASTTEXT- Common Crawl0.38 0.490.74 0.750.62 0.54
BIVEC0.400.720.57
TRANSGRAM0.420.740.59
SENT2VEC uni.0.490.580.51
BI-SENT2VEC uni.0.570.780.60
BI-SENT2VEC uni. + bi.0.600.820.66
", + "type": "table", + "image_path": "6dd016b50cfe64778b72f82487b56858d38fb6a0aec44d5ed0fc98072d36ebf1.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 185, + 107, + 425, + 120.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 185, + 120.0, + 425, + 133.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 185, + 133.0, + 425, + 146.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 185, + 146.0, + 425, + 159.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 185, + 159.0, + 425, + 172.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 185, + 172.0, + 425, + 185.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 185, + 185.0, + 425, + 198.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 185, + 198.0, + 425, + 211.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 185, + 211.0, + 425, + 224.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 185, + 224.0, + 425, + 237.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 185, + 237.0, + 425, + 250.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 3.0 + }, + { + "type": "table", + "bbox": [ + 169, + 302, + 442, + 444 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 103, + 261, + 504, + 284 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 259, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 275 + ], + "score": 1.0, + "content": "Table 8: Monolingual word similarity task performance of our methods when trained on en-es", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 271, + 332, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 332, + 285 + ], + "score": 1.0, + "content": "ParaCrawl data. We report Pearson correlation scores.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "table_body", + "bbox": [ + 169, + 302, + 442, + 444 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 169, + 302, + 442, + 444 + ], + "spans": [ + { + "bbox": [ + 169, + 302, + 442, + 444 + ], + "score": 0.981, + "html": "
Method\\DatasetSimLex-999 enWS-353 enRG-65 fr
MUSE0.380.740.72
RCSLS0.380.740.70
FASTTEXT- Common Crawl0.490.750.76
BIVEC0.400.700.74
TRANSGRAM0.390.74
SENT2VEC uni.0.72
0.460.750.71
BI-SENT2VEC uni. BI-SENT2VEC uni. + bi.0.55 0.590.78 0.790.74 0.78
", + "type": "table", + "image_path": "ab74ae0a4ec9a8376403443db7f1f5b502fc937b4df7852c62d53cae7d9000ec.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 169, + 302, + 442, + 349.3333333333333 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 169, + 349.3333333333333, + 442, + 396.66666666666663 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 169, + 396.66666666666663, + 442, + 443.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + } + ], + "index": 13.75 + }, + { + "type": "table", + "bbox": [ + 180, + 497, + 432, + 639 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 455, + 502, + 478 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 455, + 504, + 467 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 504, + 467 + ], + "score": 1.0, + "content": "Table 9: Monolingual word similarity task performance of our methods when trained on en-fr", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 466, + 332, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 332, + 478 + ], + "score": 1.0, + "content": "ParaCrawl data. We report Pearson correlation scores.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "table_body", + "bbox": [ + 180, + 497, + 432, + 639 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 180, + 497, + 432, + 639 + ], + "spans": [ + { + "bbox": [ + 180, + 497, + 432, + 639 + ], + "score": 0.98, + "html": "
Method\\DatasetSimLex-999WS-353
en deende
MUSE0.38 0.410.740.68
RCSLS0.38 0.430.740.70
FASTTEXT- Common Crawl0.49 0.390.750.64
BIVEC0.40 0.410.710.62
TRANSGRAM0.42 0.420.740.66
SENT2VEC uni.0.48 0.380.700.63
BI-SENT2VEC uni.0.56 0.470.760.68
BI-SENT2VEC uni. + bi.0.59 0.530.750.70
", + "type": "table", + "image_path": "1008cbd31964a96ebb37aa7d023437a153bbcaa3186ee3ec0be8366bf85f1dcc.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 180, + 497, + 432, + 544.3333333333334 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 180, + 544.3333333333334, + 432, + 591.6666666666667 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 180, + 591.6666666666667, + 432, + 639.0000000000001 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 105, + 650, + 501, + 672 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 649, + 503, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 503, + 662 + ], + "score": 1.0, + "content": "Table 10: Monolingual word similarity task performance of our methods when trained on en-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 660, + 344, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 344, + 673 + ], + "score": 1.0, + "content": "de ParaCrawl data. We report Pearson correlation scores.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + } + ], + "index": 20 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/SJlJSaEFwS/SJlJSaEFwS_model.json b/parse/train/SJlJSaEFwS/SJlJSaEFwS_model.json new file mode 100644 index 0000000000000000000000000000000000000000..89755b2062ab74b4831304515ae3f7b0ae6da29e --- /dev/null +++ b/parse/train/SJlJSaEFwS/SJlJSaEFwS_model.json @@ -0,0 +1,13050 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 580, + 1302, + 580, + 1302, + 1161, + 398, + 1161 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 299, + 1513, + 1402, + 1513, + 1402, + 1727, + 299, + 1727 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1252, + 1403, + 1252, + 1403, + 1498, + 298, + 1498 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1742, + 1402, + 1742, + 1402, + 1957, + 299, + 1957 + ], + "score": 0.977 + }, + { + "category_id": 0, + "poly": [ + 300, + 219, + 1244, + 219, + 1244, + 324, + 300, + 324 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 300, + 1972, + 1399, + 1972, + 1399, + 2033, + 300, + 2033 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 302, + 378, + 668, + 378, + 668, + 438, + 302, + 438 + ], + "score": 0.931 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 104, + 298, + 104 + ], + "score": 0.898 + }, + { + "category_id": 0, + "poly": [ + 301, + 1199, + 573, + 1199, + 573, + 1233, + 301, + 1233 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 773, + 520, + 926, + 520, + 926, + 553, + 773, + 553 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 842, + 2089, + 857, + 2089, + 857, + 2112, + 842, + 2112 + ], + "score": 0.633 + }, + { + "category_id": 15, + "poly": [ + 294.0, + 218.0, + 1249.0, + 218.0, + 1249.0, + 273.0, + 294.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 270.0, + 776.0, + 270.0, + 776.0, + 328.0, + 293.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1195.0, + 579.0, + 1195.0, + 579.0, + 1241.0, + 293.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 519.0, + 932.0, + 519.0, + 932.0, + 556.0, + 769.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 2088.0, + 860.0, + 2088.0, + 860.0, + 2118.0, + 841.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 580.0, + 1306.0, + 580.0, + 1306.0, + 618.0, + 393.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 614.0, + 1304.0, + 614.0, + 1304.0, + 647.0, + 393.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 642.0, + 1307.0, + 642.0, + 1307.0, + 681.0, + 393.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 674.0, + 1307.0, + 674.0, + 1307.0, + 710.0, + 392.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 705.0, + 1303.0, + 705.0, + 1303.0, + 737.0, + 394.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 734.0, + 1305.0, + 734.0, + 1305.0, + 771.0, + 393.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 767.0, + 1303.0, + 767.0, + 1303.0, + 800.0, + 393.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 798.0, + 1303.0, + 798.0, + 1303.0, + 831.0, + 393.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 827.0, + 1306.0, + 827.0, + 1306.0, + 863.0, + 392.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 854.0, + 1308.0, + 854.0, + 1308.0, + 895.0, + 390.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 889.0, + 1305.0, + 889.0, + 1305.0, + 922.0, + 393.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 917.0, + 1306.0, + 917.0, + 1306.0, + 953.0, + 394.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 947.0, + 1307.0, + 947.0, + 1307.0, + 986.0, + 392.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 979.0, + 1306.0, + 979.0, + 1306.0, + 1012.0, + 394.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1011.0, + 1306.0, + 1011.0, + 1306.0, + 1044.0, + 395.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1040.0, + 1307.0, + 1040.0, + 1307.0, + 1076.0, + 394.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1068.0, + 1305.0, + 1068.0, + 1305.0, + 1104.0, + 394.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1101.0, + 1305.0, + 1101.0, + 1305.0, + 1134.0, + 394.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1129.0, + 953.0, + 1129.0, + 953.0, + 1165.0, + 396.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1512.0, + 1404.0, + 1512.0, + 1404.0, + 1550.0, + 294.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1545.0, + 1404.0, + 1545.0, + 1404.0, + 1580.0, + 294.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1572.0, + 1406.0, + 1572.0, + 1406.0, + 1610.0, + 292.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1605.0, + 1406.0, + 1605.0, + 1406.0, + 1638.0, + 293.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1635.0, + 1406.0, + 1635.0, + 1406.0, + 1669.0, + 294.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1665.0, + 1406.0, + 1665.0, + 1406.0, + 1700.0, + 295.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1696.0, + 1054.0, + 1696.0, + 1054.0, + 1730.0, + 294.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1251.0, + 1404.0, + 1251.0, + 1404.0, + 1288.0, + 294.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1282.0, + 1404.0, + 1282.0, + 1404.0, + 1320.0, + 292.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1313.0, + 1405.0, + 1313.0, + 1405.0, + 1349.0, + 293.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1341.0, + 1407.0, + 1341.0, + 1407.0, + 1381.0, + 292.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1375.0, + 1405.0, + 1375.0, + 1405.0, + 1409.0, + 294.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1406.0, + 1405.0, + 1406.0, + 1405.0, + 1440.0, + 294.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1436.0, + 1404.0, + 1436.0, + 1404.0, + 1469.0, + 292.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1470.0, + 389.0, + 1470.0, + 389.0, + 1506.0, + 293.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1740.0, + 1406.0, + 1740.0, + 1406.0, + 1777.0, + 294.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1772.0, + 1407.0, + 1772.0, + 1407.0, + 1809.0, + 292.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1804.0, + 1404.0, + 1804.0, + 1404.0, + 1838.0, + 294.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1835.0, + 1406.0, + 1835.0, + 1406.0, + 1869.0, + 294.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1864.0, + 1404.0, + 1864.0, + 1404.0, + 1899.0, + 293.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1894.0, + 1407.0, + 1894.0, + 1407.0, + 1931.0, + 292.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1927.0, + 509.0, + 1927.0, + 509.0, + 1960.0, + 293.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2008.0, + 296.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2001.0, + 1404.0, + 2001.0, + 1404.0, + 2036.0, + 292.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 377.0, + 550.0, + 377.0, + 550.0, + 412.0, + 302.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 407.0, + 670.0, + 407.0, + 670.0, + 442.0, + 301.0, + 442.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1668, + 1403, + 1668, + 1403, + 2034, + 298, + 2034 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1178, + 1404, + 1178, + 1404, + 1392, + 298, + 1392 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1407, + 1404, + 1407, + 1404, + 1653, + 298, + 1653 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 301, + 229, + 1399, + 229, + 1399, + 322, + 301, + 322 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 301, + 337, + 1402, + 337, + 1402, + 460, + 301, + 460 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 368, + 477, + 1403, + 477, + 1403, + 1033, + 368, + 1033 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 301, + 1052, + 848, + 1052, + 848, + 1084, + 301, + 1084 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 300, + 1117, + 588, + 1117, + 588, + 1154, + 300, + 1154 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 858, + 76, + 858, + 104, + 298, + 104 + ], + "score": 0.775 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.716 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.117 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.113 + }, + { + "category_id": 13, + "poly": [ + 397, + 1003, + 463, + 1003, + 463, + 1031, + 397, + 1031 + ], + "score": 0.86, + "latex": "1 0 0 \\times" + }, + { + "category_id": 13, + "poly": [ + 1014, + 1886, + 1039, + 1886, + 1039, + 1910, + 1014, + 1910 + ], + "score": 0.65, + "latex": "^ +" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1115.0, + 593.0, + 1115.0, + 593.0, + 1159.0, + 292.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 70.0, + 859.0, + 70.0, + 859.0, + 109.0, + 296.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1668.0, + 1407.0, + 1668.0, + 1407.0, + 1705.0, + 293.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1697.0, + 1404.0, + 1697.0, + 1404.0, + 1736.0, + 293.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1728.0, + 1404.0, + 1728.0, + 1404.0, + 1765.0, + 293.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1759.0, + 1405.0, + 1759.0, + 1405.0, + 1795.0, + 292.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1787.0, + 1404.0, + 1787.0, + 1404.0, + 1828.0, + 292.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1856.0, + 292.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 1404.0, + 1851.0, + 1404.0, + 1885.0, + 294.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1881.0, + 1013.0, + 1881.0, + 1013.0, + 1917.0, + 293.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 1881.0, + 1407.0, + 1881.0, + 1407.0, + 1917.0, + 1040.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1910.0, + 1404.0, + 1910.0, + 1404.0, + 1948.0, + 294.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 1407.0, + 2001.0, + 1407.0, + 2039.0, + 294.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1175.0, + 1404.0, + 1175.0, + 1404.0, + 1214.0, + 293.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1209.0, + 1405.0, + 1209.0, + 1405.0, + 1243.0, + 294.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1235.0, + 1403.0, + 1235.0, + 1403.0, + 1278.0, + 292.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1268.0, + 1405.0, + 1268.0, + 1405.0, + 1308.0, + 292.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1297.0, + 1403.0, + 1297.0, + 1403.0, + 1337.0, + 292.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1330.0, + 1404.0, + 1330.0, + 1404.0, + 1368.0, + 293.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1357.0, + 504.0, + 1357.0, + 504.0, + 1399.0, + 292.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1405.0, + 1405.0, + 1405.0, + 1405.0, + 1445.0, + 295.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1439.0, + 1405.0, + 1439.0, + 1405.0, + 1474.0, + 293.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1466.0, + 1405.0, + 1466.0, + 1405.0, + 1506.0, + 292.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1500.0, + 1404.0, + 1500.0, + 1404.0, + 1534.0, + 293.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1528.0, + 1404.0, + 1528.0, + 1404.0, + 1564.0, + 293.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1561.0, + 1404.0, + 1561.0, + 1404.0, + 1595.0, + 296.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1588.0, + 1404.0, + 1588.0, + 1404.0, + 1627.0, + 292.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1620.0, + 854.0, + 1620.0, + 854.0, + 1657.0, + 293.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 228.0, + 1404.0, + 228.0, + 1404.0, + 265.0, + 292.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 262.0, + 1403.0, + 262.0, + 1403.0, + 296.0, + 296.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 292.0, + 699.0, + 292.0, + 699.0, + 326.0, + 296.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 337.0, + 1406.0, + 337.0, + 1406.0, + 373.0, + 296.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 366.0, + 1406.0, + 366.0, + 1406.0, + 403.0, + 294.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 398.0, + 1404.0, + 398.0, + 1404.0, + 433.0, + 295.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 431.0, + 980.0, + 431.0, + 980.0, + 464.0, + 296.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 481.0, + 1403.0, + 481.0, + 1403.0, + 515.0, + 374.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 510.0, + 1407.0, + 510.0, + 1407.0, + 548.0, + 394.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 543.0, + 1405.0, + 543.0, + 1405.0, + 577.0, + 395.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 573.0, + 1405.0, + 573.0, + 1405.0, + 607.0, + 395.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 603.0, + 1406.0, + 603.0, + 1406.0, + 638.0, + 392.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 635.0, + 491.0, + 635.0, + 491.0, + 663.0, + 393.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 673.0, + 1405.0, + 673.0, + 1405.0, + 709.0, + 391.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 707.0, + 1403.0, + 707.0, + 1403.0, + 737.0, + 395.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 733.0, + 1405.0, + 733.0, + 1405.0, + 771.0, + 394.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 766.0, + 1403.0, + 766.0, + 1403.0, + 800.0, + 394.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 797.0, + 1404.0, + 797.0, + 1404.0, + 831.0, + 394.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 830.0, + 544.0, + 830.0, + 544.0, + 861.0, + 396.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 867.0, + 1405.0, + 867.0, + 1405.0, + 905.0, + 385.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 901.0, + 1239.0, + 901.0, + 1239.0, + 935.0, + 395.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 938.0, + 1406.0, + 938.0, + 1406.0, + 977.0, + 374.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 971.0, + 1406.0, + 971.0, + 1406.0, + 1008.0, + 394.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1001.0, + 626.0, + 1001.0, + 626.0, + 1040.0, + 464.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1049.0, + 853.0, + 1049.0, + 853.0, + 1088.0, + 297.0, + 1088.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 506, + 1404, + 506, + 1404, + 843, + 298, + 843 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1097, + 1404, + 1097, + 1404, + 1320, + 297, + 1320 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1334, + 1403, + 1334, + 1403, + 1463, + 298, + 1463 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1567, + 1403, + 1567, + 1403, + 1728, + 297, + 1728 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 857, + 1403, + 857, + 1403, + 1012, + 299, + 1012 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1739, + 1403, + 1739, + 1403, + 1863, + 297, + 1863 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 337, + 1403, + 337, + 1403, + 492, + 298, + 492 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1942, + 1402, + 1942, + 1402, + 2035, + 299, + 2035 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 300, + 229, + 1402, + 229, + 1402, + 324, + 300, + 324 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 507, + 1472, + 1191, + 1472, + 1191, + 1558, + 507, + 1558 + ], + "score": 0.958 + }, + { + "category_id": 0, + "poly": [ + 299, + 1043, + 458, + 1043, + 458, + 1078, + 299, + 1078 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 104, + 298, + 104 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1871, + 1399, + 1871, + 1399, + 1901, + 1366, + 1901 + ], + "score": 0.882 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1493, + 1400, + 1493, + 1400, + 1523, + 1366, + 1523 + ], + "score": 0.881 + }, + { + "category_id": 8, + "poly": [ + 651, + 1863, + 1045, + 1863, + 1045, + 1937, + 651, + 1937 + ], + "score": 0.719 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.654 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.528 + }, + { + "category_id": 14, + "poly": [ + 654, + 1859, + 1045, + 1859, + 1045, + 1939, + 654, + 1939 + ], + "score": 0.94, + "latex": "\\ell \\big ( { \\pmb u } _ { w _ { t } } ^ { \\top } { \\pmb v } _ { S _ { l _ { 2 } } } \\big ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell \\big ( - { \\pmb u } _ { w _ { t } ^ { \\prime } } ^ { \\top } { \\pmb v } _ { S _ { l _ { 2 } } } \\big )" + }, + { + "category_id": 14, + "poly": [ + 507, + 1470, + 1193, + 1470, + 1193, + 1561, + 507, + 1561 + ], + "score": 0.94, + "latex": "\\operatorname* { m i n } _ { U , V } \\sum _ { S \\in C } \\sum _ { w _ { t } \\in S } \\left( \\ell \\big ( \\boldsymbol { u } _ { w _ { t } } ^ { \\top } \\boldsymbol { v } _ { S \\setminus \\{ w _ { t } \\} } \\big ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell \\big ( - \\boldsymbol { u } _ { w ^ { \\prime } } ^ { \\top } \\boldsymbol { v } _ { S \\setminus \\{ w _ { t } \\} } \\big ) \\right)" + }, + { + "category_id": 13, + "poly": [ + 988, + 1773, + 1151, + 1773, + 1151, + 1806, + 988, + 1806 + ], + "score": 0.92, + "latex": "S = ( S _ { l _ { 1 } } , S _ { l _ { 2 } } )" + }, + { + "category_id": 13, + "poly": [ + 1230, + 1773, + 1268, + 1773, + 1268, + 1805, + 1230, + 1805 + ], + "score": 0.92, + "latex": "S _ { l _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 725, + 1371, + 817, + 1371, + 817, + 1402, + 725, + 1402 + ], + "score": 0.91, + "latex": "{ \\pmb v } _ { S \\backslash \\{ { u v } _ { t } \\} }" + }, + { + "category_id": 13, + "poly": [ + 1267, + 1252, + 1330, + 1252, + 1330, + 1285, + 1267, + 1285 + ], + "score": 0.91, + "latex": "R ( S )" + }, + { + "category_id": 13, + "poly": [ + 794, + 1600, + 844, + 1600, + 844, + 1633, + 794, + 1633 + ], + "score": 0.91, + "latex": "N _ { w _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 349, + 1399, + 609, + 1399, + 609, + 1434, + 349, + 1434 + ], + "score": 0.9, + "latex": "\\ell : x \\mapsto \\log { ( 1 + e ^ { - x } ) }" + }, + { + "category_id": 13, + "poly": [ + 1293, + 1665, + 1334, + 1665, + 1334, + 1698, + 1293, + 1698 + ], + "score": 0.9, + "latex": "f _ { w _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 1051, + 1945, + 1088, + 1945, + 1088, + 1975, + 1051, + 1975 + ], + "score": 0.9, + "latex": "S _ { l _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 923, + 1660, + 1215, + 1660, + 1215, + 1699, + 923, + 1699 + ], + "score": 0.9, + "latex": "m i n \\{ 1 , \\sqrt { t / f _ { w _ { t } } } + t / f _ { w _ { t } } \\}" + }, + { + "category_id": 13, + "poly": [ + 1321, + 1773, + 1358, + 1773, + 1358, + 1805, + 1321, + 1805 + ], + "score": 0.89, + "latex": "S _ { l _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 606, + 1945, + 643, + 1945, + 643, + 1974, + 606, + 1974 + ], + "score": 0.89, + "latex": "S _ { l _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 903, + 1251, + 1195, + 1251, + 1195, + 1293, + 903, + 1293 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\mathbf { \\dot { \\boldsymbol { v } } } _ { s } : = \\frac { 1 } { R ( S ) } \\sum _ { w \\in R ( S ) } \\pmb { v } _ { w } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 734, + 1804, + 758, + 1804, + 758, + 1833, + 734, + 1833 + ], + "score": 0.87, + "latex": "l _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1351, + 1805, + 1374, + 1805, + 1374, + 1833, + 1351, + 1833 + ], + "score": 0.87, + "latex": "l _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 810, + 1804, + 834, + 1804, + 834, + 1833, + 810, + 1833 + ], + "score": 0.86, + "latex": "l _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 580, + 1667, + 613, + 1667, + 613, + 1695, + 580, + 1695 + ], + "score": 0.86, + "latex": "w _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1284, + 1808, + 1316, + 1808, + 1316, + 1833, + 1284, + 1833 + ], + "score": 0.85, + "latex": "w _ { t }" + }, + { + "category_id": 13, + "poly": [ + 370, + 1573, + 403, + 1573, + 403, + 1599, + 370, + 1599 + ], + "score": 0.85, + "latex": "w _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1101, + 1340, + 1134, + 1340, + 1134, + 1366, + 1101, + 1366 + ], + "score": 0.85, + "latex": "w _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1043, + 1226, + 1073, + 1226, + 1073, + 1252, + 1043, + 1252 + ], + "score": 0.84, + "latex": "{ \\pmb v } _ { s }" + }, + { + "category_id": 13, + "poly": [ + 652, + 1569, + 680, + 1569, + 680, + 1597, + 652, + 1597 + ], + "score": 0.83, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 728, + 1569, + 756, + 1569, + 756, + 1596, + 728, + 1596 + ], + "score": 0.82, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 1312, + 1338, + 1333, + 1338, + 1333, + 1363, + 1312, + 1363 + ], + "score": 0.82, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 379, + 1432, + 405, + 1432, + 405, + 1458, + 379, + 1458 + ], + "score": 0.81, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 1229, + 1223, + 1251, + 1223, + 1251, + 1249, + 1229, + 1249 + ], + "score": 0.81, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 698, + 1288, + 719, + 1288, + 719, + 1315, + 698, + 1315 + ], + "score": 0.8, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 1317, + 1978, + 1335, + 1978, + 1335, + 2001, + 1317, + 2001 + ], + "score": 0.29, + "latex": "\\mathbf { n }" + }, + { + "category_id": 13, + "poly": [ + 1080, + 1132, + 1102, + 1132, + 1102, + 1158, + 1080, + 1158 + ], + "score": 0.28, + "latex": "C" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1038.0, + 466.0, + 1038.0, + 466.0, + 1085.0, + 291.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 507.0, + 1404.0, + 507.0, + 1404.0, + 542.0, + 294.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 535.0, + 1404.0, + 535.0, + 1404.0, + 573.0, + 293.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 565.0, + 1405.0, + 565.0, + 1405.0, + 605.0, + 292.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 598.0, + 1406.0, + 598.0, + 1406.0, + 636.0, + 293.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 629.0, + 1404.0, + 629.0, + 1404.0, + 663.0, + 294.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 661.0, + 1406.0, + 661.0, + 1406.0, + 696.0, + 292.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 688.0, + 1406.0, + 688.0, + 1406.0, + 724.0, + 292.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 720.0, + 1405.0, + 720.0, + 1405.0, + 755.0, + 294.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 750.0, + 1405.0, + 750.0, + 1405.0, + 785.0, + 296.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 784.0, + 1402.0, + 784.0, + 1402.0, + 815.0, + 297.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 812.0, + 685.0, + 812.0, + 685.0, + 847.0, + 293.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1099.0, + 1405.0, + 1099.0, + 1405.0, + 1134.0, + 295.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1130.0, + 1079.0, + 1130.0, + 1079.0, + 1166.0, + 295.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1130.0, + 1405.0, + 1130.0, + 1405.0, + 1166.0, + 1103.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1161.0, + 1405.0, + 1161.0, + 1405.0, + 1196.0, + 295.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1190.0, + 1405.0, + 1190.0, + 1405.0, + 1230.0, + 292.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1215.0, + 1042.0, + 1215.0, + 1042.0, + 1263.0, + 290.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1215.0, + 1228.0, + 1215.0, + 1228.0, + 1263.0, + 1074.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1215.0, + 1406.0, + 1215.0, + 1406.0, + 1263.0, + 1252.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1244.0, + 902.0, + 1244.0, + 902.0, + 1303.0, + 287.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1244.0, + 1266.0, + 1244.0, + 1266.0, + 1303.0, + 1196.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1331.0, + 1244.0, + 1411.0, + 1244.0, + 1411.0, + 1303.0, + 1331.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1287.0, + 697.0, + 1287.0, + 697.0, + 1322.0, + 293.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1287.0, + 730.0, + 1287.0, + 730.0, + 1322.0, + 720.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1332.0, + 1100.0, + 1332.0, + 1100.0, + 1372.0, + 293.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 1332.0, + 1311.0, + 1332.0, + 1311.0, + 1372.0, + 1135.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1332.0, + 1405.0, + 1332.0, + 1405.0, + 1372.0, + 1334.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1363.0, + 724.0, + 1363.0, + 724.0, + 1407.0, + 291.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1363.0, + 1408.0, + 1363.0, + 1408.0, + 1407.0, + 818.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1398.0, + 348.0, + 1398.0, + 348.0, + 1437.0, + 292.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 1398.0, + 1405.0, + 1398.0, + 1405.0, + 1437.0, + 610.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1428.0, + 378.0, + 1428.0, + 378.0, + 1468.0, + 293.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1428.0, + 1104.0, + 1428.0, + 1104.0, + 1468.0, + 406.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1569.0, + 369.0, + 1569.0, + 369.0, + 1603.0, + 296.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1569.0, + 651.0, + 1569.0, + 651.0, + 1603.0, + 404.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1569.0, + 727.0, + 1569.0, + 727.0, + 1603.0, + 681.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1569.0, + 1404.0, + 1569.0, + 1404.0, + 1603.0, + 757.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1599.0, + 793.0, + 1599.0, + 793.0, + 1637.0, + 293.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1599.0, + 1405.0, + 1599.0, + 1405.0, + 1637.0, + 845.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1628.0, + 1404.0, + 1628.0, + 1404.0, + 1667.0, + 295.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1661.0, + 579.0, + 1661.0, + 579.0, + 1703.0, + 293.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1661.0, + 922.0, + 1661.0, + 922.0, + 1703.0, + 614.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1661.0, + 1292.0, + 1661.0, + 1292.0, + 1703.0, + 1216.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 1661.0, + 1405.0, + 1661.0, + 1405.0, + 1703.0, + 1335.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1694.0, + 703.0, + 1694.0, + 703.0, + 1730.0, + 295.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 857.0, + 1404.0, + 857.0, + 1404.0, + 895.0, + 295.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 891.0, + 1404.0, + 891.0, + 1404.0, + 924.0, + 297.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 915.0, + 1404.0, + 915.0, + 1404.0, + 960.0, + 292.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 949.0, + 1403.0, + 949.0, + 1403.0, + 986.0, + 295.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 980.0, + 815.0, + 980.0, + 815.0, + 1016.0, + 294.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1740.0, + 1404.0, + 1740.0, + 1404.0, + 1777.0, + 295.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1771.0, + 987.0, + 1771.0, + 987.0, + 1810.0, + 292.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 1771.0, + 1229.0, + 1771.0, + 1229.0, + 1810.0, + 1152.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1771.0, + 1320.0, + 1771.0, + 1320.0, + 1810.0, + 1269.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 1771.0, + 1408.0, + 1771.0, + 1408.0, + 1810.0, + 1359.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1802.0, + 733.0, + 1802.0, + 733.0, + 1838.0, + 293.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1802.0, + 809.0, + 1802.0, + 809.0, + 1838.0, + 759.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1802.0, + 1283.0, + 1802.0, + 1283.0, + 1838.0, + 835.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1802.0, + 1350.0, + 1802.0, + 1350.0, + 1838.0, + 1317.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1802.0, + 1407.0, + 1802.0, + 1407.0, + 1838.0, + 1375.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1833.0, + 402.0, + 1833.0, + 402.0, + 1871.0, + 292.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 339.0, + 1404.0, + 339.0, + 1404.0, + 372.0, + 297.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 369.0, + 1405.0, + 369.0, + 1405.0, + 403.0, + 296.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 399.0, + 1405.0, + 399.0, + 1405.0, + 436.0, + 294.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 430.0, + 1404.0, + 430.0, + 1404.0, + 464.0, + 296.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 461.0, + 1231.0, + 461.0, + 1231.0, + 498.0, + 294.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1939.0, + 605.0, + 1939.0, + 605.0, + 1978.0, + 294.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 1939.0, + 1050.0, + 1939.0, + 1050.0, + 1978.0, + 644.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1939.0, + 1406.0, + 1939.0, + 1406.0, + 1978.0, + 1089.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 1316.0, + 1973.0, + 1316.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 1973.0, + 1406.0, + 1973.0, + 1406.0, + 2007.0, + 1336.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 884.0, + 2002.0, + 884.0, + 2038.0, + 295.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 1407.0, + 229.0, + 1407.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 262.0, + 1406.0, + 262.0, + 1406.0, + 296.0, + 294.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 289.0, + 1406.0, + 289.0, + 1406.0, + 330.0, + 293.0, + 330.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1261, + 1404, + 1261, + 1404, + 1568, + 298, + 1568 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 940, + 1404, + 940, + 1404, + 1248, + 297, + 1248 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 299, + 1584, + 1404, + 1584, + 1404, + 1676, + 299, + 1676 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 833, + 1402, + 833, + 1402, + 926, + 298, + 926 + ], + "score": 0.961 + }, + { + "category_id": 2, + "poly": [ + 331, + 1944, + 778, + 1944, + 778, + 2035, + 331, + 2035 + ], + "score": 0.956 + }, + { + "category_id": 3, + "poly": [ + 308, + 229, + 1396, + 229, + 1396, + 413, + 308, + 413 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 296, + 1767, + 1401, + 1767, + 1401, + 1831, + 296, + 1831 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 297, + 593, + 1398, + 593, + 1398, + 658, + 297, + 658 + ], + "score": 0.948 + }, + { + "category_id": 4, + "poly": [ + 296, + 442, + 1406, + 442, + 1406, + 537, + 296, + 537 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 301, + 673, + 1395, + 673, + 1395, + 796, + 301, + 796 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 368, + 1859, + 855, + 1859, + 855, + 1923, + 368, + 1923 + ], + "score": 0.938 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.902 + }, + { + "category_id": 0, + "poly": [ + 300, + 1709, + 538, + 1709, + 538, + 1744, + 300, + 1744 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1366, + 794, + 1399, + 794, + 1399, + 823, + 1366, + 823 + ], + "score": 0.867 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.748 + }, + { + "category_id": 13, + "poly": [ + 776, + 836, + 833, + 836, + 833, + 863, + 776, + 863 + ], + "score": 0.86, + "latex": "\\mathrm { C } { + } { + }" + }, + { + "category_id": 14, + "poly": [ + 298, + 671, + 1390, + 671, + 1390, + 797, + 298, + 797 + ], + "score": 0.8, + "latex": "\\operatorname* { m i n } _ { U , V } \\sum _ { \\underbrace { l , l ^ { \\prime } \\in [ l _ { 1 } , l _ { 2 } ] } _ { l \\neq l ^ { \\prime } } } \\sum _ { w _ { t } \\in S _ { l } \\atop l \\neq l ^ { \\prime } } \\left( \\underbrace { \\ell ( u _ { w _ { t } } ^ { \\top } v _ { S _ { l } \\backslash \\{ w _ { t } \\} } ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell ( - u _ { w _ { t } ^ { \\prime } } v _ { S _ { l } \\backslash \\{ w _ { t } \\} } ) } _ { \\mathrm { m o n o i n g u a l ~ l o s s } } + \\underbrace { \\ell ( u _ { w _ { t } } ^ { \\top } v _ { S _ { l ^ { \\prime } } } ) + \\sum _ { w ^ { \\prime } \\in N _ { w _ { t } } } \\ell ( - u _ { w _ { t } ^ { \\prime } } v _ { S _ { l ^ { \\prime } } } ) } _ { \\mathrm { c r o s s i n g u a l ~ l o s s } } \\right)" + }, + { + "category_id": 13, + "poly": [ + 419, + 596, + 447, + 596, + 447, + 624, + 419, + 624 + ], + "score": 0.78, + "latex": "C" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1939.0, + 781.0, + 1939.0, + 781.0, + 1979.0, + 332.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1967.0, + 618.0, + 1967.0, + 618.0, + 2010.0, + 329.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1998.0, + 603.0, + 1998.0, + 603.0, + 2039.0, + 329.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 230.0, + 644.0, + 230.0, + 644.0, + 254.0, + 524.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 260.0, + 1052.0, + 260.0, + 1052.0, + 284.0, + 892.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 263.0, + 1348.0, + 263.0, + 1348.0, + 283.0, + 1187.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 298.0, + 1074.0, + 298.0, + 1074.0, + 326.0, + 1025.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 296.0, + 1242.0, + 296.0, + 1242.0, + 326.0, + 1191.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 317.0, + 436.0, + 317.0, + 436.0, + 346.0, + 385.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 318.0, + 475.0, + 318.0, + 475.0, + 343.0, + 451.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 315.0, + 644.0, + 315.0, + 644.0, + 345.0, + 583.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 318.0, + 700.0, + 318.0, + 700.0, + 346.0, + 653.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 295.5, + 949.0, + 295.5, + 949.0, + 327.5, + 896.0, + 327.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 295.0, + 1156.0, + 295.0, + 1156.0, + 328.0, + 1076.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 296.0, + 1314.0, + 296.0, + 1314.0, + 326.0, + 1245.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 295.0, + 1386.0, + 295.0, + 1386.0, + 329.5, + 1317.0, + 329.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 315.5, + 375.0, + 315.5, + 375.0, + 347.0, + 320.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 314.5, + 577.0, + 314.5, + 577.0, + 347.5, + 497.0, + 347.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.25, + 314.0, + 775.25, + 314.0, + 775.25, + 348.0, + 706.25, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 314.5, + 849.0, + 314.5, + 849.0, + 349.0, + 779.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.25, + 326.0, + 494.25, + 326.0, + 494.25, + 346.5, + 474.25, + 346.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 353.5, + 1124.0, + 353.5, + 1124.0, + 375.5, + 1087.0, + 375.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.25, + 380.5, + 1193.25, + 380.5, + 1193.25, + 412.5, + 1143.25, + 412.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.75, + 352.0, + 1264.75, + 352.0, + 1264.75, + 374.5, + 1201.75, + 374.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 443.0, + 1404.0, + 443.0, + 1404.0, + 477.0, + 295.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 474.0, + 1406.0, + 474.0, + 1406.0, + 508.0, + 295.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 505.0, + 1210.0, + 505.0, + 1210.0, + 539.0, + 295.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1705.0, + 542.0, + 1705.0, + 542.0, + 1752.0, + 292.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1262.0, + 1406.0, + 1262.0, + 1406.0, + 1297.0, + 294.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1295.0, + 1404.0, + 1295.0, + 1404.0, + 1327.0, + 296.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1320.0, + 1405.0, + 1320.0, + 1405.0, + 1361.0, + 292.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1355.0, + 1405.0, + 1355.0, + 1405.0, + 1389.0, + 292.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1387.0, + 1405.0, + 1387.0, + 1405.0, + 1418.0, + 296.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1416.0, + 1405.0, + 1416.0, + 1405.0, + 1451.0, + 294.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1445.0, + 1406.0, + 1445.0, + 1406.0, + 1481.0, + 294.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1477.0, + 1404.0, + 1477.0, + 1404.0, + 1512.0, + 294.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1505.0, + 1405.0, + 1505.0, + 1405.0, + 1542.0, + 293.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1535.0, + 1268.0, + 1535.0, + 1268.0, + 1570.0, + 295.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 937.0, + 1405.0, + 937.0, + 1405.0, + 977.0, + 294.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 972.0, + 1404.0, + 972.0, + 1404.0, + 1007.0, + 295.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1001.0, + 1405.0, + 1001.0, + 1405.0, + 1038.0, + 292.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1033.0, + 1404.0, + 1033.0, + 1404.0, + 1068.0, + 295.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1060.0, + 1407.0, + 1060.0, + 1407.0, + 1100.0, + 291.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1093.0, + 1406.0, + 1093.0, + 1406.0, + 1132.0, + 294.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1123.0, + 1407.0, + 1123.0, + 1407.0, + 1160.0, + 291.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1155.0, + 1405.0, + 1155.0, + 1405.0, + 1191.0, + 295.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1182.0, + 1405.0, + 1182.0, + 1405.0, + 1224.0, + 292.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1212.0, + 768.0, + 1212.0, + 768.0, + 1252.0, + 294.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1581.0, + 1405.0, + 1581.0, + 1405.0, + 1620.0, + 292.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1611.0, + 1405.0, + 1611.0, + 1405.0, + 1653.0, + 292.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1647.0, + 811.0, + 1647.0, + 811.0, + 1677.0, + 297.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 834.0, + 775.0, + 834.0, + 775.0, + 867.0, + 296.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 834.0, + 1401.0, + 834.0, + 1401.0, + 867.0, + 834.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 865.0, + 1403.0, + 865.0, + 1403.0, + 899.0, + 296.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 893.0, + 786.0, + 893.0, + 786.0, + 930.0, + 295.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1767.0, + 1403.0, + 1767.0, + 1403.0, + 1803.0, + 296.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1799.0, + 1184.0, + 1799.0, + 1184.0, + 1834.0, + 295.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 592.0, + 418.0, + 592.0, + 418.0, + 632.0, + 295.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 592.0, + 1402.0, + 592.0, + 1402.0, + 632.0, + 448.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 627.0, + 936.0, + 627.0, + 936.0, + 659.0, + 297.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1859.0, + 713.0, + 1859.0, + 713.0, + 1892.0, + 369.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1888.0, + 856.0, + 1888.0, + 856.0, + 1927.0, + 368.0, + 1927.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 296, + 734, + 1400, + 734, + 1400, + 1212, + 296, + 1212 + ], + "score": 0.984, + "html": "
Methoden-esen-fren-deen-ruen-itavg.
→↑→↑→↑→↑→↑
MUSE (Conneau et al.,2017)81.7 83.382.3 82.174.0 72.244.0 59.178.6 77.973.5
UMWE(Chen & Cardie,2018)82.5 83.182.5 82.174.6 72.549.5 61.778.3 77.074.4
Procrustes + refine (Conneau et al., 2017)82.4 83.982.3 83.275.3 73.250.1 63.577.5 77.674.9
RCSLS (Joulin et al., 2018)83.7 87.184.1 84.779.2 77.560.9 70.281.1 82.779.1
TRANSGRAM (Coulmance et al., 2015)91.6 88.689.1 90.187.5 87.265.6 73.788.6 89.585.2
VECMAP (unsupervised) (Artetxe et al.,2018b)87.4 87.888.3 88.584.3 87.248.6 50.587.4 86.579.6
VECMAP (supervised) (Artetxe et al.,2018a)87.2 90.287.6 90.487.3 86.849.7 65.687.2 89.282.1
BIVEC NN (Luong et al., 2015)87.4 88.686.8 89.187.5 87.264.0 59.186.8 84.081.7
BIVEC CSLS (Luong et al., 2015)87.6 89.188.8 90.386.4 87.266.1 70.687.6 87.884.3
BI-SENT2VEC uni. NN86.9 91.686.9 91.086.0 88.758.0 72.888.3 92.484.3
BI-SENT2VEC uni. + bi. NN89.4 92.989.3 92.886.7 89.359.0 70.289.5 91.885.1
BI-SENT2VEC uni.CSLS86.0 91.786.4 91.484.6 88.860.5 73.088.2 91.884.2
BI-SENT2VEC uni. + bi. CSLS89.0 92.188.9 92.486.5 89.061.0 73.589.6 91.485.3
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1638, + 1405, + 1638, + 1405, + 1945, + 298, + 1945 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1500, + 1404, + 1500, + 1404, + 1622, + 299, + 1622 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 325, + 1405, + 325, + 1405, + 417, + 299, + 417 + ], + "score": 0.97 + }, + { + "category_id": 2, + "poly": [ + 329, + 1975, + 960, + 1975, + 960, + 2034, + 329, + 2034 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 300, + 431, + 1400, + 431, + 1400, + 495, + 300, + 495 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 370, + 230, + 1013, + 230, + 1013, + 294, + 370, + 294 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 298, + 579, + 1404, + 579, + 1404, + 703, + 298, + 703 + ], + "score": 0.931 + }, + { + "category_id": 0, + "poly": [ + 300, + 525, + 623, + 525, + 623, + 558, + 300, + 558 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.85 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.755 + }, + { + "category_id": 0, + "poly": [ + 298, + 1448, + 973, + 1448, + 973, + 1479, + 298, + 1479 + ], + "score": 0.713 + }, + { + "category_id": 6, + "poly": [ + 298, + 1240, + 1405, + 1240, + 1405, + 1395, + 298, + 1395 + ], + "score": 0.595 + }, + { + "category_id": 1, + "poly": [ + 298, + 1448, + 973, + 1448, + 973, + 1479, + 298, + 1479 + ], + "score": 0.207 + }, + { + "category_id": 13, + "poly": [ + 454, + 1853, + 520, + 1853, + 520, + 1882, + 454, + 1882 + ], + "score": 0.88, + "latex": "1 0 0 \\times" + }, + { + "category_id": 13, + "poly": [ + 466, + 1365, + 502, + 1365, + 502, + 1391, + 466, + 1391 + ], + "score": 0.83, + "latex": "\\gets" + }, + { + "category_id": 13, + "poly": [ + 855, + 1335, + 891, + 1335, + 891, + 1361, + 855, + 1361 + ], + "score": 0.81, + "latex": "" + }, + { + "category_id": 13, + "poly": [ + 369, + 672, + 428, + 672, + 428, + 700, + 369, + 700 + ], + "score": 0.75, + "latex": "\\mathrm { P @ 1 }" + }, + { + "category_id": 13, + "poly": [ + 718, + 1241, + 780, + 1241, + 780, + 1271, + 718, + 1271 + ], + "score": 0.67, + "latex": "\\mathbf { P } @ \\mathbf { 1 }" + }, + { + "category_id": 13, + "poly": [ + 404, + 874, + 424, + 874, + 424, + 894, + 404, + 894 + ], + "score": 0.63, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 904, + 641, + 964, + 641, + 964, + 670, + 904, + 670 + ], + "score": 0.57, + "latex": "2 0 0 \\mathrm { k }" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1968.0, + 949.0, + 1968.0, + 949.0, + 2010.0, + 329.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1999.0, + 962.0, + 1999.0, + 962.0, + 2038.0, + 330.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 524.0, + 626.0, + 524.0, + 626.0, + 560.0, + 294.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1445.0, + 977.0, + 1445.0, + 977.0, + 1484.0, + 294.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1238.0, + 717.0, + 1238.0, + 717.0, + 1276.0, + 293.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1238.0, + 1405.0, + 1238.0, + 1405.0, + 1276.0, + 781.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1272.0, + 1403.0, + 1272.0, + 1403.0, + 1306.0, + 296.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1302.0, + 1402.0, + 1302.0, + 1402.0, + 1335.0, + 296.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1329.0, + 854.0, + 1329.0, + 854.0, + 1369.0, + 295.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1329.0, + 1406.0, + 1329.0, + 1406.0, + 1369.0, + 892.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1360.0, + 465.0, + 1360.0, + 465.0, + 1397.0, + 295.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1360.0, + 760.0, + 1360.0, + 760.0, + 1397.0, + 503.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1640.0, + 1403.0, + 1640.0, + 1403.0, + 1672.0, + 295.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1669.0, + 1405.0, + 1669.0, + 1405.0, + 1704.0, + 292.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1701.0, + 1402.0, + 1701.0, + 1402.0, + 1733.0, + 296.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1729.0, + 1406.0, + 1729.0, + 1406.0, + 1766.0, + 293.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1762.0, + 1406.0, + 1762.0, + 1406.0, + 1794.0, + 296.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1793.0, + 1403.0, + 1793.0, + 1403.0, + 1825.0, + 295.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1819.0, + 1405.0, + 1819.0, + 1405.0, + 1859.0, + 293.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1851.0, + 453.0, + 1851.0, + 453.0, + 1887.0, + 295.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1851.0, + 1403.0, + 1851.0, + 1403.0, + 1887.0, + 521.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1883.0, + 1405.0, + 1883.0, + 1405.0, + 1919.0, + 295.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1914.0, + 1313.0, + 1914.0, + 1313.0, + 1947.0, + 293.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1499.0, + 1405.0, + 1499.0, + 1405.0, + 1538.0, + 294.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1532.0, + 1404.0, + 1532.0, + 1404.0, + 1564.0, + 294.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1563.0, + 1405.0, + 1563.0, + 1405.0, + 1599.0, + 294.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1595.0, + 559.0, + 1595.0, + 559.0, + 1624.0, + 294.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 322.0, + 1405.0, + 322.0, + 1405.0, + 363.0, + 294.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 354.0, + 1405.0, + 354.0, + 1405.0, + 393.0, + 293.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 388.0, + 730.0, + 388.0, + 730.0, + 421.0, + 294.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 431.0, + 1404.0, + 431.0, + 1404.0, + 467.0, + 297.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 462.0, + 1276.0, + 462.0, + 1276.0, + 498.0, + 295.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 230.0, + 752.0, + 230.0, + 752.0, + 263.0, + 369.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 261.0, + 1015.0, + 261.0, + 1015.0, + 296.0, + 368.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 577.0, + 1404.0, + 577.0, + 1404.0, + 614.0, + 293.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 610.0, + 1405.0, + 610.0, + 1405.0, + 647.0, + 292.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 640.0, + 903.0, + 640.0, + 903.0, + 676.0, + 293.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 640.0, + 1405.0, + 640.0, + 1405.0, + 676.0, + 965.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 674.0, + 368.0, + 674.0, + 368.0, + 703.0, + 293.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 674.0, + 1392.0, + 674.0, + 1392.0, + 703.0, + 429.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1445.0, + 977.0, + 1445.0, + 977.0, + 1484.0, + 294.0, + 1484.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 500, + 212, + 1199, + 212, + 1199, + 607, + 500, + 607 + ], + "score": 0.98, + "html": "
Method\\DatasetSimLex-999WS-353
en iten it
MUSE0.380.300.74 0.64
RCSLS0.38 0.300.740.64
FASTTEXT- Common Crawl0.49 0.320.750.57
BIVEC0.40 0.360.700.60
TRANSGRAM0.43 0.370.730.63
SENT2VEC uni.0.49 0.380.730.60
BI-SENT2VEC uni.0.570.47 0.790.65
BI-SENT2VEC uni. + bi.0.580.50 0.800.69
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1757, + 1404, + 1757, + 1404, + 1880, + 298, + 1880 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 301, + 1941, + 1403, + 1941, + 1403, + 2034, + 301, + 2034 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 297, + 786, + 1404, + 786, + 1404, + 1003, + 297, + 1003 + ], + "score": 0.963 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 105, + 297, + 105 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 298, + 1708, + 963, + 1708, + 963, + 1737, + 298, + 1737 + ], + "score": 0.886 + }, + { + "category_id": 0, + "poly": [ + 300, + 1890, + 1197, + 1890, + 1197, + 1923, + 300, + 1923 + ], + "score": 0.869 + }, + { + "category_id": 6, + "poly": [ + 296, + 637, + 1397, + 637, + 1397, + 700, + 296, + 700 + ], + "score": 0.822 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.778 + }, + { + "category_id": 0, + "poly": [ + 298, + 737, + 842, + 737, + 842, + 768, + 298, + 768 + ], + "score": 0.538 + }, + { + "category_id": 7, + "poly": [ + 298, + 1566, + 1404, + 1566, + 1404, + 1658, + 298, + 1658 + ], + "score": 0.43 + }, + { + "category_id": 13, + "poly": [ + 336, + 971, + 395, + 971, + 395, + 999, + 336, + 999 + ], + "score": 0.85, + "latex": "\\mathrm { P @ 1 }" + }, + { + "category_id": 13, + "poly": [ + 894, + 1567, + 953, + 1567, + 953, + 1596, + 894, + 1596 + ], + "score": 0.79, + "latex": "\\mathrm { P @ 1 }" + }, + { + "category_id": 13, + "poly": [ + 296, + 1630, + 319, + 1630, + 319, + 1656, + 296, + 1656 + ], + "score": 0.41, + "latex": "^ +" + }, + { + "category_id": 5, + "poly": [ + 295, + 1029, + 1399, + 1029, + 1399, + 1546, + 295, + 1546 + ], + "score": 0.92, + "html": "
Methoden-esen-fren-deen-itavg.
→↑→↑→↑→↑
MUSE72.771.569.268.853.353.466.164.364.9
RCSLS26.926.719.321.28.811.315.117.618.4
TRANSGRAM83.581.480.481.664.869.977.277.977.1
VECMAP (unsupervised)81.782.179.880.462.864.669.071.174.0
VECMAP (supervised)81.38180.480.762.664.367.87173.6
BIVEC NN69.877.154.775.556.144.158.245.160.1
BIVEC CSLS81.683.478.181.671.668.174.272.476.4
BI-SENT2VEC uni. NN87.886.485.283.482.380.285.985.884.6
BI-SENT2VEC uni. + bi. NN87.987.886.183.979.579.785.185.384.4
BI-SENT2VEC uni. CSLS89.588.587.186.484.483.088.287.586.8
BI-SENT2VEC uni. + bi. CSLS89.789.687.887.484.284.087.987.687.3
Reduction in error37.5% 37.3%37.8% 31.5%44.4% 46.8%46.9% 43.9%1
" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1704.0, + 966.0, + 1704.0, + 966.0, + 1741.0, + 294.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1890.0, + 1198.0, + 1890.0, + 1198.0, + 1925.0, + 295.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 638.0, + 1402.0, + 638.0, + 1402.0, + 670.0, + 297.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 666.0, + 924.0, + 666.0, + 924.0, + 704.0, + 295.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 733.0, + 847.0, + 733.0, + 847.0, + 773.0, + 293.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1562.0, + 893.0, + 1562.0, + 893.0, + 1604.0, + 293.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1562.0, + 1405.0, + 1562.0, + 1405.0, + 1604.0, + 954.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1596.0, + 1401.0, + 1596.0, + 1401.0, + 1630.0, + 294.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1628.0, + 889.0, + 1628.0, + 889.0, + 1658.0, + 320.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1755.0, + 1404.0, + 1755.0, + 1404.0, + 1792.0, + 295.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1788.0, + 1402.0, + 1788.0, + 1402.0, + 1821.0, + 294.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1819.0, + 1404.0, + 1819.0, + 1404.0, + 1851.0, + 294.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1850.0, + 626.0, + 1850.0, + 626.0, + 1879.0, + 293.0, + 1879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1940.0, + 1405.0, + 1940.0, + 1405.0, + 1977.0, + 296.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1998.0, + 1407.0, + 1998.0, + 1407.0, + 2041.0, + 294.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 788.0, + 1404.0, + 788.0, + 1404.0, + 823.0, + 295.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 819.0, + 1406.0, + 819.0, + 1406.0, + 854.0, + 295.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 845.0, + 1406.0, + 845.0, + 1406.0, + 887.0, + 291.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 882.0, + 1402.0, + 882.0, + 1402.0, + 913.0, + 295.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 907.0, + 1404.0, + 907.0, + 1404.0, + 948.0, + 292.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 940.0, + 1405.0, + 940.0, + 1405.0, + 975.0, + 295.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 971.0, + 335.0, + 971.0, + 335.0, + 1002.0, + 295.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 971.0, + 1322.0, + 971.0, + 1322.0, + 1002.0, + 396.0, + 1002.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 295, + 223, + 1400, + 223, + 1400, + 833, + 295, + 833 + ], + "score": 0.983, + "html": "
Method word retrieval sentence retrieval
en-fi →↑en-hu → ↑→en-fi ↑en-hu →↑
MUSE48.159.553.9 64.921.729.539.146.7
RCSLS61.869.967.0 73.03.24.83.65.1
VECMAP (unsupervised)62.566.861.6 68.713.214.720.519.3
VECMAP (supervised)62.678.363.7 76.615.016.920.921.7
BIVEC NN62.155.362.1 53.714.29.726.213.7
BIVEC CSLS69.678.072.4 78.433.332.046.741.3
TRANSGRAM69.781.173.1 80.835.440.552.155
B1-SENT2VEC uni. NN71.285.475.683.963.5 64.275.2 76.2
BI-SENT2VEC uni. + bi. NN68.581.771.4 79.457.5 55.965.865.2
B1-SENT2VEC uni. CSLS72.0 86.576.3 85.170.2 69.081.480.8
BI-SENT2VEC uni. + bi. CSLS70.184.473.781.766 64.173.874.5
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1400, + 1404, + 1400, + 1404, + 1586, + 298, + 1586 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1655, + 1404, + 1655, + 1404, + 1810, + 298, + 1810 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1880, + 1401, + 1880, + 1401, + 2034, + 299, + 2034 + ], + "score": 0.976 + }, + { + "category_id": 5, + "poly": [ + 446, + 1119, + 1253, + 1119, + 1253, + 1279, + 446, + 1279 + ], + "score": 0.969, + "html": "
Methoden-esen-fren-deavg.
→↑→冏↑→↑→en-it ↑
LASER79.3 69.678.0 80.186.3 80.870.2 74.277.3
BI-SENT2VEC74.0 71.581.6 82.286.5 79.275.0 72.677.8
" + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 298, + 1608, + 1039, + 1608, + 1039, + 1641, + 298, + 1641 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 298, + 1824, + 531, + 1824, + 531, + 1860, + 298, + 1860 + ], + "score": 0.907 + }, + { + "category_id": 6, + "poly": [ + 295, + 854, + 1393, + 854, + 1393, + 918, + 295, + 918 + ], + "score": 0.861 + }, + { + "category_id": 1, + "poly": [ + 301, + 970, + 1403, + 970, + 1403, + 1096, + 301, + 1096 + ], + "score": 0.821 + }, + { + "category_id": 1, + "poly": [ + 298, + 1307, + 1402, + 1307, + 1402, + 1371, + 298, + 1371 + ], + "score": 0.754 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.678 + }, + { + "category_id": 6, + "poly": [ + 301, + 970, + 1403, + 970, + 1403, + 1096, + 301, + 1096 + ], + "score": 0.234 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 859, + 2087, + 859, + 2111, + 841, + 2111 + ], + "score": 0.221 + }, + { + "category_id": 6, + "poly": [ + 298, + 1307, + 1402, + 1307, + 1402, + 1371, + 298, + 1371 + ], + "score": 0.108 + }, + { + "category_id": 13, + "poly": [ + 1249, + 1338, + 1277, + 1338, + 1277, + 1368, + 1249, + 1368 + ], + "score": 0.84, + "latex": "\\%" + }, + { + "category_id": 13, + "poly": [ + 552, + 789, + 578, + 789, + 578, + 817, + 552, + 817 + ], + "score": 0.35, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 1334, + 855, + 1403, + 855, + 1403, + 887, + 1334, + 887 + ], + "score": 0.26, + "latex": "( \\mathbf { P } @ \\mathbf { 1 }" + }, + { + "category_id": 13, + "poly": [ + 543, + 1943, + 602, + 1943, + 602, + 1972, + 543, + 1972 + ], + "score": 0.25, + "latex": "2 \\cdot 3" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1604.0, + 1044.0, + 1604.0, + 1044.0, + 1647.0, + 293.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1818.0, + 535.0, + 1818.0, + 535.0, + 1869.0, + 290.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 853.0, + 1333.0, + 853.0, + 1333.0, + 892.0, + 294.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 885.0, + 935.0, + 885.0, + 935.0, + 921.0, + 294.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 971.0, + 1405.0, + 971.0, + 1405.0, + 1006.0, + 294.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1002.0, + 1405.0, + 1002.0, + 1405.0, + 1039.0, + 295.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1028.0, + 1405.0, + 1028.0, + 1405.0, + 1071.0, + 292.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1064.0, + 1344.0, + 1064.0, + 1344.0, + 1100.0, + 295.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1306.0, + 1404.0, + 1306.0, + 1404.0, + 1342.0, + 296.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1334.0, + 1248.0, + 1334.0, + 1248.0, + 1377.0, + 293.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1334.0, + 1391.0, + 1334.0, + 1391.0, + 1377.0, + 1278.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1400.0, + 1404.0, + 1400.0, + 1404.0, + 1437.0, + 296.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1432.0, + 1405.0, + 1432.0, + 1405.0, + 1467.0, + 294.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1462.0, + 1406.0, + 1462.0, + 1406.0, + 1498.0, + 294.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1492.0, + 1407.0, + 1492.0, + 1407.0, + 1529.0, + 292.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1522.0, + 1406.0, + 1522.0, + 1406.0, + 1559.0, + 292.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1556.0, + 1120.0, + 1556.0, + 1120.0, + 1587.0, + 296.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1655.0, + 1404.0, + 1655.0, + 1404.0, + 1689.0, + 298.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1683.0, + 1406.0, + 1683.0, + 1406.0, + 1722.0, + 293.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1718.0, + 1404.0, + 1718.0, + 1404.0, + 1751.0, + 294.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1744.0, + 1405.0, + 1744.0, + 1405.0, + 1783.0, + 295.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1778.0, + 938.0, + 1778.0, + 938.0, + 1812.0, + 293.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1880.0, + 1405.0, + 1880.0, + 1405.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1912.0, + 1403.0, + 1912.0, + 1403.0, + 1945.0, + 295.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 542.0, + 1942.0, + 542.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 1942.0, + 1402.0, + 1942.0, + 1402.0, + 1976.0, + 603.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2009.0, + 294.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 482.0, + 2003.0, + 482.0, + 2036.0, + 294.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 971.0, + 1405.0, + 971.0, + 1405.0, + 1006.0, + 294.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1002.0, + 1405.0, + 1002.0, + 1405.0, + 1039.0, + 295.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1028.0, + 1405.0, + 1028.0, + 1405.0, + 1071.0, + 292.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1064.0, + 1344.0, + 1064.0, + 1344.0, + 1100.0, + 295.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1306.0, + 1404.0, + 1306.0, + 1404.0, + 1342.0, + 296.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1334.0, + 1248.0, + 1334.0, + 1248.0, + 1377.0, + 293.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1334.0, + 1391.0, + 1334.0, + 1391.0, + 1377.0, + 1278.0, + 1377.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1191, + 1404, + 1191, + 1404, + 1467, + 298, + 1467 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1651, + 1403, + 1651, + 1403, + 1805, + 298, + 1805 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1820, + 1403, + 1820, + 1403, + 2035, + 298, + 2035 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1482, + 1402, + 1482, + 1402, + 1636, + 298, + 1636 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 309, + 244, + 1350, + 244, + 1350, + 587, + 309, + 587 + ], + "score": 0.961 + }, + { + "category_id": 3, + "poly": [ + 596, + 694, + 1071, + 694, + 1071, + 1038, + 596, + 1038 + ], + "score": 0.957 + }, + { + "category_id": 4, + "poly": [ + 292, + 1052, + 1404, + 1052, + 1404, + 1118, + 292, + 1118 + ], + "score": 0.934 + }, + { + "category_id": 4, + "poly": [ + 347, + 595, + 1345, + 595, + 1345, + 628, + 347, + 628 + ], + "score": 0.909 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 858, + 76, + 858, + 104, + 298, + 104 + ], + "score": 0.826 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.796 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 855, + 76, + 855, + 104, + 300, + 104 + ], + "score": 0.228 + }, + { + "category_id": 13, + "poly": [ + 649, + 1543, + 724, + 1543, + 724, + 1573, + 649, + 1573 + ], + "score": 0.89, + "latex": "5 5 . 1 \\%" + }, + { + "category_id": 13, + "poly": [ + 538, + 1543, + 614, + 1543, + 614, + 1574, + 538, + 1574 + ], + "score": 0.89, + "latex": "3 1 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 842, + 1574, + 914, + 1574, + 914, + 1602, + 842, + 1602 + ], + "score": 0.86, + "latex": "\\approx 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 471, + 1284, + 544, + 1284, + 544, + 1313, + 471, + 1313 + ], + "score": 0.86, + "latex": "1 1 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 635, + 1284, + 694, + 1284, + 694, + 1312, + 635, + 1312 + ], + "score": 0.84, + "latex": "\\mathrm { P @ 1 }" + }, + { + "category_id": 13, + "poly": [ + 1341, + 1514, + 1401, + 1514, + 1401, + 1543, + 1341, + 1543 + ], + "score": 0.76, + "latex": "\\mathrm { P @ 1 }" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 241.0, + 731.0, + 241.0, + 731.0, + 271.0, + 421.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 242.0, + 1349.0, + 242.0, + 1349.0, + 271.0, + 925.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 256.0, + 367.0, + 256.0, + 367.0, + 283.0, + 334.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 257.0, + 926.0, + 257.0, + 926.0, + 280.0, + 891.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 291.0, + 366.0, + 291.0, + 366.0, + 315.0, + 339.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 292.0, + 928.0, + 292.0, + 928.0, + 315.0, + 900.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 324.0, + 366.0, + 324.0, + 366.0, + 349.0, + 340.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 326.0, + 927.0, + 326.0, + 927.0, + 349.0, + 901.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 339.0, + 336.0, + 339.0, + 336.0, + 474.0, + 305.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 341.0, + 901.0, + 341.0, + 901.0, + 474.0, + 866.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 343.0, + 1267.0, + 343.0, + 1267.0, + 359.0, + 1229.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 362.0, + 364.0, + 362.0, + 364.0, + 382.0, + 341.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 362.0, + 923.0, + 362.0, + 923.0, + 382.0, + 902.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 396.0, + 365.0, + 396.0, + 365.0, + 416.0, + 342.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 396.0, + 925.0, + 396.0, + 925.0, + 416.0, + 903.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 428.0, + 366.0, + 428.0, + 366.0, + 451.0, + 341.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 429.0, + 925.0, + 429.0, + 925.0, + 451.0, + 902.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 462.0, + 372.0, + 462.0, + 372.0, + 485.0, + 342.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 459.0, + 397.0, + 459.0, + 397.0, + 479.0, + 379.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 463.0, + 926.0, + 463.0, + 926.0, + 485.0, + 903.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 488.0, + 595.0, + 488.0, + 595.0, + 501.0, + 568.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 481.0, + 780.0, + 481.0, + 780.0, + 507.0, + 601.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 481.0, + 1338.0, + 481.0, + 1338.0, + 505.0, + 1147.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 495.0, + 366.0, + 495.0, + 366.0, + 519.0, + 340.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 504.0, + 780.0, + 504.0, + 780.0, + 533.0, + 568.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 497.0, + 925.0, + 497.0, + 925.0, + 517.0, + 902.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 505.0, + 1338.0, + 505.0, + 1338.0, + 531.0, + 1147.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 532.0, + 925.0, + 532.0, + 925.0, + 552.0, + 903.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 542.0, + 480.0, + 542.0, + 480.0, + 569.0, + 443.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 540.0, + 633.0, + 540.0, + 633.0, + 570.0, + 594.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 544.0, + 782.0, + 544.0, + 782.0, + 567.0, + 748.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 542.0, + 1040.0, + 542.0, + 1040.0, + 569.0, + 1003.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 540.0, + 1194.0, + 540.0, + 1194.0, + 570.0, + 1154.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 544.0, + 1343.0, + 544.0, + 1343.0, + 570.0, + 1312.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 564.0, + 714.0, + 564.0, + 714.0, + 590.0, + 437.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 564.0, + 1275.0, + 564.0, + 1275.0, + 590.0, + 997.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 347.0, + 1333.0, + 347.0, + 1333.0, + 365.5, + 1286.0, + 365.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 416.0, + 969.0, + 416.0, + 969.0, + 428.0, + 934.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 532.5, + 372.0, + 532.5, + 372.0, + 549.5, + 334.0, + 549.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 696.0, + 1038.0, + 696.0, + 1038.0, + 720.0, + 674.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 711.0, + 654.0, + 711.0, + 654.0, + 736.0, + 617.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 724.0, + 1055.0, + 724.0, + 1055.0, + 739.0, + 1037.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 743.0, + 1054.0, + 743.0, + 1054.0, + 758.0, + 1009.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 756.0, + 647.0, + 756.0, + 647.0, + 779.0, + 618.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 777.0, + 621.0, + 777.0, + 621.0, + 938.0, + 595.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 803.0, + 643.0, + 803.0, + 643.0, + 822.0, + 620.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 832.0, + 673.0, + 832.0, + 673.0, + 844.0, + 661.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 832.0, + 1052.0, + 832.0, + 1052.0, + 839.0, + 1043.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 846.0, + 643.0, + 846.0, + 643.0, + 866.0, + 619.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 891.0, + 643.0, + 891.0, + 643.0, + 910.0, + 619.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 909.0, + 1057.0, + 909.0, + 1057.0, + 933.0, + 859.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 933.0, + 646.0, + 933.0, + 646.0, + 955.0, + 618.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 931.0, + 1026.0, + 931.0, + 1026.0, + 959.0, + 859.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 955.0, + 1013.0, + 955.0, + 1013.0, + 983.0, + 821.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 978.0, + 648.0, + 978.0, + 648.0, + 1000.0, + 618.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 997.0, + 757.0, + 997.0, + 757.0, + 1017.0, + 721.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 996.0, + 1062.0, + 996.0, + 1062.0, + 1019.0, + 1033.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1015.0, + 992.0, + 1015.0, + 992.0, + 1039.0, + 718.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 777.5, + 834.0, + 777.5, + 834.0, + 796.5, + 726.0, + 796.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 874.5, + 713.0, + 874.5, + 713.0, + 887.5, + 652.0, + 887.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 994.0, + 914.0, + 994.0, + 914.0, + 1019.5, + 873.0, + 1019.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1050.0, + 1403.0, + 1050.0, + 1403.0, + 1086.0, + 295.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1082.0, + 1069.0, + 1082.0, + 1069.0, + 1119.0, + 295.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 591.0, + 1352.0, + 591.0, + 1352.0, + 635.0, + 347.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 70.0, + 859.0, + 70.0, + 859.0, + 109.0, + 296.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1192.0, + 1404.0, + 1192.0, + 1404.0, + 1226.0, + 298.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1223.0, + 1402.0, + 1223.0, + 1402.0, + 1256.0, + 296.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1251.0, + 1404.0, + 1251.0, + 1404.0, + 1289.0, + 293.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1281.0, + 470.0, + 1281.0, + 470.0, + 1320.0, + 293.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1281.0, + 634.0, + 1281.0, + 634.0, + 1320.0, + 545.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1281.0, + 1406.0, + 1281.0, + 1406.0, + 1320.0, + 695.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1312.0, + 1406.0, + 1312.0, + 1406.0, + 1350.0, + 293.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1346.0, + 1402.0, + 1346.0, + 1402.0, + 1379.0, + 296.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1372.0, + 1405.0, + 1372.0, + 1405.0, + 1409.0, + 293.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1405.0, + 1402.0, + 1405.0, + 1402.0, + 1441.0, + 293.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1437.0, + 1053.0, + 1437.0, + 1053.0, + 1470.0, + 296.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1649.0, + 1405.0, + 1649.0, + 1405.0, + 1687.0, + 294.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1678.0, + 1405.0, + 1678.0, + 1405.0, + 1717.0, + 294.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1711.0, + 1405.0, + 1711.0, + 1405.0, + 1748.0, + 294.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1741.0, + 1405.0, + 1741.0, + 1405.0, + 1778.0, + 294.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1772.0, + 1228.0, + 1772.0, + 1228.0, + 1809.0, + 294.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1820.0, + 1404.0, + 1820.0, + 1404.0, + 1855.0, + 297.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1850.0, + 1407.0, + 1850.0, + 1407.0, + 1888.0, + 293.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1881.0, + 1408.0, + 1881.0, + 1408.0, + 1919.0, + 293.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1941.0, + 1407.0, + 1941.0, + 1407.0, + 1979.0, + 293.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2006.0, + 294.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 1218.0, + 2001.0, + 1218.0, + 2039.0, + 294.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1481.0, + 1403.0, + 1481.0, + 1403.0, + 1517.0, + 296.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1515.0, + 1340.0, + 1515.0, + 1340.0, + 1544.0, + 298.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1542.0, + 537.0, + 1542.0, + 537.0, + 1578.0, + 293.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 1542.0, + 648.0, + 1542.0, + 648.0, + 1578.0, + 615.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 1542.0, + 1406.0, + 1542.0, + 1406.0, + 1578.0, + 725.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1572.0, + 841.0, + 1572.0, + 841.0, + 1609.0, + 294.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1572.0, + 1404.0, + 1572.0, + 1404.0, + 1609.0, + 915.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1607.0, + 782.0, + 1607.0, + 782.0, + 1640.0, + 297.0, + 1640.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1150, + 1404, + 1150, + 1404, + 1485, + 297, + 1485 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 230, + 1403, + 230, + 1403, + 506, + 298, + 506 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1650, + 1403, + 1650, + 1403, + 1866, + 298, + 1866 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 781, + 1403, + 781, + 1403, + 995, + 298, + 995 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 520, + 1403, + 520, + 1403, + 766, + 298, + 766 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1881, + 1401, + 1881, + 1401, + 2034, + 299, + 2034 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1010, + 1403, + 1010, + 1403, + 1134, + 299, + 1134 + ], + "score": 0.976 + }, + { + "category_id": 0, + "poly": [ + 299, + 1562, + 842, + 1562, + 842, + 1598, + 299, + 1598 + ], + "score": 0.914 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 857, + 76, + 857, + 104, + 299, + 104 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2111, + 841, + 2111 + ], + "score": 0.787 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 856, + 76, + 856, + 104, + 300, + 104 + ], + "score": 0.121 + }, + { + "category_id": 13, + "poly": [ + 1097, + 1272, + 1172, + 1272, + 1172, + 1301, + 1097, + 1301 + ], + "score": 0.87, + "latex": "7 7 . 3 \\%" + }, + { + "category_id": 13, + "poly": [ + 672, + 1271, + 748, + 1271, + 748, + 1301, + 672, + 1301 + ], + "score": 0.87, + "latex": "7 7 . 8 \\%" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1557.0, + 847.0, + 1557.0, + 847.0, + 1603.0, + 292.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 859.0, + 73.0, + 859.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1149.0, + 1405.0, + 1149.0, + 1405.0, + 1184.0, + 296.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1178.0, + 1404.0, + 1178.0, + 1404.0, + 1214.0, + 292.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1210.0, + 1404.0, + 1210.0, + 1404.0, + 1245.0, + 292.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1241.0, + 1405.0, + 1241.0, + 1405.0, + 1277.0, + 292.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1269.0, + 671.0, + 1269.0, + 671.0, + 1308.0, + 292.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1269.0, + 1096.0, + 1269.0, + 1096.0, + 1308.0, + 749.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1269.0, + 1405.0, + 1269.0, + 1405.0, + 1308.0, + 1173.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1303.0, + 1404.0, + 1303.0, + 1404.0, + 1337.0, + 294.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1332.0, + 1405.0, + 1332.0, + 1405.0, + 1364.0, + 293.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1363.0, + 1406.0, + 1363.0, + 1406.0, + 1397.0, + 294.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1394.0, + 1404.0, + 1394.0, + 1404.0, + 1425.0, + 292.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1424.0, + 1404.0, + 1424.0, + 1404.0, + 1458.0, + 295.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1455.0, + 1389.0, + 1455.0, + 1389.0, + 1489.0, + 294.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 232.0, + 1404.0, + 232.0, + 1404.0, + 265.0, + 297.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 260.0, + 1407.0, + 260.0, + 1407.0, + 298.0, + 293.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 292.0, + 1404.0, + 292.0, + 1404.0, + 325.0, + 294.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 318.0, + 1405.0, + 318.0, + 1405.0, + 356.0, + 293.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 350.0, + 1407.0, + 350.0, + 1407.0, + 389.0, + 293.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 378.0, + 1407.0, + 378.0, + 1407.0, + 422.0, + 292.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 412.0, + 1404.0, + 412.0, + 1404.0, + 449.0, + 294.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 441.0, + 1403.0, + 441.0, + 1403.0, + 482.0, + 292.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 475.0, + 1403.0, + 475.0, + 1403.0, + 509.0, + 293.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1649.0, + 1404.0, + 1649.0, + 1404.0, + 1688.0, + 294.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1681.0, + 1404.0, + 1681.0, + 1404.0, + 1716.0, + 294.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1711.0, + 1404.0, + 1711.0, + 1404.0, + 1748.0, + 293.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1743.0, + 1403.0, + 1743.0, + 1403.0, + 1778.0, + 293.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1772.0, + 1403.0, + 1772.0, + 1403.0, + 1809.0, + 293.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1802.0, + 1407.0, + 1802.0, + 1407.0, + 1841.0, + 293.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1836.0, + 982.0, + 1836.0, + 982.0, + 1870.0, + 296.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 779.0, + 566.0, + 779.0, + 566.0, + 817.0, + 293.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 781.0, + 1404.0, + 781.0, + 1404.0, + 816.0, + 608.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 811.0, + 1407.0, + 811.0, + 1407.0, + 849.0, + 294.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 841.0, + 1406.0, + 841.0, + 1406.0, + 878.0, + 292.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 873.0, + 1404.0, + 873.0, + 1404.0, + 908.0, + 294.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 904.0, + 1405.0, + 904.0, + 1405.0, + 938.0, + 294.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 935.0, + 1404.0, + 935.0, + 1404.0, + 969.0, + 294.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 964.0, + 984.0, + 964.0, + 984.0, + 998.0, + 296.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 519.0, + 1403.0, + 519.0, + 1403.0, + 557.0, + 293.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 552.0, + 1404.0, + 552.0, + 1404.0, + 585.0, + 294.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 584.0, + 1405.0, + 584.0, + 1405.0, + 615.0, + 293.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 613.0, + 1405.0, + 613.0, + 1405.0, + 647.0, + 294.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 642.0, + 1405.0, + 642.0, + 1405.0, + 677.0, + 293.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 673.0, + 1405.0, + 673.0, + 1405.0, + 710.0, + 293.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 702.0, + 1404.0, + 702.0, + 1404.0, + 742.0, + 292.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 734.0, + 1216.0, + 734.0, + 1216.0, + 770.0, + 293.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1878.0, + 1403.0, + 1878.0, + 1403.0, + 1917.0, + 295.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1909.0, + 1402.0, + 1909.0, + 1402.0, + 1946.0, + 295.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1940.0, + 1405.0, + 1940.0, + 1405.0, + 1979.0, + 292.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1969.0, + 1405.0, + 1969.0, + 1405.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2004.0, + 971.0, + 2004.0, + 971.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1009.0, + 1405.0, + 1009.0, + 1405.0, + 1048.0, + 292.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1043.0, + 1403.0, + 1043.0, + 1403.0, + 1076.0, + 294.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1071.0, + 1405.0, + 1071.0, + 1405.0, + 1107.0, + 293.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1102.0, + 1187.0, + 1102.0, + 1187.0, + 1137.0, + 294.0, + 1137.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.892 + }, + { + "category_id": 1, + "poly": [ + 296, + 284, + 1401, + 284, + 1401, + 378, + 296, + 378 + ], + "score": 0.85 + }, + { + "category_id": 1, + "poly": [ + 293, + 1177, + 1322, + 1177, + 1322, + 1212, + 293, + 1212 + ], + "score": 0.826 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.823 + }, + { + "category_id": 1, + "poly": [ + 294, + 1229, + 1396, + 1229, + 1396, + 1293, + 294, + 1293 + ], + "score": 0.807 + }, + { + "category_id": 1, + "poly": [ + 295, + 1641, + 1399, + 1641, + 1399, + 1706, + 295, + 1706 + ], + "score": 0.788 + }, + { + "category_id": 0, + "poly": [ + 300, + 227, + 489, + 227, + 489, + 262, + 300, + 262 + ], + "score": 0.78 + }, + { + "category_id": 1, + "poly": [ + 299, + 1311, + 1398, + 1311, + 1398, + 1376, + 299, + 1376 + ], + "score": 0.78 + }, + { + "category_id": 1, + "poly": [ + 296, + 1559, + 1399, + 1559, + 1399, + 1623, + 296, + 1623 + ], + "score": 0.772 + }, + { + "category_id": 1, + "poly": [ + 297, + 1723, + 1399, + 1723, + 1399, + 1788, + 297, + 1788 + ], + "score": 0.77 + }, + { + "category_id": 1, + "poly": [ + 297, + 981, + 1395, + 981, + 1395, + 1047, + 297, + 1047 + ], + "score": 0.767 + }, + { + "category_id": 1, + "poly": [ + 296, + 1393, + 1397, + 1393, + 1397, + 1459, + 296, + 1459 + ], + "score": 0.762 + }, + { + "category_id": 1, + "poly": [ + 297, + 1805, + 1396, + 1805, + 1396, + 1871, + 297, + 1871 + ], + "score": 0.745 + }, + { + "category_id": 1, + "poly": [ + 293, + 1476, + 1400, + 1476, + 1400, + 1541, + 293, + 1541 + ], + "score": 0.74 + }, + { + "category_id": 1, + "poly": [ + 298, + 817, + 1405, + 817, + 1405, + 911, + 298, + 911 + ], + "score": 0.73 + }, + { + "category_id": 1, + "poly": [ + 296, + 1889, + 1371, + 1889, + 1371, + 1924, + 296, + 1924 + ], + "score": 0.725 + }, + { + "category_id": 1, + "poly": [ + 294, + 929, + 1361, + 929, + 1361, + 964, + 294, + 964 + ], + "score": 0.724 + }, + { + "category_id": 1, + "poly": [ + 297, + 478, + 1399, + 478, + 1399, + 575, + 297, + 575 + ], + "score": 0.707 + }, + { + "category_id": 1, + "poly": [ + 299, + 591, + 1402, + 591, + 1402, + 688, + 299, + 688 + ], + "score": 0.703 + }, + { + "category_id": 1, + "poly": [ + 297, + 704, + 1407, + 704, + 1407, + 798, + 297, + 798 + ], + "score": 0.7 + }, + { + "category_id": 1, + "poly": [ + 297, + 1063, + 1405, + 1063, + 1405, + 1161, + 297, + 1161 + ], + "score": 0.662 + }, + { + "category_id": 1, + "poly": [ + 293, + 397, + 1400, + 397, + 1400, + 460, + 293, + 460 + ], + "score": 0.653 + }, + { + "category_id": 1, + "poly": [ + 299, + 1942, + 1401, + 1942, + 1401, + 2035, + 299, + 2035 + ], + "score": 0.63 + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 490.0, + 230.0, + 490.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 285.0, + 1405.0, + 285.0, + 1405.0, + 319.0, + 295.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 313.0, + 1406.0, + 313.0, + 1406.0, + 353.0, + 320.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 347.0, + 526.0, + 347.0, + 526.0, + 378.0, + 324.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1177.0, + 1324.0, + 1177.0, + 1324.0, + 1216.0, + 294.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1228.0, + 1402.0, + 1228.0, + 1402.0, + 1267.0, + 294.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1262.0, + 642.0, + 1262.0, + 642.0, + 1292.0, + 323.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1636.0, + 1404.0, + 1636.0, + 1404.0, + 1681.0, + 293.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1674.0, + 940.0, + 1674.0, + 940.0, + 1706.0, + 323.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1314.0, + 1403.0, + 1314.0, + 1403.0, + 1346.0, + 296.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1344.0, + 1216.0, + 1344.0, + 1216.0, + 1377.0, + 323.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1559.0, + 1403.0, + 1559.0, + 1403.0, + 1595.0, + 295.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1591.0, + 816.0, + 1591.0, + 816.0, + 1623.0, + 322.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1723.0, + 1404.0, + 1723.0, + 1404.0, + 1759.0, + 295.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1757.0, + 689.0, + 1757.0, + 689.0, + 1786.0, + 323.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 983.0, + 1398.0, + 983.0, + 1398.0, + 1019.0, + 296.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1013.0, + 1099.0, + 1013.0, + 1099.0, + 1049.0, + 324.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1390.0, + 1401.0, + 1390.0, + 1401.0, + 1433.0, + 295.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1426.0, + 749.0, + 1426.0, + 749.0, + 1458.0, + 322.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1804.0, + 1400.0, + 1804.0, + 1400.0, + 1843.0, + 293.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1838.0, + 1159.0, + 1838.0, + 1159.0, + 1873.0, + 322.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1472.0, + 1405.0, + 1472.0, + 1405.0, + 1516.0, + 292.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1507.0, + 1041.0, + 1507.0, + 1041.0, + 1543.0, + 323.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 817.0, + 1403.0, + 817.0, + 1403.0, + 853.0, + 293.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 848.0, + 1402.0, + 848.0, + 1402.0, + 883.0, + 323.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 880.0, + 624.0, + 880.0, + 624.0, + 909.0, + 323.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1888.0, + 1370.0, + 1888.0, + 1370.0, + 1923.0, + 299.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 928.0, + 1364.0, + 928.0, + 1364.0, + 968.0, + 295.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 478.0, + 1403.0, + 478.0, + 1403.0, + 517.0, + 293.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 511.0, + 1403.0, + 511.0, + 1403.0, + 545.0, + 322.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 541.0, + 1234.0, + 541.0, + 1234.0, + 576.0, + 325.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 589.0, + 1404.0, + 589.0, + 1404.0, + 630.0, + 293.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 622.0, + 1407.0, + 622.0, + 1407.0, + 660.0, + 323.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 654.0, + 1369.0, + 654.0, + 1369.0, + 689.0, + 323.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 706.0, + 1404.0, + 706.0, + 1404.0, + 740.0, + 295.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 735.0, + 1405.0, + 735.0, + 1405.0, + 772.0, + 321.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 766.0, + 396.0, + 766.0, + 396.0, + 800.0, + 321.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1064.0, + 1405.0, + 1064.0, + 1405.0, + 1103.0, + 291.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1097.0, + 1403.0, + 1097.0, + 1403.0, + 1132.0, + 323.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1127.0, + 1132.0, + 1127.0, + 1132.0, + 1161.0, + 320.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 393.0, + 1404.0, + 393.0, + 1404.0, + 436.0, + 292.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 427.0, + 691.0, + 427.0, + 691.0, + 462.0, + 322.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1940.0, + 1403.0, + 1940.0, + 1403.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2009.0, + 324.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 2002.0, + 1262.0, + 2002.0, + 1262.0, + 2037.0, + 323.0, + 2037.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 300, + 229, + 1400, + 229, + 1400, + 323, + 300, + 323 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 296, + 425, + 1400, + 425, + 1400, + 490, + 296, + 490 + ], + "score": 0.907 + }, + { + "category_id": 1, + "poly": [ + 295, + 343, + 1401, + 343, + 1401, + 406, + 295, + 406 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 296, + 1256, + 1399, + 1256, + 1399, + 1320, + 296, + 1320 + ], + "score": 0.879 + }, + { + "category_id": 1, + "poly": [ + 298, + 1173, + 1398, + 1173, + 1398, + 1236, + 298, + 1236 + ], + "score": 0.875 + }, + { + "category_id": 1, + "poly": [ + 299, + 509, + 1238, + 509, + 1238, + 543, + 299, + 543 + ], + "score": 0.864 + }, + { + "category_id": 1, + "poly": [ + 296, + 1338, + 1399, + 1338, + 1399, + 1403, + 296, + 1403 + ], + "score": 0.851 + }, + { + "category_id": 1, + "poly": [ + 293, + 561, + 1400, + 561, + 1400, + 625, + 293, + 625 + ], + "score": 0.85 + }, + { + "category_id": 1, + "poly": [ + 297, + 1505, + 1398, + 1505, + 1398, + 1569, + 297, + 1569 + ], + "score": 0.841 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2113, + 836, + 2113 + ], + "score": 0.833 + }, + { + "category_id": 1, + "poly": [ + 298, + 1090, + 1397, + 1090, + 1397, + 1154, + 298, + 1154 + ], + "score": 0.823 + }, + { + "category_id": 1, + "poly": [ + 296, + 1422, + 1398, + 1422, + 1398, + 1487, + 296, + 1487 + ], + "score": 0.82 + }, + { + "category_id": 1, + "poly": [ + 301, + 643, + 1402, + 643, + 1402, + 739, + 301, + 739 + ], + "score": 0.795 + }, + { + "category_id": 1, + "poly": [ + 293, + 840, + 1400, + 840, + 1400, + 905, + 293, + 905 + ], + "score": 0.79 + }, + { + "category_id": 1, + "poly": [ + 292, + 758, + 1402, + 758, + 1402, + 822, + 292, + 822 + ], + "score": 0.784 + }, + { + "category_id": 1, + "poly": [ + 296, + 923, + 1400, + 923, + 1400, + 989, + 296, + 989 + ], + "score": 0.76 + }, + { + "category_id": 1, + "poly": [ + 297, + 1007, + 1397, + 1007, + 1397, + 1071, + 297, + 1071 + ], + "score": 0.759 + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 225.0, + 1405.0, + 225.0, + 1405.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 259.0, + 1406.0, + 259.0, + 1406.0, + 298.0, + 320.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 289.0, + 436.0, + 289.0, + 436.0, + 325.0, + 320.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 428.0, + 1401.0, + 428.0, + 1401.0, + 460.0, + 296.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 456.0, + 1273.0, + 456.0, + 1273.0, + 491.0, + 321.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 343.0, + 1405.0, + 343.0, + 1405.0, + 378.0, + 297.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 375.0, + 714.0, + 375.0, + 714.0, + 406.0, + 323.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1255.0, + 1404.0, + 1255.0, + 1404.0, + 1292.0, + 295.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1288.0, + 1405.0, + 1288.0, + 1405.0, + 1320.0, + 322.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1169.0, + 1403.0, + 1169.0, + 1403.0, + 1213.0, + 292.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1205.0, + 775.0, + 1205.0, + 775.0, + 1238.0, + 322.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 508.0, + 1239.0, + 508.0, + 1239.0, + 546.0, + 293.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1336.0, + 1403.0, + 1336.0, + 1403.0, + 1376.0, + 295.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1369.0, + 731.0, + 1369.0, + 731.0, + 1405.0, + 324.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 557.0, + 1406.0, + 557.0, + 1406.0, + 599.0, + 292.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 593.0, + 689.0, + 593.0, + 689.0, + 625.0, + 323.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1501.0, + 1403.0, + 1501.0, + 1403.0, + 1543.0, + 292.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1536.0, + 1063.0, + 1536.0, + 1063.0, + 1569.0, + 322.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1090.0, + 1401.0, + 1090.0, + 1401.0, + 1126.0, + 295.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1122.0, + 821.0, + 1122.0, + 821.0, + 1153.0, + 324.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1422.0, + 1402.0, + 1422.0, + 1402.0, + 1458.0, + 295.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1452.0, + 1393.0, + 1452.0, + 1393.0, + 1488.0, + 321.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 643.0, + 1402.0, + 643.0, + 1402.0, + 681.0, + 296.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 674.0, + 1404.0, + 674.0, + 1404.0, + 713.0, + 322.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 707.0, + 799.0, + 707.0, + 799.0, + 741.0, + 324.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 839.0, + 1402.0, + 839.0, + 1402.0, + 879.0, + 293.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 872.0, + 862.0, + 872.0, + 862.0, + 905.0, + 325.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 757.0, + 1403.0, + 757.0, + 1403.0, + 793.0, + 295.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 790.0, + 1021.0, + 790.0, + 1021.0, + 823.0, + 323.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 921.0, + 1404.0, + 921.0, + 1404.0, + 961.0, + 295.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 956.0, + 1110.0, + 956.0, + 1110.0, + 988.0, + 324.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1007.0, + 1401.0, + 1007.0, + 1401.0, + 1043.0, + 294.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1039.0, + 1071.0, + 1039.0, + 1071.0, + 1071.0, + 322.0, + 1071.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 297, + 353, + 1219, + 353, + 1219, + 992, + 297, + 992 + ], + "score": 0.985, + "html": "
DatasetNumber of sentencesNumber of tokens(English tokens if bilingual)
En-De ParaCrawl v4.017 Million308 Million
En-Es ParaCrawl v4.022 Million477 Million
En-FiParaCrawl v4.02.16 Million42 Million
En-FrParaCrawl v4.032 Million665 Million
En-Hu ParaCrawl v4.01.91Million31Million
En-It ParaCrawl v4.013 Million261 Million
En-Ru OpenSubtitles+ Tanzil27 Million363Million
Wikipedia - En70 Million1792 Million
Wikipedia - De二1384 Million
Wikipedia - Fr二1108Million
Wikipedia - Es二797 Million
Wikipedia - It二702 Million
Wikipedia - Ru二824 Million
Common Crawl - En二600 Billion
Common Crawl - De二66 Billion
Common Crawl -Fr二68 Billion
Common Crawl - It二36 Billion
Common Crawl -Es172 Billion
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1106, + 1403, + 1106, + 1403, + 1230, + 298, + 1230 + ], + "score": 0.956 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 300, + 227, + 509, + 227, + 509, + 261, + 300, + 261 + ], + "score": 0.851 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.841 + }, + { + "category_id": 6, + "poly": [ + 300, + 287, + 625, + 287, + 625, + 320, + 300, + 320 + ], + "score": 0.77 + }, + { + "category_id": 1, + "poly": [ + 297, + 1012, + 1398, + 1012, + 1398, + 1076, + 297, + 1076 + ], + "score": 0.608 + }, + { + "category_id": 7, + "poly": [ + 297, + 1012, + 1398, + 1012, + 1398, + 1076, + 297, + 1076 + ], + "score": 0.384 + }, + { + "category_id": 0, + "poly": [ + 300, + 287, + 625, + 287, + 625, + 320, + 300, + 320 + ], + "score": 0.153 + }, + { + "category_id": 13, + "poly": [ + 545, + 612, + 568, + 612, + 568, + 635, + 545, + 635 + ], + "score": 0.62, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 759, + 746, + 778, + 746, + 778, + 761, + 759, + 761 + ], + "score": 0.4, + "latex": "-" + }, + { + "category_id": 13, + "poly": [ + 532, + 1108, + 587, + 1108, + 587, + 1137, + 532, + 1137 + ], + "score": 0.33, + "latex": "\\mathrm { v } 4 . 0" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 222.0, + 513.0, + 222.0, + 513.0, + 268.0, + 295.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 285.0, + 627.0, + 285.0, + 627.0, + 323.0, + 295.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1007.0, + 1404.0, + 1007.0, + 1404.0, + 1052.0, + 293.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1041.0, + 1080.0, + 1041.0, + 1080.0, + 1080.0, + 293.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 285.0, + 627.0, + 285.0, + 627.0, + 323.0, + 295.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1106.0, + 531.0, + 1106.0, + 531.0, + 1142.0, + 294.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1106.0, + 1407.0, + 1106.0, + 1407.0, + 1142.0, + 588.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1135.0, + 1405.0, + 1135.0, + 1405.0, + 1173.0, + 294.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1168.0, + 1405.0, + 1168.0, + 1405.0, + 1203.0, + 292.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1200.0, + 733.0, + 1200.0, + 733.0, + 1232.0, + 297.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1007.0, + 1404.0, + 1007.0, + 1404.0, + 1052.0, + 293.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1041.0, + 1080.0, + 1041.0, + 1080.0, + 1080.0, + 293.0, + 1080.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 296, + 296, + 1368, + 296, + 1368, + 683, + 296, + 683 + ], + "score": 0.978, + "html": "
ModelBI-SENT2VECuni.BI-SENT2VECuni. + bi.SENT2VECuni.TRANSGRAM
Embedding dimension300300300300
Maxvocabulary size750k750k750k750k
Minimum word count5855
Initial Learning Rate0.20.20.20.025
Epochs5558
Subsampling hyper-parameter1·10-55:10-61·10-51·10-4
Word-Ngrams Bucket Size二2M二二
Word-Ngrams dropped per context141二
Window size5
Number of negatives sampled1010105
" + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.841 + }, + { + "category_id": 6, + "poly": [ + 577, + 704, + 1123, + 704, + 1123, + 737, + 577, + 737 + ], + "score": 0.639 + }, + { + "category_id": 0, + "poly": [ + 300, + 231, + 930, + 231, + 930, + 262, + 300, + 262 + ], + "score": 0.592 + }, + { + "category_id": 6, + "poly": [ + 578, + 705, + 1123, + 705, + 1123, + 737, + 578, + 737 + ], + "score": 0.148 + }, + { + "category_id": 13, + "poly": [ + 1201, + 521, + 1292, + 521, + 1292, + 547, + 1201, + 547 + ], + "score": 0.84, + "latex": "1 \\cdot 1 0 ^ { - 4 }" + }, + { + "category_id": 13, + "poly": [ + 1062, + 520, + 1152, + 520, + 1152, + 547, + 1062, + 547 + ], + "score": 0.71, + "latex": "\\overline { { 1 \\cdot 1 0 ^ { - 5 } } }" + }, + { + "category_id": 13, + "poly": [ + 703, + 519, + 795, + 519, + 795, + 547, + 703, + 547 + ], + "score": 0.43, + "latex": "\\overline { { 1 \\cdot 1 0 ^ { - 5 } } }" + }, + { + "category_id": 13, + "poly": [ + 963, + 333, + 984, + 333, + 984, + 356, + 963, + 356 + ], + "score": 0.3, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 881, + 519, + 974, + 519, + 974, + 548, + 881, + 548 + ], + "score": 0.26, + "latex": "5 \\cdot 1 0 ^ { - 6 }" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 703.0, + 1125.0, + 703.0, + 1125.0, + 738.0, + 574.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 936.0, + 230.0, + 936.0, + 264.0, + 297.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 703.0, + 1124.0, + 703.0, + 1124.0, + 740.0, + 576.0, + 740.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 471, + 840, + 1228, + 840, + 1228, + 1235, + 471, + 1235 + ], + "score": 0.981, + "html": "
Method\\DatasetSimLex-999 enWS-353 enRG-65 fr
MUSE0.380.740.72
RCSLS0.380.740.70
FASTTEXT- Common Crawl0.490.750.76
BIVEC0.400.700.74
TRANSGRAM0.390.74
SENT2VEC uni.0.72
0.460.750.71
BI-SENT2VEC uni. BI-SENT2VEC uni. + bi.0.55 0.590.78 0.790.74 0.78
" + }, + { + "category_id": 5, + "poly": [ + 500, + 1381, + 1200, + 1381, + 1200, + 1775, + 500, + 1775 + ], + "score": 0.98, + "html": "
Method\\DatasetSimLex-999WS-353
en deende
MUSE0.38 0.410.740.68
RCSLS0.38 0.430.740.70
FASTTEXT- Common Crawl0.49 0.390.750.64
BIVEC0.40 0.410.710.62
TRANSGRAM0.42 0.420.740.66
SENT2VEC uni.0.48 0.380.700.63
BI-SENT2VEC uni.0.56 0.470.760.68
BI-SENT2VEC uni. + bi.0.59 0.530.750.70
" + }, + { + "category_id": 5, + "poly": [ + 515, + 299, + 1183, + 299, + 1183, + 695, + 515, + 695 + ], + "score": 0.979, + "html": "
SimLex-999 Method\\DatasetWS-353
enenes
MUSE RCSLS0.380.740.61
FASTTEXT- Common Crawl0.38 0.490.74 0.750.62 0.54
BIVEC0.400.720.57
TRANSGRAM0.420.740.59
SENT2VEC uni.0.490.580.51
BI-SENT2VEC uni.0.570.780.60
BI-SENT2VEC uni. + bi.0.600.820.66
" + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 858, + 75, + 858, + 105, + 298, + 105 + ], + "score": 0.911 + }, + { + "category_id": 6, + "poly": [ + 295, + 1265, + 1397, + 1265, + 1397, + 1329, + 295, + 1329 + ], + "score": 0.905 + }, + { + "category_id": 6, + "poly": [ + 287, + 725, + 1401, + 725, + 1401, + 790, + 287, + 790 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.858 + }, + { + "category_id": 6, + "poly": [ + 294, + 1806, + 1393, + 1806, + 1393, + 1869, + 294, + 1869 + ], + "score": 0.817 + }, + { + "category_id": 6, + "poly": [ + 300, + 231, + 930, + 231, + 930, + 262, + 300, + 262 + ], + "score": 0.657 + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1266.0, + 1401.0, + 1266.0, + 1401.0, + 1298.0, + 296.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1296.0, + 924.0, + 1296.0, + 924.0, + 1330.0, + 295.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 721.0, + 1404.0, + 721.0, + 1404.0, + 764.0, + 294.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 755.0, + 924.0, + 755.0, + 924.0, + 792.0, + 294.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2122.0, + 832.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1803.0, + 1398.0, + 1803.0, + 1398.0, + 1841.0, + 294.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1836.0, + 956.0, + 1836.0, + 956.0, + 1871.0, + 295.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 229.0, + 935.0, + 229.0, + 935.0, + 265.0, + 297.0, + 265.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/Sk9yuql0Z/Sk9yuql0Z_middle.json b/parse/train/Sk9yuql0Z/Sk9yuql0Z_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..fbf6ec2413cfbd624963b993b207dbb176c60ed8 --- /dev/null +++ b/parse/train/Sk9yuql0Z/Sk9yuql0Z_middle.json @@ -0,0 +1,37640 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 502, + 115 + ], + "lines": [ + { + "bbox": [ + 106, + 79, + 504, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 79, + 504, + 97 + ], + "score": 1.0, + "content": "MITIGATING ADVERSARIAL EFFECTS THROUGH RAN-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 100, + 204, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 204, + 118 + ], + "score": 1.0, + "content": "DOMIZATION", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 114, + 135, + 307, + 147 + ], + "lines": [ + { + "bbox": [ + 112, + 135, + 308, + 149 + ], + "spans": [ + { + "bbox": [ + 112, + 135, + 308, + 149 + ], + "score": 1.0, + "content": "Cihang Xie, Zhishuai Zhang & Alan L. Yuille", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 113, + 148, + 435, + 190 + ], + "lines": [ + { + "bbox": [ + 111, + 145, + 249, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 249, + 158 + ], + "score": 1.0, + "content": "Department of Computer Science", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 111, + 155, + 236, + 170 + ], + "spans": [ + { + "bbox": [ + 111, + 155, + 236, + 170 + ], + "score": 1.0, + "content": "The Johns Hopkins University", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 111, + 168, + 226, + 180 + ], + "spans": [ + { + "bbox": [ + 111, + 168, + 226, + 180 + ], + "score": 1.0, + "content": "Baltimore, MD 21218 USA", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 112, + 178, + 436, + 191 + ], + "spans": [ + { + "bbox": [ + 112, + 178, + 436, + 191 + ], + "score": 1.0, + "content": "{cihangxie306, zhshuai.zhang, alan.l.yuille}@gmail.com", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 113, + 207, + 225, + 251 + ], + "lines": [ + { + "bbox": [ + 111, + 206, + 172, + 221 + ], + "spans": [ + { + "bbox": [ + 111, + 206, + 172, + 221 + ], + "score": 1.0, + "content": "Jianyu Wang", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 111, + 218, + 200, + 229 + ], + "spans": [ + { + "bbox": [ + 111, + 218, + 200, + 229 + ], + "score": 1.0, + "content": "Baidu Research USA", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 111, + 229, + 225, + 241 + ], + "spans": [ + { + "bbox": [ + 111, + 229, + 225, + 241 + ], + "score": 1.0, + "content": "Sunnyvale, CA 94089 USA", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 111, + 241, + 217, + 253 + ], + "spans": [ + { + "bbox": [ + 111, + 241, + 217, + 253 + ], + "score": 1.0, + "content": "wjyouch@gmail.com", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 112, + 268, + 240, + 312 + ], + "lines": [ + { + "bbox": [ + 112, + 267, + 158, + 280 + ], + "spans": [ + { + "bbox": [ + 112, + 267, + 158, + 280 + ], + "score": 1.0, + "content": "Zhou Ren", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 111, + 279, + 154, + 291 + ], + "spans": [ + { + "bbox": [ + 111, + 279, + 154, + 291 + ], + "score": 1.0, + "content": "Snap Inc.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 112, + 290, + 210, + 302 + ], + "spans": [ + { + "bbox": [ + 112, + 290, + 210, + 302 + ], + "score": 1.0, + "content": "Venice, CA 90291 USA", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 113, + 302, + 241, + 313 + ], + "spans": [ + { + "bbox": [ + 113, + 302, + 241, + 313 + ], + "score": 1.0, + "content": "zhou.ren@snapchat.com", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 278, + 342, + 333, + 353 + ], + "lines": [ + { + "bbox": [ + 276, + 340, + 336, + 356 + ], + "spans": [ + { + "bbox": [ + 276, + 340, + 336, + 356 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 143, + 367, + 468, + 554 + ], + "lines": [ + { + "bbox": [ + 142, + 367, + 469, + 379 + ], + "spans": [ + { + "bbox": [ + 142, + 367, + 469, + 379 + ], + "score": 1.0, + "content": "Convolutional neural networks have demonstrated high accuracy on various tasks", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 378, + 469, + 390 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 469, + 390 + ], + "score": 1.0, + "content": "in recent years. However, they are extremely vulnerable to adversarial examples.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 388, + 469, + 402 + ], + "spans": [ + { + "bbox": [ + 141, + 388, + 469, + 402 + ], + "score": 1.0, + "content": "For example, imperceptible perturbations added to clean images can cause convo-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 399, + 469, + 412 + ], + "spans": [ + { + "bbox": [ + 141, + 399, + 469, + 412 + ], + "score": 1.0, + "content": "lutional neural networks to fail. In this paper, we propose to utilize randomization", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 411, + 469, + 422 + ], + "spans": [ + { + "bbox": [ + 142, + 411, + 469, + 422 + ], + "score": 1.0, + "content": "at inference time to mitigate adversarial effects. Specifically, we use two random-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 422, + 469, + 434 + ], + "spans": [ + { + "bbox": [ + 141, + 422, + 469, + 434 + ], + "score": 1.0, + "content": "ization operations: random resizing, which resizes the input images to a random", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 432, + 469, + 445 + ], + "spans": [ + { + "bbox": [ + 141, + 432, + 469, + 445 + ], + "score": 1.0, + "content": "size, and random padding, which pads zeros around the input images in a ran-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 443, + 469, + 456 + ], + "spans": [ + { + "bbox": [ + 141, + 443, + 469, + 456 + ], + "score": 1.0, + "content": "dom manner. Extensive experiments demonstrate that the proposed randomiza-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 455, + 469, + 466 + ], + "spans": [ + { + "bbox": [ + 141, + 455, + 469, + 466 + ], + "score": 1.0, + "content": "tion method is very effective at defending against both single-step and iterative at-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 465, + 470, + 478 + ], + "spans": [ + { + "bbox": [ + 141, + 465, + 470, + 478 + ], + "score": 1.0, + "content": "tacks. Our method provides the following advantages: 1) no additional training or", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 477, + 470, + 489 + ], + "spans": [ + { + "bbox": [ + 141, + 477, + 470, + 489 + ], + "score": 1.0, + "content": "fine-tuning, 2) very few additional computations, 3) compatible with other adver-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 487, + 470, + 499 + ], + "spans": [ + { + "bbox": [ + 141, + 487, + 470, + 499 + ], + "score": 1.0, + "content": "sarial defense methods. By combining the proposed randomization method with", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 498, + 469, + 510 + ], + "spans": [ + { + "bbox": [ + 141, + 498, + 469, + 510 + ], + "score": 1.0, + "content": "an adversarially trained model, it achieves a normalized score of 0.924 (ranked", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 141, + 509, + 469, + 522 + ], + "spans": [ + { + "bbox": [ + 141, + 509, + 469, + 522 + ], + "score": 1.0, + "content": "No.2 among 107 defense teams) in the NIPS 2017 adversarial examples defense", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 142, + 520, + 469, + 532 + ], + "spans": [ + { + "bbox": [ + 142, + 520, + 469, + 532 + ], + "score": 1.0, + "content": "challenge, which is far better than using adversarial training alone with a nor-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 530, + 468, + 544 + ], + "spans": [ + { + "bbox": [ + 141, + 530, + 468, + 544 + ], + "score": 1.0, + "content": "malized score of 0.773 (ranked No.56). The code is public available at https:", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 143, + 543, + 464, + 554 + ], + "spans": [ + { + "bbox": [ + 143, + 543, + 464, + 554 + ], + "score": 1.0, + "content": "//github.com/cihangxie/NIPS2017_adv_challenge_defense.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 108, + 575, + 206, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 574, + 208, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 208, + 591 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 600, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "Convolutional Neural Networks (CNNs) have been successfully applied to a wide range of vision", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "tasks, including image classification (Krizhevsky et al., 2012; Simonyan & Zisserman, 2015; He", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 621, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 104, + 621, + 505, + 636 + ], + "score": 1.0, + "content": "et al., 2016a), object detection (Girshick, 2015; Ren et al., 2015; Zhang et al., 2017), semantic seg-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 632, + 504, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 504, + 646 + ], + "score": 1.0, + "content": "mentation (Long et al., 2015; Chen et al., 2017), visual concept discovery (Wang et al., 2017) etc.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "score": 1.0, + "content": "However, recent works show that CNNs are extremely vulnerable to small perturbations to the input", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "image. For example, adding visually imperceptible perturbations to the original image can result in", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "failures for image classification (Szegedy et al., 2014; Goodfellow et al., 2015), object detection (Xie", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "et al., 2017) and semantic segmentation (Xie et al., 2017; Fischer et al., 2017; Cisse et al., 2017).", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "These perturbed images are called adversarial examples and Figure 1 gives an example. Adversarial", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "examples pose a great security danger to the deployment of commercial machine learning systems.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "Thus, making CNNs more robust to adversarial examples is a very important yet challenging prob-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "lem. Recent works (Papernot et al., 2016b; Kurakin et al., 2017; Tramer et al., 2017; Cao & Gong, `", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 39.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 502, + 115 + ], + "lines": [ + { + "bbox": [ + 106, + 79, + 504, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 79, + 504, + 97 + ], + "score": 1.0, + "content": "MITIGATING ADVERSARIAL EFFECTS THROUGH RAN-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 100, + 204, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 204, + 118 + ], + "score": 1.0, + "content": "DOMIZATION", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 114, + 135, + 307, + 147 + ], + "lines": [ + { + "bbox": [ + 112, + 135, + 308, + 149 + ], + "spans": [ + { + "bbox": [ + 112, + 135, + 308, + 149 + ], + "score": 1.0, + "content": "Cihang Xie, Zhishuai Zhang & Alan L. Yuille", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 112, + 135, + 308, + 149 + ] + }, + { + "type": "list", + "bbox": [ + 113, + 148, + 435, + 190 + ], + "lines": [ + { + "bbox": [ + 111, + 145, + 249, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 249, + 158 + ], + "score": 1.0, + "content": "Department of Computer Science", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 155, + 236, + 170 + ], + "spans": [ + { + "bbox": [ + 111, + 155, + 236, + 170 + ], + "score": 1.0, + "content": "The Johns Hopkins University", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 168, + 226, + 180 + ], + "spans": [ + { + "bbox": [ + 111, + 168, + 226, + 180 + ], + "score": 1.0, + "content": "Baltimore, MD 21218 USA", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 112, + 178, + 436, + 191 + ], + "spans": [ + { + "bbox": [ + 112, + 178, + 436, + 191 + ], + "score": 1.0, + "content": "{cihangxie306, zhshuai.zhang, alan.l.yuille}@gmail.com", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + } + ], + "index": 4.5, + "bbox_fs": [ + 111, + 145, + 436, + 191 + ] + }, + { + "type": "text", + "bbox": [ + 113, + 207, + 225, + 251 + ], + "lines": [ + { + "bbox": [ + 111, + 206, + 172, + 221 + ], + "spans": [ + { + "bbox": [ + 111, + 206, + 172, + 221 + ], + "score": 1.0, + "content": "Jianyu Wang", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 111, + 218, + 200, + 229 + ], + "spans": [ + { + "bbox": [ + 111, + 218, + 200, + 229 + ], + "score": 1.0, + "content": "Baidu Research USA", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 111, + 229, + 225, + 241 + ], + "spans": [ + { + "bbox": [ + 111, + 229, + 225, + 241 + ], + "score": 1.0, + "content": "Sunnyvale, CA 94089 USA", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 111, + 241, + 217, + 253 + ], + "spans": [ + { + "bbox": [ + 111, + 241, + 217, + 253 + ], + "score": 1.0, + "content": "wjyouch@gmail.com", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 111, + 206, + 225, + 253 + ] + }, + { + "type": "list", + "bbox": [ + 112, + 268, + 240, + 312 + ], + "lines": [ + { + "bbox": [ + 112, + 267, + 158, + 280 + ], + "spans": [ + { + "bbox": [ + 112, + 267, + 158, + 280 + ], + "score": 1.0, + "content": "Zhou Ren", + "type": "text" + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 111, + 279, + 154, + 291 + ], + "spans": [ + { + "bbox": [ + 111, + 279, + 154, + 291 + ], + "score": 1.0, + "content": "Snap Inc.", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 112, + 290, + 210, + 302 + ], + "spans": [ + { + "bbox": [ + 112, + 290, + 210, + 302 + ], + "score": 1.0, + "content": "Venice, CA 90291 USA", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 113, + 302, + 241, + 313 + ], + "spans": [ + { + "bbox": [ + 113, + 302, + 241, + 313 + ], + "score": 1.0, + "content": "zhou.ren@snapchat.com", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + } + ], + "index": 12.5, + "bbox_fs": [ + 111, + 267, + 241, + 313 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 342, + 333, + 353 + ], + "lines": [ + { + "bbox": [ + 276, + 340, + 336, + 356 + ], + "spans": [ + { + "bbox": [ + 276, + 340, + 336, + 356 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 143, + 367, + 468, + 554 + ], + "lines": [ + { + "bbox": [ + 142, + 367, + 469, + 379 + ], + "spans": [ + { + "bbox": [ + 142, + 367, + 469, + 379 + ], + "score": 1.0, + "content": "Convolutional neural networks have demonstrated high accuracy on various tasks", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 378, + 469, + 390 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 469, + 390 + ], + "score": 1.0, + "content": "in recent years. However, they are extremely vulnerable to adversarial examples.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 388, + 469, + 402 + ], + "spans": [ + { + "bbox": [ + 141, + 388, + 469, + 402 + ], + "score": 1.0, + "content": "For example, imperceptible perturbations added to clean images can cause convo-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 399, + 469, + 412 + ], + "spans": [ + { + "bbox": [ + 141, + 399, + 469, + 412 + ], + "score": 1.0, + "content": "lutional neural networks to fail. In this paper, we propose to utilize randomization", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 411, + 469, + 422 + ], + "spans": [ + { + "bbox": [ + 142, + 411, + 469, + 422 + ], + "score": 1.0, + "content": "at inference time to mitigate adversarial effects. Specifically, we use two random-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 422, + 469, + 434 + ], + "spans": [ + { + "bbox": [ + 141, + 422, + 469, + 434 + ], + "score": 1.0, + "content": "ization operations: random resizing, which resizes the input images to a random", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 432, + 469, + 445 + ], + "spans": [ + { + "bbox": [ + 141, + 432, + 469, + 445 + ], + "score": 1.0, + "content": "size, and random padding, which pads zeros around the input images in a ran-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 443, + 469, + 456 + ], + "spans": [ + { + "bbox": [ + 141, + 443, + 469, + 456 + ], + "score": 1.0, + "content": "dom manner. Extensive experiments demonstrate that the proposed randomiza-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 455, + 469, + 466 + ], + "spans": [ + { + "bbox": [ + 141, + 455, + 469, + 466 + ], + "score": 1.0, + "content": "tion method is very effective at defending against both single-step and iterative at-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 465, + 470, + 478 + ], + "spans": [ + { + "bbox": [ + 141, + 465, + 470, + 478 + ], + "score": 1.0, + "content": "tacks. Our method provides the following advantages: 1) no additional training or", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 477, + 470, + 489 + ], + "spans": [ + { + "bbox": [ + 141, + 477, + 470, + 489 + ], + "score": 1.0, + "content": "fine-tuning, 2) very few additional computations, 3) compatible with other adver-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 487, + 470, + 499 + ], + "spans": [ + { + "bbox": [ + 141, + 487, + 470, + 499 + ], + "score": 1.0, + "content": "sarial defense methods. By combining the proposed randomization method with", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 498, + 469, + 510 + ], + "spans": [ + { + "bbox": [ + 141, + 498, + 469, + 510 + ], + "score": 1.0, + "content": "an adversarially trained model, it achieves a normalized score of 0.924 (ranked", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 141, + 509, + 469, + 522 + ], + "spans": [ + { + "bbox": [ + 141, + 509, + 469, + 522 + ], + "score": 1.0, + "content": "No.2 among 107 defense teams) in the NIPS 2017 adversarial examples defense", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 142, + 520, + 469, + 532 + ], + "spans": [ + { + "bbox": [ + 142, + 520, + 469, + 532 + ], + "score": 1.0, + "content": "challenge, which is far better than using adversarial training alone with a nor-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 530, + 468, + 544 + ], + "spans": [ + { + "bbox": [ + 141, + 530, + 468, + 544 + ], + "score": 1.0, + "content": "malized score of 0.773 (ranked No.56). The code is public available at https:", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 143, + 543, + 464, + 554 + ], + "spans": [ + { + "bbox": [ + 143, + 543, + 464, + 554 + ], + "score": 1.0, + "content": "//github.com/cihangxie/NIPS2017_adv_challenge_defense.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 24, + "bbox_fs": [ + 141, + 367, + 470, + 554 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 575, + 206, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 574, + 208, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 208, + 591 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 600, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "Convolutional Neural Networks (CNNs) have been successfully applied to a wide range of vision", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "tasks, including image classification (Krizhevsky et al., 2012; Simonyan & Zisserman, 2015; He", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 621, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 104, + 621, + 505, + 636 + ], + "score": 1.0, + "content": "et al., 2016a), object detection (Girshick, 2015; Ren et al., 2015; Zhang et al., 2017), semantic seg-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 632, + 504, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 504, + 646 + ], + "score": 1.0, + "content": "mentation (Long et al., 2015; Chen et al., 2017), visual concept discovery (Wang et al., 2017) etc.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "score": 1.0, + "content": "However, recent works show that CNNs are extremely vulnerable to small perturbations to the input", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "image. For example, adding visually imperceptible perturbations to the original image can result in", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "failures for image classification (Szegedy et al., 2014; Goodfellow et al., 2015), object detection (Xie", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "et al., 2017) and semantic segmentation (Xie et al., 2017; Fischer et al., 2017; Cisse et al., 2017).", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "These perturbed images are called adversarial examples and Figure 1 gives an example. Adversarial", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "examples pose a great security danger to the deployment of commercial machine learning systems.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "Thus, making CNNs more robust to adversarial examples is a very important yet challenging prob-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "lem. Recent works (Papernot et al., 2016b; Kurakin et al., 2017; Tramer et al., 2017; Cao & Gong, `", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 39.5, + "bbox_fs": [ + 104, + 600, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 127, + 79, + 485, + 178 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 127, + 79, + 485, + 178 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 127, + 79, + 485, + 178 + ], + "spans": [ + { + "bbox": [ + 127, + 79, + 485, + 178 + ], + "score": 0.936, + "type": "image", + "image_path": "48b3009e8ed26fd44338fb7b72d9eaf8e2ea6476d67d321696065b4e396f4087.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 127, + 79, + 485, + 112.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 127, + 112.0, + 485, + 145.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 127, + 145.0, + 485, + 178.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 190, + 505, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 190, + 504, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 504, + 201 + ], + "score": 1.0, + "content": "Figure 1: This is an adversarial example crafted for VGG (Simonyan & Zisserman, 2015). The", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 200, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 505, + 214 + ], + "score": 1.0, + "content": "left image is classified correctly as king penguin, the center image is the adversarial perturbation", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 212, + 504, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 504, + 224 + ], + "score": 1.0, + "content": "(magnified by 10 and enlarged by 128 for better visualization), and the right image is the adversarial", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 223, + 250, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 223, + 250, + 235 + ], + "score": 1.0, + "content": "example misclassfied as chihuahua.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 108, + 247, + 503, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 246, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 505, + 261 + ], + "score": 1.0, + "content": "2017; Metzen et al., 2017; Feinman et al., 2017; Meng & Chen, 2017) are making progress on this", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 258, + 173, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 173, + 270 + ], + "score": 1.0, + "content": "line of research.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 107, + 275, + 505, + 353 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "Adversarial attacks can be divided into two categories: single-step attacks, which perform only", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "one step of gradient computation, and iterative attacks, which perform multiple steps. Intuitively,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 298, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 505, + 310 + ], + "score": 1.0, + "content": "the perturbation generated by iterative methods may easily get over-fitted to the specific network", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 307, + 506, + 323 + ], + "spans": [ + { + "bbox": [ + 104, + 307, + 506, + 323 + ], + "score": 1.0, + "content": "parameters, and thus be less transferable. On the other hand, single-step methods may not be strong", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 319, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 505, + 332 + ], + "score": 1.0, + "content": "enough to fool the network. For examples, it has been demonstrated that single-step attacks, like", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "score": 1.0, + "content": "Fast Gradient Sign Method (FGSM) (Goodfellow et al., 2015), have better transferability but weaker", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 342, + 427, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 427, + 354 + ], + "score": 1.0, + "content": "attack rate than iterative attacks, like DeepFool (Moosavi-Dezfooli et al., 2016).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 358, + 505, + 468 + ], + "lines": [ + { + "bbox": [ + 105, + 356, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 506, + 372 + ], + "score": 1.0, + "content": "Due to the weak generalization of iterative attacks, low-level image transformations, e.g., resizing,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "padding, compression, etc, may probably destroy the specific structure of adversarial perturbations,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 381, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 505, + 392 + ], + "score": 1.0, + "content": "thus making it a good defense. It can even defend against white-box iterative attacks if random trans-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 391, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 506, + 405 + ], + "score": 1.0, + "content": "formations are applied. This is because each test image goes through random transformations and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "score": 1.0, + "content": "the attacker does not know the specific transformation when generating adversarial noise. Recently,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "score": 1.0, + "content": "adversarial training (Kurakin et al., 2017; Tramer et al., 2017) was developed to defend against `", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 424, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 506, + 437 + ], + "score": 1.0, + "content": "single-step attacks. Thus by adding the proposed random transformations as additional layers to", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "an adversarially trained model (Tramer et al., 2017), it is expected that the method is able to effec- `", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "tively defend against both single-step and iterative attacks, including both black-box and white-box", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 455, + 143, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 143, + 472 + ], + "score": 1.0, + "content": "settings.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 107, + 474, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "score": 1.0, + "content": "Based on the above reasoning, in this paper, we propose a defense method by randomization at", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "inference time, i.e., random resizing and random padding, to mitigate adversarial effects. To the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "score": 1.0, + "content": "best of our knowledge, this is the first work that demonstrates the effectiveness of randomization at", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 507, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 505, + 520 + ], + "score": 1.0, + "content": "inference time on mitigating adversarial effects on large-scale dataset, e.g., ImageNet (Deng et al.,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 516, + 366, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 366, + 531 + ], + "score": 1.0, + "content": "2009). The proposed method provides the following advantages:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 132, + 539, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 132, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 132, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "• Randomization at inference time makes the network much more robust to adversarial im-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 141, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "ages, especially for iterative attacks (both white-box and black box), but hardly hurts the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 141, + 561, + 506, + 575 + ], + "spans": [ + { + "bbox": [ + 141, + 561, + 506, + 575 + ], + "score": 1.0, + "content": "performance on clean (non-adversarial) images. Experiments on section 4.2 support this", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 142, + 574, + 185, + 585 + ], + "spans": [ + { + "bbox": [ + 142, + 574, + 185, + 585 + ], + "score": 1.0, + "content": "argument.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 136, + 588, + 496, + 602 + ], + "spans": [ + { + "bbox": [ + 136, + 588, + 496, + 602 + ], + "score": 1.0, + "content": "• There is no additional training or fine-tuning required which is easy for implementation.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 135, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 135, + 605, + 506, + 618 + ], + "score": 1.0, + "content": "• Very few computations are required by adding the two randomization layers, thus there is", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 142, + 617, + 256, + 629 + ], + "spans": [ + { + "bbox": [ + 142, + 617, + 256, + 629 + ], + "score": 1.0, + "content": "nearly no run time increase.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 131, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 131, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "• Randomization layers are compatible to different network structures and adversarial de-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 141, + 644, + 475, + 656 + ], + "spans": [ + { + "bbox": [ + 141, + 644, + 475, + 656 + ], + "score": 1.0, + "content": "fense methods, which can serve as a basic network module for adversarial defense.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 666, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 665, + 504, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 665, + 504, + 678 + ], + "score": 1.0, + "content": "We conduct comprehensive experiments to test the effectiveness of our defense method, using dif-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "ferent network structures, against different attack methods, and under different attack scenarios.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "The results in Section 4 demonstrate that the proposed randomization layers can significantly miti-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "gate adversarial effects, especially for iterative attack methods. Moreover, we submitted the model,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "which combines the proposed randomization layers and an adversarially trained model (Tramer`", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "et al., 2017), to the NIPS 2017 adversarial examples defense challenge. It reaches a normalized", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 127, + 79, + 485, + 178 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 127, + 79, + 485, + 178 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 127, + 79, + 485, + 178 + ], + "spans": [ + { + "bbox": [ + 127, + 79, + 485, + 178 + ], + "score": 0.936, + "type": "image", + "image_path": "48b3009e8ed26fd44338fb7b72d9eaf8e2ea6476d67d321696065b4e396f4087.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 127, + 79, + 485, + 112.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 127, + 112.0, + 485, + 145.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 127, + 145.0, + 485, + 178.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 190, + 505, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 190, + 504, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 504, + 201 + ], + "score": 1.0, + "content": "Figure 1: This is an adversarial example crafted for VGG (Simonyan & Zisserman, 2015). The", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 200, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 505, + 214 + ], + "score": 1.0, + "content": "left image is classified correctly as king penguin, the center image is the adversarial perturbation", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 212, + 504, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 504, + 224 + ], + "score": 1.0, + "content": "(magnified by 10 and enlarged by 128 for better visualization), and the right image is the adversarial", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 223, + 250, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 223, + 250, + 235 + ], + "score": 1.0, + "content": "example misclassfied as chihuahua.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 108, + 247, + 503, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 246, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 505, + 261 + ], + "score": 1.0, + "content": "2017; Metzen et al., 2017; Feinman et al., 2017; Meng & Chen, 2017) are making progress on this", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 258, + 173, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 173, + 270 + ], + "score": 1.0, + "content": "line of research.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 246, + 505, + 270 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 275, + 505, + 353 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "Adversarial attacks can be divided into two categories: single-step attacks, which perform only", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "one step of gradient computation, and iterative attacks, which perform multiple steps. Intuitively,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 298, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 505, + 310 + ], + "score": 1.0, + "content": "the perturbation generated by iterative methods may easily get over-fitted to the specific network", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 307, + 506, + 323 + ], + "spans": [ + { + "bbox": [ + 104, + 307, + 506, + 323 + ], + "score": 1.0, + "content": "parameters, and thus be less transferable. On the other hand, single-step methods may not be strong", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 319, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 505, + 332 + ], + "score": 1.0, + "content": "enough to fool the network. For examples, it has been demonstrated that single-step attacks, like", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "score": 1.0, + "content": "Fast Gradient Sign Method (FGSM) (Goodfellow et al., 2015), have better transferability but weaker", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 342, + 427, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 427, + 354 + ], + "score": 1.0, + "content": "attack rate than iterative attacks, like DeepFool (Moosavi-Dezfooli et al., 2016).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12, + "bbox_fs": [ + 104, + 276, + 506, + 354 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 358, + 505, + 468 + ], + "lines": [ + { + "bbox": [ + 105, + 356, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 506, + 372 + ], + "score": 1.0, + "content": "Due to the weak generalization of iterative attacks, low-level image transformations, e.g., resizing,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "padding, compression, etc, may probably destroy the specific structure of adversarial perturbations,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 381, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 505, + 392 + ], + "score": 1.0, + "content": "thus making it a good defense. It can even defend against white-box iterative attacks if random trans-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 391, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 506, + 405 + ], + "score": 1.0, + "content": "formations are applied. This is because each test image goes through random transformations and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "score": 1.0, + "content": "the attacker does not know the specific transformation when generating adversarial noise. Recently,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "score": 1.0, + "content": "adversarial training (Kurakin et al., 2017; Tramer et al., 2017) was developed to defend against `", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 424, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 506, + 437 + ], + "score": 1.0, + "content": "single-step attacks. Thus by adding the proposed random transformations as additional layers to", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "an adversarially trained model (Tramer et al., 2017), it is expected that the method is able to effec- `", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "tively defend against both single-step and iterative attacks, including both black-box and white-box", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 455, + 143, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 143, + 472 + ], + "score": 1.0, + "content": "settings.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 356, + 506, + 472 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 474, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "score": 1.0, + "content": "Based on the above reasoning, in this paper, we propose a defense method by randomization at", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "inference time, i.e., random resizing and random padding, to mitigate adversarial effects. To the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "score": 1.0, + "content": "best of our knowledge, this is the first work that demonstrates the effectiveness of randomization at", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 507, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 505, + 520 + ], + "score": 1.0, + "content": "inference time on mitigating adversarial effects on large-scale dataset, e.g., ImageNet (Deng et al.,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 516, + 366, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 366, + 531 + ], + "score": 1.0, + "content": "2009). The proposed method provides the following advantages:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 473, + 506, + 531 + ] + }, + { + "type": "list", + "bbox": [ + 132, + 539, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 132, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 132, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "• Randomization at inference time makes the network much more robust to adversarial im-", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 141, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "ages, especially for iterative attacks (both white-box and black box), but hardly hurts the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 141, + 561, + 506, + 575 + ], + "spans": [ + { + "bbox": [ + 141, + 561, + 506, + 575 + ], + "score": 1.0, + "content": "performance on clean (non-adversarial) images. Experiments on section 4.2 support this", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 142, + 574, + 185, + 585 + ], + "spans": [ + { + "bbox": [ + 142, + 574, + 185, + 585 + ], + "score": 1.0, + "content": "argument.", + "type": "text" + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 136, + 588, + 496, + 602 + ], + "spans": [ + { + "bbox": [ + 136, + 588, + 496, + 602 + ], + "score": 1.0, + "content": "• There is no additional training or fine-tuning required which is easy for implementation.", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 135, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 135, + 605, + 506, + 618 + ], + "score": 1.0, + "content": "• Very few computations are required by adding the two randomization layers, thus there is", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 617, + 256, + 629 + ], + "spans": [ + { + "bbox": [ + 142, + 617, + 256, + 629 + ], + "score": 1.0, + "content": "nearly no run time increase.", + "type": "text" + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 131, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "• Randomization layers are compatible to different network structures and adversarial de-", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 644, + 475, + 656 + ], + "spans": [ + { + "bbox": [ + 141, + 644, + 475, + 656 + ], + "score": 1.0, + "content": "fense methods, which can serve as a basic network module for adversarial defense.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + } + ], + "index": 35, + "bbox_fs": [ + 131, + 539, + 506, + 656 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 666, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 665, + 504, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 665, + 504, + 678 + ], + "score": 1.0, + "content": "We conduct comprehensive experiments to test the effectiveness of our defense method, using dif-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "ferent network structures, against different attack methods, and under different attack scenarios.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "The results in Section 4 demonstrate that the proposed randomization layers can significantly miti-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "gate adversarial effects, especially for iterative attack methods. Moreover, we submitted the model,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "which combines the proposed randomization layers and an adversarially trained model (Tramer`", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "et al., 2017), to the NIPS 2017 adversarial examples defense challenge. It reaches a normalized", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "score of 0.924 (ranked No.2 among 107 defense teams), which is far better than just using adversar-", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 461, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 461, + 105 + ], + "score": 1.0, + "content": "ial training (Tramer et al., 2017) alone with a normalized score of ` 0.773 (ranked No.56).", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 665, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 503, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "score of 0.924 (ranked No.2 among 107 defense teams), which is far better than just using adversar-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 461, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 461, + 105 + ], + "score": 1.0, + "content": "ial training (Tramer et al., 2017) alone with a normalized score of ` 0.773 (ranked No.56).", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "title", + "bbox": [ + 108, + 123, + 211, + 136 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 213, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 213, + 138 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 108, + 150, + 303, + 160 + ], + "lines": [ + { + "bbox": [ + 106, + 149, + 304, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 304, + 162 + ], + "score": 1.0, + "content": "2.1 GENERATING ADVERSARIAL EXAMPLES", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 171, + 505, + 325 + ], + "lines": [ + { + "bbox": [ + 106, + 171, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 506, + 184 + ], + "score": 1.0, + "content": "Generating adversarial examples has been extensively studied recently. (Szegedy et al., 2014) first", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 182, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 195 + ], + "score": 1.0, + "content": "showed that adversarial examples, computed by adding visually imperceptible perturbations to the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 193, + 506, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 506, + 206 + ], + "score": 1.0, + "content": "original images, make CNNs predict wrong labels with high confidence. (Goodfellow et al., 2015)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 204, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 506, + 217 + ], + "score": 1.0, + "content": "proposed the fast gradient sign method to generate adversarial examples based on the linear nature", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 215, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 228 + ], + "score": 1.0, + "content": "of CNNs, and also proposed adversarial training for defense. (Moosavi-Dezfooli et al., 2016) gen-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 227, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 505, + 238 + ], + "score": 1.0, + "content": "erated adversarial examples by assuming that the loss function can be linearized around the current", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "data point at each iteration. (Carlini & Wagner, 2017) developed a stronger attack to find adver-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "sarial perturbations by introducing auxiliary variables which incooperate the pixel value constrain,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "score": 1.0, + "content": "e.g., pixel intensity must be within the range [0,255], naturally into the loss function and make the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 269, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 505, + 283 + ], + "score": 1.0, + "content": "optimization process easier. (Liu et al., 2017) proposed an ensemble-based approaches to gener-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "ate adversarial examples with stronger transferability. Unlike the works above, (Biggio & Laskov,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 292, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 505, + 305 + ], + "score": 1.0, + "content": "2012; Koh & Liang, 2017) showed that manipulating only a small fraction of the training data can", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "score": 1.0, + "content": "significantly increase the number of misclassified samples at test time for learning algorithms, and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 315, + 272, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 272, + 326 + ], + "score": 1.0, + "content": "such attacks are called poisoning attacks.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 107, + 341, + 340, + 352 + ], + "lines": [ + { + "bbox": [ + 106, + 341, + 341, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 341, + 354 + ], + "score": 1.0, + "content": "2.2 DEFENDING AGAINST ADVERSARIAL EXAMPLES", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 363, + 505, + 516 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "Opposite to generating adversarial examples, there is also progress on reducing the effects of ad-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "versarial examples. (Papernot et al., 2016b) showed networks trained using defensive distillation", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "can effectively defend against adversarial examples. (Kurakin et al., 2017) proposed to replace the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "original clean images with a mixture of clean images and corresponding adversarial images in each", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 407, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 505, + 419 + ], + "score": 1.0, + "content": "training batch to improve the network robustness. (Tramer et al., 2017) improved the robustness `", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "score": 1.0, + "content": "further by training the network on an ensemble of adversarial images generated from the trained", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "score": 1.0, + "content": "model itself and from a number of other pre-trained models. Cao & Gong (2017) proposed a region-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 439, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 505, + 452 + ], + "score": 1.0, + "content": "based classification to let models be robust to adversarial examples. (Metzen et al., 2017) trained", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 451, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 462 + ], + "score": 1.0, + "content": "a detector on the inner layer of the classifier to detect adversarial examples. (Feinman et al., 2017)", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 460, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 474 + ], + "score": 1.0, + "content": "detected adversarial examples by looking at the Bayesian uncertainty estimates of the input images", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "in dropout neural networks and by performing density estimation in the subspace of deep features", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "learned by the model. MagNet (Meng & Chen, 2017) detected adversarial examples with large per-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "turbation using detector networks, and pushed adversarial examples with small perturbation towards", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 504, + 227, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 227, + 518 + ], + "score": 1.0, + "content": "the manifold of clean images.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 25.5 + }, + { + "type": "title", + "bbox": [ + 108, + 535, + 182, + 547 + ], + "lines": [ + { + "bbox": [ + 105, + 533, + 185, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 185, + 550 + ], + "score": 1.0, + "content": "3 APPROACH", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "title", + "bbox": [ + 108, + 561, + 383, + 573 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 384, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 384, + 574 + ], + "score": 1.0, + "content": "3.1 AN OVERVIEW OF GENERATING ADVERSARIAL EXAMPLES", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 661 + ], + "lines": [ + { + "bbox": [ + 105, + 581, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 505, + 596 + ], + "score": 1.0, + "content": "Before introducing the proposed adversarial defense method, we give an overview of generating", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 217, + 606 + ], + "score": 1.0, + "content": "adversarial examples. Let", + "type": "text" + }, + { + "bbox": [ + 218, + 594, + 232, + 605 + ], + "score": 0.9, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 594, + 281, + 606 + ], + "score": 1.0, + "content": "denote the", + "type": "text" + }, + { + "bbox": [ + 281, + 596, + 288, + 604 + ], + "score": 0.79, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 594, + 426, + 606 + ], + "score": 1.0, + "content": "-th image in a dataset containing", + "type": "text" + }, + { + "bbox": [ + 427, + 594, + 437, + 604 + ], + "score": 0.8, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "images, and let", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 601, + 508, + 622 + ], + "spans": [ + { + "bbox": [ + 107, + 605, + 124, + 617 + ], + "score": 0.91, + "content": "y _ { n } ^ { \\mathrm { t r u e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 601, + 341, + 622 + ], + "score": 1.0, + "content": "denote the corresponding ground-truth label. We use", + "type": "text" + }, + { + "bbox": [ + 341, + 605, + 348, + 615 + ], + "score": 0.78, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 601, + 508, + 622 + ], + "score": 1.0, + "content": "to denote the network parameters, and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 107, + 614, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 107, + 616, + 166, + 628 + ], + "score": 0.9, + "content": "L ( X _ { n } , y _ { n } ^ { \\mathrm { t r u e } } ; \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 614, + 505, + 630 + ], + "score": 1.0, + "content": "to denote the loss. For the adversarial example generation, the goal is to maximize", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 103, + 623, + 507, + 643 + ], + "spans": [ + { + "bbox": [ + 103, + 623, + 140, + 643 + ], + "score": 1.0, + "content": "the loss", + "type": "text" + }, + { + "bbox": [ + 140, + 627, + 221, + 639 + ], + "score": 0.88, + "content": "L ( X _ { n } + r _ { n } , y _ { n } ^ { \\mathrm { t r u e } } ; \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 623, + 284, + 643 + ], + "score": 1.0, + "content": "for each image", + "type": "text" + }, + { + "bbox": [ + 285, + 627, + 299, + 638 + ], + "score": 0.89, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 623, + 507, + 643 + ], + "score": 1.0, + "content": ", under the constraint that the generated adversarial", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 635, + 507, + 653 + ], + "spans": [ + { + "bbox": [ + 104, + 635, + 142, + 653 + ], + "score": 1.0, + "content": "example", + "type": "text" + }, + { + "bbox": [ + 142, + 639, + 211, + 650 + ], + "score": 0.9, + "content": "X _ { n } ^ { \\mathrm { a d v } } = X _ { n } + r _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 635, + 409, + 653 + ], + "score": 1.0, + "content": "should look visually similar to the original image", + "type": "text" + }, + { + "bbox": [ + 410, + 639, + 424, + 649 + ], + "score": 0.89, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 635, + 445, + 653 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + }, + { + "bbox": [ + 446, + 638, + 485, + 650 + ], + "score": 0.92, + "content": "| | r _ { n } | | \\leq \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 635, + 507, + 653 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 645, + 294, + 665 + ], + "spans": [ + { + "bbox": [ + 104, + 645, + 242, + 665 + ], + "score": 1.0, + "content": "the corresponding predicted label", + "type": "text" + }, + { + "bbox": [ + 242, + 649, + 289, + 661 + ], + "score": 0.92, + "content": "y _ { n } ^ { \\mathrm { a d v } } \\neq y _ { n } ^ { \\mathrm { t r u e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 645, + 294, + 665 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 666, + 504, + 699 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "In our experiment, we consider three different attack methods, including one single-step attack", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 675, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 505, + 689 + ], + "score": 1.0, + "content": "method and two iterative attack methods. We use the cleverhans library (Papernot et al., 2016a)", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 687, + 497, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 497, + 700 + ], + "score": 1.0, + "content": "to generate adversarial examples, where all these attacks have been implemented via TensorFlow.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 132, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 132, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 132, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "• Fast Gradient Sign Method (FGSM): FGSM (Goodfellow et al., 2015) is a single-step at-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 141, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 141, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "tack method. It finds the adversarial perturbation that yields the highest increase of the", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 503, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 505, + 105 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 123, + 211, + 136 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 213, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 213, + 138 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 108, + 150, + 303, + 160 + ], + "lines": [ + { + "bbox": [ + 106, + 149, + 304, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 304, + 162 + ], + "score": 1.0, + "content": "2.1 GENERATING ADVERSARIAL EXAMPLES", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 171, + 505, + 325 + ], + "lines": [ + { + "bbox": [ + 106, + 171, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 506, + 184 + ], + "score": 1.0, + "content": "Generating adversarial examples has been extensively studied recently. (Szegedy et al., 2014) first", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 182, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 195 + ], + "score": 1.0, + "content": "showed that adversarial examples, computed by adding visually imperceptible perturbations to the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 193, + 506, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 506, + 206 + ], + "score": 1.0, + "content": "original images, make CNNs predict wrong labels with high confidence. (Goodfellow et al., 2015)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 204, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 506, + 217 + ], + "score": 1.0, + "content": "proposed the fast gradient sign method to generate adversarial examples based on the linear nature", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 215, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 228 + ], + "score": 1.0, + "content": "of CNNs, and also proposed adversarial training for defense. (Moosavi-Dezfooli et al., 2016) gen-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 227, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 505, + 238 + ], + "score": 1.0, + "content": "erated adversarial examples by assuming that the loss function can be linearized around the current", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "data point at each iteration. (Carlini & Wagner, 2017) developed a stronger attack to find adver-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "sarial perturbations by introducing auxiliary variables which incooperate the pixel value constrain,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "score": 1.0, + "content": "e.g., pixel intensity must be within the range [0,255], naturally into the loss function and make the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 269, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 505, + 283 + ], + "score": 1.0, + "content": "optimization process easier. (Liu et al., 2017) proposed an ensemble-based approaches to gener-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "ate adversarial examples with stronger transferability. Unlike the works above, (Biggio & Laskov,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 292, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 505, + 305 + ], + "score": 1.0, + "content": "2012; Koh & Liang, 2017) showed that manipulating only a small fraction of the training data can", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "score": 1.0, + "content": "significantly increase the number of misclassified samples at test time for learning algorithms, and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 315, + 272, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 272, + 326 + ], + "score": 1.0, + "content": "such attacks are called poisoning attacks.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 171, + 506, + 326 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 341, + 340, + 352 + ], + "lines": [ + { + "bbox": [ + 106, + 341, + 341, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 341, + 354 + ], + "score": 1.0, + "content": "2.2 DEFENDING AGAINST ADVERSARIAL EXAMPLES", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 363, + 505, + 516 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "Opposite to generating adversarial examples, there is also progress on reducing the effects of ad-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "versarial examples. (Papernot et al., 2016b) showed networks trained using defensive distillation", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "can effectively defend against adversarial examples. (Kurakin et al., 2017) proposed to replace the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "original clean images with a mixture of clean images and corresponding adversarial images in each", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 407, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 505, + 419 + ], + "score": 1.0, + "content": "training batch to improve the network robustness. (Tramer et al., 2017) improved the robustness `", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 506, + 430 + ], + "score": 1.0, + "content": "further by training the network on an ensemble of adversarial images generated from the trained", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "score": 1.0, + "content": "model itself and from a number of other pre-trained models. Cao & Gong (2017) proposed a region-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 439, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 505, + 452 + ], + "score": 1.0, + "content": "based classification to let models be robust to adversarial examples. (Metzen et al., 2017) trained", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 451, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 462 + ], + "score": 1.0, + "content": "a detector on the inner layer of the classifier to detect adversarial examples. (Feinman et al., 2017)", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 460, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 474 + ], + "score": 1.0, + "content": "detected adversarial examples by looking at the Bayesian uncertainty estimates of the input images", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "in dropout neural networks and by performing density estimation in the subspace of deep features", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "learned by the model. MagNet (Meng & Chen, 2017) detected adversarial examples with large per-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "turbation using detector networks, and pushed adversarial examples with small perturbation towards", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 504, + 227, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 227, + 518 + ], + "score": 1.0, + "content": "the manifold of clean images.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 362, + 506, + 518 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 535, + 182, + 547 + ], + "lines": [ + { + "bbox": [ + 105, + 533, + 185, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 185, + 550 + ], + "score": 1.0, + "content": "3 APPROACH", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "title", + "bbox": [ + 108, + 561, + 383, + 573 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 384, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 384, + 574 + ], + "score": 1.0, + "content": "3.1 AN OVERVIEW OF GENERATING ADVERSARIAL EXAMPLES", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 661 + ], + "lines": [ + { + "bbox": [ + 105, + 581, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 505, + 596 + ], + "score": 1.0, + "content": "Before introducing the proposed adversarial defense method, we give an overview of generating", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 217, + 606 + ], + "score": 1.0, + "content": "adversarial examples. Let", + "type": "text" + }, + { + "bbox": [ + 218, + 594, + 232, + 605 + ], + "score": 0.9, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 594, + 281, + 606 + ], + "score": 1.0, + "content": "denote the", + "type": "text" + }, + { + "bbox": [ + 281, + 596, + 288, + 604 + ], + "score": 0.79, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 594, + 426, + 606 + ], + "score": 1.0, + "content": "-th image in a dataset containing", + "type": "text" + }, + { + "bbox": [ + 427, + 594, + 437, + 604 + ], + "score": 0.8, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "images, and let", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 601, + 508, + 622 + ], + "spans": [ + { + "bbox": [ + 107, + 605, + 124, + 617 + ], + "score": 0.91, + "content": "y _ { n } ^ { \\mathrm { t r u e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 601, + 341, + 622 + ], + "score": 1.0, + "content": "denote the corresponding ground-truth label. We use", + "type": "text" + }, + { + "bbox": [ + 341, + 605, + 348, + 615 + ], + "score": 0.78, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 601, + 508, + 622 + ], + "score": 1.0, + "content": "to denote the network parameters, and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 107, + 614, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 107, + 616, + 166, + 628 + ], + "score": 0.9, + "content": "L ( X _ { n } , y _ { n } ^ { \\mathrm { t r u e } } ; \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 614, + 505, + 630 + ], + "score": 1.0, + "content": "to denote the loss. For the adversarial example generation, the goal is to maximize", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 103, + 623, + 507, + 643 + ], + "spans": [ + { + "bbox": [ + 103, + 623, + 140, + 643 + ], + "score": 1.0, + "content": "the loss", + "type": "text" + }, + { + "bbox": [ + 140, + 627, + 221, + 639 + ], + "score": 0.88, + "content": "L ( X _ { n } + r _ { n } , y _ { n } ^ { \\mathrm { t r u e } } ; \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 623, + 284, + 643 + ], + "score": 1.0, + "content": "for each image", + "type": "text" + }, + { + "bbox": [ + 285, + 627, + 299, + 638 + ], + "score": 0.89, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 623, + 507, + 643 + ], + "score": 1.0, + "content": ", under the constraint that the generated adversarial", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 635, + 507, + 653 + ], + "spans": [ + { + "bbox": [ + 104, + 635, + 142, + 653 + ], + "score": 1.0, + "content": "example", + "type": "text" + }, + { + "bbox": [ + 142, + 639, + 211, + 650 + ], + "score": 0.9, + "content": "X _ { n } ^ { \\mathrm { a d v } } = X _ { n } + r _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 635, + 409, + 653 + ], + "score": 1.0, + "content": "should look visually similar to the original image", + "type": "text" + }, + { + "bbox": [ + 410, + 639, + 424, + 649 + ], + "score": 0.89, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 635, + 445, + 653 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + }, + { + "bbox": [ + 446, + 638, + 485, + 650 + ], + "score": 0.92, + "content": "| | r _ { n } | | \\leq \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 635, + 507, + 653 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 645, + 294, + 665 + ], + "spans": [ + { + "bbox": [ + 104, + 645, + 242, + 665 + ], + "score": 1.0, + "content": "the corresponding predicted label", + "type": "text" + }, + { + "bbox": [ + 242, + 649, + 289, + 661 + ], + "score": 0.92, + "content": "y _ { n } ^ { \\mathrm { a d v } } \\neq y _ { n } ^ { \\mathrm { t r u e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 645, + 294, + 665 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38, + "bbox_fs": [ + 103, + 581, + 508, + 665 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 666, + 504, + 699 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "In our experiment, we consider three different attack methods, including one single-step attack", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 675, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 505, + 689 + ], + "score": 1.0, + "content": "method and two iterative attack methods. We use the cleverhans library (Papernot et al., 2016a)", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 687, + 497, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 497, + 700 + ], + "score": 1.0, + "content": "to generate adversarial examples, where all these attacks have been implemented via TensorFlow.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 666, + 505, + 700 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 132, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 132, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "• Fast Gradient Sign Method (FGSM): FGSM (Goodfellow et al., 2015) is a single-step at-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 141, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 141, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "tack method. It finds the adversarial perturbation that yields the highest increase of the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 141, + 81, + 382, + 96 + ], + "spans": [ + { + "bbox": [ + 141, + 81, + 246, + 96 + ], + "score": 1.0, + "content": "linear cost function under", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 247, + 83, + 259, + 93 + ], + "score": 0.89, + "content": "l _ { \\infty }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 259, + 81, + 382, + 96 + ], + "score": 1.0, + "content": "-norm. The update equation is", + "type": "text", + "cross_page": true + } + ], + "index": 0 + } + ], + "index": 45.5, + "bbox_fs": [ + 132, + 708, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 142, + 82, + 381, + 93 + ], + "lines": [ + { + "bbox": [ + 141, + 81, + 382, + 96 + ], + "spans": [ + { + "bbox": [ + 141, + 81, + 246, + 96 + ], + "score": 1.0, + "content": "linear cost function under", + "type": "text" + }, + { + "bbox": [ + 247, + 83, + 259, + 93 + ], + "score": 0.89, + "content": "l _ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 81, + 382, + 96 + ], + "score": 1.0, + "content": "-norm. The update equation is", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 96, + 412, + 110 + ], + "lines": [ + { + "bbox": [ + 233, + 96, + 412, + 110 + ], + "spans": [ + { + "bbox": [ + 233, + 96, + 412, + 110 + ], + "score": 0.93, + "content": "X _ { n } ^ { \\mathrm { a d v } } = X _ { n } + \\epsilon \\cdot s i g n \\big ( \\nabla _ { X _ { n } } L ( X _ { n } , y _ { n } ^ { \\mathrm { t r u e } } ; \\theta ) \\big ) ,", + "type": "interline_equation", + "image_path": "256d0580ab719538c36dbc65ddaa18b6508c027919ce8cd831ad834e72a8a2b4.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 233, + 96, + 412, + 110 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 141, + 111, + 504, + 144 + ], + "lines": [ + { + "bbox": [ + 142, + 110, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 142, + 110, + 169, + 124 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 169, + 113, + 175, + 121 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 110, + 505, + 124 + ], + "score": 1.0, + "content": "controls the magnitude of adversarial perturbation. In the experiment, we choose", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 142, + 122, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 142, + 122, + 203, + 134 + ], + "score": 0.93, + "content": "\\epsilon = \\{ 2 , 5 , 1 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 122, + 505, + 135 + ], + "score": 1.0, + "content": ", which corresponds to small, medium and high magnitude of adversarial", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 141, + 133, + 252, + 146 + ], + "spans": [ + { + "bbox": [ + 141, + 133, + 252, + 146 + ], + "score": 1.0, + "content": "perturbations, respectively.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 134, + 146, + 505, + 191 + ], + "lines": [ + { + "bbox": [ + 133, + 146, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 133, + 146, + 505, + 159 + ], + "score": 1.0, + "content": "• DeepFool: DeepFool (Moosavi-Dezfooli et al., 2016) is an iterative attack method which", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 158, + 506, + 170 + ], + "spans": [ + { + "bbox": [ + 141, + 158, + 506, + 170 + ], + "score": 1.0, + "content": "finds the minimal perturbation to cross the decision boundary based on the linearization of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 168, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 141, + 168, + 289, + 182 + ], + "score": 1.0, + "content": "the classifier at each iteration. Any", + "type": "text" + }, + { + "bbox": [ + 290, + 169, + 298, + 181 + ], + "score": 0.88, + "content": "l _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 168, + 506, + 182 + ], + "score": 1.0, + "content": "-norm can be used with DeepFool, and we choose", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 142, + 179, + 283, + 194 + ], + "spans": [ + { + "bbox": [ + 142, + 180, + 151, + 191 + ], + "score": 0.85, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 179, + 283, + 194 + ], + "score": 1.0, + "content": "-norm for the study in this paper.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 132, + 194, + 504, + 226 + ], + "lines": [ + { + "bbox": [ + 131, + 193, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 131, + 193, + 173, + 207 + ], + "score": 1.0, + "content": "• Carlini", + "type": "text" + }, + { + "bbox": [ + 173, + 194, + 182, + 204 + ], + "score": 0.25, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 193, + 505, + 207 + ], + "score": 1.0, + "content": "Wagner (C&W): C&W (Carlini & Wagner, 2017) is a stronger iterative attack", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 204, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 141, + 204, + 402, + 218 + ], + "score": 1.0, + "content": "method proposed recently. It finds the adversarial perturbation", + "type": "text" + }, + { + "bbox": [ + 402, + 207, + 414, + 216 + ], + "score": 0.86, + "content": "r _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 204, + 505, + 218 + ], + "score": 1.0, + "content": "by using an auxiliary", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 215, + 203, + 229 + ], + "spans": [ + { + "bbox": [ + 141, + 215, + 176, + 229 + ], + "score": 1.0, + "content": "variable", + "type": "text" + }, + { + "bbox": [ + 177, + 217, + 189, + 227 + ], + "score": 0.86, + "content": "\\omega _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 215, + 203, + 229 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 260, + 225, + 387, + 248 + ], + "lines": [ + { + "bbox": [ + 260, + 225, + 387, + 248 + ], + "spans": [ + { + "bbox": [ + 260, + 225, + 387, + 248 + ], + "score": 0.95, + "content": "r _ { n } = { \\frac { 1 } { 2 } } ( t a n h ( \\omega _ { n } + 1 ) ) - X _ { n } .", + "type": "interline_equation", + "image_path": "6ac2e21a49ac9a59dcee26a8cc88dbc8ae952e60add70940c62a298ad305ebe3.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 260, + 225, + 387, + 248 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 143, + 249, + 375, + 261 + ], + "lines": [ + { + "bbox": [ + 141, + 246, + 374, + 264 + ], + "spans": [ + { + "bbox": [ + 141, + 246, + 362, + 264 + ], + "score": 1.0, + "content": "Then the loss function optimizes the auxiliary variable", + "type": "text" + }, + { + "bbox": [ + 362, + 251, + 374, + 260 + ], + "score": 0.85, + "content": "\\omega _ { n }", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 201, + 262, + 444, + 285 + ], + "lines": [ + { + "bbox": [ + 201, + 262, + 444, + 285 + ], + "spans": [ + { + "bbox": [ + 201, + 262, + 444, + 285 + ], + "score": 0.93, + "content": "\\operatorname* { m i n } _ { \\omega _ { n } } | | \\frac { 1 } { 2 } ( t a n h ( \\omega _ { n } ) + 1 ) - X _ { n } | | + c \\cdot f ( \\frac { 1 } { 2 } ( t a n h ( \\omega _ { n } ) + 1 ) ) .", + "type": "interline_equation", + "image_path": "9a4c56ddb6d5a120595154f56ce8c6cdd23b4680a466e4b911993433976f94cb.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 201, + 262, + 444, + 285 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 143, + 289, + 267, + 300 + ], + "lines": [ + { + "bbox": [ + 142, + 287, + 267, + 302 + ], + "spans": [ + { + "bbox": [ + 142, + 287, + 196, + 302 + ], + "score": 1.0, + "content": "The function", + "type": "text" + }, + { + "bbox": [ + 196, + 289, + 214, + 301 + ], + "score": 0.91, + "content": "f ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 287, + 267, + 302 + ], + "score": 1.0, + "content": "is defined as", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 302, + 433, + 315 + ], + "lines": [ + { + "bbox": [ + 212, + 302, + 433, + 315 + ], + "spans": [ + { + "bbox": [ + 212, + 302, + 433, + 315 + ], + "score": 0.86, + "content": "f ( x ) = \\operatorname* { m a x } ( Z ( x ) _ { y ^ { \\mathrm { t u e } } } - \\operatorname* { m a x } \\{ Z ( x ) _ { i } : i \\neq y ^ { \\mathrm { t r u e } } \\} , - k ) ,", + "type": "interline_equation", + "image_path": "471d0561206caf18e7dae8a27a418ea0baf79b3fc6adce078bd4d7d503c07252.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 212, + 302, + 433, + 315 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 144, + 317, + 504, + 350 + ], + "lines": [ + { + "bbox": [ + 142, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 142, + 316, + 169, + 329 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 169, + 316, + 195, + 329 + ], + "score": 0.92, + "content": "Z ( x ) _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 316, + 309, + 329 + ], + "score": 1.0, + "content": "is the logits output for class", + "type": "text" + }, + { + "bbox": [ + 309, + 318, + 313, + 327 + ], + "score": 0.75, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 316, + 334, + 329 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 335, + 317, + 342, + 326 + ], + "score": 0.83, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 316, + 505, + 329 + ], + "score": 1.0, + "content": "controls the confidence gap between the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 142, + 327, + 406, + 340 + ], + "score": 1.0, + "content": "adversarial class and true class. C&W can also work with various", + "type": "text" + }, + { + "bbox": [ + 407, + 328, + 415, + 340 + ], + "score": 0.87, + "content": "l _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "-norm, and we choose", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 142, + 339, + 255, + 351 + ], + "spans": [ + { + "bbox": [ + 142, + 339, + 151, + 349 + ], + "score": 0.86, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 339, + 255, + 351 + ], + "score": 1.0, + "content": "-norm in the experiments.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 107, + 363, + 341, + 374 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 341, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 341, + 375 + ], + "score": 1.0, + "content": "3.2 DEFENDING AGAINST ADVERSARIAL EXAMPLES", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 383, + 505, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 383, + 504, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 504, + 395 + ], + "score": 1.0, + "content": "The goal of defense is to build a network that is robust to adversarial examples, i.e., it can classify", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 394, + 504, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 504, + 406 + ], + "score": 1.0, + "content": "adversarial images correctly with little performance loss on non-adversarial (clean) images. Towards", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "score": 1.0, + "content": "this goal, we propose a randomization-based method, as shown in Figure 2, which adds a random", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "resizing layer and a random padding layer to the beginning of the classification networks. There is", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 428, + 489, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 489, + 440 + ], + "score": 1.0, + "content": "no re-training or fine-tuning needed which makes the proposed method very easy to implement.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 108, + 450, + 252, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 253, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 253, + 462 + ], + "score": 1.0, + "content": "3.2.1 RANDOMIZATION LAYERS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 469, + 505, + 547 + ], + "lines": [ + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 468, + 482 + ], + "score": 1.0, + "content": "The first randomization layer is a random resizing layer, which resizes the original image", + "type": "text" + }, + { + "bbox": [ + 469, + 470, + 483, + 480 + ], + "score": 0.87, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 469, + 505, + 482 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 479, + 504, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 141, + 493 + ], + "score": 1.0, + "content": "the size", + "type": "text" + }, + { + "bbox": [ + 141, + 480, + 194, + 491 + ], + "score": 0.91, + "content": "W \\times H \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 479, + 261, + 493 + ], + "score": 1.0, + "content": "to a new image", + "type": "text" + }, + { + "bbox": [ + 261, + 480, + 276, + 492 + ], + "score": 0.92, + "content": "X _ { n } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 479, + 351, + 493 + ], + "score": 1.0, + "content": "with random size", + "type": "text" + }, + { + "bbox": [ + 352, + 480, + 410, + 491 + ], + "score": 0.93, + "content": "W ^ { \\prime } \\times H ^ { \\prime } \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 479, + 460, + 493 + ], + "score": 1.0, + "content": ". Note that,", + "type": "text" + }, + { + "bbox": [ + 460, + 480, + 504, + 492 + ], + "score": 0.91, + "content": "| W ^ { \\prime } - W |", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 490, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 124, + 504 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 491, + 164, + 503 + ], + "score": 0.92, + "content": "| H ^ { \\prime } - H |", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 490, + 505, + 504 + ], + "score": 1.0, + "content": "should be within a reasonablely small range, otherwise the network performance on", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "score": 1.0, + "content": "non-adversarial images would significantly drop. Taking Inception-ResNet network (Szegedy et al.,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 512, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 315, + 526 + ], + "score": 1.0, + "content": "2017) as an example, the original data input size is", + "type": "text" + }, + { + "bbox": [ + 315, + 513, + 376, + 524 + ], + "score": 0.92, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 512, + 506, + 526 + ], + "score": 1.0, + "content": ". Empirically we found that the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 466, + 537 + ], + "score": 1.0, + "content": "network performance hardly drops if we control the height and width of the resized image", + "type": "text" + }, + { + "bbox": [ + 467, + 524, + 482, + 536 + ], + "score": 0.93, + "content": "X _ { n } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "to be", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 535, + 220, + 548 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 220, + 548 + ], + "score": 1.0, + "content": "within the range [299, 331).", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 551, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "The second randomization layer is the random padding layer, which pads zeros around the resized", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 559, + 504, + 579 + ], + "spans": [ + { + "bbox": [ + 104, + 559, + 398, + 579 + ], + "score": 1.0, + "content": "image in a random manner. Specifically, by padding the resized image", + "type": "text" + }, + { + "bbox": [ + 398, + 563, + 413, + 574 + ], + "score": 0.9, + "content": "X _ { n } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 559, + 488, + 579 + ], + "score": 1.0, + "content": "into a new image", + "type": "text" + }, + { + "bbox": [ + 488, + 563, + 504, + 575 + ], + "score": 0.89, + "content": "X _ { n } ^ { \\prime \\prime }", + "type": "inline_equation" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 572, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 160, + 587 + ], + "score": 1.0, + "content": "with the size", + "type": "text" + }, + { + "bbox": [ + 160, + 574, + 222, + 585 + ], + "score": 0.92, + "content": "W ^ { \\prime \\prime } \\times H ^ { \\prime \\prime } \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 572, + 316, + 587 + ], + "score": 1.0, + "content": ", we can choose to pad", + "type": "text" + }, + { + "bbox": [ + 317, + 576, + 325, + 584 + ], + "score": 0.67, + "content": "w", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 572, + 420, + 587 + ], + "score": 1.0, + "content": "zero pixels on the left,", + "type": "text" + }, + { + "bbox": [ + 420, + 574, + 484, + 585 + ], + "score": 0.91, + "content": "W ^ { \\prime \\prime } - W ^ { \\prime } - w", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 572, + 506, + 587 + ], + "score": 1.0, + "content": "zero", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 584, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 182, + 597 + ], + "score": 1.0, + "content": "pixels on the right,", + "type": "text" + }, + { + "bbox": [ + 183, + 586, + 190, + 595 + ], + "score": 0.76, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 584, + 295, + 597 + ], + "score": 1.0, + "content": "zero pixels on the top and", + "type": "text" + }, + { + "bbox": [ + 295, + 585, + 350, + 595 + ], + "score": 0.91, + "content": "H ^ { \\prime \\prime } - H ^ { \\prime } - h", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 584, + 505, + 597 + ], + "score": 1.0, + "content": "zero pixels on the bottom. This results", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 595, + 475, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 187, + 609 + ], + "score": 1.0, + "content": "in a total number of", + "type": "text" + }, + { + "bbox": [ + 188, + 596, + 331, + 608 + ], + "score": 0.92, + "content": "\\left( W ^ { \\prime \\prime } - W ^ { \\prime } + 1 \\right) \\times \\left( H ^ { \\prime \\prime } - H ^ { \\prime } + 1 \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 595, + 475, + 609 + ], + "score": 1.0, + "content": "different possible padding patterns.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 613, + 505, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 505, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 505, + 626 + ], + "score": 1.0, + "content": "During implementation, the original image first goes through two randomization layers, and then", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "we pass the transformed image to the original CNN for classification. The pipeline is illustrated in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 634, + 146, + 648 + ], + "spans": [ + { + "bbox": [ + 104, + 634, + 146, + 648 + ], + "score": 1.0, + "content": "Figure 2.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40 + }, + { + "type": "title", + "bbox": [ + 107, + 657, + 372, + 669 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 373, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 252, + 670 + ], + "score": 1.0, + "content": "3.2.2 RANDOMIZATION LAYERS", + "type": "text" + }, + { + "bbox": [ + 252, + 659, + 261, + 667 + ], + "score": 0.53, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 657, + 373, + 670 + ], + "score": 1.0, + "content": "ADVERSARIAL TRAINING", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Note that our randomization-based method is good at defending against iterative attacks, and adver-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "sarial training (Kurakin et al., 2017; Tramer et al., 2017) can effectively increase the robustness of `", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "neural networks to single-step attacks. Thus, to make the best of both worlds, we can combine the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "proposed randomization layers and an adversarially trained model (Tramer et al., 2017) together to `", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 721, + 316, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 316, + 732 + ], + "score": 1.0, + "content": "defend against both single-step and iterative attacks.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 142, + 82, + 381, + 93 + ], + "lines": [], + "index": 0, + "bbox_fs": [ + 141, + 81, + 382, + 96 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 96, + 412, + 110 + ], + "lines": [ + { + "bbox": [ + 233, + 96, + 412, + 110 + ], + "spans": [ + { + "bbox": [ + 233, + 96, + 412, + 110 + ], + "score": 0.93, + "content": "X _ { n } ^ { \\mathrm { a d v } } = X _ { n } + \\epsilon \\cdot s i g n \\big ( \\nabla _ { X _ { n } } L ( X _ { n } , y _ { n } ^ { \\mathrm { t r u e } } ; \\theta ) \\big ) ,", + "type": "interline_equation", + "image_path": "256d0580ab719538c36dbc65ddaa18b6508c027919ce8cd831ad834e72a8a2b4.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 233, + 96, + 412, + 110 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 141, + 111, + 504, + 144 + ], + "lines": [ + { + "bbox": [ + 142, + 110, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 142, + 110, + 169, + 124 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 169, + 113, + 175, + 121 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 110, + 505, + 124 + ], + "score": 1.0, + "content": "controls the magnitude of adversarial perturbation. In the experiment, we choose", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 142, + 122, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 142, + 122, + 203, + 134 + ], + "score": 0.93, + "content": "\\epsilon = \\{ 2 , 5 , 1 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 122, + 505, + 135 + ], + "score": 1.0, + "content": ", which corresponds to small, medium and high magnitude of adversarial", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 141, + 133, + 252, + 146 + ], + "spans": [ + { + "bbox": [ + 141, + 133, + 252, + 146 + ], + "score": 1.0, + "content": "perturbations, respectively.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 141, + 110, + 505, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 134, + 146, + 505, + 191 + ], + "lines": [ + { + "bbox": [ + 133, + 146, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 133, + 146, + 505, + 159 + ], + "score": 1.0, + "content": "• DeepFool: DeepFool (Moosavi-Dezfooli et al., 2016) is an iterative attack method which", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 158, + 506, + 170 + ], + "spans": [ + { + "bbox": [ + 141, + 158, + 506, + 170 + ], + "score": 1.0, + "content": "finds the minimal perturbation to cross the decision boundary based on the linearization of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 168, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 141, + 168, + 289, + 182 + ], + "score": 1.0, + "content": "the classifier at each iteration. Any", + "type": "text" + }, + { + "bbox": [ + 290, + 169, + 298, + 181 + ], + "score": 0.88, + "content": "l _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 168, + 506, + 182 + ], + "score": 1.0, + "content": "-norm can be used with DeepFool, and we choose", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 142, + 179, + 283, + 194 + ], + "spans": [ + { + "bbox": [ + 142, + 180, + 151, + 191 + ], + "score": 0.85, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 179, + 283, + 194 + ], + "score": 1.0, + "content": "-norm for the study in this paper.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 133, + 146, + 506, + 194 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 194, + 504, + 226 + ], + "lines": [ + { + "bbox": [ + 131, + 193, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 131, + 193, + 173, + 207 + ], + "score": 1.0, + "content": "• Carlini", + "type": "text" + }, + { + "bbox": [ + 173, + 194, + 182, + 204 + ], + "score": 0.25, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 193, + 505, + 207 + ], + "score": 1.0, + "content": "Wagner (C&W): C&W (Carlini & Wagner, 2017) is a stronger iterative attack", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 204, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 141, + 204, + 402, + 218 + ], + "score": 1.0, + "content": "method proposed recently. It finds the adversarial perturbation", + "type": "text" + }, + { + "bbox": [ + 402, + 207, + 414, + 216 + ], + "score": 0.86, + "content": "r _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 204, + 505, + 218 + ], + "score": 1.0, + "content": "by using an auxiliary", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 215, + 203, + 229 + ], + "spans": [ + { + "bbox": [ + 141, + 215, + 176, + 229 + ], + "score": 1.0, + "content": "variable", + "type": "text" + }, + { + "bbox": [ + 177, + 217, + 189, + 227 + ], + "score": 0.86, + "content": "\\omega _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 215, + 203, + 229 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 131, + 193, + 505, + 229 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 260, + 225, + 387, + 248 + ], + "lines": [ + { + "bbox": [ + 260, + 225, + 387, + 248 + ], + "spans": [ + { + "bbox": [ + 260, + 225, + 387, + 248 + ], + "score": 0.95, + "content": "r _ { n } = { \\frac { 1 } { 2 } } ( t a n h ( \\omega _ { n } + 1 ) ) - X _ { n } .", + "type": "interline_equation", + "image_path": "6ac2e21a49ac9a59dcee26a8cc88dbc8ae952e60add70940c62a298ad305ebe3.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 260, + 225, + 387, + 248 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 143, + 249, + 375, + 261 + ], + "lines": [ + { + "bbox": [ + 141, + 246, + 374, + 264 + ], + "spans": [ + { + "bbox": [ + 141, + 246, + 362, + 264 + ], + "score": 1.0, + "content": "Then the loss function optimizes the auxiliary variable", + "type": "text" + }, + { + "bbox": [ + 362, + 251, + 374, + 260 + ], + "score": 0.85, + "content": "\\omega _ { n }", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 141, + 246, + 374, + 264 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 201, + 262, + 444, + 285 + ], + "lines": [ + { + "bbox": [ + 201, + 262, + 444, + 285 + ], + "spans": [ + { + "bbox": [ + 201, + 262, + 444, + 285 + ], + "score": 0.93, + "content": "\\operatorname* { m i n } _ { \\omega _ { n } } | | \\frac { 1 } { 2 } ( t a n h ( \\omega _ { n } ) + 1 ) - X _ { n } | | + c \\cdot f ( \\frac { 1 } { 2 } ( t a n h ( \\omega _ { n } ) + 1 ) ) .", + "type": "interline_equation", + "image_path": "9a4c56ddb6d5a120595154f56ce8c6cdd23b4680a466e4b911993433976f94cb.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 201, + 262, + 444, + 285 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 143, + 289, + 267, + 300 + ], + "lines": [ + { + "bbox": [ + 142, + 287, + 267, + 302 + ], + "spans": [ + { + "bbox": [ + 142, + 287, + 196, + 302 + ], + "score": 1.0, + "content": "The function", + "type": "text" + }, + { + "bbox": [ + 196, + 289, + 214, + 301 + ], + "score": 0.91, + "content": "f ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 287, + 267, + 302 + ], + "score": 1.0, + "content": "is defined as", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 142, + 287, + 267, + 302 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 302, + 433, + 315 + ], + "lines": [ + { + "bbox": [ + 212, + 302, + 433, + 315 + ], + "spans": [ + { + "bbox": [ + 212, + 302, + 433, + 315 + ], + "score": 0.86, + "content": "f ( x ) = \\operatorname* { m a x } ( Z ( x ) _ { y ^ { \\mathrm { t u e } } } - \\operatorname* { m a x } \\{ Z ( x ) _ { i } : i \\neq y ^ { \\mathrm { t r u e } } \\} , - k ) ,", + "type": "interline_equation", + "image_path": "471d0561206caf18e7dae8a27a418ea0baf79b3fc6adce078bd4d7d503c07252.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 212, + 302, + 433, + 315 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 144, + 317, + 504, + 350 + ], + "lines": [ + { + "bbox": [ + 142, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 142, + 316, + 169, + 329 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 169, + 316, + 195, + 329 + ], + "score": 0.92, + "content": "Z ( x ) _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 316, + 309, + 329 + ], + "score": 1.0, + "content": "is the logits output for class", + "type": "text" + }, + { + "bbox": [ + 309, + 318, + 313, + 327 + ], + "score": 0.75, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 316, + 334, + 329 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 335, + 317, + 342, + 326 + ], + "score": 0.83, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 316, + 505, + 329 + ], + "score": 1.0, + "content": "controls the confidence gap between the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 142, + 327, + 406, + 340 + ], + "score": 1.0, + "content": "adversarial class and true class. C&W can also work with various", + "type": "text" + }, + { + "bbox": [ + 407, + 328, + 415, + 340 + ], + "score": 0.87, + "content": "l _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "-norm, and we choose", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 142, + 339, + 255, + 351 + ], + "spans": [ + { + "bbox": [ + 142, + 339, + 151, + 349 + ], + "score": 0.86, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 339, + 255, + 351 + ], + "score": 1.0, + "content": "-norm in the experiments.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 142, + 316, + 505, + 351 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 363, + 341, + 374 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 341, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 341, + 375 + ], + "score": 1.0, + "content": "3.2 DEFENDING AGAINST ADVERSARIAL EXAMPLES", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 383, + 505, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 383, + 504, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 504, + 395 + ], + "score": 1.0, + "content": "The goal of defense is to build a network that is robust to adversarial examples, i.e., it can classify", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 394, + 504, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 504, + 406 + ], + "score": 1.0, + "content": "adversarial images correctly with little performance loss on non-adversarial (clean) images. Towards", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "score": 1.0, + "content": "this goal, we propose a randomization-based method, as shown in Figure 2, which adds a random", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "resizing layer and a random padding layer to the beginning of the classification networks. There is", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 428, + 489, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 489, + 440 + ], + "score": 1.0, + "content": "no re-training or fine-tuning needed which makes the proposed method very easy to implement.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 383, + 505, + 440 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 450, + 252, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 253, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 253, + 462 + ], + "score": 1.0, + "content": "3.2.1 RANDOMIZATION LAYERS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 469, + 505, + 547 + ], + "lines": [ + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 468, + 482 + ], + "score": 1.0, + "content": "The first randomization layer is a random resizing layer, which resizes the original image", + "type": "text" + }, + { + "bbox": [ + 469, + 470, + 483, + 480 + ], + "score": 0.87, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 469, + 505, + 482 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 479, + 504, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 141, + 493 + ], + "score": 1.0, + "content": "the size", + "type": "text" + }, + { + "bbox": [ + 141, + 480, + 194, + 491 + ], + "score": 0.91, + "content": "W \\times H \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 479, + 261, + 493 + ], + "score": 1.0, + "content": "to a new image", + "type": "text" + }, + { + "bbox": [ + 261, + 480, + 276, + 492 + ], + "score": 0.92, + "content": "X _ { n } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 479, + 351, + 493 + ], + "score": 1.0, + "content": "with random size", + "type": "text" + }, + { + "bbox": [ + 352, + 480, + 410, + 491 + ], + "score": 0.93, + "content": "W ^ { \\prime } \\times H ^ { \\prime } \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 479, + 460, + 493 + ], + "score": 1.0, + "content": ". Note that,", + "type": "text" + }, + { + "bbox": [ + 460, + 480, + 504, + 492 + ], + "score": 0.91, + "content": "| W ^ { \\prime } - W |", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 490, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 124, + 504 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 491, + 164, + 503 + ], + "score": 0.92, + "content": "| H ^ { \\prime } - H |", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 490, + 505, + 504 + ], + "score": 1.0, + "content": "should be within a reasonablely small range, otherwise the network performance on", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "score": 1.0, + "content": "non-adversarial images would significantly drop. Taking Inception-ResNet network (Szegedy et al.,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 512, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 315, + 526 + ], + "score": 1.0, + "content": "2017) as an example, the original data input size is", + "type": "text" + }, + { + "bbox": [ + 315, + 513, + 376, + 524 + ], + "score": 0.92, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 512, + 506, + 526 + ], + "score": 1.0, + "content": ". Empirically we found that the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 466, + 537 + ], + "score": 1.0, + "content": "network performance hardly drops if we control the height and width of the resized image", + "type": "text" + }, + { + "bbox": [ + 467, + 524, + 482, + 536 + ], + "score": 0.93, + "content": "X _ { n } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "to be", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 535, + 220, + 548 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 220, + 548 + ], + "score": 1.0, + "content": "within the range [299, 331).", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 469, + 506, + 548 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 551, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "The second randomization layer is the random padding layer, which pads zeros around the resized", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 559, + 504, + 579 + ], + "spans": [ + { + "bbox": [ + 104, + 559, + 398, + 579 + ], + "score": 1.0, + "content": "image in a random manner. Specifically, by padding the resized image", + "type": "text" + }, + { + "bbox": [ + 398, + 563, + 413, + 574 + ], + "score": 0.9, + "content": "X _ { n } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 559, + 488, + 579 + ], + "score": 1.0, + "content": "into a new image", + "type": "text" + }, + { + "bbox": [ + 488, + 563, + 504, + 575 + ], + "score": 0.89, + "content": "X _ { n } ^ { \\prime \\prime }", + "type": "inline_equation" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 572, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 160, + 587 + ], + "score": 1.0, + "content": "with the size", + "type": "text" + }, + { + "bbox": [ + 160, + 574, + 222, + 585 + ], + "score": 0.92, + "content": "W ^ { \\prime \\prime } \\times H ^ { \\prime \\prime } \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 572, + 316, + 587 + ], + "score": 1.0, + "content": ", we can choose to pad", + "type": "text" + }, + { + "bbox": [ + 317, + 576, + 325, + 584 + ], + "score": 0.67, + "content": "w", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 572, + 420, + 587 + ], + "score": 1.0, + "content": "zero pixels on the left,", + "type": "text" + }, + { + "bbox": [ + 420, + 574, + 484, + 585 + ], + "score": 0.91, + "content": "W ^ { \\prime \\prime } - W ^ { \\prime } - w", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 572, + 506, + 587 + ], + "score": 1.0, + "content": "zero", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 584, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 182, + 597 + ], + "score": 1.0, + "content": "pixels on the right,", + "type": "text" + }, + { + "bbox": [ + 183, + 586, + 190, + 595 + ], + "score": 0.76, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 584, + 295, + 597 + ], + "score": 1.0, + "content": "zero pixels on the top and", + "type": "text" + }, + { + "bbox": [ + 295, + 585, + 350, + 595 + ], + "score": 0.91, + "content": "H ^ { \\prime \\prime } - H ^ { \\prime } - h", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 584, + 505, + 597 + ], + "score": 1.0, + "content": "zero pixels on the bottom. This results", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 595, + 475, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 187, + 609 + ], + "score": 1.0, + "content": "in a total number of", + "type": "text" + }, + { + "bbox": [ + 188, + 596, + 331, + 608 + ], + "score": 0.92, + "content": "\\left( W ^ { \\prime \\prime } - W ^ { \\prime } + 1 \\right) \\times \\left( H ^ { \\prime \\prime } - H ^ { \\prime } + 1 \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 595, + 475, + 609 + ], + "score": 1.0, + "content": "different possible padding patterns.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36, + "bbox_fs": [ + 104, + 552, + 506, + 609 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 613, + 505, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 505, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 505, + 626 + ], + "score": 1.0, + "content": "During implementation, the original image first goes through two randomization layers, and then", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "we pass the transformed image to the original CNN for classification. The pipeline is illustrated in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 634, + 146, + 648 + ], + "spans": [ + { + "bbox": [ + 104, + 634, + 146, + 648 + ], + "score": 1.0, + "content": "Figure 2.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40, + "bbox_fs": [ + 104, + 612, + 505, + 648 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 657, + 372, + 669 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 373, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 252, + 670 + ], + "score": 1.0, + "content": "3.2.2 RANDOMIZATION LAYERS", + "type": "text" + }, + { + "bbox": [ + 252, + 659, + 261, + 667 + ], + "score": 0.53, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 657, + 373, + 670 + ], + "score": 1.0, + "content": "ADVERSARIAL TRAINING", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Note that our randomization-based method is good at defending against iterative attacks, and adver-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "sarial training (Kurakin et al., 2017; Tramer et al., 2017) can effectively increase the robustness of `", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "neural networks to single-step attacks. Thus, to make the best of both worlds, we can combine the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "proposed randomization layers and an adversarially trained model (Tramer et al., 2017) together to `", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 721, + 316, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 316, + 732 + ], + "score": 1.0, + "content": "defend against both single-step and iterative attacks.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 677, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 84, + 506, + 268 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 84, + 506, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 84, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 112, + 84, + 506, + 268 + ], + "score": 0.972, + "type": "image", + "image_path": "b391dc2283db9234f0f293be29409bb96e04b1cf038128ded1f2d00537f1f162.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 84, + 506, + 145.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 145.33333333333334, + 506, + 206.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 206.66666666666669, + 506, + 268.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 288, + 506, + 333 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 288, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 470, + 301 + ], + "score": 1.0, + "content": "Figure 2: The pipeline of our randomization-based defense mechanism. The input image", + "type": "text" + }, + { + "bbox": [ + 470, + 289, + 485, + 300 + ], + "score": 0.89, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 288, + 506, + 301 + ], + "score": 1.0, + "content": "first", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 299, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 104, + 299, + 506, + 313 + ], + "score": 1.0, + "content": "goes through the random resizing layer with a random scale applied. Then the random padding", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 309, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 223, + 324 + ], + "score": 1.0, + "content": "layer pads the resized image", + "type": "text" + }, + { + "bbox": [ + 224, + 311, + 239, + 323 + ], + "score": 0.9, + "content": "X _ { n } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 309, + 443, + 324 + ], + "score": 1.0, + "content": "in a random manner. The resulting padded image", + "type": "text" + }, + { + "bbox": [ + 443, + 311, + 459, + 323 + ], + "score": 0.91, + "content": "X _ { n } ^ { \\prime \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 309, + 506, + 324 + ], + "score": 1.0, + "content": "is used for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 321, + 163, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 163, + 334 + ], + "score": 1.0, + "content": "classification.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "title", + "bbox": [ + 108, + 344, + 200, + 356 + ], + "lines": [ + { + "bbox": [ + 105, + 343, + 202, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 202, + 358 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 108, + 370, + 218, + 381 + ], + "lines": [ + { + "bbox": [ + 106, + 370, + 219, + 382 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 219, + 382 + ], + "score": 1.0, + "content": "4.1 EXPERIMENT SETUP", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 505, + 425 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 405 + ], + "score": 1.0, + "content": "Dataset: It is less meaningful to attack the images that are already classified wrongly. Therefore,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "we randomly choose 5000 images from the ImageNet validation set that are classified correctly by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 414, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 442, + 426 + ], + "score": 1.0, + "content": "all the considered networks to form our test dataset. All these images are of the size", + "type": "text" + }, + { + "bbox": [ + 442, + 414, + 501, + 424 + ], + "score": 0.91, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 414, + 505, + 426 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 430, + 505, + 486 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 430, + 445 + ], + "score": 1.0, + "content": "Networks: We test with four publicly available networks1,2, including Inception-", + "type": "text" + }, + { + "bbox": [ + 430, + 431, + 441, + 441 + ], + "score": 0.56, + "content": "\\nu 3", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 428, + 506, + 445 + ], + "score": 1.0, + "content": "(Szegedy et al.,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 441, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 166, + 454 + ], + "score": 1.0, + "content": "2016), ResNet-", + "type": "text" + }, + { + "bbox": [ + 167, + 442, + 178, + 452 + ], + "score": 0.45, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 441, + 383, + 454 + ], + "score": 1.0, + "content": "(He et al., 2016b) of 101 layers, Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 383, + 442, + 394, + 452 + ], + "score": 0.54, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 441, + 505, + 454 + ], + "score": 1.0, + "content": "(Szegedy et al., 2017), and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 452, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 213, + 465 + ], + "score": 1.0, + "content": "ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 214, + 453, + 224, + 462 + ], + "score": 0.42, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 452, + 505, + 465 + ], + "score": 1.0, + "content": "which applies the ensemble adversarial training (Tramer et al., 2017) `", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 463, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 192, + 476 + ], + "score": 1.0, + "content": "on Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 192, + 464, + 203, + 474 + ], + "score": 0.56, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 463, + 505, + 476 + ], + "score": 1.0, + "content": ". These networks have been trained on ImageNet, and we do not perform", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 474, + 386, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 386, + 487 + ], + "score": 1.0, + "content": "any re-training or fine-tuning on them for the whole experiments.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 505, + 503 + ], + "score": 1.0, + "content": "Defense Models: The defense models consist of the original networks (i.e., four above-mentioned", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 503, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 503, + 505, + 514 + ], + "score": 1.0, + "content": "networks) and two additional randomization layers. For the random resizing layer, it changes the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 180, + 525 + ], + "score": 1.0, + "content": "input shape from", + "type": "text" + }, + { + "bbox": [ + 180, + 513, + 245, + 524 + ], + "score": 0.93, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 513, + 258, + 525 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 259, + 513, + 325, + 524 + ], + "score": 0.92, + "content": "r n d \\times r n d \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 513, + 505, + 525 + ], + "score": 1.0, + "content": ", where rnd is a integer randomly sampled", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "score": 1.0, + "content": "from the range [299, 331). For the random padding layer, it pads the resized image to the shape of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 534, + 505, + 548 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 169, + 546 + ], + "score": 0.9, + "content": "3 3 1 \\times 3 3 1 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 534, + 505, + 548 + ], + "score": 1.0, + "content": "in a random manner. By applying these two randomization layers, we can create", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 545, + 132, + 556 + ], + "spans": [ + { + "bbox": [ + 115, + 545, + 132, + 556 + ], + "score": 1.0, + "content": "330", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 551, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 254, + 572 + ], + "score": 0.75, + "content": "\\sum _ { r n d = 2 9 9 } ( 3 3 1 - r n d + 1 ) ^ { 2 } = 1 2 5 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 552, + 505, + 566 + ], + "score": 1.0, + "content": "different patterns for a single image. Since there exists small", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 572, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 583 + ], + "score": 1.0, + "content": "variance on model performance w.r.t. different random patterns, we run the defense model three", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 581, + 320, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 320, + 595 + ], + "score": 1.0, + "content": "times independently and report the average accuracy.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 505, + 665 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 505, + 610 + ], + "score": 1.0, + "content": "Target Models under Different Attack Scenarios: The strongest attack would be that the attackers", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 610, + 504, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 504, + 622 + ], + "score": 1.0, + "content": "consider ALL possible patterns of the defense models when generating the adversarial examples.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "score": 1.0, + "content": "However, this is computationally impossible, because failing a large number of patterns (e.g., 12528", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "here) at the same time takes extremely long time, and may not even converge. Thus, we let attackers", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 643, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 643, + 505, + 655 + ], + "score": 1.0, + "content": "use the target models to generate adversarial examples instead, and consider the following three", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 653, + 211, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 653, + 211, + 666 + ], + "score": 1.0, + "content": "different attack scenarios.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 127, + 675, + 504, + 698 + ], + "lines": [ + { + "bbox": [ + 133, + 675, + 505, + 687 + ], + "spans": [ + { + "bbox": [ + 133, + 675, + 505, + 687 + ], + "score": 1.0, + "content": "• Vanilla Attack: The attackers do not know the existence of the randomization layers and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 141, + 686, + 324, + 699 + ], + "spans": [ + { + "bbox": [ + 141, + 686, + 324, + 699 + ], + "score": 1.0, + "content": "the target model is just the original network.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 710, + 435, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 709, + 352, + 722 + ], + "spans": [ + { + "bbox": [ + 119, + 709, + 352, + 722 + ], + "score": 1.0, + "content": "1https://github.com/tensorflow/models/tree/master/research/slim", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 720, + 435, + 733 + ], + "spans": [ + { + "bbox": [ + 118, + 720, + 435, + 733 + ], + "score": 1.0, + "content": "2http://download.tensorflow.org/models/ens adv inception resnet v2 2017 08 18.tar.gz", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 84, + 506, + 268 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 84, + 506, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 84, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 112, + 84, + 506, + 268 + ], + "score": 0.972, + "type": "image", + "image_path": "b391dc2283db9234f0f293be29409bb96e04b1cf038128ded1f2d00537f1f162.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 84, + 506, + 145.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 145.33333333333334, + 506, + 206.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 206.66666666666669, + 506, + 268.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 288, + 506, + 333 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 288, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 470, + 301 + ], + "score": 1.0, + "content": "Figure 2: The pipeline of our randomization-based defense mechanism. The input image", + "type": "text" + }, + { + "bbox": [ + 470, + 289, + 485, + 300 + ], + "score": 0.89, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 288, + 506, + 301 + ], + "score": 1.0, + "content": "first", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 299, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 104, + 299, + 506, + 313 + ], + "score": 1.0, + "content": "goes through the random resizing layer with a random scale applied. Then the random padding", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 309, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 223, + 324 + ], + "score": 1.0, + "content": "layer pads the resized image", + "type": "text" + }, + { + "bbox": [ + 224, + 311, + 239, + 323 + ], + "score": 0.9, + "content": "X _ { n } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 309, + 443, + 324 + ], + "score": 1.0, + "content": "in a random manner. The resulting padded image", + "type": "text" + }, + { + "bbox": [ + 443, + 311, + 459, + 323 + ], + "score": 0.91, + "content": "X _ { n } ^ { \\prime \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 309, + 506, + 324 + ], + "score": 1.0, + "content": "is used for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 321, + 163, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 163, + 334 + ], + "score": 1.0, + "content": "classification.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "title", + "bbox": [ + 108, + 344, + 200, + 356 + ], + "lines": [ + { + "bbox": [ + 105, + 343, + 202, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 202, + 358 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 108, + 370, + 218, + 381 + ], + "lines": [ + { + "bbox": [ + 106, + 370, + 219, + 382 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 219, + 382 + ], + "score": 1.0, + "content": "4.1 EXPERIMENT SETUP", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 505, + 425 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 405 + ], + "score": 1.0, + "content": "Dataset: It is less meaningful to attack the images that are already classified wrongly. Therefore,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "we randomly choose 5000 images from the ImageNet validation set that are classified correctly by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 414, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 442, + 426 + ], + "score": 1.0, + "content": "all the considered networks to form our test dataset. All these images are of the size", + "type": "text" + }, + { + "bbox": [ + 442, + 414, + 501, + 424 + ], + "score": 0.91, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 414, + 505, + 426 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 390, + 505, + 426 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 430, + 505, + 486 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 430, + 445 + ], + "score": 1.0, + "content": "Networks: We test with four publicly available networks1,2, including Inception-", + "type": "text" + }, + { + "bbox": [ + 430, + 431, + 441, + 441 + ], + "score": 0.56, + "content": "\\nu 3", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 428, + 506, + 445 + ], + "score": 1.0, + "content": "(Szegedy et al.,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 441, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 166, + 454 + ], + "score": 1.0, + "content": "2016), ResNet-", + "type": "text" + }, + { + "bbox": [ + 167, + 442, + 178, + 452 + ], + "score": 0.45, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 441, + 383, + 454 + ], + "score": 1.0, + "content": "(He et al., 2016b) of 101 layers, Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 383, + 442, + 394, + 452 + ], + "score": 0.54, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 441, + 505, + 454 + ], + "score": 1.0, + "content": "(Szegedy et al., 2017), and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 452, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 213, + 465 + ], + "score": 1.0, + "content": "ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 214, + 453, + 224, + 462 + ], + "score": 0.42, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 452, + 505, + 465 + ], + "score": 1.0, + "content": "which applies the ensemble adversarial training (Tramer et al., 2017) `", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 463, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 192, + 476 + ], + "score": 1.0, + "content": "on Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 192, + 464, + 203, + 474 + ], + "score": 0.56, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 463, + 505, + 476 + ], + "score": 1.0, + "content": ". These networks have been trained on ImageNet, and we do not perform", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 474, + 386, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 386, + 487 + ], + "score": 1.0, + "content": "any re-training or fine-tuning on them for the whole experiments.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 428, + 506, + 487 + ] + }, + { + "type": "list", + "bbox": [ + 106, + 491, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 505, + 503 + ], + "score": 1.0, + "content": "Defense Models: The defense models consist of the original networks (i.e., four above-mentioned", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 503, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 503, + 505, + 514 + ], + "score": 1.0, + "content": "networks) and two additional randomization layers. For the random resizing layer, it changes the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 180, + 525 + ], + "score": 1.0, + "content": "input shape from", + "type": "text" + }, + { + "bbox": [ + 180, + 513, + 245, + 524 + ], + "score": 0.93, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 513, + 258, + 525 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 259, + 513, + 325, + 524 + ], + "score": 0.92, + "content": "r n d \\times r n d \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 513, + 505, + 525 + ], + "score": 1.0, + "content": ", where rnd is a integer randomly sampled", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "score": 1.0, + "content": "from the range [299, 331). For the random padding layer, it pads the resized image to the shape of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 534, + 505, + 548 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 169, + 546 + ], + "score": 0.9, + "content": "3 3 1 \\times 3 3 1 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 534, + 505, + 548 + ], + "score": 1.0, + "content": "in a random manner. By applying these two randomization layers, we can create", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 545, + 132, + 556 + ], + "spans": [ + { + "bbox": [ + 115, + 545, + 132, + 556 + ], + "score": 1.0, + "content": "330", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 551, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 254, + 572 + ], + "score": 0.75, + "content": "\\sum _ { r n d = 2 9 9 } ( 3 3 1 - r n d + 1 ) ^ { 2 } = 1 2 5 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 552, + 505, + 566 + ], + "score": 1.0, + "content": "different patterns for a single image. Since there exists small", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 572, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 583 + ], + "score": 1.0, + "content": "variance on model performance w.r.t. different random patterns, we run the defense model three", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 581, + 320, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 320, + 595 + ], + "score": 1.0, + "content": "times independently and report the average accuracy.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + } + ], + "index": 21, + "bbox_fs": [ + 105, + 491, + 506, + 595 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 505, + 665 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 505, + 610 + ], + "score": 1.0, + "content": "Target Models under Different Attack Scenarios: The strongest attack would be that the attackers", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 610, + 504, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 504, + 622 + ], + "score": 1.0, + "content": "consider ALL possible patterns of the defense models when generating the adversarial examples.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "score": 1.0, + "content": "However, this is computationally impossible, because failing a large number of patterns (e.g., 12528", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "here) at the same time takes extremely long time, and may not even converge. Thus, we let attackers", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 643, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 643, + 505, + 655 + ], + "score": 1.0, + "content": "use the target models to generate adversarial examples instead, and consider the following three", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 653, + 211, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 653, + 211, + 666 + ], + "score": 1.0, + "content": "different attack scenarios.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 599, + 505, + 666 + ] + }, + { + "type": "text", + "bbox": [ + 127, + 675, + 504, + 698 + ], + "lines": [ + { + "bbox": [ + 133, + 675, + 505, + 687 + ], + "spans": [ + { + "bbox": [ + 133, + 675, + 505, + 687 + ], + "score": 1.0, + "content": "• Vanilla Attack: The attackers do not know the existence of the randomization layers and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 141, + 686, + 324, + 699 + ], + "spans": [ + { + "bbox": [ + 141, + 686, + 324, + 699 + ], + "score": 1.0, + "content": "the target model is just the original network.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 133, + 675, + 505, + 699 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 109, + 112, + 503, + 170 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 504, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 89, + 505, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 89, + 505, + 101 + ], + "score": 1.0, + "content": "Table 1: Top-1 classification accuracy on the clean images. We see that adding random resizing and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 100, + 438, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 438, + 113 + ], + "score": 1.0, + "content": "random padding cause very little accuracy drop on clean (non-adversarial) images.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 109, + 112, + 503, + 170 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 112, + 503, + 170 + ], + "spans": [ + { + "bbox": [ + 109, + 112, + 503, + 170 + ], + "score": 0.98, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
w/o randomization layers100%100%100%100%
wrandomization layers97.3%98.3%99.3%99.2%
", + "type": "table", + "image_path": "f7a87e7b923a88f8e0adc99a1b2c5c1e87d84b1e032095c951694d1d5697a82d.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 109, + 112, + 503, + 131.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 109, + 131.33333333333334, + 503, + 150.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 150.66666666666669, + 503, + 170.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 133, + 191, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 133, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 133, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "• Single-Pattern Attack: The attackers know the existence of the randomization layers. In", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 141, + 203, + 505, + 215 + ], + "score": 1.0, + "content": "order to mimic the structures of defense models, the target model is chosen as the original", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 212, + 406, + 227 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 176, + 227 + ], + "score": 1.0, + "content": "network", + "type": "text" + }, + { + "bbox": [ + 177, + 214, + 185, + 223 + ], + "score": 0.73, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 212, + 406, + 227 + ], + "score": 1.0, + "content": "randomization layers with only one predefined pattern.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 133, + 228, + 505, + 273 + ], + "lines": [ + { + "bbox": [ + 132, + 228, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 132, + 228, + 505, + 241 + ], + "score": 1.0, + "content": "• Ensemble-Pattern Attack: The attackers know the existence of the randomization layers.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 240, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 141, + 240, + 505, + 252 + ], + "score": 1.0, + "content": "In order to mimic the structures of defense models in a more representative way, the tar-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 251, + 506, + 263 + ], + "spans": [ + { + "bbox": [ + 141, + 251, + 323, + 263 + ], + "score": 1.0, + "content": "get model is chosen as the original network", + "type": "text" + }, + { + "bbox": [ + 323, + 252, + 332, + 261 + ], + "score": 0.78, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 251, + 506, + 263 + ], + "score": 1.0, + "content": "randomization layers with an ensemble of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 263, + 223, + 275 + ], + "spans": [ + { + "bbox": [ + 141, + 263, + 223, + 275 + ], + "score": 1.0, + "content": "predefined patterns.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 283, + 505, + 371 + ], + "lines": [ + { + "bbox": [ + 107, + 284, + 504, + 295 + ], + "spans": [ + { + "bbox": [ + 107, + 284, + 504, + 295 + ], + "score": 1.0, + "content": "Target Models and Defense Models: The target models and the defense models are exactly the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "score": 1.0, + "content": "same except for the parameter settings of the randomization layers, i.e., the randomization param-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 306, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 317 + ], + "score": 1.0, + "content": "eters at the target models are predefined while randomization parameters at the defense models are", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 402, + 329 + ], + "score": 1.0, + "content": "randomly generated at test time. The original networks (e.g., Inception-", + "type": "text" + }, + { + "bbox": [ + 402, + 317, + 412, + 326 + ], + "score": 0.3, + "content": "\\nu 3", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 316, + 505, + 329 + ], + "score": 1.0, + "content": ") utilized by the target", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 326, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 506, + 341 + ], + "score": 1.0, + "content": "models and the defense models are the same. The attackers first generate adversarial examples using", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "the target models, and then evaluate the classification accuracy of these generated adversarial exam-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "ples on both the target and defense models. A low accuracy of the target model indicates that the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 360, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 372 + ], + "score": 1.0, + "content": "attack is successful, and a high accuracy of the defense model indicates that the defense is effective.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 107, + 386, + 198, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 384, + 200, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 200, + 399 + ], + "score": 1.0, + "content": "4.2 CLEAN IMAGES", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 407, + 505, + 484 + ], + "lines": [ + { + "bbox": [ + 105, + 406, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 505, + 420 + ], + "score": 1.0, + "content": "Table 1 shows the top-1 accuracy of networks with and without randomization layers on the clean", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "score": 1.0, + "content": "images. We can see that randomization layers introduce negligible performance degradation on", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 430, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 505, + 441 + ], + "score": 1.0, + "content": "clean images. Specifically, we can observe that: (1) models with more advanced architectures tend", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 356, + 452 + ], + "score": 1.0, + "content": "to have less performance degradation, e.g., Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 356, + 441, + 367, + 450 + ], + "score": 0.54, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 440, + 406, + 452 + ], + "score": 1.0, + "content": "only has", + "type": "text" + }, + { + "bbox": [ + 406, + 440, + 429, + 451 + ], + "score": 0.87, + "content": "0 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "degradation while", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 175, + 464 + ], + "score": 1.0, + "content": "Inception-v3 has", + "type": "text" + }, + { + "bbox": [ + 175, + 451, + 198, + 462 + ], + "score": 0.87, + "content": "2 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "degradation; (2) ensemble adversarial training brings nearly no performance", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "score": 1.0, + "content": "degradation to the models, e.g., Inception-ResNet-v2 and ens-adv-Inception-ResNet-v2 have nearly", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 473, + 249, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 249, + 486 + ], + "score": 1.0, + "content": "the same performance degradation.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 108, + 498, + 255, + 510 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 256, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 256, + 511 + ], + "score": 1.0, + "content": "4.3 VANILLA ATTACK SCENARIO", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 520, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 519, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 532 + ], + "score": 1.0, + "content": "For the vanilla attack scenario, the attackers are not aware of randomization layers, and directly use", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "the original networks as the target model to generate adversarial examples. The attack ability on the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "defense models mostly rely on the transferability of adversarial examples to different resizing and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 552, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 567 + ], + "score": 1.0, + "content": "padding. From the top-1 accuracy presented in Table 2, we observe that randomization layers can", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 563, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 577 + ], + "score": 1.0, + "content": "mitigate the adversarial effects for both single-step and iterative attacks significantly. As for single-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 225, + 587 + ], + "score": 1.0, + "content": "step attacks FGSM-\u000f, larger", + "type": "text" + }, + { + "bbox": [ + 225, + 577, + 231, + 585 + ], + "score": 0.66, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 575, + 505, + 587 + ], + "score": 1.0, + "content": "indicates stronger transferability, thus making it harder to defend.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 585, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 599 + ], + "score": 1.0, + "content": "However, we can still get satisfactory accuracy of the defense model on single-step attacks (even", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 597, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 597, + 150, + 610 + ], + "score": 1.0, + "content": "with large", + "type": "text" + }, + { + "bbox": [ + 150, + 599, + 156, + 607 + ], + "score": 0.51, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 597, + 292, + 610 + ], + "score": 1.0, + "content": ") using ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 293, + 597, + 303, + 607 + ], + "score": 0.54, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 597, + 308, + 610 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 308, + 597, + 335, + 608 + ], + "score": 0.82, + "content": "9 4 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 597, + 505, + 610 + ], + "score": 1.0, + "content": "top-1 accuracy). As for iterative attacks,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "attackers always reach a very high attack rate on target model, but have almost no impact on models", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 618, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 506, + 632 + ], + "score": 1.0, + "content": "after randomization layers are applied. This is because iterative attack methods are over-fitted to the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 630, + 286, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 286, + 642 + ], + "score": 1.0, + "content": "target models thus have weak transferability.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 34 + }, + { + "type": "title", + "bbox": [ + 108, + 655, + 291, + 667 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 292, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 292, + 668 + ], + "score": 1.0, + "content": "4.4 SINGLE-PATTERN ATTACK SCENARIO", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "For the single-pattern attack scenario, the attackers are aware of the existence of randomization", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 687, + 504, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 459, + 699 + ], + "score": 1.0, + "content": "layers and also the parameters of the random resizing and random padding (i.e., from", + "type": "text" + }, + { + "bbox": [ + 459, + 687, + 504, + 698 + ], + "score": 0.87, + "content": "2 9 9 \\times 2 9 9", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 118, + 712 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 118, + 699, + 163, + 710 + ], + "score": 0.88, + "content": "3 3 1 \\times 3 3 1", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "), but they do not know the specific randomization patterns utilized by the defense", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "models (even the defense models themselves do not know these specific randomization patterns", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "since they are randomly instantiated at test time). In order to generate adversarial examples, the", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 109, + 112, + 503, + 170 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 504, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 89, + 505, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 89, + 505, + 101 + ], + "score": 1.0, + "content": "Table 1: Top-1 classification accuracy on the clean images. We see that adding random resizing and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 100, + 438, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 438, + 113 + ], + "score": 1.0, + "content": "random padding cause very little accuracy drop on clean (non-adversarial) images.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 109, + 112, + 503, + 170 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 112, + 503, + 170 + ], + "spans": [ + { + "bbox": [ + 109, + 112, + 503, + 170 + ], + "score": 0.98, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
w/o randomization layers100%100%100%100%
wrandomization layers97.3%98.3%99.3%99.2%
", + "type": "table", + "image_path": "f7a87e7b923a88f8e0adc99a1b2c5c1e87d84b1e032095c951694d1d5697a82d.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 109, + 112, + 503, + 131.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 109, + 131.33333333333334, + 503, + 150.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 150.66666666666669, + 503, + 170.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 133, + 191, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 133, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 133, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "• Single-Pattern Attack: The attackers know the existence of the randomization layers. In", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 141, + 203, + 505, + 215 + ], + "score": 1.0, + "content": "order to mimic the structures of defense models, the target model is chosen as the original", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 212, + 406, + 227 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 176, + 227 + ], + "score": 1.0, + "content": "network", + "type": "text" + }, + { + "bbox": [ + 177, + 214, + 185, + 223 + ], + "score": 0.73, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 212, + 406, + 227 + ], + "score": 1.0, + "content": "randomization layers with only one predefined pattern.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 133, + 191, + 505, + 227 + ] + }, + { + "type": "text", + "bbox": [ + 133, + 228, + 505, + 273 + ], + "lines": [ + { + "bbox": [ + 132, + 228, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 132, + 228, + 505, + 241 + ], + "score": 1.0, + "content": "• Ensemble-Pattern Attack: The attackers know the existence of the randomization layers.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 240, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 141, + 240, + 505, + 252 + ], + "score": 1.0, + "content": "In order to mimic the structures of defense models in a more representative way, the tar-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 251, + 506, + 263 + ], + "spans": [ + { + "bbox": [ + 141, + 251, + 323, + 263 + ], + "score": 1.0, + "content": "get model is chosen as the original network", + "type": "text" + }, + { + "bbox": [ + 323, + 252, + 332, + 261 + ], + "score": 0.78, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 251, + 506, + 263 + ], + "score": 1.0, + "content": "randomization layers with an ensemble of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 263, + 223, + 275 + ], + "spans": [ + { + "bbox": [ + 141, + 263, + 223, + 275 + ], + "score": 1.0, + "content": "predefined patterns.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 132, + 228, + 506, + 275 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 283, + 505, + 371 + ], + "lines": [ + { + "bbox": [ + 107, + 284, + 504, + 295 + ], + "spans": [ + { + "bbox": [ + 107, + 284, + 504, + 295 + ], + "score": 1.0, + "content": "Target Models and Defense Models: The target models and the defense models are exactly the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 307 + ], + "score": 1.0, + "content": "same except for the parameter settings of the randomization layers, i.e., the randomization param-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 306, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 317 + ], + "score": 1.0, + "content": "eters at the target models are predefined while randomization parameters at the defense models are", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 402, + 329 + ], + "score": 1.0, + "content": "randomly generated at test time. The original networks (e.g., Inception-", + "type": "text" + }, + { + "bbox": [ + 402, + 317, + 412, + 326 + ], + "score": 0.3, + "content": "\\nu 3", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 316, + 505, + 329 + ], + "score": 1.0, + "content": ") utilized by the target", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 326, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 506, + 341 + ], + "score": 1.0, + "content": "models and the defense models are the same. The attackers first generate adversarial examples using", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "the target models, and then evaluate the classification accuracy of these generated adversarial exam-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "ples on both the target and defense models. A low accuracy of the target model indicates that the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 360, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 372 + ], + "score": 1.0, + "content": "attack is successful, and a high accuracy of the defense model indicates that the defense is effective.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 284, + 506, + 372 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 386, + 198, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 384, + 200, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 200, + 399 + ], + "score": 1.0, + "content": "4.2 CLEAN IMAGES", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 407, + 505, + 484 + ], + "lines": [ + { + "bbox": [ + 105, + 406, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 505, + 420 + ], + "score": 1.0, + "content": "Table 1 shows the top-1 accuracy of networks with and without randomization layers on the clean", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "score": 1.0, + "content": "images. We can see that randomization layers introduce negligible performance degradation on", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 430, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 505, + 441 + ], + "score": 1.0, + "content": "clean images. Specifically, we can observe that: (1) models with more advanced architectures tend", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 356, + 452 + ], + "score": 1.0, + "content": "to have less performance degradation, e.g., Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 356, + 441, + 367, + 450 + ], + "score": 0.54, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 440, + 406, + 452 + ], + "score": 1.0, + "content": "only has", + "type": "text" + }, + { + "bbox": [ + 406, + 440, + 429, + 451 + ], + "score": 0.87, + "content": "0 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "degradation while", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 175, + 464 + ], + "score": 1.0, + "content": "Inception-v3 has", + "type": "text" + }, + { + "bbox": [ + 175, + 451, + 198, + 462 + ], + "score": 0.87, + "content": "2 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "degradation; (2) ensemble adversarial training brings nearly no performance", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "score": 1.0, + "content": "degradation to the models, e.g., Inception-ResNet-v2 and ens-adv-Inception-ResNet-v2 have nearly", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 473, + 249, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 249, + 486 + ], + "score": 1.0, + "content": "the same performance degradation.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 406, + 505, + 486 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 498, + 255, + 510 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 256, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 256, + 511 + ], + "score": 1.0, + "content": "4.3 VANILLA ATTACK SCENARIO", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 520, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 519, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 532 + ], + "score": 1.0, + "content": "For the vanilla attack scenario, the attackers are not aware of randomization layers, and directly use", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "the original networks as the target model to generate adversarial examples. The attack ability on the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "defense models mostly rely on the transferability of adversarial examples to different resizing and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 552, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 567 + ], + "score": 1.0, + "content": "padding. From the top-1 accuracy presented in Table 2, we observe that randomization layers can", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 563, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 577 + ], + "score": 1.0, + "content": "mitigate the adversarial effects for both single-step and iterative attacks significantly. As for single-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 225, + 587 + ], + "score": 1.0, + "content": "step attacks FGSM-\u000f, larger", + "type": "text" + }, + { + "bbox": [ + 225, + 577, + 231, + 585 + ], + "score": 0.66, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 575, + 505, + 587 + ], + "score": 1.0, + "content": "indicates stronger transferability, thus making it harder to defend.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 585, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 599 + ], + "score": 1.0, + "content": "However, we can still get satisfactory accuracy of the defense model on single-step attacks (even", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 597, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 597, + 150, + 610 + ], + "score": 1.0, + "content": "with large", + "type": "text" + }, + { + "bbox": [ + 150, + 599, + 156, + 607 + ], + "score": 0.51, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 597, + 292, + 610 + ], + "score": 1.0, + "content": ") using ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 293, + 597, + 303, + 607 + ], + "score": 0.54, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 597, + 308, + 610 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 308, + 597, + 335, + 608 + ], + "score": 0.82, + "content": "9 4 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 597, + 505, + 610 + ], + "score": 1.0, + "content": "top-1 accuracy). As for iterative attacks,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "attackers always reach a very high attack rate on target model, but have almost no impact on models", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 618, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 506, + 632 + ], + "score": 1.0, + "content": "after randomization layers are applied. This is because iterative attack methods are over-fitted to the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 630, + 286, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 286, + 642 + ], + "score": 1.0, + "content": "target models thus have weak transferability.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 519, + 506, + 642 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 655, + 291, + 667 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 292, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 292, + 668 + ], + "score": 1.0, + "content": "4.4 SINGLE-PATTERN ATTACK SCENARIO", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "For the single-pattern attack scenario, the attackers are aware of the existence of randomization", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 687, + 504, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 459, + 699 + ], + "score": 1.0, + "content": "layers and also the parameters of the random resizing and random padding (i.e., from", + "type": "text" + }, + { + "bbox": [ + 459, + 687, + 504, + 698 + ], + "score": 0.87, + "content": "2 9 9 \\times 2 9 9", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 118, + 712 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 118, + 699, + 163, + 710 + ], + "score": 0.88, + "content": "3 3 1 \\times 3 3 1", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "), but they do not know the specific randomization patterns utilized by the defense", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "models (even the defense models themselves do not know these specific randomization patterns", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "since they are randomly instantiated at test time). In order to generate adversarial examples, the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 451, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 350, + 466 + ], + "score": 1.0, + "content": "attackers choose the target models as the original networks", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 350, + 454, + 359, + 463 + ], + "score": 0.79, + "content": "^ +", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 359, + 451, + 505, + 466 + ], + "score": 1.0, + "content": "randomization layers but with only", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "one specific pattern to compute the gradient. In this experiment, the specific pattern that we use is to", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 473, + 507, + 490 + ], + "spans": [ + { + "bbox": [ + 104, + 473, + 200, + 490 + ], + "score": 1.0, + "content": "place the original input", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 201, + 475, + 215, + 486 + ], + "score": 0.89, + "content": "X _ { n }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 216, + 473, + 350, + 490 + ], + "score": 1.0, + "content": "at the center of the padded image", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 351, + 475, + 365, + 487 + ], + "score": 0.9, + "content": "X _ { n } ^ { \\prime \\prime }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 366, + 473, + 507, + 490 + ], + "score": 1.0, + "content": ", i.e., no resizing is applied, and 16", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 485, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 506, + 498 + ], + "score": 1.0, + "content": "zeros pixels are padded on the left, right, top and bottom on the input images, respectively. Table 3", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 496, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 506, + 509 + ], + "score": 1.0, + "content": "shows the top-1 accuracy of both target models and defense models, and similar results to vanilla", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 507, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 506, + 520 + ], + "score": 1.0, + "content": "attack scenario are observed: (1) for single-step attacks, randomization layers are less effective on", + "type": "text", + "cross_page": true + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 270, + 530 + ], + "score": 1.0, + "content": "mitigating adversarial effects for a larger", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 270, + 521, + 275, + 528 + ], + "score": 0.35, + "content": "\\epsilon", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 276, + 518, + 505, + 530 + ], + "score": 1.0, + "content": ", while the adversarially trained models are able to defend", + "type": "text", + "cross_page": true + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "score": 1.0, + "content": "against such attacks; (2) for iterative attacks, they reach high attack rates on target models, while", + "type": "text", + "cross_page": true + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 541, + 275, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 275, + 553 + ], + "score": 1.0, + "content": "have nearly no impact on defense models.", + "type": "text", + "cross_page": true + } + ], + "index": 22 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 677, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 133, + 505, + 249 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 505, + 132 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 89, + 505, + 102 + ], + "spans": [ + { + "bbox": [ + 106, + 89, + 505, + 102 + ], + "score": 1.0, + "content": "Table 2: Top-1 classification accuracy under the vanilla attack scenario. We see that randomization", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 99, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 505, + 114 + ], + "score": 1.0, + "content": "layers effectively mitigate adversarial effects for all attacks and all networks. Particularly, combining", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 110, + 504, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 504, + 124 + ], + "score": 1.0, + "content": "randomization layers with ensemble adversarial training (ens-adv-Inception-ResNet-v2) performs", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 122, + 203, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 203, + 135 + ], + "score": 1.0, + "content": "very well on all attacks.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 133, + 505, + 249 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 133, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 505, + 249 + ], + "score": 0.985, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-233.2%65.1%26.3%71.8%65.3%81.0%84.4%95.7%
FGSM-531.1%54.5%20.4%54.3%61.7%74.1%87.4%94.5%
FGSM-1033.0%52.4%20.4%46.1%61.2%71.3%90.2%94.3%
DeepFool0%98.3%0%97.7%0%98.2%0.2%99.1%
C&W0%96.9%0%97.1%0.3%97.7%0.9%98.8%
", + "type": "table", + "image_path": "1850526665cf95de722fac8eb55948dd851b7674e5ed589cdafa394a060b212b.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 106, + 133, + 505, + 171.66666666666666 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 171.66666666666666, + 505, + 210.33333333333331 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 210.33333333333331, + 505, + 248.99999999999997 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "table", + "bbox": [ + 106, + 315, + 506, + 430 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 270, + 505, + 313 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 270, + 504, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 504, + 283 + ], + "score": 1.0, + "content": "Table 3: Top-1 classification accuracy under the single-pattern attack scenario. We see that random-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "ization layers effectively mitigate adversarial effects for all attacks and all networks. Particularly,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 291, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 489, + 306 + ], + "score": 1.0, + "content": "combining randomization layers with ensemble adversarial training (ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 490, + 293, + 500, + 302 + ], + "score": 0.28, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 291, + 505, + 306 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 303, + 242, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 242, + 316 + ], + "score": 1.0, + "content": "performs very well on all attacks.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 315, + 506, + 430 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 315, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 506, + 430 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-235.1%63.8%29.5%70.1%71.6%83.4%86.3%96.4%
FGSM-532.4%53.9%23.2%52.3%68.3%78.2%88.4%95.4%
FGSM-1034.7%51.8%22.4%43.8%66.8%75.6%90.7%95.2%
DeepFool1.1%98.2%1.7%97.8%0.6%98.4%1.0%99.2%
C&W1.1%97.4%1.7%97.0%0.8%97.9%1.6%99.1%
", + "type": "table", + "image_path": "2d308f4aca054eed2b4b010547f1a317397847f183c5cd9f607fd3bdbab1d6af.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 106, + 315, + 506, + 353.3333333333333 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 106, + 353.3333333333333, + 506, + 391.66666666666663 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 106, + 391.66666666666663, + 506, + 429.99999999999994 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 10.25 + }, + { + "type": "text", + "bbox": [ + 106, + 452, + 505, + 552 + ], + "lines": [ + { + "bbox": [ + 105, + 451, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 350, + 466 + ], + "score": 1.0, + "content": "attackers choose the target models as the original networks", + "type": "text" + }, + { + "bbox": [ + 350, + 454, + 359, + 463 + ], + "score": 0.79, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 451, + 505, + 466 + ], + "score": 1.0, + "content": "randomization layers but with only", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "one specific pattern to compute the gradient. In this experiment, the specific pattern that we use is to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 473, + 507, + 490 + ], + "spans": [ + { + "bbox": [ + 104, + 473, + 200, + 490 + ], + "score": 1.0, + "content": "place the original input", + "type": "text" + }, + { + "bbox": [ + 201, + 475, + 215, + 486 + ], + "score": 0.89, + "content": "X _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 473, + 350, + 490 + ], + "score": 1.0, + "content": "at the center of the padded image", + "type": "text" + }, + { + "bbox": [ + 351, + 475, + 365, + 487 + ], + "score": 0.9, + "content": "X _ { n } ^ { \\prime \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 473, + 507, + 490 + ], + "score": 1.0, + "content": ", i.e., no resizing is applied, and 16", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 485, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 506, + 498 + ], + "score": 1.0, + "content": "zeros pixels are padded on the left, right, top and bottom on the input images, respectively. Table 3", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 496, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 506, + 509 + ], + "score": 1.0, + "content": "shows the top-1 accuracy of both target models and defense models, and similar results to vanilla", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 507, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 506, + 520 + ], + "score": 1.0, + "content": "attack scenario are observed: (1) for single-step attacks, randomization layers are less effective on", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 270, + 530 + ], + "score": 1.0, + "content": "mitigating adversarial effects for a larger", + "type": "text" + }, + { + "bbox": [ + 270, + 521, + 275, + 528 + ], + "score": 0.35, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 518, + 505, + 530 + ], + "score": 1.0, + "content": ", while the adversarially trained models are able to defend", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "score": 1.0, + "content": "against such attacks; (2) for iterative attacks, they reach high attack rates on target models, while", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 541, + 275, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 275, + 553 + ], + "score": 1.0, + "content": "have nearly no impact on defense models.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 108, + 568, + 306, + 579 + ], + "lines": [ + { + "bbox": [ + 105, + 567, + 307, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 307, + 580 + ], + "score": 1.0, + "content": "4.5 ENSEMBLE-PATTERN ATTACK SCENARIO", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 589, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 506, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 602 + ], + "score": 1.0, + "content": "For the ensemble-pattern attack scenario, similar to single-pattern attack scenario, the attackers are", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 599, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 104, + 599, + 505, + 614 + ], + "score": 1.0, + "content": "aware of the randomization layers and the parameters of the random resizing and random padding", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 611, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 180, + 624 + ], + "score": 1.0, + "content": "(i.e., starting from", + "type": "text" + }, + { + "bbox": [ + 180, + 611, + 222, + 622 + ], + "score": 0.9, + "content": "2 9 9 \\times 2 9 9", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 611, + 232, + 624 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 232, + 611, + 274, + 622 + ], + "score": 0.88, + "content": "3 3 1 \\times 3 3 1", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 611, + 505, + 624 + ], + "score": 1.0, + "content": "), but they do not know the specific patterns utilized by the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 620, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 104, + 620, + 505, + 636 + ], + "score": 1.0, + "content": "defense models at test time. The target models thus are constructed in a more representative way:", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "let randomization layers choose an ensemble of predefined patterns, and the goal of the attackers is", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "score": 1.0, + "content": "to let all chosen patterns fail on classification. In this experiment, the specific ensemble patterns that", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 653, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 104, + 653, + 383, + 669 + ], + "score": 1.0, + "content": "we choose are: (1) first resize the input image to five different scales", + "type": "text" + }, + { + "bbox": [ + 384, + 655, + 487, + 667 + ], + "score": 0.74, + "content": "\\{ 2 9 9 , 3 0 7 , 3 1 5 , 3 2 3 , 3 3 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 653, + 505, + 669 + ], + "score": 1.0, + "content": "; (2)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 664, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 104, + 664, + 506, + 680 + ], + "score": 1.0, + "content": "then pad each resized image to five different patterns, where the resized image is placed at the top", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "left, top right, bottom left, bottom right, and center of the padded image, respectively. Since there is", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 688, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 405, + 701 + ], + "score": 1.0, + "content": "only one padding pattern for the resized image with size 331, we can obtain", + "type": "text" + }, + { + "bbox": [ + 406, + 688, + 460, + 698 + ], + "score": 0.93, + "content": "4 * 5 + 1 = 2 1", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 688, + 506, + 701 + ], + "score": 1.0, + "content": "patterns in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "total. Due to the large computation amounts introduced by the ensemble-pattern attack scenario, we", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "score": 1.0, + "content": "randomly choose 500 images out of the entire test dataset for this experiment. The top-1 accuracy for", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "the target model here is calculated by summing up the number of correctly classified patterns of each", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 133, + 505, + 249 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 505, + 132 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 89, + 505, + 102 + ], + "spans": [ + { + "bbox": [ + 106, + 89, + 505, + 102 + ], + "score": 1.0, + "content": "Table 2: Top-1 classification accuracy under the vanilla attack scenario. We see that randomization", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 99, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 505, + 114 + ], + "score": 1.0, + "content": "layers effectively mitigate adversarial effects for all attacks and all networks. Particularly, combining", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 110, + 504, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 504, + 124 + ], + "score": 1.0, + "content": "randomization layers with ensemble adversarial training (ens-adv-Inception-ResNet-v2) performs", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 122, + 203, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 203, + 135 + ], + "score": 1.0, + "content": "very well on all attacks.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 133, + 505, + 249 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 133, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 505, + 249 + ], + "score": 0.985, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-233.2%65.1%26.3%71.8%65.3%81.0%84.4%95.7%
FGSM-531.1%54.5%20.4%54.3%61.7%74.1%87.4%94.5%
FGSM-1033.0%52.4%20.4%46.1%61.2%71.3%90.2%94.3%
DeepFool0%98.3%0%97.7%0%98.2%0.2%99.1%
C&W0%96.9%0%97.1%0.3%97.7%0.9%98.8%
", + "type": "table", + "image_path": "1850526665cf95de722fac8eb55948dd851b7674e5ed589cdafa394a060b212b.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 106, + 133, + 505, + 171.66666666666666 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 171.66666666666666, + 505, + 210.33333333333331 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 210.33333333333331, + 505, + 248.99999999999997 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "table", + "bbox": [ + 106, + 315, + 506, + 430 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 270, + 505, + 313 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 270, + 504, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 504, + 283 + ], + "score": 1.0, + "content": "Table 3: Top-1 classification accuracy under the single-pattern attack scenario. We see that random-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "ization layers effectively mitigate adversarial effects for all attacks and all networks. Particularly,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 291, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 489, + 306 + ], + "score": 1.0, + "content": "combining randomization layers with ensemble adversarial training (ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 490, + 293, + 500, + 302 + ], + "score": 0.28, + "content": "\\cdot \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 291, + 505, + 306 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 303, + 242, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 242, + 316 + ], + "score": 1.0, + "content": "performs very well on all attacks.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 315, + 506, + 430 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 315, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 506, + 430 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-235.1%63.8%29.5%70.1%71.6%83.4%86.3%96.4%
FGSM-532.4%53.9%23.2%52.3%68.3%78.2%88.4%95.4%
FGSM-1034.7%51.8%22.4%43.8%66.8%75.6%90.7%95.2%
DeepFool1.1%98.2%1.7%97.8%0.6%98.4%1.0%99.2%
C&W1.1%97.4%1.7%97.0%0.8%97.9%1.6%99.1%
", + "type": "table", + "image_path": "2d308f4aca054eed2b4b010547f1a317397847f183c5cd9f607fd3bdbab1d6af.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 106, + 315, + 506, + 353.3333333333333 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 106, + 353.3333333333333, + 506, + 391.66666666666663 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 106, + 391.66666666666663, + 506, + 429.99999999999994 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 10.25 + }, + { + "type": "text", + "bbox": [ + 106, + 452, + 505, + 552 + ], + "lines": [], + "index": 18, + "bbox_fs": [ + 104, + 451, + 507, + 553 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 568, + 306, + 579 + ], + "lines": [ + { + "bbox": [ + 105, + 567, + 307, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 307, + 580 + ], + "score": 1.0, + "content": "4.5 ENSEMBLE-PATTERN ATTACK SCENARIO", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 589, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 506, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 602 + ], + "score": 1.0, + "content": "For the ensemble-pattern attack scenario, similar to single-pattern attack scenario, the attackers are", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 599, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 104, + 599, + 505, + 614 + ], + "score": 1.0, + "content": "aware of the randomization layers and the parameters of the random resizing and random padding", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 611, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 180, + 624 + ], + "score": 1.0, + "content": "(i.e., starting from", + "type": "text" + }, + { + "bbox": [ + 180, + 611, + 222, + 622 + ], + "score": 0.9, + "content": "2 9 9 \\times 2 9 9", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 611, + 232, + 624 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 232, + 611, + 274, + 622 + ], + "score": 0.88, + "content": "3 3 1 \\times 3 3 1", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 611, + 505, + 624 + ], + "score": 1.0, + "content": "), but they do not know the specific patterns utilized by the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 620, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 104, + 620, + 505, + 636 + ], + "score": 1.0, + "content": "defense models at test time. The target models thus are constructed in a more representative way:", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "let randomization layers choose an ensemble of predefined patterns, and the goal of the attackers is", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "score": 1.0, + "content": "to let all chosen patterns fail on classification. In this experiment, the specific ensemble patterns that", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 653, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 104, + 653, + 383, + 669 + ], + "score": 1.0, + "content": "we choose are: (1) first resize the input image to five different scales", + "type": "text" + }, + { + "bbox": [ + 384, + 655, + 487, + 667 + ], + "score": 0.74, + "content": "\\{ 2 9 9 , 3 0 7 , 3 1 5 , 3 2 3 , 3 3 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 653, + 505, + 669 + ], + "score": 1.0, + "content": "; (2)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 664, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 104, + 664, + 506, + 680 + ], + "score": 1.0, + "content": "then pad each resized image to five different patterns, where the resized image is placed at the top", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "left, top right, bottom left, bottom right, and center of the padded image, respectively. Since there is", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 688, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 405, + 701 + ], + "score": 1.0, + "content": "only one padding pattern for the resized image with size 331, we can obtain", + "type": "text" + }, + { + "bbox": [ + 406, + 688, + 460, + 698 + ], + "score": 0.93, + "content": "4 * 5 + 1 = 2 1", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 688, + 506, + 701 + ], + "score": 1.0, + "content": "patterns in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "total. Due to the large computation amounts introduced by the ensemble-pattern attack scenario, we", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "score": 1.0, + "content": "randomly choose 500 images out of the entire test dataset for this experiment. The top-1 accuracy for", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "the target model here is calculated by summing up the number of correctly classified patterns of each", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 285, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 505, + 296 + ], + "score": 1.0, + "content": "image over the entire pattern number of all images. For the results presented in Table 4, we can see", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 294, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 308 + ], + "score": 1.0, + "content": "that the adversarial examples generated under ensemble-pattern attack scenario are much stronger.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 307, + 504, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 504, + 318 + ], + "score": 1.0, + "content": "For single-step attacks, the generated adversarial examples can let the performance of the defense", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 317, + 504, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 340, + 329 + ], + "score": 1.0, + "content": "model with an adversarially trained network drop around", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 341, + 317, + 356, + 328 + ], + "score": 0.86, + "content": "8 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 356, + 318, + 504, + 329 + ], + "score": 1.0, + "content": "compared to the performance under", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "vanilla attack and single-pattern attack scenarios, and drop much more on other defense models.", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "For iterative attacks, we observe that the adversarial examples generated by C&W are stronger than", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 350, + 504, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 387, + 363 + ], + "score": 1.0, + "content": "those generated by DeepFool, e.g., the defense model with Inception-", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 387, + 351, + 398, + 361 + ], + "score": 0.37, + "content": "\\nu 3", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 398, + 350, + 476, + 363 + ], + "score": 1.0, + "content": "has an accuracy of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 477, + 350, + 504, + 361 + ], + "score": 0.87, + "content": "8 1 . 3 \\%", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 285, + 374 + ], + "score": 1.0, + "content": "on DeepFool, while only has an accuracy of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 286, + 361, + 313, + 372 + ], + "score": 0.88, + "content": "6 2 . 9 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 314, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "on C&W. We argue that this is due to the more", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 372, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 505, + 385 + ], + "score": 1.0, + "content": "advanced loss function (i.e., introduction of auxiliary variable for pixel value control) utilized by", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "C&W than DeepFool. Additionally, the accuracy of defense model on C&W can be improved by", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 394, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 407 + ], + "score": 1.0, + "content": "utilizing more advanced architecture (e.g., ResNet-v2-101 has higher accuracy than Inception-v3)", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 405, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 411, + 419 + ], + "score": 1.0, + "content": "and applying ensemble adversarial training (e.g., ens-adv-Inception-ResNet-", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 411, + 406, + 422, + 415 + ], + "score": 0.5, + "content": "\\nu 2", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 422, + 405, + 506, + 419 + ], + "score": 1.0, + "content": "has higher accuracy", + "type": "text", + "cross_page": true + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 416, + 504, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 493, + 428 + ], + "score": 1.0, + "content": "than Inception-ResNet-v2). For the best defense model that we have, ens-adv-Inception-ResNet-", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 493, + 416, + 504, + 426 + ], + "score": 0.37, + "content": "\\nu 2", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 427, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 115, + 437 + ], + "score": 0.7, + "content": "^ +", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 115, + 427, + 329, + 441 + ], + "score": 1.0, + "content": "randomization layers reaches the top-1 accuracy of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 329, + 427, + 357, + 438 + ], + "score": 0.89, + "content": "9 3 . 5 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 358, + 427, + 433, + 441 + ], + "score": 1.0, + "content": "on DeepFool and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 434, + 427, + 461, + 438 + ], + "score": 0.88, + "content": "8 6 . 1 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 462, + 427, + 506, + 441 + ], + "score": 1.0, + "content": "on C&W,", + "type": "text", + "cross_page": true + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 438, + 159, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 159, + 452 + ], + "score": 1.0, + "content": "respectively.", + "type": "text", + "cross_page": true + } + ], + "index": 22 + } + ], + "index": 30, + "bbox_fs": [ + 104, + 588, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 143, + 505, + 258 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 505, + 142 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 90, + 505, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 90, + 505, + 101 + ], + "score": 1.0, + "content": "Table 4: Top-1 classification accuracy under the ensemble-pattern attack scenario. Similar to vanilla", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "score": 1.0, + "content": "attack and single-pattern attack scenarios, we see that randomization layers increase the accuracy", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 111, + 504, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 504, + 124 + ], + "score": 1.0, + "content": "under all attacks and networks. This clearly demonstrates the effectiveness of the proposed ran-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 121, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 506, + 135 + ], + "score": 1.0, + "content": "domization method on defending against adversarial examples, even under this very strong attack", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 134, + 144, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 134, + 144, + 145 + ], + "score": 1.0, + "content": "scenario.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 106, + 143, + 505, + 258 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 143, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 258 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-237.3%41.2%39.2%44.9%71.5%74.3%86.2%88.9%
FGSM-531.7%34.0%24.6%29.7%65.2%67.3%85.8%87.5%
FGSM-1030.4%32.8%18.6%21.7%62.9%64.5%86.6%87.9%
DeepFool0.6%81.3%0.9%80.5%0.9%69.4%1.6%93.5%
C&W0.6%62.9%1.0%74.3%1.6%68.3%5.8%86.1%
", + "type": "table", + "image_path": "a5f5f049a290593d5c99320c3903b6b4ea7a50d09c509f730261fde2a47ca05a.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 143, + 505, + 181.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 181.33333333333334, + 505, + 219.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 219.66666666666669, + 505, + 258.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "text", + "bbox": [ + 106, + 284, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 106, + 285, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 505, + 296 + ], + "score": 1.0, + "content": "image over the entire pattern number of all images. For the results presented in Table 4, we can see", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 294, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 308 + ], + "score": 1.0, + "content": "that the adversarial examples generated under ensemble-pattern attack scenario are much stronger.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 307, + 504, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 504, + 318 + ], + "score": 1.0, + "content": "For single-step attacks, the generated adversarial examples can let the performance of the defense", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 317, + 504, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 340, + 329 + ], + "score": 1.0, + "content": "model with an adversarially trained network drop around", + "type": "text" + }, + { + "bbox": [ + 341, + 317, + 356, + 328 + ], + "score": 0.86, + "content": "8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 318, + 504, + 329 + ], + "score": 1.0, + "content": "compared to the performance under", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "vanilla attack and single-pattern attack scenarios, and drop much more on other defense models.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "For iterative attacks, we observe that the adversarial examples generated by C&W are stronger than", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 350, + 504, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 387, + 363 + ], + "score": 1.0, + "content": "those generated by DeepFool, e.g., the defense model with Inception-", + "type": "text" + }, + { + "bbox": [ + 387, + 351, + 398, + 361 + ], + "score": 0.37, + "content": "\\nu 3", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 350, + 476, + 363 + ], + "score": 1.0, + "content": "has an accuracy of", + "type": "text" + }, + { + "bbox": [ + 477, + 350, + 504, + 361 + ], + "score": 0.87, + "content": "8 1 . 3 \\%", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 285, + 374 + ], + "score": 1.0, + "content": "on DeepFool, while only has an accuracy of", + "type": "text" + }, + { + "bbox": [ + 286, + 361, + 313, + 372 + ], + "score": 0.88, + "content": "6 2 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "on C&W. We argue that this is due to the more", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 372, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 505, + 385 + ], + "score": 1.0, + "content": "advanced loss function (i.e., introduction of auxiliary variable for pixel value control) utilized by", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "C&W than DeepFool. Additionally, the accuracy of defense model on C&W can be improved by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 394, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 407 + ], + "score": 1.0, + "content": "utilizing more advanced architecture (e.g., ResNet-v2-101 has higher accuracy than Inception-v3)", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 405, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 411, + 419 + ], + "score": 1.0, + "content": "and applying ensemble adversarial training (e.g., ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 411, + 406, + 422, + 415 + ], + "score": 0.5, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 405, + 506, + 419 + ], + "score": 1.0, + "content": "has higher accuracy", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 416, + 504, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 493, + 428 + ], + "score": 1.0, + "content": "than Inception-ResNet-v2). For the best defense model that we have, ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 493, + 416, + 504, + 426 + ], + "score": 0.37, + "content": "\\nu 2", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 427, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 115, + 437 + ], + "score": 0.7, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 427, + 329, + 441 + ], + "score": 1.0, + "content": "randomization layers reaches the top-1 accuracy of", + "type": "text" + }, + { + "bbox": [ + 329, + 427, + 357, + 438 + ], + "score": 0.89, + "content": "9 3 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 427, + 433, + 441 + ], + "score": 1.0, + "content": "on DeepFool and", + "type": "text" + }, + { + "bbox": [ + 434, + 427, + 461, + 438 + ], + "score": 0.88, + "content": "8 6 . 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 427, + 506, + 441 + ], + "score": 1.0, + "content": "on C&W,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 438, + 159, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 159, + 452 + ], + "score": 1.0, + "content": "respectively.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 108, + 469, + 245, + 480 + ], + "lines": [ + { + "bbox": [ + 106, + 469, + 247, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 247, + 481 + ], + "score": 1.0, + "content": "4.6 DIAGNOSTIC EXPERIMENT", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 559 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "score": 1.0, + "content": "Due to the large amount of possible patterns introduced by randomization layers, it is hard to analyze", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "the effectiveness of random resizing and random padding precisely. In this section, we limit the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "score": 1.0, + "content": "freedom of randomization to be a small number (i.e., 4 in random padding and 1 in random resizing)", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 525, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 538 + ], + "score": 1.0, + "content": "and analyze the effectiveness of these two operations separately. The same 500 images in section", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 537, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 505, + 549 + ], + "score": 1.0, + "content": "4.5 are used in this experiment. In addition, the input images for target models and defense models", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 548, + 313, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 548, + 202, + 559 + ], + "score": 1.0, + "content": "are resized to the shape", + "type": "text" + }, + { + "bbox": [ + 202, + 548, + 263, + 558 + ], + "score": 0.92, + "content": "3 3 0 \\times 3 3 0 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 548, + 313, + 559 + ], + "score": 1.0, + "content": "beforehand.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5 + }, + { + "type": "title", + "bbox": [ + 108, + 578, + 230, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 232, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 232, + 591 + ], + "score": 1.0, + "content": "4.6.1 ONE PIXEL PADDING", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 599, + 504, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 449, + 613 + ], + "score": 1.0, + "content": "For the random padding, there are only 4 patterns when padding the input images from", + "type": "text" + }, + { + "bbox": [ + 449, + 600, + 504, + 611 + ], + "score": 0.89, + "content": "3 3 0 \\times 3 3 0 \\times 3", + "type": "inline_equation" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 610, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 116, + 624 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 611, + 170, + 622 + ], + "score": 0.9, + "content": "3 3 1 \\times 3 3 1 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 610, + 506, + 624 + ], + "score": 1.0, + "content": ". In order to construct a stronger attack, we follow the experiment setup in section 4.5", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "where 3 chosen patterns are ensembled. Specifically, the target model takes an ensemble of patterns", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "where the original images are at the top left, top right and bottom left (3 patterns) of the padded", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "score": 1.0, + "content": "images, and the defense model takes the last pattern where the original images are at the bottom", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 655, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 506, + 667 + ], + "score": 1.0, + "content": "right of the padded images. Note that, since there is no randomization in the defense model, we", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "only run the defense model once. Table 5 summaries the results, and we can see that: (1) adversarial", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "examples generated by single-step attacks have strong transferability, but still cannot attack the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "defense model with an adversarially trained model successfully (i.e., the defense model with ens-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 197, + 711 + ], + "score": 1.0, + "content": "adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 197, + 700, + 208, + 709 + ], + "score": 0.29, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "); (2) adversarial examples generated by iterative attacks are much less", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "transferable between different padding patterns even when only 4 different patterns exist. The results", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 720, + 488, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 488, + 733 + ], + "score": 1.0, + "content": "demonstrate that creating different padding patterns can effectively mitigate adversarial effects.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 36.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 143, + 505, + 258 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 505, + 142 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 90, + 505, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 90, + 505, + 101 + ], + "score": 1.0, + "content": "Table 4: Top-1 classification accuracy under the ensemble-pattern attack scenario. Similar to vanilla", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "score": 1.0, + "content": "attack and single-pattern attack scenarios, we see that randomization layers increase the accuracy", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 111, + 504, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 504, + 124 + ], + "score": 1.0, + "content": "under all attacks and networks. This clearly demonstrates the effectiveness of the proposed ran-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 121, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 506, + 135 + ], + "score": 1.0, + "content": "domization method on defending against adversarial examples, even under this very strong attack", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 134, + 144, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 134, + 144, + 145 + ], + "score": 1.0, + "content": "scenario.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 106, + 143, + 505, + 258 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 143, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 258 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-237.3%41.2%39.2%44.9%71.5%74.3%86.2%88.9%
FGSM-531.7%34.0%24.6%29.7%65.2%67.3%85.8%87.5%
FGSM-1030.4%32.8%18.6%21.7%62.9%64.5%86.6%87.9%
DeepFool0.6%81.3%0.9%80.5%0.9%69.4%1.6%93.5%
C&W0.6%62.9%1.0%74.3%1.6%68.3%5.8%86.1%
", + "type": "table", + "image_path": "a5f5f049a290593d5c99320c3903b6b4ea7a50d09c509f730261fde2a47ca05a.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 143, + 505, + 181.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 181.33333333333334, + 505, + 219.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 219.66666666666669, + 505, + 258.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "text", + "bbox": [ + 106, + 284, + 505, + 449 + ], + "lines": [], + "index": 15, + "bbox_fs": [ + 105, + 285, + 506, + 452 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 469, + 245, + 480 + ], + "lines": [ + { + "bbox": [ + 106, + 469, + 247, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 247, + 481 + ], + "score": 1.0, + "content": "4.6 DIAGNOSTIC EXPERIMENT", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 559 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "score": 1.0, + "content": "Due to the large amount of possible patterns introduced by randomization layers, it is hard to analyze", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "the effectiveness of random resizing and random padding precisely. In this section, we limit the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "score": 1.0, + "content": "freedom of randomization to be a small number (i.e., 4 in random padding and 1 in random resizing)", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 525, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 538 + ], + "score": 1.0, + "content": "and analyze the effectiveness of these two operations separately. The same 500 images in section", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 537, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 505, + 549 + ], + "score": 1.0, + "content": "4.5 are used in this experiment. In addition, the input images for target models and defense models", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 548, + 313, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 548, + 202, + 559 + ], + "score": 1.0, + "content": "are resized to the shape", + "type": "text" + }, + { + "bbox": [ + 202, + 548, + 263, + 558 + ], + "score": 0.92, + "content": "3 3 0 \\times 3 3 0 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 548, + 313, + 559 + ], + "score": 1.0, + "content": "beforehand.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 492, + 506, + 559 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 578, + 230, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 232, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 232, + 591 + ], + "score": 1.0, + "content": "4.6.1 ONE PIXEL PADDING", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 599, + 504, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 449, + 613 + ], + "score": 1.0, + "content": "For the random padding, there are only 4 patterns when padding the input images from", + "type": "text" + }, + { + "bbox": [ + 449, + 600, + 504, + 611 + ], + "score": 0.89, + "content": "3 3 0 \\times 3 3 0 \\times 3", + "type": "inline_equation" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 610, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 116, + 624 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 611, + 170, + 622 + ], + "score": 0.9, + "content": "3 3 1 \\times 3 3 1 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 610, + 506, + 624 + ], + "score": 1.0, + "content": ". In order to construct a stronger attack, we follow the experiment setup in section 4.5", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "where 3 chosen patterns are ensembled. Specifically, the target model takes an ensemble of patterns", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "where the original images are at the top left, top right and bottom left (3 patterns) of the padded", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "score": 1.0, + "content": "images, and the defense model takes the last pattern where the original images are at the bottom", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 655, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 506, + 667 + ], + "score": 1.0, + "content": "right of the padded images. Note that, since there is no randomization in the defense model, we", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "only run the defense model once. Table 5 summaries the results, and we can see that: (1) adversarial", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "examples generated by single-step attacks have strong transferability, but still cannot attack the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "defense model with an adversarially trained model successfully (i.e., the defense model with ens-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 197, + 711 + ], + "score": 1.0, + "content": "adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 197, + 700, + 208, + 709 + ], + "score": 0.29, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "); (2) adversarial examples generated by iterative attacks are much less", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "transferable between different padding patterns even when only 4 different patterns exist. The results", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 720, + 488, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 488, + 733 + ], + "score": 1.0, + "content": "demonstrate that creating different padding patterns can effectively mitigate adversarial effects.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 599, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 112, + 505, + 227 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 502, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 88, + 504, + 102 + ], + "spans": [ + { + "bbox": [ + 106, + 88, + 504, + 102 + ], + "score": 1.0, + "content": "Table 5: Top-1 classification accuracy under one pixel padding scenario. This table shows that cre-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 100, + 500, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 500, + 113 + ], + "score": 1.0, + "content": "ating different padding patterns (even 1-pixel padding) can effectively mitigate adversarial effects.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 112, + 505, + 227 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 112, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 227 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-236.4%39.6%29.8%34.4%71.3%74.0%88.2%94.8%
FGSM-533.5%36.2%22.2%26.2%68.4%71.0%92.1%94.4%
FGSM-1034.5%38.8%21.3%23.6%67.4%70.4%93.7%94.0%
DeepFool0.9%97.2%0.9%95.2%0.9%87.6%1.5%99.2%
C&W0.8%70.2%0.9%76.8%1.0%79.4%2.4%98.2%
", + "type": "table", + "image_path": "ddbf82dab98d11b4ec804e9536e64fc830a6f8fc88e592a97f1d5f2ea68f8304.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 106, + 112, + 505, + 150.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 150.33333333333334, + 505, + 188.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 188.66666666666669, + 505, + 227.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "table", + "bbox": [ + 106, + 271, + 505, + 386 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 248, + 504, + 270 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 247, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 505, + 260 + ], + "score": 1.0, + "content": "Table 6: Top-1 classification accuracy under one pixel resizing scenario. This table shows that", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 260, + 497, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 497, + 271 + ], + "score": 1.0, + "content": "resizing image to a different scale (even 1-pixel scale) can effectively mitigate adversarial effects.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 271, + 505, + 386 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 271, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 386 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-230.8%56.2%31.6%44.6%66.2%75.0%87.6%97.2%
FGSM-531.2%48.8%25.6%35.8%61.4%70.2%91.2%96.6%
FGSM-1036.4%51.0%23.8%32.6%62.8%68.2%94.8%95.2%
DeepFool2.6%99.4%1.0%98.6%1.2%97.4%1.2%99.4%
C&W2.6%97.8%1.0%94.8%2.0%94.8%1.8%99.6%
", + "type": "table", + "image_path": "db4fc58373e021e6bbc56ab326d0e43d13af8c9ff40303a58ff718a293bda7bd.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 106, + 271, + 505, + 309.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 106, + 309.3333333333333, + 505, + 347.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 106, + 347.66666666666663, + 505, + 385.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "title", + "bbox": [ + 107, + 407, + 234, + 419 + ], + "lines": [ + { + "bbox": [ + 106, + 407, + 235, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 235, + 420 + ], + "score": 1.0, + "content": "4.6.2 ONE-PIXEL RESIZING", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 427, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 427, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 505, + 439 + ], + "score": 1.0, + "content": "For the random resizing, there is only 1 pattern that exists when the input images are resized from", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 107, + 438, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 107, + 438, + 168, + 449 + ], + "score": 0.91, + "content": "3 3 0 \\times 3 3 0 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 438, + 180, + 451 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 181, + 438, + 243, + 449 + ], + "score": 0.92, + "content": "3 3 1 \\times 3 3 1 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 438, + 505, + 451 + ], + "score": 1.0, + "content": ". The results in Table 6 indicate that resizing the images by only", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 449, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 505, + 462 + ], + "score": 1.0, + "content": "1 pixel can effectively destroy the transferability of adversarial examples by both single-step and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 460, + 174, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 174, + 472 + ], + "score": 1.0, + "content": "iterative attacks.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 107, + 488, + 437, + 502 + ], + "lines": [ + { + "bbox": [ + 105, + 489, + 439, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 439, + 504 + ], + "score": 1.0, + "content": "5 NIPS 2017 ADVERSARIAL EXAMPLES DEFENSE CHALLENGE", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 505, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 513, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 506, + 528 + ], + "score": 1.0, + "content": "We submitted our model to the NIPS 2017 adversarial examples defense challenge3 for a more", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 525, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 538 + ], + "score": 1.0, + "content": "comprehensive performance evaluation. The test dataset contains 5000 images which are all of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 537, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 124, + 550 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 125, + 537, + 185, + 548 + ], + "score": 0.91, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 537, + 505, + 550 + ], + "score": 1.0, + "content": ", and their corresponding labels are the same as the ImageNet 1000-class labels.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 548, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 548, + 505, + 561 + ], + "score": 1.0, + "content": "Each defense method are run on all 5000 adversarial images generated against all adversarial attacks.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 559, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 506, + 572 + ], + "score": 1.0, + "content": "For each correctly classified image, the defense method gets one point. The normalized score for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 570, + 359, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 359, + 582 + ], + "score": 1.0, + "content": "each defense method is computed using the following formula:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 191, + 588, + 417, + 623 + ], + "lines": [ + { + "bbox": [ + 191, + 588, + 417, + 623 + ], + "spans": [ + { + "bbox": [ + 191, + 588, + 417, + 623 + ], + "score": 0.93, + "content": "\\mathrm { s c o r e } = \\frac { 1 } { M } \\sum _ { \\mathrm { a t t a c k } \\in A } \\sum _ { n = 1 } ^ { 5 0 0 0 } \\left[ \\mathrm { d e f e n s e } ( \\mathrm { a t t a c k } ( X _ { n } ) ) = y _ { n } ^ { \\mathrm { t r u e } } \\right] ,", + "type": "interline_equation", + "image_path": "bb66dddbf90ac17217662fd9866c1a815b58375accf48ddf4f55b248371231ed.jpg" + } + ] + } + ], + "index": 22.5, + "virtual_lines": [ + { + "bbox": [ + 191, + 588, + 417, + 605.5 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 191, + 605.5, + 417, + 623.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 628, + 505, + 652 + ], + "lines": [ + { + "bbox": [ + 106, + 629, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 134, + 642 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 630, + 143, + 639 + ], + "score": 0.8, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 629, + 243, + 642 + ], + "score": 1.0, + "content": "is the set of all attacks,", + "type": "text" + }, + { + "bbox": [ + 244, + 629, + 255, + 639 + ], + "score": 0.82, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 629, + 505, + 642 + ], + "score": 1.0, + "content": "is the total number of generated adversarial examples by all", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 639, + 498, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 207, + 653 + ], + "score": 1.0, + "content": "attacks, and the function", + "type": "text" + }, + { + "bbox": [ + 207, + 640, + 216, + 652 + ], + "score": 0.71, + "content": "[ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 639, + 498, + 653 + ], + "score": 1.0, + "content": "is the indicator function which equals to 1 when the prediction is true.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 107, + 666, + 227, + 678 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 227, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 227, + 680 + ], + "score": 1.0, + "content": "5.1 CHALLENGE RESULTS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 108, + 687, + 504, + 711 + ], + "lines": [ + { + "bbox": [ + 106, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 686, + 388, + 700 + ], + "score": 1.0, + "content": "The best defense model in our experiments, i.e., randomization layers", + "type": "text" + }, + { + "bbox": [ + 389, + 690, + 397, + 698 + ], + "score": 0.73, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "ens-adv-Inception-Resnet-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 117, + 709 + ], + "score": 0.7, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 698, + 505, + 712 + ], + "score": 1.0, + "content": ", was submitted to the challenge. To increase the classification accuracy, we (1) changed the", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 721, + 382, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 719, + 383, + 734 + ], + "spans": [ + { + "bbox": [ + 119, + 719, + 383, + 734 + ], + "score": 1.0, + "content": "3https://www.kaggle.com/c/nips-2017-defense-against-adversarial-attack", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 112, + 505, + 227 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 502, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 88, + 504, + 102 + ], + "spans": [ + { + "bbox": [ + 106, + 88, + 504, + 102 + ], + "score": 1.0, + "content": "Table 5: Top-1 classification accuracy under one pixel padding scenario. This table shows that cre-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 100, + 500, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 500, + 113 + ], + "score": 1.0, + "content": "ating different padding patterns (even 1-pixel padding) can effectively mitigate adversarial effects.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 112, + 505, + 227 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 112, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 227 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-236.4%39.6%29.8%34.4%71.3%74.0%88.2%94.8%
FGSM-533.5%36.2%22.2%26.2%68.4%71.0%92.1%94.4%
FGSM-1034.5%38.8%21.3%23.6%67.4%70.4%93.7%94.0%
DeepFool0.9%97.2%0.9%95.2%0.9%87.6%1.5%99.2%
C&W0.8%70.2%0.9%76.8%1.0%79.4%2.4%98.2%
", + "type": "table", + "image_path": "ddbf82dab98d11b4ec804e9536e64fc830a6f8fc88e592a97f1d5f2ea68f8304.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 106, + 112, + 505, + 150.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 150.33333333333334, + 505, + 188.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 188.66666666666669, + 505, + 227.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "table", + "bbox": [ + 106, + 271, + 505, + 386 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 248, + 504, + 270 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 247, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 505, + 260 + ], + "score": 1.0, + "content": "Table 6: Top-1 classification accuracy under one pixel resizing scenario. This table shows that", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 260, + 497, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 497, + 271 + ], + "score": 1.0, + "content": "resizing image to a different scale (even 1-pixel scale) can effectively mitigate adversarial effects.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 271, + 505, + 386 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 271, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 386 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
targetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodeltargetmodeldefensemodel
FGSM-230.8%56.2%31.6%44.6%66.2%75.0%87.6%97.2%
FGSM-531.2%48.8%25.6%35.8%61.4%70.2%91.2%96.6%
FGSM-1036.4%51.0%23.8%32.6%62.8%68.2%94.8%95.2%
DeepFool2.6%99.4%1.0%98.6%1.2%97.4%1.2%99.4%
C&W2.6%97.8%1.0%94.8%2.0%94.8%1.8%99.6%
", + "type": "table", + "image_path": "db4fc58373e021e6bbc56ab326d0e43d13af8c9ff40303a58ff718a293bda7bd.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 106, + 271, + 505, + 309.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 106, + 309.3333333333333, + 505, + 347.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 106, + 347.66666666666663, + 505, + 385.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "title", + "bbox": [ + 107, + 407, + 234, + 419 + ], + "lines": [ + { + "bbox": [ + 106, + 407, + 235, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 235, + 420 + ], + "score": 1.0, + "content": "4.6.2 ONE-PIXEL RESIZING", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 427, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 427, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 505, + 439 + ], + "score": 1.0, + "content": "For the random resizing, there is only 1 pattern that exists when the input images are resized from", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 107, + 438, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 107, + 438, + 168, + 449 + ], + "score": 0.91, + "content": "3 3 0 \\times 3 3 0 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 438, + 180, + 451 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 181, + 438, + 243, + 449 + ], + "score": 0.92, + "content": "3 3 1 \\times 3 3 1 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 438, + 505, + 451 + ], + "score": 1.0, + "content": ". The results in Table 6 indicate that resizing the images by only", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 449, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 505, + 462 + ], + "score": 1.0, + "content": "1 pixel can effectively destroy the transferability of adversarial examples by both single-step and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 460, + 174, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 174, + 472 + ], + "score": 1.0, + "content": "iterative attacks.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 427, + 505, + 472 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 488, + 437, + 502 + ], + "lines": [ + { + "bbox": [ + 105, + 489, + 439, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 439, + 504 + ], + "score": 1.0, + "content": "5 NIPS 2017 ADVERSARIAL EXAMPLES DEFENSE CHALLENGE", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 505, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 513, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 506, + 528 + ], + "score": 1.0, + "content": "We submitted our model to the NIPS 2017 adversarial examples defense challenge3 for a more", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 525, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 538 + ], + "score": 1.0, + "content": "comprehensive performance evaluation. The test dataset contains 5000 images which are all of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 537, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 124, + 550 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 125, + 537, + 185, + 548 + ], + "score": 0.91, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 537, + 505, + 550 + ], + "score": 1.0, + "content": ", and their corresponding labels are the same as the ImageNet 1000-class labels.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 548, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 548, + 505, + 561 + ], + "score": 1.0, + "content": "Each defense method are run on all 5000 adversarial images generated against all adversarial attacks.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 559, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 506, + 572 + ], + "score": 1.0, + "content": "For each correctly classified image, the defense method gets one point. The normalized score for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 570, + 359, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 359, + 582 + ], + "score": 1.0, + "content": "each defense method is computed using the following formula:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 513, + 506, + 582 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 191, + 588, + 417, + 623 + ], + "lines": [ + { + "bbox": [ + 191, + 588, + 417, + 623 + ], + "spans": [ + { + "bbox": [ + 191, + 588, + 417, + 623 + ], + "score": 0.93, + "content": "\\mathrm { s c o r e } = \\frac { 1 } { M } \\sum _ { \\mathrm { a t t a c k } \\in A } \\sum _ { n = 1 } ^ { 5 0 0 0 } \\left[ \\mathrm { d e f e n s e } ( \\mathrm { a t t a c k } ( X _ { n } ) ) = y _ { n } ^ { \\mathrm { t r u e } } \\right] ,", + "type": "interline_equation", + "image_path": "bb66dddbf90ac17217662fd9866c1a815b58375accf48ddf4f55b248371231ed.jpg" + } + ] + } + ], + "index": 22.5, + "virtual_lines": [ + { + "bbox": [ + 191, + 588, + 417, + 605.5 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 191, + 605.5, + 417, + 623.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 628, + 505, + 652 + ], + "lines": [ + { + "bbox": [ + 106, + 629, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 134, + 642 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 630, + 143, + 639 + ], + "score": 0.8, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 629, + 243, + 642 + ], + "score": 1.0, + "content": "is the set of all attacks,", + "type": "text" + }, + { + "bbox": [ + 244, + 629, + 255, + 639 + ], + "score": 0.82, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 629, + 505, + 642 + ], + "score": 1.0, + "content": "is the total number of generated adversarial examples by all", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 639, + 498, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 207, + 653 + ], + "score": 1.0, + "content": "attacks, and the function", + "type": "text" + }, + { + "bbox": [ + 207, + 640, + 216, + 652 + ], + "score": 0.71, + "content": "[ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 639, + 498, + 653 + ], + "score": 1.0, + "content": "is the indicator function which equals to 1 when the prediction is true.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 629, + 505, + 653 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 666, + 227, + 678 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 227, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 227, + 680 + ], + "score": 1.0, + "content": "5.1 CHALLENGE RESULTS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 108, + 687, + 504, + 711 + ], + "lines": [ + { + "bbox": [ + 106, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 686, + 388, + 700 + ], + "score": 1.0, + "content": "The best defense model in our experiments, i.e., randomization layers", + "type": "text" + }, + { + "bbox": [ + 389, + 690, + 397, + 698 + ], + "score": 0.73, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "ens-adv-Inception-Resnet-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 117, + 709 + ], + "score": 0.7, + "content": "\\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 698, + 505, + 712 + ], + "score": 1.0, + "content": ", was submitted to the challenge. To increase the classification accuracy, we (1) changed the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "resizing range from [299, 331) to [310, 331); (2) averaged the prediction results over 30 randomiza-", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "tion patterns for each image; (3) flipped the input image with probability 0.5 for each randomization", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 105, + 139, + 117 + ], + "spans": [ + { + "bbox": [ + 104, + 105, + 139, + 117 + ], + "score": 1.0, + "content": "pattern.", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 686, + 505, + 712 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 503, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "resizing range from [299, 331) to [310, 331); (2) averaged the prediction results over 30 randomiza-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "tion patterns for each image; (3) flipped the input image with probability 0.5 for each randomization", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 105, + 139, + 117 + ], + "spans": [ + { + "bbox": [ + 104, + 105, + 139, + 117 + ], + "score": 1.0, + "content": "pattern.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 121, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "score": 1.0, + "content": "By evaluating our model against 156 different attacks, it reaches a normalized score of 0.924", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "(ranked No.2 among 107 defense models), which is far better than using ensemble adversarial train-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "ing (Tramer et al., 2017) alone with a normalized score of ` 0.773 (ranked No.56). This result further", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "score": 1.0, + "content": "demonstrates that the proposed randomization method effectively make deep networks much more", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 166, + 222, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 222, + 176 + ], + "score": 1.0, + "content": "robust to adversarial attacks.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "title", + "bbox": [ + 107, + 193, + 195, + 206 + ], + "lines": [ + { + "bbox": [ + 105, + 191, + 197, + 209 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 197, + 209 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 218, + 505, + 350 + ], + "lines": [ + { + "bbox": [ + 106, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "In this paper, we propose a randomization-based mechanism to mitigate adversarial effects. We", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 231, + 504, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 504, + 241 + ], + "score": 1.0, + "content": "conduct comprehensive experiments to validate the effectiveness of our defense method, using dif-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 240, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 505, + 253 + ], + "score": 1.0, + "content": "ferent network structures, against different attack methods, and under different attack scenarios.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 252, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 505, + 264 + ], + "score": 1.0, + "content": "The experimental results show that adversarial examples rarely transfer between different random-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 263, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 505, + 275 + ], + "score": 1.0, + "content": "ization patterns, especially for iterative attacks. In addition, the proposed randomization layers are", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 273, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 506, + 286 + ], + "score": 1.0, + "content": "compatible to different network structures and adversarial defense methods, which can serve as a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 284, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 297 + ], + "score": 1.0, + "content": "basic module for defense against adversarial examples. By adding the proposed randomization lay-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 295, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 506, + 307 + ], + "score": 1.0, + "content": "ers to an adversarially trained model (Tramer et al., 2017), it achieves a normalized score of ` 0.924", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 306, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 320 + ], + "score": 1.0, + "content": "(ranked No.2 among 107 defense models) in the NIPS 2017 adversarial examples defense challenge,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "which is far better than using adversarial training alone with a normalized score of 0.773 (ranked", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "score": 1.0, + "content": "No.56). The code is public available at https://github.com/cihangxie/NIPS2017_", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 340, + 237, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 237, + 351 + ], + "score": 1.0, + "content": "adv_challenge_defense.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 108, + 368, + 225, + 380 + ], + "lines": [ + { + "bbox": [ + 107, + 366, + 226, + 383 + ], + "spans": [ + { + "bbox": [ + 107, + 366, + 226, + 383 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENTS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 503, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 505, + 405 + ], + "score": 1.0, + "content": "This work is supported by a gift grant from SNAP Research, ONR–N00014-15-1-2356 and NSF", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 402, + 268, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 268, + 415 + ], + "score": 1.0, + "content": "Visual Cortex on Silicon CCF-1317560.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "title", + "bbox": [ + 108, + 432, + 175, + 444 + ], + "lines": [ + { + "bbox": [ + 106, + 432, + 176, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 176, + 446 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 451, + 503, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 504, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 504, + 463 + ], + "score": 1.0, + "content": "Battista Biggio and Pavel Laskov. Poisoning attacks against support vector machines. In Interna-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 462, + 306, + 474 + ], + "spans": [ + { + "bbox": [ + 115, + 462, + 306, + 474 + ], + "score": 1.0, + "content": "tional Conference on Machine Learning, 2012.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 481, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 505, + 495 + ], + "score": 1.0, + "content": "Xiaoyu Cao and Neil Zhenqiang Gong. Mitigating evasion attacks to deep neural networks via", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 493, + 504, + 506 + ], + "spans": [ + { + "bbox": [ + 116, + 493, + 504, + 506 + ], + "score": 1.0, + "content": "region-based classification. In Proceedings of the 33rd Annual Computer Security Applications", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 117, + 504, + 220, + 516 + ], + "spans": [ + { + "bbox": [ + 117, + 504, + 220, + 516 + ], + "score": 1.0, + "content": "Conference. ACM, 2017.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 523, + 504, + 546 + ], + "lines": [ + { + "bbox": [ + 106, + 523, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 506, + 536 + ], + "score": 1.0, + "content": "Nicholas Carlini and David Wagner. Towards evaluating the robustness of neural networks. In IEEE", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 535, + 316, + 547 + ], + "spans": [ + { + "bbox": [ + 116, + 535, + 316, + 547 + ], + "score": 1.0, + "content": "Symposium on Security and Privacy. IEEE, 2017.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 554, + 506, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "score": 1.0, + "content": "Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L Yuille.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 115, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 577, + 491, + 589 + ], + "spans": [ + { + "bbox": [ + 116, + 577, + 491, + 589 + ], + "score": 1.0, + "content": "fully connected crfs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 105, + 596, + 503, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "score": 1.0, + "content": "Moustapha Cisse, Yossi Adi, Natalia Neverova, and Joseph Keshet. Houdini: Fooling deep struc-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 608, + 380, + 619 + ], + "spans": [ + { + "bbox": [ + 116, + 608, + 380, + 619 + ], + "score": 1.0, + "content": "tured prediction models. arXiv preprint arXiv:1707.05373, 2017.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 105, + 627, + 504, + 650 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 505, + 641 + ], + "score": 1.0, + "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 638, + 467, + 651 + ], + "spans": [ + { + "bbox": [ + 115, + 638, + 467, + 651 + ], + "score": 1.0, + "content": "hierarchical image database. In Computer Vision and Pattern Recognition. IEEE, 2009.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 105, + 658, + 504, + 681 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "Reuben Feinman, Ryan R Curtin, Saurabh Shintre, and Andrew B Gardner. Detecting adversarial", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 669, + 374, + 682 + ], + "spans": [ + { + "bbox": [ + 115, + 669, + 374, + 682 + ], + "score": 1.0, + "content": "samples from artifacts. arXiv preprint arXiv:1703.00410, 2017.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 107, + 689, + 505, + 712 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 702 + ], + "score": 1.0, + "content": "Volker Fischer, Mummadi Chaithanya Kumar, Jan Hendrik Metzen, and Thomas Brox. Adversarial", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 700, + 457, + 713 + ], + "spans": [ + { + "bbox": [ + 116, + 700, + 457, + 713 + ], + "score": 1.0, + "content": "examples for semantic image segmentation. arXiv preprint arXiv:1703.01101, 2017.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 104, + 720, + 464, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 720, + 465, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 465, + 733 + ], + "score": 1.0, + "content": "Ross Girshick. Fast r-cnn. In International Conference on Computer Vision. IEEE, 2015.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 503, + 116 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 104, + 82, + 505, + 117 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 121, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "score": 1.0, + "content": "By evaluating our model against 156 different attacks, it reaches a normalized score of 0.924", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "(ranked No.2 among 107 defense models), which is far better than using ensemble adversarial train-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "ing (Tramer et al., 2017) alone with a normalized score of ` 0.773 (ranked No.56). This result further", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "score": 1.0, + "content": "demonstrates that the proposed randomization method effectively make deep networks much more", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 166, + 222, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 222, + 176 + ], + "score": 1.0, + "content": "robust to adversarial attacks.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 121, + 505, + 176 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 193, + 195, + 206 + ], + "lines": [ + { + "bbox": [ + 105, + 191, + 197, + 209 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 197, + 209 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 218, + 505, + 350 + ], + "lines": [ + { + "bbox": [ + 106, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "In this paper, we propose a randomization-based mechanism to mitigate adversarial effects. We", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 231, + 504, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 504, + 241 + ], + "score": 1.0, + "content": "conduct comprehensive experiments to validate the effectiveness of our defense method, using dif-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 240, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 505, + 253 + ], + "score": 1.0, + "content": "ferent network structures, against different attack methods, and under different attack scenarios.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 252, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 505, + 264 + ], + "score": 1.0, + "content": "The experimental results show that adversarial examples rarely transfer between different random-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 263, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 505, + 275 + ], + "score": 1.0, + "content": "ization patterns, especially for iterative attacks. In addition, the proposed randomization layers are", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 273, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 506, + 286 + ], + "score": 1.0, + "content": "compatible to different network structures and adversarial defense methods, which can serve as a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 284, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 297 + ], + "score": 1.0, + "content": "basic module for defense against adversarial examples. By adding the proposed randomization lay-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 295, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 506, + 307 + ], + "score": 1.0, + "content": "ers to an adversarially trained model (Tramer et al., 2017), it achieves a normalized score of ` 0.924", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 306, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 320 + ], + "score": 1.0, + "content": "(ranked No.2 among 107 defense models) in the NIPS 2017 adversarial examples defense challenge,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "which is far better than using adversarial training alone with a normalized score of 0.773 (ranked", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "score": 1.0, + "content": "No.56). The code is public available at https://github.com/cihangxie/NIPS2017_", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 340, + 237, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 237, + 351 + ], + "score": 1.0, + "content": "adv_challenge_defense.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 219, + 506, + 351 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 368, + 225, + 380 + ], + "lines": [ + { + "bbox": [ + 107, + 366, + 226, + 383 + ], + "spans": [ + { + "bbox": [ + 107, + 366, + 226, + 383 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENTS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 503, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 505, + 405 + ], + "score": 1.0, + "content": "This work is supported by a gift grant from SNAP Research, ONR–N00014-15-1-2356 and NSF", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 402, + 268, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 268, + 415 + ], + "score": 1.0, + "content": "Visual Cortex on Silicon CCF-1317560.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 392, + 505, + 415 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 432, + 175, + 444 + ], + "lines": [ + { + "bbox": [ + 106, + 432, + 176, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 176, + 446 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 451, + 503, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 504, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 504, + 463 + ], + "score": 1.0, + "content": "Battista Biggio and Pavel Laskov. Poisoning attacks against support vector machines. In Interna-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 462, + 306, + 474 + ], + "spans": [ + { + "bbox": [ + 115, + 462, + 306, + 474 + ], + "score": 1.0, + "content": "tional Conference on Machine Learning, 2012.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 450, + 504, + 474 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 481, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 505, + 495 + ], + "score": 1.0, + "content": "Xiaoyu Cao and Neil Zhenqiang Gong. Mitigating evasion attacks to deep neural networks via", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 493, + 504, + 506 + ], + "spans": [ + { + "bbox": [ + 116, + 493, + 504, + 506 + ], + "score": 1.0, + "content": "region-based classification. In Proceedings of the 33rd Annual Computer Security Applications", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 117, + 504, + 220, + 516 + ], + "spans": [ + { + "bbox": [ + 117, + 504, + 220, + 516 + ], + "score": 1.0, + "content": "Conference. ACM, 2017.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 481, + 505, + 516 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 523, + 504, + 546 + ], + "lines": [ + { + "bbox": [ + 106, + 523, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 506, + 536 + ], + "score": 1.0, + "content": "Nicholas Carlini and David Wagner. Towards evaluating the robustness of neural networks. In IEEE", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 535, + 316, + 547 + ], + "spans": [ + { + "bbox": [ + 116, + 535, + 316, + 547 + ], + "score": 1.0, + "content": "Symposium on Security and Privacy. IEEE, 2017.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 106, + 523, + 506, + 547 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 554, + 506, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 567 + ], + "score": 1.0, + "content": "Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L Yuille.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 115, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 577, + 491, + 589 + ], + "spans": [ + { + "bbox": [ + 116, + 577, + 491, + 589 + ], + "score": 1.0, + "content": "fully connected crfs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 555, + 506, + 589 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 596, + 503, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "score": 1.0, + "content": "Moustapha Cisse, Yossi Adi, Natalia Neverova, and Joseph Keshet. Houdini: Fooling deep struc-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 608, + 380, + 619 + ], + "spans": [ + { + "bbox": [ + 116, + 608, + 380, + 619 + ], + "score": 1.0, + "content": "tured prediction models. arXiv preprint arXiv:1707.05373, 2017.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 595, + 505, + 619 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 627, + 504, + 650 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 505, + 641 + ], + "score": 1.0, + "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 638, + 467, + 651 + ], + "spans": [ + { + "bbox": [ + 115, + 638, + 467, + 651 + ], + "score": 1.0, + "content": "hierarchical image database. In Computer Vision and Pattern Recognition. IEEE, 2009.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 628, + 505, + 651 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 658, + 504, + 681 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "Reuben Feinman, Ryan R Curtin, Saurabh Shintre, and Andrew B Gardner. Detecting adversarial", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 669, + 374, + 682 + ], + "spans": [ + { + "bbox": [ + 115, + 669, + 374, + 682 + ], + "score": 1.0, + "content": "samples from artifacts. arXiv preprint arXiv:1703.00410, 2017.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 106, + 658, + 505, + 682 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 689, + 505, + 712 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 702 + ], + "score": 1.0, + "content": "Volker Fischer, Mummadi Chaithanya Kumar, Jan Hendrik Metzen, and Thomas Brox. Adversarial", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 700, + 457, + 713 + ], + "spans": [ + { + "bbox": [ + 116, + 700, + 457, + 713 + ], + "score": 1.0, + "content": "examples for semantic image segmentation. arXiv preprint arXiv:1703.01101, 2017.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 106, + 688, + 505, + 713 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 720, + 464, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 720, + 465, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 465, + 733 + ], + "score": 1.0, + "content": "Ross Girshick. Fast r-cnn. In International Conference on Computer Vision. IEEE, 2015.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 720, + 465, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 94, + 420, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 420, + 106 + ], + "score": 1.0, + "content": "examples. In International Conference on Learning Representations, 2015.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 111, + 503, + 135 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 126 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 384, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 384, + 136 + ], + "score": 1.0, + "content": "nition. In Computer Vision and Pattern Recognition. IEEE, 2016a.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 105, + 141, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 505, + 154 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 505, + 154 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 412, + 165 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 412, + 165 + ], + "score": 1.0, + "content": "networks. In European Conference on Computer Vision. Springer, 2016b.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 171, + 505, + 194 + ], + "lines": [ + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "score": 1.0, + "content": "Pang Wei Koh and Percy Liang. Understanding black-box predictions via influence functions. arXiv", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 254, + 193 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 254, + 193 + ], + "score": 1.0, + "content": "preprint arXiv:1703.04730, 2017.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 504, + 224 + ], + "lines": [ + { + "bbox": [ + 106, + 200, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 505, + 213 + ], + "score": 1.0, + "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convo-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 211, + 468, + 224 + ], + "spans": [ + { + "bbox": [ + 116, + 211, + 468, + 224 + ], + "score": 1.0, + "content": "lutional neural networks. In Advances in Neural Information Processing Systems, 2012.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 229, + 504, + 253 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 505, + 243 + ], + "score": 1.0, + "content": "Alexey Kurakin, Ian Goodfellow, and Samy Bengio. Adversarial machine learning at scale. In", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 241, + 365, + 253 + ], + "spans": [ + { + "bbox": [ + 115, + 241, + 365, + 253 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 107, + 259, + 504, + 283 + ], + "lines": [ + { + "bbox": [ + 106, + 259, + 504, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 504, + 272 + ], + "score": 1.0, + "content": "Yanpei Liu, Xinyun Chen, Chang Liu, and Dawn Song. Delving into transferable adversarial exam-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 271, + 487, + 283 + ], + "spans": [ + { + "bbox": [ + 115, + 271, + 487, + 283 + ], + "score": 1.0, + "content": "ples and black-box attacks. In International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 105, + 289, + 504, + 312 + ], + "lines": [ + { + "bbox": [ + 106, + 290, + 504, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 504, + 302 + ], + "score": 1.0, + "content": "Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 300, + 410, + 312 + ], + "spans": [ + { + "bbox": [ + 115, + 300, + 410, + 312 + ], + "score": 1.0, + "content": "segmentation. In Computer Vision and Pattern Recognition. IEEE, 2015.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 318, + 505, + 341 + ], + "lines": [ + { + "bbox": [ + 106, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "Dongyu Meng and Hao Chen. Magnet: a two-pronged defense against adversarial examples. arXiv", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 330, + 254, + 341 + ], + "spans": [ + { + "bbox": [ + 115, + 330, + 254, + 341 + ], + "score": 1.0, + "content": "preprint arXiv:1705.09064, 2017.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 105, + 347, + 504, + 371 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 505, + 361 + ], + "score": 1.0, + "content": "Jan Hendrik Metzen, Tim Genewein, Volker Fischer, and Bastian Bischoff. On detecting adversarial", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 360, + 434, + 371 + ], + "spans": [ + { + "bbox": [ + 115, + 360, + 434, + 371 + ], + "score": 1.0, + "content": "perturbations. In International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 377, + 504, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "Seyed-Mohsen Moosavi-Dezfooli, Alhussein Fawzi, and Pascal Frossard. Deepfool: a simple and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 115, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "accurate method to fool deep neural networks. In Computer Vision and Pattern Recognition.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 400, + 168, + 411 + ], + "spans": [ + { + "bbox": [ + 116, + 400, + 168, + 411 + ], + "score": 1.0, + "content": "IEEE, 2016.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 104, + 417, + 504, + 442 + ], + "lines": [ + { + "bbox": [ + 106, + 418, + 504, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 504, + 430 + ], + "score": 1.0, + "content": "Nicolas Papernot, Ian Goodfellow, Ryan Sheatsley, Reuben Feinman, and Patrick McDaniel. clev-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 429, + 504, + 442 + ], + "spans": [ + { + "bbox": [ + 116, + 429, + 504, + 442 + ], + "score": 1.0, + "content": "erhans v1.0.0: an adversarial machine learning library. arXiv preprint arXiv:1610.00768, 2016a.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 504, + 481 + ], + "lines": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "Nicolas Papernot, Patrick McDaniel, Xi Wu, Somesh Jha, and Ananthram Swami. Distillation as", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 115, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "a defense to adversarial perturbations against deep neural networks. In IEEE Symposium on", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 117, + 470, + 261, + 482 + ], + "spans": [ + { + "bbox": [ + 117, + 470, + 261, + 482 + ], + "score": 1.0, + "content": "Security and Privacy. IEEE, 2016b.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 488, + 504, + 522 + ], + "lines": [ + { + "bbox": [ + 106, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 498, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 115, + 498, + 505, + 513 + ], + "score": 1.0, + "content": "detection with region proposal networks. In Advances in Neural Information Processing Systems,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 509, + 143, + 523 + ], + "spans": [ + { + "bbox": [ + 115, + 509, + 143, + 523 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 105, + 528, + 504, + 552 + ], + "lines": [ + { + "bbox": [ + 105, + 527, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 543 + ], + "score": 1.0, + "content": "Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 540, + 427, + 552 + ], + "spans": [ + { + "bbox": [ + 115, + 540, + 427, + 552 + ], + "score": 1.0, + "content": "recognition. In International Conference on Learning Representations, 2015.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 109, + 558, + 504, + 592 + ], + "lines": [ + { + "bbox": [ + 107, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 107, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfel-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 569, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 115, + 569, + 505, + 582 + ], + "score": 1.0, + "content": "low, and Rob Fergus. Intriguing properties of neural networks. In International Conference on", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 581, + 250, + 593 + ], + "spans": [ + { + "bbox": [ + 116, + 581, + 250, + 593 + ], + "score": 1.0, + "content": "Learning Representations, 2014.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 598, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 597, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 612 + ], + "score": 1.0, + "content": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 610, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 116, + 610, + 505, + 622 + ], + "score": 1.0, + "content": "the inception architecture for computer vision. In Computer Vision and Pattern Recognition.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 621, + 168, + 632 + ], + "spans": [ + { + "bbox": [ + 116, + 621, + 168, + 632 + ], + "score": 1.0, + "content": "IEEE, 2016.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 639, + 503, + 662 + ], + "lines": [ + { + "bbox": [ + 106, + 638, + 504, + 651 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 504, + 651 + ], + "score": 1.0, + "content": "Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, and Alexander A Alemi. Inception-v4,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 650, + 453, + 663 + ], + "spans": [ + { + "bbox": [ + 116, + 650, + 453, + 663 + ], + "score": 1.0, + "content": "inception-resnet and the impact of residual connections on learning. In AAAI, 2017.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 106, + 668, + 503, + 692 + ], + "lines": [ + { + "bbox": [ + 106, + 668, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 668, + 505, + 681 + ], + "score": 1.0, + "content": "Florian Tramer, Alexey Kurakin, Nicolas Papernot, Dan Boneh, and Patrick McDaniel. Ensemble `", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 680, + 452, + 692 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 452, + 692 + ], + "score": 1.0, + "content": "adversarial training: Attacks and defenses. arXiv preprint arXiv:1705.07204, 2017.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Jianyu Wang, Zhishuai Zhang, Cihang Xie, Yuyin Zhou, Vittal Premachandran, Jun Zhu, Lingxi Xie,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "and Alan Yuille. Visual concepts and compositional voting. arXiv preprint arXiv:1711.04451,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 719, + 143, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 719, + 143, + 732 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 94, + 420, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 420, + 106 + ], + "score": 1.0, + "content": "examples. In International Conference on Learning Representations, 2015.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 505, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 111, + 503, + 135 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 126 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 384, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 384, + 136 + ], + "score": 1.0, + "content": "nition. In Computer Vision and Pattern Recognition. IEEE, 2016a.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 110, + 505, + 136 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 141, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 505, + 154 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 505, + 154 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 412, + 165 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 412, + 165 + ], + "score": 1.0, + "content": "networks. In European Conference on Computer Vision. Springer, 2016b.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 141, + 505, + 165 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 171, + 505, + 194 + ], + "lines": [ + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 183 + ], + "score": 1.0, + "content": "Pang Wei Koh and Percy Liang. Understanding black-box predictions via influence functions. arXiv", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 254, + 193 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 254, + 193 + ], + "score": 1.0, + "content": "preprint arXiv:1703.04730, 2017.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 171, + 505, + 193 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 504, + 224 + ], + "lines": [ + { + "bbox": [ + 106, + 200, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 505, + 213 + ], + "score": 1.0, + "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convo-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 211, + 468, + 224 + ], + "spans": [ + { + "bbox": [ + 116, + 211, + 468, + 224 + ], + "score": 1.0, + "content": "lutional neural networks. In Advances in Neural Information Processing Systems, 2012.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 106, + 200, + 505, + 224 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 229, + 504, + 253 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 505, + 243 + ], + "score": 1.0, + "content": "Alexey Kurakin, Ian Goodfellow, and Samy Bengio. Adversarial machine learning at scale. In", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 241, + 365, + 253 + ], + "spans": [ + { + "bbox": [ + 115, + 241, + 365, + 253 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 106, + 230, + 505, + 253 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 259, + 504, + 283 + ], + "lines": [ + { + "bbox": [ + 106, + 259, + 504, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 504, + 272 + ], + "score": 1.0, + "content": "Yanpei Liu, Xinyun Chen, Chang Liu, and Dawn Song. Delving into transferable adversarial exam-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 271, + 487, + 283 + ], + "spans": [ + { + "bbox": [ + 115, + 271, + 487, + 283 + ], + "score": 1.0, + "content": "ples and black-box attacks. In International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 259, + 504, + 283 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 289, + 504, + 312 + ], + "lines": [ + { + "bbox": [ + 106, + 290, + 504, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 504, + 302 + ], + "score": 1.0, + "content": "Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 300, + 410, + 312 + ], + "spans": [ + { + "bbox": [ + 115, + 300, + 410, + 312 + ], + "score": 1.0, + "content": "segmentation. In Computer Vision and Pattern Recognition. IEEE, 2015.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 106, + 290, + 504, + 312 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 318, + 505, + 341 + ], + "lines": [ + { + "bbox": [ + 106, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "Dongyu Meng and Hao Chen. Magnet: a two-pronged defense against adversarial examples. arXiv", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 330, + 254, + 341 + ], + "spans": [ + { + "bbox": [ + 115, + 330, + 254, + 341 + ], + "score": 1.0, + "content": "preprint arXiv:1705.09064, 2017.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 106, + 318, + 505, + 341 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 347, + 504, + 371 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 505, + 361 + ], + "score": 1.0, + "content": "Jan Hendrik Metzen, Tim Genewein, Volker Fischer, and Bastian Bischoff. On detecting adversarial", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 360, + 434, + 371 + ], + "spans": [ + { + "bbox": [ + 115, + 360, + 434, + 371 + ], + "score": 1.0, + "content": "perturbations. In International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 106, + 348, + 505, + 371 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 377, + 504, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "Seyed-Mohsen Moosavi-Dezfooli, Alhussein Fawzi, and Pascal Frossard. Deepfool: a simple and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 115, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "accurate method to fool deep neural networks. In Computer Vision and Pattern Recognition.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 400, + 168, + 411 + ], + "spans": [ + { + "bbox": [ + 116, + 400, + 168, + 411 + ], + "score": 1.0, + "content": "IEEE, 2016.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 378, + 505, + 411 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 417, + 504, + 442 + ], + "lines": [ + { + "bbox": [ + 106, + 418, + 504, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 504, + 430 + ], + "score": 1.0, + "content": "Nicolas Papernot, Ian Goodfellow, Ryan Sheatsley, Reuben Feinman, and Patrick McDaniel. clev-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 429, + 504, + 442 + ], + "spans": [ + { + "bbox": [ + 116, + 429, + 504, + 442 + ], + "score": 1.0, + "content": "erhans v1.0.0: an adversarial machine learning library. arXiv preprint arXiv:1610.00768, 2016a.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 418, + 504, + 442 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 504, + 481 + ], + "lines": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "Nicolas Papernot, Patrick McDaniel, Xi Wu, Somesh Jha, and Ananthram Swami. Distillation as", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 115, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "a defense to adversarial perturbations against deep neural networks. In IEEE Symposium on", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 117, + 470, + 261, + 482 + ], + "spans": [ + { + "bbox": [ + 117, + 470, + 261, + 482 + ], + "score": 1.0, + "content": "Security and Privacy. IEEE, 2016b.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 447, + 505, + 482 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 488, + 504, + 522 + ], + "lines": [ + { + "bbox": [ + 106, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 498, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 115, + 498, + 505, + 513 + ], + "score": 1.0, + "content": "detection with region proposal networks. In Advances in Neural Information Processing Systems,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 509, + 143, + 523 + ], + "spans": [ + { + "bbox": [ + 115, + 509, + 143, + 523 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 488, + 506, + 523 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 528, + 504, + 552 + ], + "lines": [ + { + "bbox": [ + 105, + 527, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 543 + ], + "score": 1.0, + "content": "Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 540, + 427, + 552 + ], + "spans": [ + { + "bbox": [ + 115, + 540, + 427, + 552 + ], + "score": 1.0, + "content": "recognition. In International Conference on Learning Representations, 2015.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 527, + 505, + 552 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 558, + 504, + 592 + ], + "lines": [ + { + "bbox": [ + 107, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 107, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfel-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 569, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 115, + 569, + 505, + 582 + ], + "score": 1.0, + "content": "low, and Rob Fergus. Intriguing properties of neural networks. In International Conference on", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 581, + 250, + 593 + ], + "spans": [ + { + "bbox": [ + 116, + 581, + 250, + 593 + ], + "score": 1.0, + "content": "Learning Representations, 2014.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 107, + 558, + 505, + 593 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 598, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 597, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 612 + ], + "score": 1.0, + "content": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 610, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 116, + 610, + 505, + 622 + ], + "score": 1.0, + "content": "the inception architecture for computer vision. In Computer Vision and Pattern Recognition.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 621, + 168, + 632 + ], + "spans": [ + { + "bbox": [ + 116, + 621, + 168, + 632 + ], + "score": 1.0, + "content": "IEEE, 2016.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 597, + 506, + 632 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 639, + 503, + 662 + ], + "lines": [ + { + "bbox": [ + 106, + 638, + 504, + 651 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 504, + 651 + ], + "score": 1.0, + "content": "Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, and Alexander A Alemi. Inception-v4,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 650, + 453, + 663 + ], + "spans": [ + { + "bbox": [ + 116, + 650, + 453, + 663 + ], + "score": 1.0, + "content": "inception-resnet and the impact of residual connections on learning. In AAAI, 2017.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 106, + 638, + 504, + 663 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 668, + 503, + 692 + ], + "lines": [ + { + "bbox": [ + 106, + 668, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 668, + 505, + 681 + ], + "score": 1.0, + "content": "Florian Tramer, Alexey Kurakin, Nicolas Papernot, Dan Boneh, and Patrick McDaniel. Ensemble `", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 680, + 452, + 692 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 452, + 692 + ], + "score": 1.0, + "content": "adversarial training: Attacks and defenses. arXiv preprint arXiv:1705.07204, 2017.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 106, + 668, + 505, + 692 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Jianyu Wang, Zhishuai Zhang, Cihang Xie, Yuyin Zhou, Vittal Premachandran, Jun Zhu, Lingxi Xie,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "and Alan Yuille. Visual concepts and compositional voting. arXiv preprint arXiv:1711.04451,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 719, + 143, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 719, + 143, + 732 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 698, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 503, + 115 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Cihang Xie, Jianyu Wang, Zhishuai Zhang, Yuyin Zhou, Lingxi Xie, and Alan Yuille. Adversar-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "ial Examples for Semantic Segmentation and Object Detection. In International Conference on", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 117, + 105, + 241, + 117 + ], + "spans": [ + { + "bbox": [ + 117, + 105, + 241, + 117 + ], + "score": 1.0, + "content": "Computer Vision. IEEE, 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 505, + 136 + ], + "score": 1.0, + "content": "Zhishuai Zhang, Siyuan Qiao, Cihang Xie, Wei Shen, Bo Wang, and Alan L Yuille. Single-shot", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 134, + 448, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 134, + 448, + 146 + ], + "score": 1.0, + "content": "object detection with enriched semantics. arXiv preprint arXiv:1712.00433, 2017.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 503, + 115 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Cihang Xie, Jianyu Wang, Zhishuai Zhang, Yuyin Zhou, Lingxi Xie, and Alan Yuille. Adversar-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "ial Examples for Semantic Segmentation and Object Detection. In International Conference on", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 117, + 105, + 241, + 117 + ], + "spans": [ + { + "bbox": [ + 117, + 105, + 241, + 117 + ], + "score": 1.0, + "content": "Computer Vision. IEEE, 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 107, + 82, + 505, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 505, + 136 + ], + "score": 1.0, + "content": "Zhishuai Zhang, Siyuan Qiao, Cihang Xie, Wei Shen, Bo Wang, and Alan L Yuille. Single-shot", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 134, + 448, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 134, + 448, + 146 + ], + "score": 1.0, + "content": "object detection with enriched semantics. arXiv preprint arXiv:1712.00433, 2017.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 122, + 505, + 146 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 370, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 371, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 371, + 96 + ], + "score": 1.0, + "content": "APPENDIX A OTHER RANDOMIZATION METHODS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 105, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "Besides random resizing and random padding, we further evaluate the effectiveness of four other", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "score": 1.0, + "content": "randomization methods against adversarial examples. All these four methods are used as data-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 127, + 313, + 141 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 313, + 141 + ], + "score": 1.0, + "content": "augmentation during the standard network training.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 132, + 147, + 505, + 266 + ], + "lines": [ + { + "bbox": [ + 131, + 146, + 504, + 162 + ], + "spans": [ + { + "bbox": [ + 131, + 146, + 302, + 162 + ], + "score": 1.0, + "content": "• Random Brightness: a brightness factor", + "type": "text" + }, + { + "bbox": [ + 303, + 149, + 309, + 158 + ], + "score": 0.79, + "content": "\\delta", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 146, + 446, + 162 + ], + "score": 1.0, + "content": "is randomly picked in the interval", + "type": "text" + }, + { + "bbox": [ + 446, + 147, + 504, + 160 + ], + "score": 0.9, + "content": "[ - \\delta _ { \\mathrm { m a x } } , \\delta _ { \\mathrm { m a x } } ]", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 140, + 158, + 453, + 176 + ], + "spans": [ + { + "bbox": [ + 140, + 158, + 336, + 176 + ], + "score": 1.0, + "content": "to adjust the brightness of the normalized image", + "type": "text" + }, + { + "bbox": [ + 336, + 159, + 351, + 172 + ], + "score": 0.9, + "content": "\\hat { X _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 158, + 401, + 176 + ], + "score": 1.0, + "content": ". We choose", + "type": "text" + }, + { + "bbox": [ + 402, + 160, + 450, + 174 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\delta _ { \\mathrm { m a x } } = \\frac { 3 2 } { 2 5 5 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 165, + 453, + 176 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 132, + 174, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 132, + 174, + 329, + 188 + ], + "score": 1.0, + "content": "• Random Saturation: a saturation factor", + "type": "text" + }, + { + "bbox": [ + 330, + 177, + 338, + 185 + ], + "score": 0.77, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 174, + 505, + 188 + ], + "score": 1.0, + "content": "is randomly picked in the interval", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 143, + 186, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 143, + 187, + 208, + 200 + ], + "score": 0.89, + "content": "[ \\alpha _ { l o w e r } , \\alpha _ { u p p e r } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 186, + 400, + 202 + ], + "score": 1.0, + "content": "to adjust the saturation of the normalized image", + "type": "text" + }, + { + "bbox": [ + 401, + 186, + 415, + 199 + ], + "score": 0.9, + "content": "\\hat { X _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 186, + 465, + 202 + ], + "score": 1.0, + "content": ". We choose", + "type": "text" + }, + { + "bbox": [ + 465, + 189, + 505, + 199 + ], + "score": 0.83, + "content": "\\alpha _ { l o w e r } =", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 199, + 234, + 213 + ], + "spans": [ + { + "bbox": [ + 141, + 199, + 174, + 213 + ], + "score": 1.0, + "content": "0.5 and", + "type": "text" + }, + { + "bbox": [ + 175, + 200, + 229, + 211 + ], + "score": 0.89, + "content": "\\alpha _ { u p p e r } = 1 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 199, + 234, + 213 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 131, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 131, + 213, + 250, + 226 + ], + "score": 1.0, + "content": "• Random Hue: a hue factor", + "type": "text" + }, + { + "bbox": [ + 250, + 214, + 257, + 223 + ], + "score": 0.8, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 213, + 394, + 226 + ], + "score": 1.0, + "content": "is randomly picked in the interval", + "type": "text" + }, + { + "bbox": [ + 394, + 213, + 452, + 225 + ], + "score": 0.92, + "content": "[ - \\theta _ { \\mathrm { m a x } } , \\theta _ { \\mathrm { m a x } } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "to adjust the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 224, + 375, + 239 + ], + "spans": [ + { + "bbox": [ + 141, + 225, + 259, + 239 + ], + "score": 1.0, + "content": "hue of the normalized image", + "type": "text" + }, + { + "bbox": [ + 259, + 224, + 273, + 237 + ], + "score": 0.91, + "content": "{ \\hat { X _ { n } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 225, + 324, + 239 + ], + "score": 1.0, + "content": ". We choose", + "type": "text" + }, + { + "bbox": [ + 324, + 226, + 371, + 237 + ], + "score": 0.91, + "content": "\\theta _ { \\mathrm { m a x } } = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 225, + 375, + 239 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 131, + 237, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 131, + 237, + 284, + 254 + ], + "score": 1.0, + "content": "• Random Contrast: a contrast factor", + "type": "text" + }, + { + "bbox": [ + 285, + 240, + 292, + 252 + ], + "score": 0.85, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 237, + 430, + 254 + ], + "score": 1.0, + "content": "is randomly picked in the interval", + "type": "text" + }, + { + "bbox": [ + 430, + 240, + 493, + 253 + ], + "score": 0.91, + "content": "[ \\beta _ { l o w e r } , \\beta _ { u p p e r } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 237, + 506, + 254 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 252, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 141, + 252, + 313, + 267 + ], + "score": 1.0, + "content": "adjust the contrast of the normalized image", + "type": "text" + }, + { + "bbox": [ + 313, + 252, + 327, + 264 + ], + "score": 0.9, + "content": "\\hat { X _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 252, + 376, + 267 + ], + "score": 1.0, + "content": ". We choose", + "type": "text" + }, + { + "bbox": [ + 377, + 253, + 430, + 265 + ], + "score": 0.93, + "content": "\\beta _ { l o w e r } = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 252, + 447, + 267 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 447, + 254, + 501, + 266 + ], + "score": 0.89, + "content": "\\beta _ { u p p e r } = 1 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 252, + 506, + 267 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 108, + 274, + 505, + 309 + ], + "lines": [ + { + "bbox": [ + 105, + 273, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 505, + 287 + ], + "score": 1.0, + "content": "Note that, (1) the parameters chosen above are the same as the ones used during network training", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 285, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 325, + 299 + ], + "score": 1.0, + "content": "process4; (2) the pixel value of the normalized image", + "type": "text" + }, + { + "bbox": [ + 326, + 285, + 340, + 298 + ], + "score": 0.91, + "content": "\\hat { X _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 286, + 448, + 299 + ], + "score": 1.0, + "content": "are all within the interval", + "type": "text" + }, + { + "bbox": [ + 448, + 286, + 469, + 299 + ], + "score": 0.74, + "content": "[ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 286, + 505, + 299 + ], + "score": 1.0, + "content": ", and we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 297, + 407, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 407, + 311 + ], + "score": 1.0, + "content": "also use this range to clip the pixel value of the image after pre-processing.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 314, + 505, + 380 + ], + "lines": [ + { + "bbox": [ + 105, + 314, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 505, + 327 + ], + "score": 1.0, + "content": "Following the experiment setup in section 4, we first evaluate the effectiveness of each of these", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 326, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 505, + 337 + ], + "score": 1.0, + "content": "randomization methods on the 5000 clean images. The results are shown in the Table 7. We can see", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 335, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 349 + ], + "score": 1.0, + "content": "that these methods hardly hurt the performance on clean images. We further combine the proposed", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "score": 1.0, + "content": "randomization layers, i.e., random resizing layer and random padding layer, with each of these", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 252, + 371 + ], + "score": 1.0, + "content": "randomization methods (denoted as", + "type": "text" + }, + { + "bbox": [ + 252, + 359, + 274, + 369 + ], + "score": 0.83, + "content": "^ { 6 6 } + + ^ { 7 }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 358, + 505, + 371 + ], + "score": 1.0, + "content": "). We see that the combined randomization modules only", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 369, + 299, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 299, + 382 + ], + "score": 1.0, + "content": "cause very little accuracy drop on clean images.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18.5 + }, + { + "type": "table", + "bbox": [ + 108, + 443, + 502, + 571 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 398, + 505, + 443 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "score": 1.0, + "content": "Table 7: Top-1 classification accuracy on clean images. We see that these four randomization meth-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 410, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 352, + 422 + ], + "score": 1.0, + "content": "ods hardly hurt the performance on clean images. We use", + "type": "text" + }, + { + "bbox": [ + 352, + 410, + 376, + 420 + ], + "score": 0.86, + "content": "^ { 6 6 } { + + ^ { 9 9 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 410, + 505, + 422 + ], + "score": 1.0, + "content": "to denote the addition of the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 421, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 433 + ], + "score": 1.0, + "content": "proposed randomization layers, i.e., random resizing and random padding, and the results indi-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 431, + 389, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 389, + 445 + ], + "score": 1.0, + "content": "cate that combined models still performs pretty good on clean images.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 443, + 502, + 571 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 443, + 502, + 571 + ], + "spans": [ + { + "bbox": [ + 108, + 443, + 502, + 571 + ], + "score": 0.985, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
random brightness99.6%99.7%99.8%99.8%
random brightness ++98.6%98.1%99.1%99.2%
random saturation99.6%99.7%99.9%99.9%
random saturation ++98.6%98.3%99.3%99.3%
random hue99.4%99.6%99.7%99.4%
random hue ++98.6%98.3%99.2%99.1%
random contrast99.5%99.6%99.7%99.6%
random contrast ++98.6%98.2%99.3%99.1%
", + "type": "table", + "image_path": "ee1124c3f76bde054f599243676b17e54b17d22111dd0bac1b90566e8bc3c191.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 108, + 443, + 502, + 485.6666666666667 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 108, + 485.6666666666667, + 502, + 528.3333333333334 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 108, + 528.3333333333334, + 502, + 571.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "index": 25.25 + }, + { + "type": "text", + "bbox": [ + 106, + 587, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 504, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 504, + 599 + ], + "score": 1.0, + "content": "We then evaluate the effectiveness of these randomization methods against the adversarial examples", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "generated under the vanilla attack scenario. The results are shown in the Tables 8 - 11. Compared to", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "score": 1.0, + "content": "the results in Table 2, all these four methods are much less effective than the proposed randomization", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "score": 1.0, + "content": "layers. By combining the proposed randomization layers with each of these four randomization", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 631, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 195, + 644 + ], + "score": 1.0, + "content": "methods (denoted as", + "type": "text" + }, + { + "bbox": [ + 195, + 632, + 219, + 642 + ], + "score": 0.84, + "content": "^ { 6 6 } { + } { + } ^ { , 9 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 631, + 506, + 644 + ], + "score": 1.0, + "content": ", the performance can be slightly improved than using the proposed", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 642, + 220, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 220, + 655 + ], + "score": 1.0, + "content": "randomization layers alone.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 658, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 106, + 659, + 504, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 504, + 671 + ], + "score": 1.0, + "content": "Since each of these four randomization methods alone are not as effective as our proposed random-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "score": 1.0, + "content": "ization methods against adversarial examples generated under the vanilla attack scenario, we do not", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "further investigate their effectiveness under single-pattern attack and ensemble-pattern attack sce-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 692, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 505, + 705 + ], + "score": 1.0, + "content": "narios. However, combining these randomization methods with our proposed randomization layers", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 703, + 394, + 715 + ], + "spans": [ + { + "bbox": [ + 106, + 703, + 394, + 715 + ], + "score": 1.0, + "content": "together provides a way to build a slightly stronger defense mechanism.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 116, + 721, + 507, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 719, + 508, + 734 + ], + "spans": [ + { + "bbox": [ + 119, + 719, + 508, + 734 + ], + "score": 1.0, + "content": "4https://github.com/tensorflow/models/blob/master/research/inception/inception/image processing.py#L182", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 370, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 371, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 371, + 96 + ], + "score": 1.0, + "content": "APPENDIX A OTHER RANDOMIZATION METHODS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 105, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "Besides random resizing and random padding, we further evaluate the effectiveness of four other", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 129 + ], + "score": 1.0, + "content": "randomization methods against adversarial examples. All these four methods are used as data-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 127, + 313, + 141 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 313, + 141 + ], + "score": 1.0, + "content": "augmentation during the standard network training.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 105, + 505, + 141 + ] + }, + { + "type": "list", + "bbox": [ + 132, + 147, + 505, + 266 + ], + "lines": [ + { + "bbox": [ + 131, + 146, + 504, + 162 + ], + "spans": [ + { + "bbox": [ + 131, + 146, + 302, + 162 + ], + "score": 1.0, + "content": "• Random Brightness: a brightness factor", + "type": "text" + }, + { + "bbox": [ + 303, + 149, + 309, + 158 + ], + "score": 0.79, + "content": "\\delta", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 146, + 446, + 162 + ], + "score": 1.0, + "content": "is randomly picked in the interval", + "type": "text" + }, + { + "bbox": [ + 446, + 147, + 504, + 160 + ], + "score": 0.9, + "content": "[ - \\delta _ { \\mathrm { m a x } } , \\delta _ { \\mathrm { m a x } } ]", + "type": "inline_equation" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 140, + 158, + 453, + 176 + ], + "spans": [ + { + "bbox": [ + 140, + 158, + 336, + 176 + ], + "score": 1.0, + "content": "to adjust the brightness of the normalized image", + "type": "text" + }, + { + "bbox": [ + 336, + 159, + 351, + 172 + ], + "score": 0.9, + "content": "\\hat { X _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 158, + 401, + 176 + ], + "score": 1.0, + "content": ". We choose", + "type": "text" + }, + { + "bbox": [ + 402, + 160, + 450, + 174 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\delta _ { \\mathrm { m a x } } = \\frac { 3 2 } { 2 5 5 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 165, + 453, + 176 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 174, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 132, + 174, + 329, + 188 + ], + "score": 1.0, + "content": "• Random Saturation: a saturation factor", + "type": "text" + }, + { + "bbox": [ + 330, + 177, + 338, + 185 + ], + "score": 0.77, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 174, + 505, + 188 + ], + "score": 1.0, + "content": "is randomly picked in the interval", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 143, + 186, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 143, + 187, + 208, + 200 + ], + "score": 0.89, + "content": "[ \\alpha _ { l o w e r } , \\alpha _ { u p p e r } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 186, + 400, + 202 + ], + "score": 1.0, + "content": "to adjust the saturation of the normalized image", + "type": "text" + }, + { + "bbox": [ + 401, + 186, + 415, + 199 + ], + "score": 0.9, + "content": "\\hat { X _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 186, + 465, + 202 + ], + "score": 1.0, + "content": ". We choose", + "type": "text" + }, + { + "bbox": [ + 465, + 189, + 505, + 199 + ], + "score": 0.83, + "content": "\\alpha _ { l o w e r } =", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 199, + 234, + 213 + ], + "spans": [ + { + "bbox": [ + 141, + 199, + 174, + 213 + ], + "score": 1.0, + "content": "0.5 and", + "type": "text" + }, + { + "bbox": [ + 175, + 200, + 229, + 211 + ], + "score": 0.89, + "content": "\\alpha _ { u p p e r } = 1 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 199, + 234, + 213 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 131, + 213, + 250, + 226 + ], + "score": 1.0, + "content": "• Random Hue: a hue factor", + "type": "text" + }, + { + "bbox": [ + 250, + 214, + 257, + 223 + ], + "score": 0.8, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 213, + 394, + 226 + ], + "score": 1.0, + "content": "is randomly picked in the interval", + "type": "text" + }, + { + "bbox": [ + 394, + 213, + 452, + 225 + ], + "score": 0.92, + "content": "[ - \\theta _ { \\mathrm { m a x } } , \\theta _ { \\mathrm { m a x } } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "to adjust the", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 224, + 375, + 239 + ], + "spans": [ + { + "bbox": [ + 141, + 225, + 259, + 239 + ], + "score": 1.0, + "content": "hue of the normalized image", + "type": "text" + }, + { + "bbox": [ + 259, + 224, + 273, + 237 + ], + "score": 0.91, + "content": "{ \\hat { X _ { n } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 225, + 324, + 239 + ], + "score": 1.0, + "content": ". We choose", + "type": "text" + }, + { + "bbox": [ + 324, + 226, + 371, + 237 + ], + "score": 0.91, + "content": "\\theta _ { \\mathrm { m a x } } = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 225, + 375, + 239 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 237, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 131, + 237, + 284, + 254 + ], + "score": 1.0, + "content": "• Random Contrast: a contrast factor", + "type": "text" + }, + { + "bbox": [ + 285, + 240, + 292, + 252 + ], + "score": 0.85, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 237, + 430, + 254 + ], + "score": 1.0, + "content": "is randomly picked in the interval", + "type": "text" + }, + { + "bbox": [ + 430, + 240, + 493, + 253 + ], + "score": 0.91, + "content": "[ \\beta _ { l o w e r } , \\beta _ { u p p e r } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 237, + 506, + 254 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 252, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 141, + 252, + 313, + 267 + ], + "score": 1.0, + "content": "adjust the contrast of the normalized image", + "type": "text" + }, + { + "bbox": [ + 313, + 252, + 327, + 264 + ], + "score": 0.9, + "content": "\\hat { X _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 252, + 376, + 267 + ], + "score": 1.0, + "content": ". We choose", + "type": "text" + }, + { + "bbox": [ + 377, + 253, + 430, + 265 + ], + "score": 0.93, + "content": "\\beta _ { l o w e r } = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 252, + 447, + 267 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 447, + 254, + 501, + 266 + ], + "score": 0.89, + "content": "\\beta _ { u p p e r } = 1 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 252, + 506, + 267 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8, + "bbox_fs": [ + 131, + 146, + 506, + 267 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 274, + 505, + 309 + ], + "lines": [ + { + "bbox": [ + 105, + 273, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 505, + 287 + ], + "score": 1.0, + "content": "Note that, (1) the parameters chosen above are the same as the ones used during network training", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 285, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 325, + 299 + ], + "score": 1.0, + "content": "process4; (2) the pixel value of the normalized image", + "type": "text" + }, + { + "bbox": [ + 326, + 285, + 340, + 298 + ], + "score": 0.91, + "content": "\\hat { X _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 286, + 448, + 299 + ], + "score": 1.0, + "content": "are all within the interval", + "type": "text" + }, + { + "bbox": [ + 448, + 286, + 469, + 299 + ], + "score": 0.74, + "content": "[ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 286, + 505, + 299 + ], + "score": 1.0, + "content": ", and we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 297, + 407, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 407, + 311 + ], + "score": 1.0, + "content": "also use this range to clip the pixel value of the image after pre-processing.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 273, + 505, + 311 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 314, + 505, + 380 + ], + "lines": [ + { + "bbox": [ + 105, + 314, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 505, + 327 + ], + "score": 1.0, + "content": "Following the experiment setup in section 4, we first evaluate the effectiveness of each of these", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 326, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 505, + 337 + ], + "score": 1.0, + "content": "randomization methods on the 5000 clean images. The results are shown in the Table 7. We can see", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 335, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 349 + ], + "score": 1.0, + "content": "that these methods hardly hurt the performance on clean images. We further combine the proposed", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "score": 1.0, + "content": "randomization layers, i.e., random resizing layer and random padding layer, with each of these", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 252, + 371 + ], + "score": 1.0, + "content": "randomization methods (denoted as", + "type": "text" + }, + { + "bbox": [ + 252, + 359, + 274, + 369 + ], + "score": 0.83, + "content": "^ { 6 6 } + + ^ { 7 }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 358, + 505, + 371 + ], + "score": 1.0, + "content": "). We see that the combined randomization modules only", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 369, + 299, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 299, + 382 + ], + "score": 1.0, + "content": "cause very little accuracy drop on clean images.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 314, + 505, + 382 + ] + }, + { + "type": "table", + "bbox": [ + 108, + 443, + 502, + 571 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 398, + 505, + 443 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "score": 1.0, + "content": "Table 7: Top-1 classification accuracy on clean images. We see that these four randomization meth-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 410, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 352, + 422 + ], + "score": 1.0, + "content": "ods hardly hurt the performance on clean images. We use", + "type": "text" + }, + { + "bbox": [ + 352, + 410, + 376, + 420 + ], + "score": 0.86, + "content": "^ { 6 6 } { + + ^ { 9 9 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 410, + 505, + 422 + ], + "score": 1.0, + "content": "to denote the addition of the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 421, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 433 + ], + "score": 1.0, + "content": "proposed randomization layers, i.e., random resizing and random padding, and the results indi-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 431, + 389, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 389, + 445 + ], + "score": 1.0, + "content": "cate that combined models still performs pretty good on clean images.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 443, + 502, + 571 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 443, + 502, + 571 + ], + "spans": [ + { + "bbox": [ + 108, + 443, + 502, + 571 + ], + "score": 0.985, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
random brightness99.6%99.7%99.8%99.8%
random brightness ++98.6%98.1%99.1%99.2%
random saturation99.6%99.7%99.9%99.9%
random saturation ++98.6%98.3%99.3%99.3%
random hue99.4%99.6%99.7%99.4%
random hue ++98.6%98.3%99.2%99.1%
random contrast99.5%99.6%99.7%99.6%
random contrast ++98.6%98.2%99.3%99.1%
", + "type": "table", + "image_path": "ee1124c3f76bde054f599243676b17e54b17d22111dd0bac1b90566e8bc3c191.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 108, + 443, + 502, + 485.6666666666667 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 108, + 485.6666666666667, + 502, + 528.3333333333334 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 108, + 528.3333333333334, + 502, + 571.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "index": 25.25 + }, + { + "type": "text", + "bbox": [ + 106, + 587, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 504, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 504, + 599 + ], + "score": 1.0, + "content": "We then evaluate the effectiveness of these randomization methods against the adversarial examples", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "generated under the vanilla attack scenario. The results are shown in the Tables 8 - 11. Compared to", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 621 + ], + "score": 1.0, + "content": "the results in Table 2, all these four methods are much less effective than the proposed randomization", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 505, + 632 + ], + "score": 1.0, + "content": "layers. By combining the proposed randomization layers with each of these four randomization", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 631, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 195, + 644 + ], + "score": 1.0, + "content": "methods (denoted as", + "type": "text" + }, + { + "bbox": [ + 195, + 632, + 219, + 642 + ], + "score": 0.84, + "content": "^ { 6 6 } { + } { + } ^ { , 9 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 631, + 506, + 644 + ], + "score": 1.0, + "content": ", the performance can be slightly improved than using the proposed", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 642, + 220, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 220, + 655 + ], + "score": 1.0, + "content": "randomization layers alone.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 587, + 506, + 655 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 658, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 106, + 659, + 504, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 504, + 671 + ], + "score": 1.0, + "content": "Since each of these four randomization methods alone are not as effective as our proposed random-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "score": 1.0, + "content": "ization methods against adversarial examples generated under the vanilla attack scenario, we do not", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "further investigate their effectiveness under single-pattern attack and ensemble-pattern attack sce-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 692, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 505, + 705 + ], + "score": 1.0, + "content": "narios. However, combining these randomization methods with our proposed randomization layers", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 703, + 394, + 715 + ], + "spans": [ + { + "bbox": [ + 106, + 703, + 394, + 715 + ], + "score": 1.0, + "content": "together provides a way to build a slightly stronger defense mechanism.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 659, + 505, + 715 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 159, + 505, + 286 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 103, + 505, + 158 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 104, + 504, + 115 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 504, + 115 + ], + "score": 1.0, + "content": "Table 8: Top-1 classification accuracy by using random brightness under the vanilla attack scenario.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "Compared to the results in Table 2, random brightness is much less effective than the proposed ran-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 125, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 505, + 138 + ], + "score": 1.0, + "content": "domization layers. By combing random brightness and the proposed randomization layers (denoted", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 199, + 150 + ], + "score": 1.0, + "content": "as random brightness", + "type": "text" + }, + { + "bbox": [ + 199, + 138, + 211, + 147 + ], + "score": 0.51, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "), it reaches slightly better performance than using the proposed random-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 147, + 189, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 147, + 189, + 160 + ], + "score": 1.0, + "content": "ization layers alone.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 106, + 159, + 505, + 286 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 159, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 286 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
randombright-nessrandombright-ness++randombright-nessrandombright-ness++randombright-nessrandombright-ness++randombright-nessrandombright-ness++
FGSM-234.9%67.0%28.5%73.5%66.4%81.3%85.0%95.8%
FGSM-531.9%55.5%21.6%55.7%62.4%74.7%87.6%95.0%
FGSM-1033.2%52.9%20.9%47.2%61.8%71.5%90.4%94.5%
DeepFool79.4%98.1%82.3%97.5%62.8%98.3%79.0%99.1%
C&W34.5%96.9%47.7%97.2%51.3%98.0%42.3%98.6%
", + "type": "table", + "image_path": "241df516881557f904257b857685708c093f5a5bcd788734712e22bdd8aa507d.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 159, + 505, + 201.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 201.33333333333334, + 505, + 243.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 243.66666666666669, + 505, + 286.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "table", + "bbox": [ + 106, + 385, + 505, + 511 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 329, + 505, + 384 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 330, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 505, + 342 + ], + "score": 1.0, + "content": "Table 9: Top-1 classification accuracy by using random saturation under the vanilla attack scenario.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 341, + 504, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 504, + 353 + ], + "score": 1.0, + "content": "Compared to the results in Table 2, random saturation is much less effective than the proposed ran-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 351, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 365 + ], + "score": 1.0, + "content": "domization layers. By combing random saturation and the proposed randomization layers (denoted", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 362, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 198, + 376 + ], + "score": 1.0, + "content": "as random saturation", + "type": "text" + }, + { + "bbox": [ + 199, + 364, + 210, + 373 + ], + "score": 0.64, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 362, + 505, + 376 + ], + "score": 1.0, + "content": "), it reaches slightly better performance than using the proposed random-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 374, + 189, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 189, + 386 + ], + "score": 1.0, + "content": "ization layers alone.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10 + }, + { + "type": "table_body", + "bbox": [ + 106, + 385, + 505, + 511 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 385, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 505, + 511 + ], + "score": 0.985, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
randomsatura-tionrandomsatura-tion++randomrandomrandomrandom satura-tion++randomsatura-tionrandom satura-tion++
satura-tionsatura-tion++satura-tion
FGSM-234.2%66.5%27.9%73.6%66.3%81.5%85.2%95.7%
FGSM-531.8%55.2%21.1%55.7%62.1%74.6%87.0%95.0%
FGSM-1033.5%52.2%20.7%46.5%61.7%71.4%90.1%93.9%
DeepFool82.6%98.1%79.6%97.6%64.7%98.2%78.5%99.0%
C&W39.2%97.2%47.5%96.9%51.7%97.7%50.9%99.1%
", + "type": "table", + "image_path": "a77bd7ea9d65edad3499f047bba7c530008b294dd86121fa2def44e3afd05811.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 106, + 385, + 505, + 427.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 427.0, + 505, + 469.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 106, + 469.0, + 505, + 511.0 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 12.0 + }, + { + "type": "table", + "bbox": [ + 106, + 600, + 505, + 716 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 555, + 505, + 600 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 555, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 568 + ], + "score": 1.0, + "content": "Table 10: Top-1 classification accuracy by using random hue under the vanilla attack scenario.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "Compared to the results in Table 2, random hue is much less effective than the proposed randomiza-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "tion layers. By combing random hue and the proposed randomization layers (denoted as random", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 588, + 504, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 136, + 600 + ], + "score": 0.53, + "content": "\\mathbf { h u e + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 588, + 504, + 601 + ], + "score": 1.0, + "content": "), it reaches slightly better performance than using the proposed randomization layers alone.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 600, + 505, + 716 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 600, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 716 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
randomhuerandomhue++randomhuerandomhue++randomhuerandomhue++randomhuerandomhue++
FGSM-238.1%69.0%32.0%74.9%68.6%83.0%87.4%95.8%
FGSM-533.9%57.2%23.0%57.6%64.0%76.1%86.7%93.9%
FGSM-1036.4%54.2%22.1%48.4%63.0%72.5%88.0%91.3%
DeepFool95.0%97.9%91.2%97.6%86.5%98.4%96.8%99.1%
C&W72.1%97.3%74.0%97.0%77.4%98.2%81.1%98.8%
", + "type": "table", + "image_path": "159990e9477e81f3b8b7168e7a5a656d363edfa0315ec3fe5ab3ce9c887e700d.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 106, + 600, + 505, + 638.6666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 106, + 638.6666666666666, + 505, + 677.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 106, + 677.3333333333333, + 505, + 715.9999999999999 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 19.25 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 159, + 505, + 286 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 103, + 505, + 158 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 104, + 504, + 115 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 504, + 115 + ], + "score": 1.0, + "content": "Table 8: Top-1 classification accuracy by using random brightness under the vanilla attack scenario.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "Compared to the results in Table 2, random brightness is much less effective than the proposed ran-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 125, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 505, + 138 + ], + "score": 1.0, + "content": "domization layers. By combing random brightness and the proposed randomization layers (denoted", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 199, + 150 + ], + "score": 1.0, + "content": "as random brightness", + "type": "text" + }, + { + "bbox": [ + 199, + 138, + 211, + 147 + ], + "score": 0.51, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "), it reaches slightly better performance than using the proposed random-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 147, + 189, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 147, + 189, + 160 + ], + "score": 1.0, + "content": "ization layers alone.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 106, + 159, + 505, + 286 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 159, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 286 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
randombright-nessrandombright-ness++randombright-nessrandombright-ness++randombright-nessrandombright-ness++randombright-nessrandombright-ness++
FGSM-234.9%67.0%28.5%73.5%66.4%81.3%85.0%95.8%
FGSM-531.9%55.5%21.6%55.7%62.4%74.7%87.6%95.0%
FGSM-1033.2%52.9%20.9%47.2%61.8%71.5%90.4%94.5%
DeepFool79.4%98.1%82.3%97.5%62.8%98.3%79.0%99.1%
C&W34.5%96.9%47.7%97.2%51.3%98.0%42.3%98.6%
", + "type": "table", + "image_path": "241df516881557f904257b857685708c093f5a5bcd788734712e22bdd8aa507d.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 159, + 505, + 201.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 201.33333333333334, + 505, + 243.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 243.66666666666669, + 505, + 286.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "table", + "bbox": [ + 106, + 385, + 505, + 511 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 329, + 505, + 384 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 330, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 505, + 342 + ], + "score": 1.0, + "content": "Table 9: Top-1 classification accuracy by using random saturation under the vanilla attack scenario.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 341, + 504, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 504, + 353 + ], + "score": 1.0, + "content": "Compared to the results in Table 2, random saturation is much less effective than the proposed ran-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 351, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 365 + ], + "score": 1.0, + "content": "domization layers. By combing random saturation and the proposed randomization layers (denoted", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 362, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 198, + 376 + ], + "score": 1.0, + "content": "as random saturation", + "type": "text" + }, + { + "bbox": [ + 199, + 364, + 210, + 373 + ], + "score": 0.64, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 362, + 505, + 376 + ], + "score": 1.0, + "content": "), it reaches slightly better performance than using the proposed random-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 374, + 189, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 189, + 386 + ], + "score": 1.0, + "content": "ization layers alone.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10 + }, + { + "type": "table_body", + "bbox": [ + 106, + 385, + 505, + 511 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 385, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 505, + 511 + ], + "score": 0.985, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
randomsatura-tionrandomsatura-tion++randomrandomrandomrandom satura-tion++randomsatura-tionrandom satura-tion++
satura-tionsatura-tion++satura-tion
FGSM-234.2%66.5%27.9%73.6%66.3%81.5%85.2%95.7%
FGSM-531.8%55.2%21.1%55.7%62.1%74.6%87.0%95.0%
FGSM-1033.5%52.2%20.7%46.5%61.7%71.4%90.1%93.9%
DeepFool82.6%98.1%79.6%97.6%64.7%98.2%78.5%99.0%
C&W39.2%97.2%47.5%96.9%51.7%97.7%50.9%99.1%
", + "type": "table", + "image_path": "a77bd7ea9d65edad3499f047bba7c530008b294dd86121fa2def44e3afd05811.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 106, + 385, + 505, + 427.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 427.0, + 505, + 469.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 106, + 469.0, + 505, + 511.0 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 12.0 + }, + { + "type": "table", + "bbox": [ + 106, + 600, + 505, + 716 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 555, + 505, + 600 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 555, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 568 + ], + "score": 1.0, + "content": "Table 10: Top-1 classification accuracy by using random hue under the vanilla attack scenario.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "Compared to the results in Table 2, random hue is much less effective than the proposed randomiza-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "tion layers. By combing random hue and the proposed randomization layers (denoted as random", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 588, + 504, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 136, + 600 + ], + "score": 0.53, + "content": "\\mathbf { h u e + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 588, + 504, + 601 + ], + "score": 1.0, + "content": "), it reaches slightly better performance than using the proposed randomization layers alone.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 600, + 505, + 716 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 600, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 716 + ], + "score": 0.984, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
randomhuerandomhue++randomhuerandomhue++randomhuerandomhue++randomhuerandomhue++
FGSM-238.1%69.0%32.0%74.9%68.6%83.0%87.4%95.8%
FGSM-533.9%57.2%23.0%57.6%64.0%76.1%86.7%93.9%
FGSM-1036.4%54.2%22.1%48.4%63.0%72.5%88.0%91.3%
DeepFool95.0%97.9%91.2%97.6%86.5%98.4%96.8%99.1%
C&W72.1%97.3%74.0%97.0%77.4%98.2%81.1%98.8%
", + "type": "table", + "image_path": "159990e9477e81f3b8b7168e7a5a656d363edfa0315ec3fe5ab3ce9c887e700d.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 106, + 600, + 505, + 638.6666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 106, + 638.6666666666666, + 505, + 677.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 106, + 677.3333333333333, + 505, + 715.9999999999999 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 19.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 144, + 506, + 271 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 505, + 144 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 89, + 505, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 89, + 505, + 101 + ], + "score": 1.0, + "content": "Table 11: Top-1 classification accuracy by using random contrast under the vanilla attack scenario.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "score": 1.0, + "content": "Compared to the results in Table 2, random contrast is much less effective than the proposed ran-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 110, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 124 + ], + "score": 1.0, + "content": "domization layers. By combing random contrast and the proposed randomization layers (denoted as", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 121, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 178, + 135 + ], + "score": 1.0, + "content": "random contrast", + "type": "text" + }, + { + "bbox": [ + 178, + 123, + 190, + 132 + ], + "score": 0.55, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 121, + 505, + 135 + ], + "score": 1.0, + "content": "), it reaches slightly better performance than using the proposed randomization", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 133, + 158, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 158, + 145 + ], + "score": 1.0, + "content": "layers alone.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 106, + 144, + 506, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 144, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 506, + 271 + ], + "score": 0.985, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
randomcon-trastrandomcon-trast++randomcon-trastrandomcon-trast++randomcon-trastrandomcon-trast++randomcon-trastrandomcon-trast++
FGSM-237.0%68.0%29.4%74.1%67.2%82.5%85.9%96.0%
FGSM-532.7%56.3%22.7%57.0%63.1%74.9%88.1%94.9%
FGSM-1034.4%53.5%21.2%47.0%62.1%72.0%90.6%94.3%
DeepFool90.4%98.1%87.5%97.5%73.8%98.1%90.8%99.0%
C&W56.8%97.0%57.1%96.7%63.7%97.9%67.6%98.8%
", + "type": "table", + "image_path": "6f3b024c88dd017504069a106ef4000d10bd683f52dadbc7e9bf4c4674b24fb9.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 144, + 506, + 186.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 186.33333333333334, + 506, + 228.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 228.66666666666669, + 506, + 271.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "title", + "bbox": [ + 108, + 289, + 429, + 302 + ], + "lines": [ + { + "bbox": [ + 106, + 288, + 431, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 431, + 304 + ], + "score": 1.0, + "content": "APPENDIX B RANDOMIZATION LAYERS WITH SMALLER SIZE", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 313, + 505, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 313, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 505, + 327 + ], + "score": 1.0, + "content": "Instead of resizing the input image to a larger size, we here resize the input image to a smaller size,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 324, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 104, + 324, + 506, + 339 + ], + "score": 1.0, + "content": "i.e., the resizing parameter is randomly sampled from the range [267, 299). The random padding", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 335, + 506, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 302, + 348 + ], + "score": 1.0, + "content": "layer then pads the resized image to the shape of", + "type": "text" + }, + { + "bbox": [ + 302, + 336, + 362, + 347 + ], + "score": 0.92, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 335, + 506, + 348 + ], + "score": 1.0, + "content": "in a random manner. Note that, the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 347, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 506, + 360 + ], + "score": 1.0, + "content": "random resizing layer and the random padding layer here have the same freedom as the ones used", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 358, + 506, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 506, + 371 + ], + "score": 1.0, + "content": "in the paper, i.e., they create the same number, 12528, of different patterns for a single image. We", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 369, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 505, + 381 + ], + "score": 1.0, + "content": "evaluate the effectiveness of this parameter setting on both the 5000 clean images and the adversarial", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 379, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 506, + 393 + ], + "score": 1.0, + "content": "examples generated under the vanilla attack scenario. The results are shown in the Table 12. We see", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 389, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 506, + 405 + ], + "score": 1.0, + "content": "that randomization layers still work well with smaller size images, but is slightly worse than using", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "larger size images as in the paper. This is because resizing to a smaller size loses certain information", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 411, + 195, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 195, + 426 + ], + "score": 1.0, + "content": "of the original image.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 13.5 + }, + { + "type": "table", + "bbox": [ + 108, + 498, + 503, + 603 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 443, + 505, + 497 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "score": 1.0, + "content": "Table 12: Top-1 classification accuracy on the clean images and the adversarial examples gener-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 455, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 505, + 466 + ], + "score": 1.0, + "content": "ated under the vanilla attack scenario. Compared to the results in Tables 1 and 2, randomization", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 464, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 104, + 464, + 374, + 479 + ], + "score": 1.0, + "content": "parameters applied here (i.e., resize between [267, 299), and pad to", + "type": "text" + }, + { + "bbox": [ + 375, + 465, + 435, + 476 + ], + "score": 0.91, + "content": "2 9 9 \\times 2 9 9 \\times 3 )", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 464, + 506, + 479 + ], + "score": 1.0, + "content": ") is slightly worse", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 476, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 505, + 489 + ], + "score": 1.0, + "content": "than the randomization parameters applied in the paper (i.e., resize between [299, 331), and pad to", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 486, + 174, + 499 + ], + "spans": [ + { + "bbox": [ + 107, + 487, + 169, + 498 + ], + "score": 0.84, + "content": "3 3 1 \\times 3 3 1 \\times 3 )", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 486, + 174, + 499 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "table_body", + "bbox": [ + 108, + 498, + 503, + 603 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 108, + 498, + 503, + 603 + ], + "spans": [ + { + "bbox": [ + 108, + 498, + 503, + 603 + ], + "score": 0.982, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
clean images98.2%97.5%99.1%98.7%
FGSM-263.1%65.0%79.9%95.0%
FGSM-553.4%48.3%73.3%94.0%
FGSM-1050.8%40.5%70.6%93.4%
DeepFool97.2%96.5%96.0%98.6%
C&W95.2%95.2%97.2%97.8%
", + "type": "table", + "image_path": "b63d8a7250398b8bd06b39e706ab64735a9a5b36a087184c416f91f69b0751b1.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 108, + 498, + 503, + 533.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 108, + 533.0, + 503, + 568.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 108, + 568.0, + 503, + 603.0 + ], + "spans": [], + "index": 26 + } + ] + } + ], + "index": 23.0 + }, + { + "type": "title", + "bbox": [ + 106, + 623, + 471, + 637 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 471, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 471, + 638 + ], + "score": 1.0, + "content": "APPENDIX C RANDOMIZATION LAYERS WITH MULTIPLE ITERATIONS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "In this section, we show the relationship between the top-1 accuracy of the defense model and the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 485, + 673 + ], + "score": 1.0, + "content": "iteration number performed on each image. Specifically, we choose ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 485, + 660, + 505, + 670 + ], + "score": 0.8, + "content": "\\nu 2 +", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "randomization layers as the defense model for the experiment. The same trend can be observed for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 682, + 196, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 196, + 693 + ], + "score": 1.0, + "content": "other defense models.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 354, + 711 + ], + "score": 1.0, + "content": "For the defense model, the iteration number is chosen to be", + "type": "text" + }, + { + "bbox": [ + 355, + 699, + 423, + 711 + ], + "score": 0.91, + "content": "\\{ 1 , 5 , 1 0 , 2 0 , 3 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 697, + 506, + 711 + ], + "score": 1.0, + "content": ", and it is evaluated", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "on the 5000 clean test images and the adversarial examples generated under all three attack scenar-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 721, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 506, + 733 + ], + "score": 1.0, + "content": "ios. The results are shown in the Figures 3 - 5. We can observe that (1) increasing the number of", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 144, + 506, + 271 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 505, + 144 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 89, + 505, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 89, + 505, + 101 + ], + "score": 1.0, + "content": "Table 11: Top-1 classification accuracy by using random contrast under the vanilla attack scenario.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "score": 1.0, + "content": "Compared to the results in Table 2, random contrast is much less effective than the proposed ran-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 110, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 124 + ], + "score": 1.0, + "content": "domization layers. By combing random contrast and the proposed randomization layers (denoted as", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 121, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 178, + 135 + ], + "score": 1.0, + "content": "random contrast", + "type": "text" + }, + { + "bbox": [ + 178, + 123, + 190, + 132 + ], + "score": 0.55, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 121, + 505, + 135 + ], + "score": 1.0, + "content": "), it reaches slightly better performance than using the proposed randomization", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 133, + 158, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 158, + 145 + ], + "score": 1.0, + "content": "layers alone.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 106, + 144, + 506, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 144, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 506, + 271 + ], + "score": 0.985, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
randomcon-trastrandomcon-trast++randomcon-trastrandomcon-trast++randomcon-trastrandomcon-trast++randomcon-trastrandomcon-trast++
FGSM-237.0%68.0%29.4%74.1%67.2%82.5%85.9%96.0%
FGSM-532.7%56.3%22.7%57.0%63.1%74.9%88.1%94.9%
FGSM-1034.4%53.5%21.2%47.0%62.1%72.0%90.6%94.3%
DeepFool90.4%98.1%87.5%97.5%73.8%98.1%90.8%99.0%
C&W56.8%97.0%57.1%96.7%63.7%97.9%67.6%98.8%
", + "type": "table", + "image_path": "6f3b024c88dd017504069a106ef4000d10bd683f52dadbc7e9bf4c4674b24fb9.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 144, + 506, + 186.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 186.33333333333334, + 506, + 228.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 228.66666666666669, + 506, + 271.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "title", + "bbox": [ + 108, + 289, + 429, + 302 + ], + "lines": [ + { + "bbox": [ + 106, + 288, + 431, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 431, + 304 + ], + "score": 1.0, + "content": "APPENDIX B RANDOMIZATION LAYERS WITH SMALLER SIZE", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 313, + 505, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 313, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 505, + 327 + ], + "score": 1.0, + "content": "Instead of resizing the input image to a larger size, we here resize the input image to a smaller size,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 324, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 104, + 324, + 506, + 339 + ], + "score": 1.0, + "content": "i.e., the resizing parameter is randomly sampled from the range [267, 299). The random padding", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 335, + 506, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 302, + 348 + ], + "score": 1.0, + "content": "layer then pads the resized image to the shape of", + "type": "text" + }, + { + "bbox": [ + 302, + 336, + 362, + 347 + ], + "score": 0.92, + "content": "2 9 9 \\times 2 9 9 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 335, + 506, + 348 + ], + "score": 1.0, + "content": "in a random manner. Note that, the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 347, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 506, + 360 + ], + "score": 1.0, + "content": "random resizing layer and the random padding layer here have the same freedom as the ones used", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 358, + 506, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 506, + 371 + ], + "score": 1.0, + "content": "in the paper, i.e., they create the same number, 12528, of different patterns for a single image. We", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 369, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 505, + 381 + ], + "score": 1.0, + "content": "evaluate the effectiveness of this parameter setting on both the 5000 clean images and the adversarial", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 379, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 506, + 393 + ], + "score": 1.0, + "content": "examples generated under the vanilla attack scenario. The results are shown in the Table 12. We see", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 389, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 506, + 405 + ], + "score": 1.0, + "content": "that randomization layers still work well with smaller size images, but is slightly worse than using", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "larger size images as in the paper. This is because resizing to a smaller size loses certain information", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 411, + 195, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 195, + 426 + ], + "score": 1.0, + "content": "of the original image.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 13.5, + "bbox_fs": [ + 104, + 313, + 506, + 426 + ] + }, + { + "type": "table", + "bbox": [ + 108, + 498, + 503, + 603 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 443, + 505, + 497 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "score": 1.0, + "content": "Table 12: Top-1 classification accuracy on the clean images and the adversarial examples gener-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 455, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 505, + 466 + ], + "score": 1.0, + "content": "ated under the vanilla attack scenario. Compared to the results in Tables 1 and 2, randomization", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 464, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 104, + 464, + 374, + 479 + ], + "score": 1.0, + "content": "parameters applied here (i.e., resize between [267, 299), and pad to", + "type": "text" + }, + { + "bbox": [ + 375, + 465, + 435, + 476 + ], + "score": 0.91, + "content": "2 9 9 \\times 2 9 9 \\times 3 )", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 464, + 506, + 479 + ], + "score": 1.0, + "content": ") is slightly worse", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 476, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 505, + 489 + ], + "score": 1.0, + "content": "than the randomization parameters applied in the paper (i.e., resize between [299, 331), and pad to", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 486, + 174, + 499 + ], + "spans": [ + { + "bbox": [ + 107, + 487, + 169, + 498 + ], + "score": 0.84, + "content": "3 3 1 \\times 3 3 1 \\times 3 )", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 486, + 174, + 499 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "table_body", + "bbox": [ + 108, + 498, + 503, + 603 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 108, + 498, + 503, + 603 + ], + "spans": [ + { + "bbox": [ + 108, + 498, + 503, + 603 + ], + "score": 0.982, + "html": "
ModelsInception-v3ResNet-v2-101Inception-ResNet-v2ens-adv-Inception-ResNet-v2
clean images98.2%97.5%99.1%98.7%
FGSM-263.1%65.0%79.9%95.0%
FGSM-553.4%48.3%73.3%94.0%
FGSM-1050.8%40.5%70.6%93.4%
DeepFool97.2%96.5%96.0%98.6%
C&W95.2%95.2%97.2%97.8%
", + "type": "table", + "image_path": "b63d8a7250398b8bd06b39e706ab64735a9a5b36a087184c416f91f69b0751b1.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 108, + 498, + 503, + 533.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 108, + 533.0, + 503, + 568.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 108, + 568.0, + 503, + 603.0 + ], + "spans": [], + "index": 26 + } + ] + } + ], + "index": 23.0 + }, + { + "type": "title", + "bbox": [ + 106, + 623, + 471, + 637 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 471, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 471, + 638 + ], + "score": 1.0, + "content": "APPENDIX C RANDOMIZATION LAYERS WITH MULTIPLE ITERATIONS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "In this section, we show the relationship between the top-1 accuracy of the defense model and the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 485, + 673 + ], + "score": 1.0, + "content": "iteration number performed on each image. Specifically, we choose ens-adv-Inception-ResNet-", + "type": "text" + }, + { + "bbox": [ + 485, + 660, + 505, + 670 + ], + "score": 0.8, + "content": "\\nu 2 +", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "randomization layers as the defense model for the experiment. The same trend can be observed for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 682, + 196, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 196, + 693 + ], + "score": 1.0, + "content": "other defense models.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 648, + 505, + 693 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 354, + 711 + ], + "score": 1.0, + "content": "For the defense model, the iteration number is chosen to be", + "type": "text" + }, + { + "bbox": [ + 355, + 699, + 423, + 711 + ], + "score": 0.91, + "content": "\\{ 1 , 5 , 1 0 , 2 0 , 3 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 697, + 506, + 711 + ], + "score": 1.0, + "content": ", and it is evaluated", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "on the 5000 clean test images and the adversarial examples generated under all three attack scenar-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 721, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 506, + 733 + ], + "score": 1.0, + "content": "ios. The results are shown in the Figures 3 - 5. We can observe that (1) increasing the number of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "iteration can slightly improve the top-1 classification accuracy of the defense model on clean im-", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "ages and adversarial examples generated under both the vanilla attack and the single-pattern attack", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 504, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 504, + 117 + ], + "score": 1.0, + "content": "scenarios; (2) increasing the number of iteration has nearly no improvement for the top-1 classifica-", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "tion accuracy of the defense model on adversarial examples generated by single-step attacks under", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 505, + 139 + ], + "score": 1.0, + "content": "the ensemble-pattern attack scenario; (3) increasing the number of iteration can improve the top-1", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "classification accuracy of the defense model on adversarial examples generated by iterative attacks", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 149, + 281, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 281, + 160 + ], + "score": 1.0, + "content": "under the ensemble-pattern attack scenario.", + "type": "text", + "cross_page": true + } + ], + "index": 6 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 697, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "iteration can slightly improve the top-1 classification accuracy of the defense model on clean im-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "ages and adversarial examples generated under both the vanilla attack and the single-pattern attack", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 504, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 504, + 117 + ], + "score": 1.0, + "content": "scenarios; (2) increasing the number of iteration has nearly no improvement for the top-1 classifica-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "tion accuracy of the defense model on adversarial examples generated by single-step attacks under", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 505, + 139 + ], + "score": 1.0, + "content": "the ensemble-pattern attack scenario; (3) increasing the number of iteration can improve the top-1", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "classification accuracy of the defense model on adversarial examples generated by iterative attacks", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 149, + 281, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 281, + 160 + ], + "score": 1.0, + "content": "under the ensemble-pattern attack scenario.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "image", + "bbox": [ + 113, + 178, + 223, + 300 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 178, + 223, + 300 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 178, + 223, + 300 + ], + "spans": [ + { + "bbox": [ + 113, + 178, + 223, + 300 + ], + "score": 0.956, + "type": "image", + "image_path": "a25aa63356ef141cc900583f46c838547d3cb72eeae8c3fce9b665da139db3ec.jpg" + } + ] + } + ], + "index": 8.0, + "virtual_lines": [ + { + "bbox": [ + 113, + 178, + 223, + 239.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 239.0, + 223, + 300.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 317, + 231, + 372 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 317, + 231, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 231, + 329 + ], + "score": 1.0, + "content": "Figure 3: Top-1 classification", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 329, + 231, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 231, + 340 + ], + "score": 1.0, + "content": "accuracy on the clean images", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 339, + 231, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 231, + 351 + ], + "score": 1.0, + "content": "and the adversarial examples", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 350, + 232, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 232, + 361 + ], + "score": 1.0, + "content": "generated under the vanilla at-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 361, + 159, + 372 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 159, + 372 + ], + "score": 1.0, + "content": "tack scenrio.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + } + ], + "index": 11.5 + }, + { + "type": "image", + "bbox": [ + 248, + 178, + 360, + 300 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 248, + 178, + 360, + 300 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 248, + 178, + 360, + 300 + ], + "spans": [ + { + "bbox": [ + 248, + 178, + 360, + 300 + ], + "score": 0.965, + "type": "image", + "image_path": "0cf83e4d60ebaeaa48670a3eae350647157c26303fc23d006fea2afbff49f186.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 248, + 178, + 360, + 239.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 248, + 239.0, + 360, + 300.0 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 242, + 317, + 367, + 372 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 243, + 316, + 367, + 329 + ], + "spans": [ + { + "bbox": [ + 243, + 316, + 367, + 329 + ], + "score": 1.0, + "content": "Figure 4: Top-1 classifica-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 243, + 328, + 368, + 340 + ], + "spans": [ + { + "bbox": [ + 243, + 328, + 368, + 340 + ], + "score": 1.0, + "content": "tion accuracy on the adversar-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 242, + 339, + 369, + 351 + ], + "spans": [ + { + "bbox": [ + 242, + 339, + 369, + 351 + ], + "score": 1.0, + "content": "ial examples generated under", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 241, + 349, + 368, + 362 + ], + "spans": [ + { + "bbox": [ + 241, + 349, + 368, + 362 + ], + "score": 1.0, + "content": "the single-pattern attack scen-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 241, + 360, + 260, + 373 + ], + "spans": [ + { + "bbox": [ + 241, + 360, + 260, + 373 + ], + "score": 1.0, + "content": "rio.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + } + ], + "index": 14.75 + }, + { + "type": "image", + "bbox": [ + 385, + 178, + 496, + 300 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 385, + 178, + 496, + 300 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 385, + 178, + 496, + 300 + ], + "spans": [ + { + "bbox": [ + 385, + 178, + 496, + 300 + ], + "score": 0.959, + "type": "image", + "image_path": "49c765887d9ca814549732caf7e88c202b46e49cbcf0f3ce57891fd62a33e450.jpg" + } + ] + } + ], + "index": 11.0, + "virtual_lines": [ + { + "bbox": [ + 385, + 178, + 496, + 239.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 385, + 239.0, + 496, + 300.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 379, + 317, + 505, + 372 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 379, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 379, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "Figure 5: Top-1 classification", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 379, + 329, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 379, + 329, + 505, + 339 + ], + "score": 1.0, + "content": "accuracy on the adversarial ex-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 379, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 379, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "amples generated under the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 379, + 351, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 379, + 351, + 505, + 361 + ], + "score": 1.0, + "content": "ensemble-pattern attack scen-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 378, + 361, + 396, + 373 + ], + "spans": [ + { + "bbox": [ + 378, + 361, + 396, + 373 + ], + "score": 1.0, + "content": "rio.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + } + ], + "index": 18.0 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2018", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 160 + ], + "lines": [], + "index": 3, + "bbox_fs": [ + 105, + 81, + 505, + 160 + ], + "lines_deleted": true + }, + { + "type": "image", + "bbox": [ + 113, + 178, + 223, + 300 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 178, + 223, + 300 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 178, + 223, + 300 + ], + "spans": [ + { + "bbox": [ + 113, + 178, + 223, + 300 + ], + "score": 0.956, + "type": "image", + "image_path": "a25aa63356ef141cc900583f46c838547d3cb72eeae8c3fce9b665da139db3ec.jpg" + } + ] + } + ], + "index": 8.0, + "virtual_lines": [ + { + "bbox": [ + 113, + 178, + 223, + 239.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 239.0, + 223, + 300.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 317, + 231, + 372 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 317, + 231, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 231, + 329 + ], + "score": 1.0, + "content": "Figure 3: Top-1 classification", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 329, + 231, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 231, + 340 + ], + "score": 1.0, + "content": "accuracy on the clean images", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 339, + 231, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 231, + 351 + ], + "score": 1.0, + "content": "and the adversarial examples", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 350, + 232, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 232, + 361 + ], + "score": 1.0, + "content": "generated under the vanilla at-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 361, + 159, + 372 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 159, + 372 + ], + "score": 1.0, + "content": "tack scenrio.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + } + ], + "index": 11.5 + }, + { + "type": "image", + "bbox": [ + 248, + 178, + 360, + 300 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 248, + 178, + 360, + 300 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 248, + 178, + 360, + 300 + ], + "spans": [ + { + "bbox": [ + 248, + 178, + 360, + 300 + ], + "score": 0.965, + "type": "image", + "image_path": "0cf83e4d60ebaeaa48670a3eae350647157c26303fc23d006fea2afbff49f186.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 248, + 178, + 360, + 239.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 248, + 239.0, + 360, + 300.0 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 242, + 317, + 367, + 372 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 243, + 316, + 367, + 329 + ], + "spans": [ + { + "bbox": [ + 243, + 316, + 367, + 329 + ], + "score": 1.0, + "content": "Figure 4: Top-1 classifica-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 243, + 328, + 368, + 340 + ], + "spans": [ + { + "bbox": [ + 243, + 328, + 368, + 340 + ], + "score": 1.0, + "content": "tion accuracy on the adversar-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 242, + 339, + 369, + 351 + ], + "spans": [ + { + "bbox": [ + 242, + 339, + 369, + 351 + ], + "score": 1.0, + "content": "ial examples generated under", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 241, + 349, + 368, + 362 + ], + "spans": [ + { + "bbox": [ + 241, + 349, + 368, + 362 + ], + "score": 1.0, + "content": "the single-pattern attack scen-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 241, + 360, + 260, + 373 + ], + "spans": [ + { + "bbox": [ + 241, + 360, + 260, + 373 + ], + "score": 1.0, + "content": "rio.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + } + ], + "index": 14.75 + }, + { + "type": "image", + "bbox": [ + 385, + 178, + 496, + 300 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 385, + 178, + 496, + 300 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 385, + 178, + 496, + 300 + ], + "spans": [ + { + "bbox": [ + 385, + 178, + 496, + 300 + ], + "score": 0.959, + "type": "image", + "image_path": "49c765887d9ca814549732caf7e88c202b46e49cbcf0f3ce57891fd62a33e450.jpg" + } + ] + } + ], + "index": 11.0, + "virtual_lines": [ + { + "bbox": [ + 385, + 178, + 496, + 239.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 385, + 239.0, + 496, + 300.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 379, + 317, + 505, + 372 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 379, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 379, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "Figure 5: Top-1 classification", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 379, + 329, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 379, + 329, + 505, + 339 + ], + "score": 1.0, + "content": "accuracy on the adversarial ex-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 379, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 379, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "amples generated under the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 379, + 351, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 379, + 351, + 505, + 361 + ], + "score": 1.0, + "content": "ensemble-pattern attack scen-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 378, + 361, + 396, + 373 + ], + "spans": [ + { + "bbox": [ + 378, + 361, + 396, + 373 + ], + "score": 1.0, + "content": "rio.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + } + ], + "index": 18.0 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/SkgC6TNFvr/images/019c005408e6ebec987244e1d6a537172ea85946059cd775443d3f29074dde5f.jpg b/parse/train/SkgC6TNFvr/images/019c005408e6ebec987244e1d6a537172ea85946059cd775443d3f29074dde5f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3678235b6bdd573e0df0d33552d83e0e35f196d1 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/019c005408e6ebec987244e1d6a537172ea85946059cd775443d3f29074dde5f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8ef33e6937e900d160c8779c9549a4a7a46ce08bc79cb229d7cc4739126eaad +size 24365 diff --git a/parse/train/SkgC6TNFvr/images/04b67ac28c5a652a557d086ebf99cc4067ce968989b526c6895ca6bd368e8393.jpg b/parse/train/SkgC6TNFvr/images/04b67ac28c5a652a557d086ebf99cc4067ce968989b526c6895ca6bd368e8393.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5444ce48845b3d6e0fa12cf64790b7f38d709270 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/04b67ac28c5a652a557d086ebf99cc4067ce968989b526c6895ca6bd368e8393.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:086e3e9c7ec350a4c06de09698f76759a3e39b3c283828d467e48f8abf3e6a37 +size 61227 diff --git a/parse/train/SkgC6TNFvr/images/157cb863bc53fa0501b39cc603bd48c18c162643aca4a103df1d9c26c9af047b.jpg b/parse/train/SkgC6TNFvr/images/157cb863bc53fa0501b39cc603bd48c18c162643aca4a103df1d9c26c9af047b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7099ff10ec08da6fbb13e13eda60c7a594fb13a6 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/157cb863bc53fa0501b39cc603bd48c18c162643aca4a103df1d9c26c9af047b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89f17a4821002fb576e8d40f2e9fa9fd47c546479c44e1872a0e7732f8c6e886 +size 4402 diff --git a/parse/train/SkgC6TNFvr/images/569ce9f506eecf6f50ededaec76c14f0980114b549f3f65795fa3f12c03ace54.jpg b/parse/train/SkgC6TNFvr/images/569ce9f506eecf6f50ededaec76c14f0980114b549f3f65795fa3f12c03ace54.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f2945897c8e489c36ba2667ffdd7f690563e63e --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/569ce9f506eecf6f50ededaec76c14f0980114b549f3f65795fa3f12c03ace54.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bd37be950c6649bf21eac367c136360b503e930a456c0f18e5ccc79b2e7b7ae +size 67240 diff --git a/parse/train/SkgC6TNFvr/images/64faba326ae8f09ec41e220352db37da0a6f3519a7400c031f0c1307bcb5df8c.jpg b/parse/train/SkgC6TNFvr/images/64faba326ae8f09ec41e220352db37da0a6f3519a7400c031f0c1307bcb5df8c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bf185c2be623e299068ab5b4f7dfd032795be118 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/64faba326ae8f09ec41e220352db37da0a6f3519a7400c031f0c1307bcb5df8c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9954d1f0c7481f87bbafc94d7372896f39e132e53856c0980a64423ef08d8ac0 +size 51665 diff --git a/parse/train/SkgC6TNFvr/images/8337d9a96cb14857c33a3b828c2e86e6ecc47aab5071754921c5c30c7f0db782.jpg b/parse/train/SkgC6TNFvr/images/8337d9a96cb14857c33a3b828c2e86e6ecc47aab5071754921c5c30c7f0db782.jpg new file mode 100644 index 0000000000000000000000000000000000000000..88a467798ad5c41143bd2bf6c4f327c1dfb676f3 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/8337d9a96cb14857c33a3b828c2e86e6ecc47aab5071754921c5c30c7f0db782.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5a1b94a396325671624b3e24cfb1c6db43f03731831b164216b2967cc8accb6 +size 4963 diff --git a/parse/train/SkgC6TNFvr/images/97599a01622a8beb5c6e72a5a916d832ba85563d43f410aed308835e5de73da9.jpg b/parse/train/SkgC6TNFvr/images/97599a01622a8beb5c6e72a5a916d832ba85563d43f410aed308835e5de73da9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9cf701d8a9c9f8fbe66d2d1c56d4663b777c777c --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/97599a01622a8beb5c6e72a5a916d832ba85563d43f410aed308835e5de73da9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cc5872091c10ad135f94ea3290820578c0dee2d79da24f6913fe480e51d5bcb +size 58242 diff --git a/parse/train/SkgC6TNFvr/images/a3f9bf9173e3bc56b54ea0cbae3e08b4846b0e38036fc1d324bef0d0a1d2fa95.jpg b/parse/train/SkgC6TNFvr/images/a3f9bf9173e3bc56b54ea0cbae3e08b4846b0e38036fc1d324bef0d0a1d2fa95.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3b4a47fe46e2bce4b4b9dd217144d6346e4d74dc --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/a3f9bf9173e3bc56b54ea0cbae3e08b4846b0e38036fc1d324bef0d0a1d2fa95.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb38faf8532ce23fb42ce597b9256f91ed690f662d9a0bc6a241cc43e8e88fe3 +size 70445 diff --git a/parse/train/SkgC6TNFvr/images/a4e54814eb9de81352c6274e854d6500883df78773da4f79a52ff553811c005b.jpg b/parse/train/SkgC6TNFvr/images/a4e54814eb9de81352c6274e854d6500883df78773da4f79a52ff553811c005b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..448dc48a3b6c973adbc6dae0bc01594802954b32 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/a4e54814eb9de81352c6274e854d6500883df78773da4f79a52ff553811c005b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89437f9f6a5d485923264bc7c47332bc4f3c4b5260b7d6c0e45f5ab1aeab5aa8 +size 33814 diff --git a/parse/train/SkgC6TNFvr/images/ac2b81f4e4b95b2145d33d67171de96120dc4a9fe838c9c5f6be340a70ce866f.jpg b/parse/train/SkgC6TNFvr/images/ac2b81f4e4b95b2145d33d67171de96120dc4a9fe838c9c5f6be340a70ce866f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf8e74023cdba2eddb4fb457ab3accd1df486e91 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/ac2b81f4e4b95b2145d33d67171de96120dc4a9fe838c9c5f6be340a70ce866f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eafedcf4753836d99c934b996859fa79c8fc299497d97b2d7d635d00fc90c4c0 +size 169018 diff --git a/parse/train/SkgC6TNFvr/images/c2582ac7d7dac6b40fb7499f678a27945b6a55db17f63bec16fa397fb67a71ee.jpg b/parse/train/SkgC6TNFvr/images/c2582ac7d7dac6b40fb7499f678a27945b6a55db17f63bec16fa397fb67a71ee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..03cecb6dacb3fe11322600a6299990a5eee5a657 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/c2582ac7d7dac6b40fb7499f678a27945b6a55db17f63bec16fa397fb67a71ee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:93c439b55c8fcb17a532c77fdfa23796fb2fbd61942bc1c2d09333ef38a5f4a9 +size 73160 diff --git a/parse/train/SkgC6TNFvr/images/c49abc9a0b68205a8fcf0fc323330ed2d1c1e47f9b6a90236bf94ac40ee70f77.jpg b/parse/train/SkgC6TNFvr/images/c49abc9a0b68205a8fcf0fc323330ed2d1c1e47f9b6a90236bf94ac40ee70f77.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7ecafce1f2c83ae3754c2e248d01e09ad7a9122c --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/c49abc9a0b68205a8fcf0fc323330ed2d1c1e47f9b6a90236bf94ac40ee70f77.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3acc76000c6cb776d8f6c606da476c730acb1edaf594e194443740b1e6529bd2 +size 5523 diff --git a/parse/train/SkgC6TNFvr/images/cd695bc69edc0e4d86e52895a51017a2de68f0f2a6f9a0c8d0de8cc90dcc4a69.jpg b/parse/train/SkgC6TNFvr/images/cd695bc69edc0e4d86e52895a51017a2de68f0f2a6f9a0c8d0de8cc90dcc4a69.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5ffcb5003e6fe79733437196bbd08acfc5064744 --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/cd695bc69edc0e4d86e52895a51017a2de68f0f2a6f9a0c8d0de8cc90dcc4a69.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:591d2381e61a109f71c8ab4f73fb966412942ac2d97db2876ff49e8b5c6eba1a +size 19147 diff --git a/parse/train/SkgC6TNFvr/images/d012cafdad02bae560a810cb1f412d8c57946e28e0cd701c49531a18ee2e6703.jpg b/parse/train/SkgC6TNFvr/images/d012cafdad02bae560a810cb1f412d8c57946e28e0cd701c49531a18ee2e6703.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3e998e31e8d5f0473645e5d1922518a38463cf8f --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/d012cafdad02bae560a810cb1f412d8c57946e28e0cd701c49531a18ee2e6703.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4aa8fbb1c5b031aba19ff7e306298f8a10b34b5a8031e3191d0f5eb887fcd93d +size 44556 diff --git a/parse/train/SkgC6TNFvr/images/dea64fe0b04ddff679237bc9e2c1967b82f363d3eba77a7d648013ebce7e3c48.jpg b/parse/train/SkgC6TNFvr/images/dea64fe0b04ddff679237bc9e2c1967b82f363d3eba77a7d648013ebce7e3c48.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c95f3025033629f53a139b82aec7a1bf1ecdcc8b --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/dea64fe0b04ddff679237bc9e2c1967b82f363d3eba77a7d648013ebce7e3c48.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aedf1c5076d524499186e8c1b3b971f6a02a7cb36ec1e95d450299531dadc2a4 +size 15541 diff --git a/parse/train/SkgC6TNFvr/images/e75325e9e68273a4e78cc62e0c3b8ff13f0475c2729371d333b211099f32e9fd.jpg b/parse/train/SkgC6TNFvr/images/e75325e9e68273a4e78cc62e0c3b8ff13f0475c2729371d333b211099f32e9fd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8fd6f2b0839799c1e2ea10ea73309199e68cbe9a --- /dev/null +++ b/parse/train/SkgC6TNFvr/images/e75325e9e68273a4e78cc62e0c3b8ff13f0475c2729371d333b211099f32e9fd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e76d8a6d8500f8677c0fd576ecdd89b4fa6ef486b3dd3fbba61c0ad78a5264de +size 7752 diff --git a/parse/train/SylzhkBtDB/images/01c26c50f25dc8fc5f4d775aa9d443a3b8219ec826481c2269715d3153b97069.jpg b/parse/train/SylzhkBtDB/images/01c26c50f25dc8fc5f4d775aa9d443a3b8219ec826481c2269715d3153b97069.jpg new file mode 100644 index 0000000000000000000000000000000000000000..32dcd2e6af32a64511f58c5ed716a79045f75d80 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/01c26c50f25dc8fc5f4d775aa9d443a3b8219ec826481c2269715d3153b97069.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02224486542acc6bcef6b499641ef45621c9363ab726a78ee02058326e4475ba +size 20230 diff --git a/parse/train/SylzhkBtDB/images/06f84f2f7c917d5b4711a549a8a82450b30fc0fbdfb27a2c0a6151a653d42785.jpg b/parse/train/SylzhkBtDB/images/06f84f2f7c917d5b4711a549a8a82450b30fc0fbdfb27a2c0a6151a653d42785.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e223932030d3c31cbd47f5571a596a270c8a2ba9 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/06f84f2f7c917d5b4711a549a8a82450b30fc0fbdfb27a2c0a6151a653d42785.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2844d3cc0b78452f06e4106401300cd0435df523a5290ef73cfe3c9f89197886 +size 69751 diff --git a/parse/train/SylzhkBtDB/images/0e1cfe2824816d986b03ed84166f9d8eafd2a9adbca94176bf67a8b7bf88f7e5.jpg b/parse/train/SylzhkBtDB/images/0e1cfe2824816d986b03ed84166f9d8eafd2a9adbca94176bf67a8b7bf88f7e5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d8b490236f84428ede0202a2a18ffb8cabcdd7c1 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/0e1cfe2824816d986b03ed84166f9d8eafd2a9adbca94176bf67a8b7bf88f7e5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f7e0a107a2ba2e575f31ea67db598217ad9fdca063c6df9d1e569707d523921 +size 44804 diff --git a/parse/train/SylzhkBtDB/images/1052f4bdd0f3f6ffe3e688628fa98a499a0b31d4fed473e7c6a6a83d8c4488f1.jpg b/parse/train/SylzhkBtDB/images/1052f4bdd0f3f6ffe3e688628fa98a499a0b31d4fed473e7c6a6a83d8c4488f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1cc581012b3dd10fc6767ab61835a7cc75c2111c --- /dev/null +++ b/parse/train/SylzhkBtDB/images/1052f4bdd0f3f6ffe3e688628fa98a499a0b31d4fed473e7c6a6a83d8c4488f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed24aac89e825d42b70e4788410f6a215d65a2afac6bd3cefbbb40f411f68aae +size 23134 diff --git a/parse/train/SylzhkBtDB/images/1265067e12279ba1311e637db5e6cdb4feb067631c7faec01094b7e79476e1a6.jpg b/parse/train/SylzhkBtDB/images/1265067e12279ba1311e637db5e6cdb4feb067631c7faec01094b7e79476e1a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d7eab1a45112316826b7e635da38d6000ff6e02b --- /dev/null +++ b/parse/train/SylzhkBtDB/images/1265067e12279ba1311e637db5e6cdb4feb067631c7faec01094b7e79476e1a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9507b02a05c4aedf8412624e6dc306e67dcfafc83dd5ecc0a0dcfb61764cfd88 +size 10335 diff --git a/parse/train/SylzhkBtDB/images/136d00318e3fe1d6a44084c2009f245339789f3489c472a8ea66b452c80ec2f8.jpg b/parse/train/SylzhkBtDB/images/136d00318e3fe1d6a44084c2009f245339789f3489c472a8ea66b452c80ec2f8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..495ae0ce2597a9467f2022633f19bbb151d0c81f --- /dev/null +++ b/parse/train/SylzhkBtDB/images/136d00318e3fe1d6a44084c2009f245339789f3489c472a8ea66b452c80ec2f8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ae9a2fb29611b2479091cab0e060b2a987027bce7af47e01c9c8e2604d84661 +size 2285 diff --git a/parse/train/SylzhkBtDB/images/1bcbe7afdf2993a9e5f2e6fa11b517f1b12e64a974c3ee286f6cf62d73f294bf.jpg b/parse/train/SylzhkBtDB/images/1bcbe7afdf2993a9e5f2e6fa11b517f1b12e64a974c3ee286f6cf62d73f294bf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..067d3e79803b8715dd4850973c0059699a5549b3 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/1bcbe7afdf2993a9e5f2e6fa11b517f1b12e64a974c3ee286f6cf62d73f294bf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2d81ea5de0dfc744a08a9852f3ffa20d682d731f7e933ef5179efe865f86b0f +size 12042 diff --git a/parse/train/SylzhkBtDB/images/2179c45727139cc439a82dc7ecfdf25e8c20065d850bc74018a3787a65548574.jpg b/parse/train/SylzhkBtDB/images/2179c45727139cc439a82dc7ecfdf25e8c20065d850bc74018a3787a65548574.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b3f23631c0955253836ff6473a96bce4db32b957 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/2179c45727139cc439a82dc7ecfdf25e8c20065d850bc74018a3787a65548574.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22beb344ac816c854cabdcee94d0dd56505f55d9275fa0bd7d04644acf034fa0 +size 6646 diff --git a/parse/train/SylzhkBtDB/images/2478eec2e7c46c0c7471daeefb3c793c3ebca470681809224e929f909ba71303.jpg b/parse/train/SylzhkBtDB/images/2478eec2e7c46c0c7471daeefb3c793c3ebca470681809224e929f909ba71303.jpg new file mode 100644 index 0000000000000000000000000000000000000000..399386056043ed21dba8e9ba7875dd8367aca390 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/2478eec2e7c46c0c7471daeefb3c793c3ebca470681809224e929f909ba71303.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dca98249608918a303dbc7fa1e54fd0ee8a796d393d5537894d5949f5fbe890e +size 13274 diff --git a/parse/train/SylzhkBtDB/images/2c3119c6f7dbbb5d9d38281ebc9f492af6c43598b77ca43065483b5361b117f6.jpg b/parse/train/SylzhkBtDB/images/2c3119c6f7dbbb5d9d38281ebc9f492af6c43598b77ca43065483b5361b117f6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9ae226c280c73fd92be4da19ed25d931a6349131 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/2c3119c6f7dbbb5d9d38281ebc9f492af6c43598b77ca43065483b5361b117f6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a37eac97d551d660200e1f2d012c63cf9c3aa98364444d108a8e3b699a1a40c +size 13578 diff --git a/parse/train/SylzhkBtDB/images/331feaf72a7dc0705355a7bc78000fc7c6ff78adf0c2d0fe9b205e4e41e5bc47.jpg b/parse/train/SylzhkBtDB/images/331feaf72a7dc0705355a7bc78000fc7c6ff78adf0c2d0fe9b205e4e41e5bc47.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0f0fcc1969dd33834afcdac5e62a20d895d71d50 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/331feaf72a7dc0705355a7bc78000fc7c6ff78adf0c2d0fe9b205e4e41e5bc47.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:81dfc85cccb37f741fc57acf1b26a832939774579ce5e6e525119bde13b10ae2 +size 15500 diff --git a/parse/train/SylzhkBtDB/images/409adb75cc0f593df018a96691db4b2a73a5b8d62fc9948c9914654434713b51.jpg b/parse/train/SylzhkBtDB/images/409adb75cc0f593df018a96691db4b2a73a5b8d62fc9948c9914654434713b51.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b343fcd92f435ed277cc1f7fd3077e273e54ad2c --- /dev/null +++ b/parse/train/SylzhkBtDB/images/409adb75cc0f593df018a96691db4b2a73a5b8d62fc9948c9914654434713b51.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e59834ced624f2ba9798def0801eb5f5e45deb70c46fee10a164c5715f9cf8d8 +size 61514 diff --git a/parse/train/SylzhkBtDB/images/424c89f310e1119870696e57c57b8eea3f5244978625b8d0396caf02301c1bde.jpg b/parse/train/SylzhkBtDB/images/424c89f310e1119870696e57c57b8eea3f5244978625b8d0396caf02301c1bde.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6d71a5e1e422a9e097af2acd719ef34abe99b8c0 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/424c89f310e1119870696e57c57b8eea3f5244978625b8d0396caf02301c1bde.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c451fb8e115f7e1387fd8eee117199b7af90153c367448fbda636508638c1aad +size 15212 diff --git a/parse/train/SylzhkBtDB/images/4a4cbbd0c8ce458856f5e18a4ed2ce1c6ca6276040e7ba4aa95b6411c819fdf6.jpg b/parse/train/SylzhkBtDB/images/4a4cbbd0c8ce458856f5e18a4ed2ce1c6ca6276040e7ba4aa95b6411c819fdf6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..78be5ff94efe1dcadba4e88bb909deff86a28917 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/4a4cbbd0c8ce458856f5e18a4ed2ce1c6ca6276040e7ba4aa95b6411c819fdf6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e5fe6a81c9c728a0751a01fab4fca194a7018b4680bdba5eca03b5fc9bdf7812 +size 28003 diff --git a/parse/train/SylzhkBtDB/images/4ea7c670f5c2dff371664ded10d53cca93690cccb08226ebcc99eacb63d23ccb.jpg b/parse/train/SylzhkBtDB/images/4ea7c670f5c2dff371664ded10d53cca93690cccb08226ebcc99eacb63d23ccb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e90496c2d19a885aaba1e535a8cdeab40108cfc4 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/4ea7c670f5c2dff371664ded10d53cca93690cccb08226ebcc99eacb63d23ccb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3aa7714acfcc19083ed5582175b656c4b6109982538c56bceaa2bf5164a50c0c +size 7628 diff --git a/parse/train/SylzhkBtDB/images/537c2d12833f37adfe34301597d4456222a507d71c992e7d46ea2b8d7f34d90e.jpg b/parse/train/SylzhkBtDB/images/537c2d12833f37adfe34301597d4456222a507d71c992e7d46ea2b8d7f34d90e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..75b28970fcdc937ba505299c279cb994f8491e17 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/537c2d12833f37adfe34301597d4456222a507d71c992e7d46ea2b8d7f34d90e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:616f13f468b3604cd3b070dce914ace601404233532eee1d27bc8f0d2f75553c +size 9928 diff --git a/parse/train/SylzhkBtDB/images/5460973e541c1c7b8e01bc7e799a8a687ea0e8d27a402f378ae12618141dfb15.jpg b/parse/train/SylzhkBtDB/images/5460973e541c1c7b8e01bc7e799a8a687ea0e8d27a402f378ae12618141dfb15.jpg new file mode 100644 index 0000000000000000000000000000000000000000..968705824a1ad6e2c0f0296d6658164a5de8daee --- /dev/null +++ b/parse/train/SylzhkBtDB/images/5460973e541c1c7b8e01bc7e799a8a687ea0e8d27a402f378ae12618141dfb15.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d704d6b4cac14427cc9f7b2898ce2b0aa83927e579df0e1f5576d87c11c309a2 +size 11270 diff --git a/parse/train/SylzhkBtDB/images/547f5291075a7c1696553ad2984a7b8ca97db74e01282a68ce6f6afdedb79abf.jpg b/parse/train/SylzhkBtDB/images/547f5291075a7c1696553ad2984a7b8ca97db74e01282a68ce6f6afdedb79abf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f84afbf42692ea8b1e3f8095abc4253521ef22bc --- /dev/null +++ b/parse/train/SylzhkBtDB/images/547f5291075a7c1696553ad2984a7b8ca97db74e01282a68ce6f6afdedb79abf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82beea43b6342e775979149859ec8f663a40856b204156ae8a172c6ac781b030 +size 4455 diff --git a/parse/train/SylzhkBtDB/images/555c4d9490c55a0bace7076eef4f427ead0bb731ed66d95cff6d3ddd1df59b16.jpg b/parse/train/SylzhkBtDB/images/555c4d9490c55a0bace7076eef4f427ead0bb731ed66d95cff6d3ddd1df59b16.jpg new file mode 100644 index 0000000000000000000000000000000000000000..25d4b3b6c5842208d070895417bf3ad4fc07a64d --- /dev/null +++ b/parse/train/SylzhkBtDB/images/555c4d9490c55a0bace7076eef4f427ead0bb731ed66d95cff6d3ddd1df59b16.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbd119ca1fafb1abd97c159a8b62ba924da9ed312b308419a5474dd08765aea7 +size 14822 diff --git a/parse/train/SylzhkBtDB/images/5819b7d36c90f66bb69c9a8bac924e258999184d467eaaef44af7e9fe99f9fdb.jpg b/parse/train/SylzhkBtDB/images/5819b7d36c90f66bb69c9a8bac924e258999184d467eaaef44af7e9fe99f9fdb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..05d2d92f348d08d4f06350051128ce27b6713631 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/5819b7d36c90f66bb69c9a8bac924e258999184d467eaaef44af7e9fe99f9fdb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:55b3c6e2d9fdc335aa9d5758d66f726dd9fa63c3fec26d4689a841fc014d20b2 +size 14485 diff --git a/parse/train/SylzhkBtDB/images/76a67ca6ff94f3bf2c24e7b00257e17e704b7b1f43884022580a9703bf40ece4.jpg b/parse/train/SylzhkBtDB/images/76a67ca6ff94f3bf2c24e7b00257e17e704b7b1f43884022580a9703bf40ece4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..74fd706b8afa942e1bb9fb69387352d8cb445473 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/76a67ca6ff94f3bf2c24e7b00257e17e704b7b1f43884022580a9703bf40ece4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa7b234390a82309cd1d5c0fa59aeec2a9dd75c8f2f59afaaaf137a9efdbb62f +size 12093 diff --git a/parse/train/SylzhkBtDB/images/778b9dd9c57652cc34e19ac16e1b1478f377195f761640003dd18ddba2bd9b13.jpg b/parse/train/SylzhkBtDB/images/778b9dd9c57652cc34e19ac16e1b1478f377195f761640003dd18ddba2bd9b13.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a15c2136e02c3520546e734f10529f29ca82fa44 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/778b9dd9c57652cc34e19ac16e1b1478f377195f761640003dd18ddba2bd9b13.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:253b62c3c63bbe94aaee7d96dd3759ff31d5b86c2886b7eafb1bf014cde0dbe1 +size 16542 diff --git a/parse/train/SylzhkBtDB/images/79c9d288625d84b09a735a3f3b6166b7671b646fd425a723fa7ce77c8a633ede.jpg b/parse/train/SylzhkBtDB/images/79c9d288625d84b09a735a3f3b6166b7671b646fd425a723fa7ce77c8a633ede.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a86cf973a3d6f3f090a1e44216cfe52ab8e24297 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/79c9d288625d84b09a735a3f3b6166b7671b646fd425a723fa7ce77c8a633ede.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcfc1855d1a6de6f91d7ffe5ba229c158e3e5704ec2ad4d15be7763230246057 +size 9701 diff --git a/parse/train/SylzhkBtDB/images/8288118373304628edbb8c3a109559c520573a4f5686a3a510adb3123bea4d4c.jpg b/parse/train/SylzhkBtDB/images/8288118373304628edbb8c3a109559c520573a4f5686a3a510adb3123bea4d4c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ced2820c939eb9a6ad69b550c619c3afa6cae2e2 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/8288118373304628edbb8c3a109559c520573a4f5686a3a510adb3123bea4d4c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a494206fa88664311b6e8b9e67e71259e1aa2a371d990b58feb088b5357a6d78 +size 6629 diff --git a/parse/train/SylzhkBtDB/images/8559aa086edf8e2dc0856ea2d4046794c5be807406e434b39f96313a564d95f2.jpg b/parse/train/SylzhkBtDB/images/8559aa086edf8e2dc0856ea2d4046794c5be807406e434b39f96313a564d95f2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a2f3a66a6ebbed35c345597830319c8c9a41c60d --- /dev/null +++ b/parse/train/SylzhkBtDB/images/8559aa086edf8e2dc0856ea2d4046794c5be807406e434b39f96313a564d95f2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58fd538382a6dd5922043849d74ec70177ea1bc70641a2aa1740e31342f70b83 +size 5378 diff --git a/parse/train/SylzhkBtDB/images/8663dc9f667369dce0825923d28eed7eeb9b5b560be817f836e69adc1c52a2be.jpg b/parse/train/SylzhkBtDB/images/8663dc9f667369dce0825923d28eed7eeb9b5b560be817f836e69adc1c52a2be.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5b0d306d5b108f45ab3a1de4d7c8d91ac940ce5 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/8663dc9f667369dce0825923d28eed7eeb9b5b560be817f836e69adc1c52a2be.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:425afc16f694f692d757a07aff53d3c9190a8254c37600171c6885bc15ed1dd4 +size 56370 diff --git a/parse/train/SylzhkBtDB/images/872064d322dbc97b9800ba46a0b15df073b87771a77259f8e2eab170fd5cbdcd.jpg b/parse/train/SylzhkBtDB/images/872064d322dbc97b9800ba46a0b15df073b87771a77259f8e2eab170fd5cbdcd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2ef718d710cb69b2aec89bbaaf7aebdd4a5032d9 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/872064d322dbc97b9800ba46a0b15df073b87771a77259f8e2eab170fd5cbdcd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cdf51b1adb949aa1ca0ec2aaad7267be4bfa49e3844b0e7fc66e1dd1f1e8eae9 +size 84323 diff --git a/parse/train/SylzhkBtDB/images/8800c4a052dcc9039a39f30d4701cdd0c8edd8274e6c74141861bfbf1fdb4dac.jpg b/parse/train/SylzhkBtDB/images/8800c4a052dcc9039a39f30d4701cdd0c8edd8274e6c74141861bfbf1fdb4dac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1758898c734dfca1d93b6628ae7028b23cc73019 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/8800c4a052dcc9039a39f30d4701cdd0c8edd8274e6c74141861bfbf1fdb4dac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4bbb289429c8d3e964aa94d315672e78071951139944519c7ed094d004e9ce47 +size 89713 diff --git a/parse/train/SylzhkBtDB/images/8a38cc322546c899d515fe8834c32b9238ff7051c23776ce7df9f096a78b43d6.jpg b/parse/train/SylzhkBtDB/images/8a38cc322546c899d515fe8834c32b9238ff7051c23776ce7df9f096a78b43d6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b2b3ef6213c7175cf7c8fad4ce573118557d4351 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/8a38cc322546c899d515fe8834c32b9238ff7051c23776ce7df9f096a78b43d6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39e3c3e44c70aaba3a8f49350712aa71997af48ef66719b1824298c68200e86e +size 31748 diff --git a/parse/train/SylzhkBtDB/images/8be52cb6cbb116edff9ab311c81a1dc9c2af8134a484facaebadd03a4b8d0d69.jpg b/parse/train/SylzhkBtDB/images/8be52cb6cbb116edff9ab311c81a1dc9c2af8134a484facaebadd03a4b8d0d69.jpg new file mode 100644 index 0000000000000000000000000000000000000000..72a9bd60a895903f8fc6a22588b8a6b42c158c00 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/8be52cb6cbb116edff9ab311c81a1dc9c2af8134a484facaebadd03a4b8d0d69.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad13fba541ad2db20159ddf1c9179c4441add8791c765dde9676924da420d7dd +size 6620 diff --git a/parse/train/SylzhkBtDB/images/8f5e67c48e7a328f71e0cbd03a237756132562ff114a806938ab730f8be8003d.jpg b/parse/train/SylzhkBtDB/images/8f5e67c48e7a328f71e0cbd03a237756132562ff114a806938ab730f8be8003d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..84cafe98e82128a8a3ec8b7c8e5ccd57e0f7fdce --- /dev/null +++ b/parse/train/SylzhkBtDB/images/8f5e67c48e7a328f71e0cbd03a237756132562ff114a806938ab730f8be8003d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e5b4804b646fc29820f49be9a49f276b57a1447301595f958a9ea3250020767 +size 12005 diff --git a/parse/train/SylzhkBtDB/images/9a05cfb6e92e8252ee550b7f75ea37e50558582c4e84c6b172292092e1020453.jpg b/parse/train/SylzhkBtDB/images/9a05cfb6e92e8252ee550b7f75ea37e50558582c4e84c6b172292092e1020453.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c22ba5a3aae9fd5e1fb60c3a855decc953e6524a --- /dev/null +++ b/parse/train/SylzhkBtDB/images/9a05cfb6e92e8252ee550b7f75ea37e50558582c4e84c6b172292092e1020453.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09d72905dbeb9555e395c006ab51591c811b044dcdb93ec45c64928a301899e1 +size 8120 diff --git a/parse/train/SylzhkBtDB/images/a2230a0464edae9189c13a8bae4c6aa24cf185252c9ba2e25509933b30bf8bac.jpg b/parse/train/SylzhkBtDB/images/a2230a0464edae9189c13a8bae4c6aa24cf185252c9ba2e25509933b30bf8bac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f5402bb8ba902bae4be13033adcc62519cd0be0d --- /dev/null +++ b/parse/train/SylzhkBtDB/images/a2230a0464edae9189c13a8bae4c6aa24cf185252c9ba2e25509933b30bf8bac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:35d7b47ba88a3e7c51e8c4481439ca714278c139c4186ffb1108e868f4bfbaf4 +size 85827 diff --git a/parse/train/SylzhkBtDB/images/a31f12c10831f8a99fbec36239c5b6ae473b79d173881ecf49804e042f833175.jpg b/parse/train/SylzhkBtDB/images/a31f12c10831f8a99fbec36239c5b6ae473b79d173881ecf49804e042f833175.jpg new file mode 100644 index 0000000000000000000000000000000000000000..be9ff874b046967a256707b514d1245f6b63e652 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/a31f12c10831f8a99fbec36239c5b6ae473b79d173881ecf49804e042f833175.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e1e08705b0ce6c7069be67abe577117f8f0da688ba9ed1b44c7e7d1e14b9f46 +size 6832 diff --git a/parse/train/SylzhkBtDB/images/b209d0896e9e3bae8cfbdbc299f356db8708c5e7971930ae6fe80a9ced005c1b.jpg b/parse/train/SylzhkBtDB/images/b209d0896e9e3bae8cfbdbc299f356db8708c5e7971930ae6fe80a9ced005c1b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3918adc11fc7beaabf88c17cb289cb088d28864d --- /dev/null +++ b/parse/train/SylzhkBtDB/images/b209d0896e9e3bae8cfbdbc299f356db8708c5e7971930ae6fe80a9ced005c1b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d66996ce5f2a9a9bf09d8ef4011ec4855edfea54fabf2b1a6a3aba276165034 +size 8408 diff --git a/parse/train/SylzhkBtDB/images/b462276de8443b1fa57442aa54b577156fa19869cec52e851d603702d0be77ed.jpg b/parse/train/SylzhkBtDB/images/b462276de8443b1fa57442aa54b577156fa19869cec52e851d603702d0be77ed.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ae5f3e100aa775de01278d348c4e45638a6f1666 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/b462276de8443b1fa57442aa54b577156fa19869cec52e851d603702d0be77ed.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4dbbb18cfd5e0adbfcfb4891a3e668cd997ba192154799e99b810c9ebd171aa4 +size 20616 diff --git a/parse/train/SylzhkBtDB/images/b48820e0b44d6949a08ab8b584d69f14792a26e88d9b5990d03e8af2f3c6b53b.jpg b/parse/train/SylzhkBtDB/images/b48820e0b44d6949a08ab8b584d69f14792a26e88d9b5990d03e8af2f3c6b53b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..76a57c8c7e8008c2c3a2ef381e3705f6024e9d42 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/b48820e0b44d6949a08ab8b584d69f14792a26e88d9b5990d03e8af2f3c6b53b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01f8ed2ec48af2c5763864ef626a11bd338d702d3f3aa25833ee13e827ee2f3d +size 5209 diff --git a/parse/train/SylzhkBtDB/images/ba143d87f5241d2cca1b7992e7f6f7e7958cde7c57abd15e015ad1553cedb65a.jpg b/parse/train/SylzhkBtDB/images/ba143d87f5241d2cca1b7992e7f6f7e7958cde7c57abd15e015ad1553cedb65a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..19fbea27b32c197ba94080a7323106926a911e1d --- /dev/null +++ b/parse/train/SylzhkBtDB/images/ba143d87f5241d2cca1b7992e7f6f7e7958cde7c57abd15e015ad1553cedb65a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6a729ebee9bbde49569cf078c204a82621bb9d3e73b21086dffb3dc05d31070 +size 3779 diff --git a/parse/train/SylzhkBtDB/images/c6d16097f6a81d4d70fbec1c64a278d9b04af6d5190afbc99db77f8eff724672.jpg b/parse/train/SylzhkBtDB/images/c6d16097f6a81d4d70fbec1c64a278d9b04af6d5190afbc99db77f8eff724672.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5759269274892c40d77350b36f594b38d31844ba --- /dev/null +++ b/parse/train/SylzhkBtDB/images/c6d16097f6a81d4d70fbec1c64a278d9b04af6d5190afbc99db77f8eff724672.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a2537df68089167d93c1fc1db10f6156605301b0911d82fc34e8290e01c66ad +size 103561 diff --git a/parse/train/SylzhkBtDB/images/d132ea01545111dd1f93d2e205d3820881ff60318d53fd614172820d24bd8640.jpg b/parse/train/SylzhkBtDB/images/d132ea01545111dd1f93d2e205d3820881ff60318d53fd614172820d24bd8640.jpg new file mode 100644 index 0000000000000000000000000000000000000000..58b54b913d692542e096416c90305a235554e9f8 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/d132ea01545111dd1f93d2e205d3820881ff60318d53fd614172820d24bd8640.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2953de4911dedd9a9e16a44d0a3d1ff34ee6253ee412634ba60becb5ca391f12 +size 5054 diff --git a/parse/train/SylzhkBtDB/images/d4883375098b9d390a1d74e7a07d0768d2280caa329d6a71954320f20b1ca684.jpg b/parse/train/SylzhkBtDB/images/d4883375098b9d390a1d74e7a07d0768d2280caa329d6a71954320f20b1ca684.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dc9abdc57773c6115824f3ba4c2d144719c90cd5 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/d4883375098b9d390a1d74e7a07d0768d2280caa329d6a71954320f20b1ca684.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42d0426c703f750a7faa57d130f51af80cfd6818f5f889194cd270aba3e6ea86 +size 8378 diff --git a/parse/train/SylzhkBtDB/images/d844b6ab41f0aac17a03fdd2536bd6cfaed1d59efbc6eec9f5fa4382240a9293.jpg b/parse/train/SylzhkBtDB/images/d844b6ab41f0aac17a03fdd2536bd6cfaed1d59efbc6eec9f5fa4382240a9293.jpg new file mode 100644 index 0000000000000000000000000000000000000000..30ff52883db91f268eddc1284bcc223f76fae8b1 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/d844b6ab41f0aac17a03fdd2536bd6cfaed1d59efbc6eec9f5fa4382240a9293.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:440df6d64ad7c9a60d54b7579707a654a2cbd648a2496f3a7b6bf97614473082 +size 6812 diff --git a/parse/train/SylzhkBtDB/images/e8101e890ade49385a09291a61517bca41f93ea277b7e89ae616e9a0b8bb4bf8.jpg b/parse/train/SylzhkBtDB/images/e8101e890ade49385a09291a61517bca41f93ea277b7e89ae616e9a0b8bb4bf8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..97ebc0cf3a0ca280adc7513aa9fbbdd6550b58e4 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/e8101e890ade49385a09291a61517bca41f93ea277b7e89ae616e9a0b8bb4bf8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5621a68e08a25de22041b127c3db23bca1347edda2157a2563dd3342fcc47640 +size 17014 diff --git a/parse/train/SylzhkBtDB/images/fb5f848ba2f7e226e92f2b70843af03d1c321d584bd81b99fe32ee9699ac6b45.jpg b/parse/train/SylzhkBtDB/images/fb5f848ba2f7e226e92f2b70843af03d1c321d584bd81b99fe32ee9699ac6b45.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eca3bb840c8d192a8efb078c86d271145867a065 --- /dev/null +++ b/parse/train/SylzhkBtDB/images/fb5f848ba2f7e226e92f2b70843af03d1c321d584bd81b99fe32ee9699ac6b45.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:667c235123df97ea4b1699279129701347b8850b710d92ac8c0f256fb7368e2a +size 8846 diff --git a/parse/train/Ua6zuk0WRH/images/078976a0f499bdd0cb05b7ea0602bacb211cfd378bf6b09022d12879c09c7e3e.jpg b/parse/train/Ua6zuk0WRH/images/078976a0f499bdd0cb05b7ea0602bacb211cfd378bf6b09022d12879c09c7e3e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a02a9fe3f8d2ebe8c4deb3bcf62ddd2f0f63374c --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/078976a0f499bdd0cb05b7ea0602bacb211cfd378bf6b09022d12879c09c7e3e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c6acf54fd26492596dc0d46ecc0c0612f10b1a747108d2893f7ebb318ac1088 +size 11799 diff --git a/parse/train/Ua6zuk0WRH/images/0a0828266dd58606f1f51ff4f00b8d5be82cce989e3bb3b144f7746856538b0b.jpg b/parse/train/Ua6zuk0WRH/images/0a0828266dd58606f1f51ff4f00b8d5be82cce989e3bb3b144f7746856538b0b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3c1e5d73584ad44873079cf6ef2f0698951db6ae --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/0a0828266dd58606f1f51ff4f00b8d5be82cce989e3bb3b144f7746856538b0b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b368f27f88ce309cf31a4047b41c803d065ac718e91d342cbf1a8c8f0533b67 +size 8467 diff --git a/parse/train/Ua6zuk0WRH/images/136897e184b6579546e0f53d3b33f825bf2a0eff2233875384d9a83b817cbc1f.jpg b/parse/train/Ua6zuk0WRH/images/136897e184b6579546e0f53d3b33f825bf2a0eff2233875384d9a83b817cbc1f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7657d0ca038400a4251b7d21aed019002c30a642 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/136897e184b6579546e0f53d3b33f825bf2a0eff2233875384d9a83b817cbc1f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d179ee55c6b6b5875eaf24fe38d22815ef0a0c43724d068d8a69567aab12940 +size 11595 diff --git a/parse/train/Ua6zuk0WRH/images/22107213203cecd44294d1040700af553b789b46f0600023057b504f77988c61.jpg b/parse/train/Ua6zuk0WRH/images/22107213203cecd44294d1040700af553b789b46f0600023057b504f77988c61.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bb7af5485c4c740cb3b1412e909b293a06166035 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/22107213203cecd44294d1040700af553b789b46f0600023057b504f77988c61.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96fedbef98d32073f1ee60e9f7a7ee94d1bcf706033c9f0a53ea10aac60d3599 +size 3824 diff --git a/parse/train/Ua6zuk0WRH/images/2618018843a1345e38095525ffcad1784597054f878f031754b95ee7671c7ebe.jpg b/parse/train/Ua6zuk0WRH/images/2618018843a1345e38095525ffcad1784597054f878f031754b95ee7671c7ebe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dacd859a0c6ac3d5fde24c2b530da318bc4761f1 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/2618018843a1345e38095525ffcad1784597054f878f031754b95ee7671c7ebe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00ac1d9878aa35a5006c85fbb0fd1e8b7508c3730620ebcdfe990573854c6c6e +size 11731 diff --git a/parse/train/Ua6zuk0WRH/images/31d4aa15952a521c2c07a1ea46c52fd6b57a08ef04c41d14e8c4cbc19a65aaca.jpg b/parse/train/Ua6zuk0WRH/images/31d4aa15952a521c2c07a1ea46c52fd6b57a08ef04c41d14e8c4cbc19a65aaca.jpg new file mode 100644 index 0000000000000000000000000000000000000000..94c284b4e840e3814607764f1635165691a0e90c --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/31d4aa15952a521c2c07a1ea46c52fd6b57a08ef04c41d14e8c4cbc19a65aaca.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23fdfabfb60320ee8a614fdd752b645b21933149dc12c6e214acf680dc0d620b +size 55861 diff --git a/parse/train/Ua6zuk0WRH/images/3a08d3d40bc5ff0e22d340d417c4b03a4cf7bf3770236fcfa3bafe676dc5b2ee.jpg b/parse/train/Ua6zuk0WRH/images/3a08d3d40bc5ff0e22d340d417c4b03a4cf7bf3770236fcfa3bafe676dc5b2ee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8e7a00451a8b2e44aa5d28a15f529b0fdf6cb09a --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/3a08d3d40bc5ff0e22d340d417c4b03a4cf7bf3770236fcfa3bafe676dc5b2ee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d68e7cc31d1cac5112b0ab78453920fc25d30e4edbe6b343925aa37f051dd39d +size 44324 diff --git a/parse/train/Ua6zuk0WRH/images/3c71a0b8239d3985443de2a7e5c3ee40eeda8d7b1bcc23ce989757f36b7c7129.jpg b/parse/train/Ua6zuk0WRH/images/3c71a0b8239d3985443de2a7e5c3ee40eeda8d7b1bcc23ce989757f36b7c7129.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b4a651d5177ae28b491ab0a1f731f73a9d5ee663 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/3c71a0b8239d3985443de2a7e5c3ee40eeda8d7b1bcc23ce989757f36b7c7129.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1d7964e930d977b1bc7d4d3e86831e8638dc200eaa9764d9eacff5bba995174 +size 13184 diff --git a/parse/train/Ua6zuk0WRH/images/4d2d7d2488e675e5b4bb184fc491a5533780523f0b42637d8263b0fbf1355752.jpg b/parse/train/Ua6zuk0WRH/images/4d2d7d2488e675e5b4bb184fc491a5533780523f0b42637d8263b0fbf1355752.jpg new file mode 100644 index 0000000000000000000000000000000000000000..78e6032d3d570832c529be2111c3e7e82e722ef2 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/4d2d7d2488e675e5b4bb184fc491a5533780523f0b42637d8263b0fbf1355752.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4cac3327600408175ba4c01fcd81d68bcd6d4db0329c45d69f1cace7f36d2c39 +size 12800 diff --git a/parse/train/Ua6zuk0WRH/images/53cbdf0285bbcbf1548b823008788eabc833ee1f64540ae48ecca11daace21b7.jpg b/parse/train/Ua6zuk0WRH/images/53cbdf0285bbcbf1548b823008788eabc833ee1f64540ae48ecca11daace21b7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..423c574ea6a9377e934d599905caf7df4b0105d8 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/53cbdf0285bbcbf1548b823008788eabc833ee1f64540ae48ecca11daace21b7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87857dd5763a5ef6b62a7d1fc4440899a132b97d7d87e6a6a67c2e5b4c5b1c13 +size 12422 diff --git a/parse/train/Ua6zuk0WRH/images/57cc0c79e37543f8a417f1e15efb7acdc4a6284e97f932e7184030dccb7d1620.jpg b/parse/train/Ua6zuk0WRH/images/57cc0c79e37543f8a417f1e15efb7acdc4a6284e97f932e7184030dccb7d1620.jpg new file mode 100644 index 0000000000000000000000000000000000000000..391af383cfd7af029d4e6a7ac95a70d2f215121e --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/57cc0c79e37543f8a417f1e15efb7acdc4a6284e97f932e7184030dccb7d1620.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f72883dbeba7fb875e7289279868adbac0526731e10efb471ea0a4fd05af2530 +size 10744 diff --git a/parse/train/Ua6zuk0WRH/images/597752c8656e4e600e394bcc97f76ad2fbfdad68739ed28a6d59262b28b01743.jpg b/parse/train/Ua6zuk0WRH/images/597752c8656e4e600e394bcc97f76ad2fbfdad68739ed28a6d59262b28b01743.jpg new file mode 100644 index 0000000000000000000000000000000000000000..288f5e7e6ea91be5f0ceb675d3fc3f3555fb4ba6 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/597752c8656e4e600e394bcc97f76ad2fbfdad68739ed28a6d59262b28b01743.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff4c0dde04557643ecd07cd898655845acbae1c5eda8aeb72f96dfbf28f877a3 +size 20605 diff --git a/parse/train/Ua6zuk0WRH/images/5e8da28d0ea10e1fc8f82b0958dd9cee1564e6f67d6dd34750243ab7231d423a.jpg b/parse/train/Ua6zuk0WRH/images/5e8da28d0ea10e1fc8f82b0958dd9cee1564e6f67d6dd34750243ab7231d423a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..136efa3cffd61a1756a2968ec91171fdcb3a91e5 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/5e8da28d0ea10e1fc8f82b0958dd9cee1564e6f67d6dd34750243ab7231d423a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4733e1ba110d6ac9e27b0c116b108a1a928596bf3d9c868c1cbebca2fe173e5 +size 10037 diff --git a/parse/train/Ua6zuk0WRH/images/63d3b8b75f241b2e23665c44c93be89d5ce52e0ade957cb90cc5d6283df2fed1.jpg b/parse/train/Ua6zuk0WRH/images/63d3b8b75f241b2e23665c44c93be89d5ce52e0ade957cb90cc5d6283df2fed1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6155916d6904833f05518787368018384ea433a5 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/63d3b8b75f241b2e23665c44c93be89d5ce52e0ade957cb90cc5d6283df2fed1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f42eb77b807b1437ef57453042feccc096352b36b3657e51c9aa98c00a169e30 +size 18299 diff --git a/parse/train/Ua6zuk0WRH/images/66f462e5c2d830968dc794ea7d883010c79f477a02e17d13797952acdbab0a7f.jpg b/parse/train/Ua6zuk0WRH/images/66f462e5c2d830968dc794ea7d883010c79f477a02e17d13797952acdbab0a7f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a16992485e8b828024c711ba6312ea8ef53b4557 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/66f462e5c2d830968dc794ea7d883010c79f477a02e17d13797952acdbab0a7f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b0b69b0d56c5abe609ba217a3683f8266d98b8270fed83e6dc91f0d06bb3e38 +size 56257 diff --git a/parse/train/Ua6zuk0WRH/images/6bd38350b189b3aa51c9e3e8d4218240dad297b3e1a3cdf218d283746de14cbb.jpg b/parse/train/Ua6zuk0WRH/images/6bd38350b189b3aa51c9e3e8d4218240dad297b3e1a3cdf218d283746de14cbb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..44d671cbb644eb6408e71f1b40d13af261f36958 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/6bd38350b189b3aa51c9e3e8d4218240dad297b3e1a3cdf218d283746de14cbb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6cee1c337eb7d482a9ca150e7535bd1d6672303079b5f563b3f1848f1e127476 +size 5205 diff --git a/parse/train/Ua6zuk0WRH/images/6bd6fff5a67548512ca4f2f95f4f5873b94cb97ac4e708d2c24754e903403294.jpg b/parse/train/Ua6zuk0WRH/images/6bd6fff5a67548512ca4f2f95f4f5873b94cb97ac4e708d2c24754e903403294.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1be1888b8e19193687d9beb3785a44050bef17b8 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/6bd6fff5a67548512ca4f2f95f4f5873b94cb97ac4e708d2c24754e903403294.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32ca8480285cd780c0a7bfe8a2953c543980225d7520bf5a5463362ade8abf3f +size 11669 diff --git a/parse/train/Ua6zuk0WRH/images/6c93b24d4829ff5bad8de2aa59ed0c36bc1f6d11b497124e503dea4e95cd79b5.jpg b/parse/train/Ua6zuk0WRH/images/6c93b24d4829ff5bad8de2aa59ed0c36bc1f6d11b497124e503dea4e95cd79b5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..61e98bdb1fff895598c0314a855b36c62eb6e973 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/6c93b24d4829ff5bad8de2aa59ed0c36bc1f6d11b497124e503dea4e95cd79b5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8747b906dad85022f74866e72f965b29cd92c8409c7ac67be86589bc98367b46 +size 5639 diff --git a/parse/train/Ua6zuk0WRH/images/71e46bc8024ceafe9a0dd09c82b47016a44360c3ecebd14be8cd616b41c752bb.jpg b/parse/train/Ua6zuk0WRH/images/71e46bc8024ceafe9a0dd09c82b47016a44360c3ecebd14be8cd616b41c752bb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..32e6a0fa983e4c9edab4c6e77ed1a7c164df23b2 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/71e46bc8024ceafe9a0dd09c82b47016a44360c3ecebd14be8cd616b41c752bb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c7940ab29236a691ed324ee36ccd4e7ab3ad8b4c4f241159177be1e4e34196c +size 9355 diff --git a/parse/train/Ua6zuk0WRH/images/77219576a9331e37609cf6c74a27395d846410bd816343cb2837b4a58211c224.jpg b/parse/train/Ua6zuk0WRH/images/77219576a9331e37609cf6c74a27395d846410bd816343cb2837b4a58211c224.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e668b39defd38faa312be973436878c72fac2572 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/77219576a9331e37609cf6c74a27395d846410bd816343cb2837b4a58211c224.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c56d70869d08eac71375896ba84d9e7f3079dee53bbcfe03a317b2f38a5a1b64 +size 61333 diff --git a/parse/train/Ua6zuk0WRH/images/79774ee5d6c572caea7b3a9fa07325e5712bee70c1126a6a2a929b18d307205d.jpg b/parse/train/Ua6zuk0WRH/images/79774ee5d6c572caea7b3a9fa07325e5712bee70c1126a6a2a929b18d307205d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..29ddeefeb46eae405478287ecf9da8fe6af702cc --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/79774ee5d6c572caea7b3a9fa07325e5712bee70c1126a6a2a929b18d307205d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03237981920a1a2ccb15289cc1f08ea8cf8a694172e022157a90e2cc13f067c8 +size 5817 diff --git a/parse/train/Ua6zuk0WRH/images/870ad0ec02820c247e3a8b0b40405199d3cae8f955fdcd1788dfd7c116918f3f.jpg b/parse/train/Ua6zuk0WRH/images/870ad0ec02820c247e3a8b0b40405199d3cae8f955fdcd1788dfd7c116918f3f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..043580f6d177e58fb3a3854d726303f043775eb6 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/870ad0ec02820c247e3a8b0b40405199d3cae8f955fdcd1788dfd7c116918f3f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ea4da4c0e8243fb02cbd8ea222bb53e08e115851d8d46dd02798c9dc4b7db08 +size 10169 diff --git a/parse/train/Ua6zuk0WRH/images/8a77079935ad622cf5bf447aca66a4c02af3d74e5125879074238b3477a385d7.jpg b/parse/train/Ua6zuk0WRH/images/8a77079935ad622cf5bf447aca66a4c02af3d74e5125879074238b3477a385d7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c3ae7cd7e06244b0b3e192ba9c289469859a3b7b --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/8a77079935ad622cf5bf447aca66a4c02af3d74e5125879074238b3477a385d7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:152ba6d500fb2b4c6f1d9777d7f0ff06e153a70a8b540426285ebdb39f1c25a1 +size 32404 diff --git a/parse/train/Ua6zuk0WRH/images/8f1f5b4b166db41bf7c8ad37ca5a504652274e1a656619d1a3dc6c15f047bfd0.jpg b/parse/train/Ua6zuk0WRH/images/8f1f5b4b166db41bf7c8ad37ca5a504652274e1a656619d1a3dc6c15f047bfd0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b9dff2867f572d6374bb628db821d239cd39d88 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/8f1f5b4b166db41bf7c8ad37ca5a504652274e1a656619d1a3dc6c15f047bfd0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9f6646ee9660e40294801b6a258a2510a3345a9471b7f73f3f4e1975929aeb8 +size 27909 diff --git a/parse/train/Ua6zuk0WRH/images/8f95a8b3dd1215644e3bcf1e8041e13f67d34f40d8a3b6c17f7beaee84bd861b.jpg b/parse/train/Ua6zuk0WRH/images/8f95a8b3dd1215644e3bcf1e8041e13f67d34f40d8a3b6c17f7beaee84bd861b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93902fedf8508c42c25e9dd6655e6118ae311988 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/8f95a8b3dd1215644e3bcf1e8041e13f67d34f40d8a3b6c17f7beaee84bd861b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:356e8997512695cc2cbc982aeacdbb25d38a8264216e8a1e126627735683c714 +size 42238 diff --git a/parse/train/Ua6zuk0WRH/images/9b3820d44f87186a131d1945ae248a75e9f39a0b800de46274d274bacb1f53d6.jpg b/parse/train/Ua6zuk0WRH/images/9b3820d44f87186a131d1945ae248a75e9f39a0b800de46274d274bacb1f53d6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9c1e1ec2883727c199e4d8ae89a16e3473c76aad --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/9b3820d44f87186a131d1945ae248a75e9f39a0b800de46274d274bacb1f53d6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1120d6c39926e7b29e32060847eebba2fef20880bf79b37d3a76262ac848952d +size 16286 diff --git a/parse/train/Ua6zuk0WRH/images/aa7a9c0568d7b5cd6f5c9c0368e924c1642ead31f4603c9204fea5dd8f0897bf.jpg b/parse/train/Ua6zuk0WRH/images/aa7a9c0568d7b5cd6f5c9c0368e924c1642ead31f4603c9204fea5dd8f0897bf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..74e90343af9150fcd0a726a4395a5ee75ca9c1a5 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/aa7a9c0568d7b5cd6f5c9c0368e924c1642ead31f4603c9204fea5dd8f0897bf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2838bd54b746a8762edc0ece0d76a3379e27dadb85d1fab38aed208758cda8d9 +size 4519 diff --git a/parse/train/Ua6zuk0WRH/images/ad6767a2c1e3093cfde9def47c71621fd93bcef37325d03a4075ee7032ec62bf.jpg b/parse/train/Ua6zuk0WRH/images/ad6767a2c1e3093cfde9def47c71621fd93bcef37325d03a4075ee7032ec62bf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..99bfc05b60b79b70c06c3872ba3787e09966d59c --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/ad6767a2c1e3093cfde9def47c71621fd93bcef37325d03a4075ee7032ec62bf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a319b1dea97841d7619ec3d5cd273ecd1af7af5f2fbd95ccdcd050e7d3b22e6 +size 56179 diff --git a/parse/train/Ua6zuk0WRH/images/b103fe97d538c8abba8d469bf14a095c30a2f6d9e10cc9390f37520b90aad6de.jpg b/parse/train/Ua6zuk0WRH/images/b103fe97d538c8abba8d469bf14a095c30a2f6d9e10cc9390f37520b90aad6de.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c0d360687b4b41512b95ab459beb14c307f03ba4 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/b103fe97d538c8abba8d469bf14a095c30a2f6d9e10cc9390f37520b90aad6de.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77cd67dd44330beb5aa49c14fbce44ff51e98a4d49db9c22f9de0e155fc14b1d +size 8081 diff --git a/parse/train/Ua6zuk0WRH/images/bccc452054e55eaa2768674c7dcec1db9dc695a38fa8a3787456a330c1c4c1f5.jpg b/parse/train/Ua6zuk0WRH/images/bccc452054e55eaa2768674c7dcec1db9dc695a38fa8a3787456a330c1c4c1f5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8bccf719444463f120c1f302e795cfe1870e2f26 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/bccc452054e55eaa2768674c7dcec1db9dc695a38fa8a3787456a330c1c4c1f5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc3b272ed3f7a9ef5c03627751578f9467fba270083c9655dea9c68bbad0b405 +size 43787 diff --git a/parse/train/Ua6zuk0WRH/images/c2d76116d2bee6d46fd8f3edbb104e4469f5e56eaebc88c1276d4fcb7f402e92.jpg b/parse/train/Ua6zuk0WRH/images/c2d76116d2bee6d46fd8f3edbb104e4469f5e56eaebc88c1276d4fcb7f402e92.jpg new file mode 100644 index 0000000000000000000000000000000000000000..76739a53d7544bf9b9777f4d82dc551c11997956 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/c2d76116d2bee6d46fd8f3edbb104e4469f5e56eaebc88c1276d4fcb7f402e92.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f267121927eaad4985936d30cad93a5535aebc0acf4b15ecd35ae792a729351 +size 63278 diff --git a/parse/train/Ua6zuk0WRH/images/c53a325aa3887cc783d4a9653bf3c8a3f1bc72df1f11536b7fe2343e679c3117.jpg b/parse/train/Ua6zuk0WRH/images/c53a325aa3887cc783d4a9653bf3c8a3f1bc72df1f11536b7fe2343e679c3117.jpg new file mode 100644 index 0000000000000000000000000000000000000000..57c6160569d0e450544ed9c78243ceedd03b5193 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/c53a325aa3887cc783d4a9653bf3c8a3f1bc72df1f11536b7fe2343e679c3117.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9bb5ddd318f5951546585e546415b57de5edd9ff7d52d28182dd963ce294463 +size 73930 diff --git a/parse/train/Ua6zuk0WRH/images/c839af29231462a5bc93edefd17ab13375d987c9c6149e906fa1b727d245a791.jpg b/parse/train/Ua6zuk0WRH/images/c839af29231462a5bc93edefd17ab13375d987c9c6149e906fa1b727d245a791.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f3c68bccbe4d828c469de63f7c91584acca44a60 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/c839af29231462a5bc93edefd17ab13375d987c9c6149e906fa1b727d245a791.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d01f45223f29d87614e13effa90e67cc0bae6e27d2b3a04dcb840e87350847a +size 11801 diff --git a/parse/train/Ua6zuk0WRH/images/cbb7b7d2530599f473ee2ad85ed3d39e3fe2ae0b8fd4354af78270c577c3165a.jpg b/parse/train/Ua6zuk0WRH/images/cbb7b7d2530599f473ee2ad85ed3d39e3fe2ae0b8fd4354af78270c577c3165a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9ad14aeb64d47b723c643cf31fd1fbd84eb03ea0 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/cbb7b7d2530599f473ee2ad85ed3d39e3fe2ae0b8fd4354af78270c577c3165a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cfd614b2ca30a1473bbf24fed53bc22628ba8551a1f1ba5e8d9d5bc5423c74c7 +size 5774 diff --git a/parse/train/Ua6zuk0WRH/images/d1606bcfa6b6d271698262010fa04ffd72ebeba156d5aa10967713ea5e245f97.jpg b/parse/train/Ua6zuk0WRH/images/d1606bcfa6b6d271698262010fa04ffd72ebeba156d5aa10967713ea5e245f97.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2017562ef69888a72806ef8430ce53cf77d57621 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/d1606bcfa6b6d271698262010fa04ffd72ebeba156d5aa10967713ea5e245f97.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0258d350edb6cd2e3831f297b46df44573608eb9989e20e6b2523cc000e6db4 +size 36735 diff --git a/parse/train/Ua6zuk0WRH/images/d974be3d29ab50070ff31d4a145002ef35cbc7f1e6ad48a75b66bf5fb4e91d31.jpg b/parse/train/Ua6zuk0WRH/images/d974be3d29ab50070ff31d4a145002ef35cbc7f1e6ad48a75b66bf5fb4e91d31.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7c19c5173328d18f5c7b3a42c1dbcefbceb6fbdd --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/d974be3d29ab50070ff31d4a145002ef35cbc7f1e6ad48a75b66bf5fb4e91d31.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5233e01da0f7756625135754af4c47c4ad691d77e270d1e852129eb20566fe7d +size 4450 diff --git a/parse/train/Ua6zuk0WRH/images/dc7fc3aa5db61f6d8826df2411fec35a3bcc7226e199afba7083690026f0320a.jpg b/parse/train/Ua6zuk0WRH/images/dc7fc3aa5db61f6d8826df2411fec35a3bcc7226e199afba7083690026f0320a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..16d31a317bbe4cec59e19fb2c3b840bd5e5cb541 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/dc7fc3aa5db61f6d8826df2411fec35a3bcc7226e199afba7083690026f0320a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ced04a55b2e3ba451f7c3c76b9cfa2835d656eb29cfd6c262d82f5fc629d183 +size 5114 diff --git a/parse/train/Ua6zuk0WRH/images/dd1b1e265db507fa0f2eafd99c27a4c58f063de8a4c3533de9ad027a0e1d4964.jpg b/parse/train/Ua6zuk0WRH/images/dd1b1e265db507fa0f2eafd99c27a4c58f063de8a4c3533de9ad027a0e1d4964.jpg new file mode 100644 index 0000000000000000000000000000000000000000..231b1f2a6fc7130a140223f42a8f4c851b764085 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/dd1b1e265db507fa0f2eafd99c27a4c58f063de8a4c3533de9ad027a0e1d4964.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd3f45ff65b5b0a23c20daee989992a05d7e64e33123a4fefcb284f623d9a795 +size 9405 diff --git a/parse/train/Ua6zuk0WRH/images/ea7f1045ceddddfeefaab39ca9f1395f6eb387aa6bc9a085ff7b744e2447629b.jpg b/parse/train/Ua6zuk0WRH/images/ea7f1045ceddddfeefaab39ca9f1395f6eb387aa6bc9a085ff7b744e2447629b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c3c9fd778e395d53a58cce25a340f4e8d06cd6a3 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/ea7f1045ceddddfeefaab39ca9f1395f6eb387aa6bc9a085ff7b744e2447629b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1589d18fc8c015aa6f15818bf9d4ae68c1d066ef1de71c93fd75232a983efeb +size 7906 diff --git a/parse/train/Ua6zuk0WRH/images/eddd0f654c077bc68d235566413552443f89b9977d9f2d0fb443b70f60068523.jpg b/parse/train/Ua6zuk0WRH/images/eddd0f654c077bc68d235566413552443f89b9977d9f2d0fb443b70f60068523.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7bad67eb277a28855a616ab5cc20382dd0b504e2 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/eddd0f654c077bc68d235566413552443f89b9977d9f2d0fb443b70f60068523.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ed8576013f72e2486a4c64a110533f4ecb1659c3aab14a66d31de7624b8c6da +size 10587 diff --git a/parse/train/Ua6zuk0WRH/images/fbc39406e04ae12ea1472981deed8820d3940cda82d56a24bb61aaa100facb57.jpg b/parse/train/Ua6zuk0WRH/images/fbc39406e04ae12ea1472981deed8820d3940cda82d56a24bb61aaa100facb57.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6f80b312b817217d954e6b37abb0daf8c6db7531 --- /dev/null +++ b/parse/train/Ua6zuk0WRH/images/fbc39406e04ae12ea1472981deed8820d3940cda82d56a24bb61aaa100facb57.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c48b694bf583b09a23bb49bc7c2cad5e9e237b6115b737852452a7604da31b2 +size 58764 diff --git a/parse/train/WigDnV-_Gq/images/190ec5391113ac944665522e1adf4c8cf8fc31bf8e9a0938f191d278b12a49dd.jpg b/parse/train/WigDnV-_Gq/images/190ec5391113ac944665522e1adf4c8cf8fc31bf8e9a0938f191d278b12a49dd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f8cb87e96e6ef86165a89aee70e9af031777a138 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/190ec5391113ac944665522e1adf4c8cf8fc31bf8e9a0938f191d278b12a49dd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e3ab4308f3e878c70513a9f2cb70157bbb7c421a5fe7f5bc6a7296628a4771f +size 41619 diff --git a/parse/train/WigDnV-_Gq/images/24f28c6a27df4ff8cc35201b1407d7e14288356e7100dfbc628bca44193ac547.jpg b/parse/train/WigDnV-_Gq/images/24f28c6a27df4ff8cc35201b1407d7e14288356e7100dfbc628bca44193ac547.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aeb9ad2dd5bff65227e749dd90c75856b93d4b9f --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/24f28c6a27df4ff8cc35201b1407d7e14288356e7100dfbc628bca44193ac547.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:644c06789adc9552f0a0daa3b5a6232aa559a8b82370b1817042e2f841d99e0a +size 11266 diff --git a/parse/train/WigDnV-_Gq/images/307cc9a3d1fe81593f26a9044a69a609c09b6a1c5b9c38117f8c94f9bdb2fda0.jpg b/parse/train/WigDnV-_Gq/images/307cc9a3d1fe81593f26a9044a69a609c09b6a1c5b9c38117f8c94f9bdb2fda0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..843c72e336d1e31a5c2d5a91fd7a4e57e08618b4 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/307cc9a3d1fe81593f26a9044a69a609c09b6a1c5b9c38117f8c94f9bdb2fda0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:306580dbb3842db202a8df38692492c037255e834c14fe56b08c663edf812cfb +size 7170 diff --git a/parse/train/WigDnV-_Gq/images/4d70723a86e3d339b5e2fdb626e31bd8a97cd63885ffc673851aedeaeeeb5383.jpg b/parse/train/WigDnV-_Gq/images/4d70723a86e3d339b5e2fdb626e31bd8a97cd63885ffc673851aedeaeeeb5383.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d0651b3a923fd690840a3045657bdaa0b5aac7bd --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/4d70723a86e3d339b5e2fdb626e31bd8a97cd63885ffc673851aedeaeeeb5383.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6a0266b3897086a70c3761872d978f734ae7fe41be16e92b98b93aa0a1ff63f +size 8269 diff --git a/parse/train/WigDnV-_Gq/images/4d8c2a5ca52d36a788530f215dc6aeff5dce30ab2fd1299b32b0c16440bfe80b.jpg b/parse/train/WigDnV-_Gq/images/4d8c2a5ca52d36a788530f215dc6aeff5dce30ab2fd1299b32b0c16440bfe80b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a0d9bda67041a569b2c20124d352a7eec6f8aa96 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/4d8c2a5ca52d36a788530f215dc6aeff5dce30ab2fd1299b32b0c16440bfe80b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d851ecef188219e4a0b1cd7806995e8ea00672713e01cd4e204057e7c96a72f +size 12652 diff --git a/parse/train/WigDnV-_Gq/images/4e3f9aef828ce8cb7c13260da80aba665fdf99a1d9f1071787d95b5cba395d94.jpg b/parse/train/WigDnV-_Gq/images/4e3f9aef828ce8cb7c13260da80aba665fdf99a1d9f1071787d95b5cba395d94.jpg new file mode 100644 index 0000000000000000000000000000000000000000..335f1aaba94732e4b671cf3f49d8130272ad01dd --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/4e3f9aef828ce8cb7c13260da80aba665fdf99a1d9f1071787d95b5cba395d94.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bb0b4a8357d1bdc66646450c5b2a976bfbec9534a8e4132859e8fd89bbd1b78 +size 10261 diff --git a/parse/train/WigDnV-_Gq/images/58c0408e57fa0739cc2955cd29474512c1d5bf68fe4e57a05facb7cf5597abee.jpg b/parse/train/WigDnV-_Gq/images/58c0408e57fa0739cc2955cd29474512c1d5bf68fe4e57a05facb7cf5597abee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b0cffdf8ce2a28c165cb59501f8bc0626696ea36 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/58c0408e57fa0739cc2955cd29474512c1d5bf68fe4e57a05facb7cf5597abee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c58ae82e5fc4e7adf97b4f21029b9ca716ea1f1504851484eefc4b88b76984c8 +size 8235 diff --git a/parse/train/WigDnV-_Gq/images/63e83abd549a8e0b4e6706eb749430e54629fffaf1ae38ab3e1947f15f70b7fe.jpg b/parse/train/WigDnV-_Gq/images/63e83abd549a8e0b4e6706eb749430e54629fffaf1ae38ab3e1947f15f70b7fe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cd1f79f54c61bcb64038c868d6af29bf82f35ec3 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/63e83abd549a8e0b4e6706eb749430e54629fffaf1ae38ab3e1947f15f70b7fe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73c772cf3d1b0445d8a87f65f221c5e4d3f09c799bb0bc2d0ecf5ecc5b5beb28 +size 92912 diff --git a/parse/train/WigDnV-_Gq/images/6426e4cd56a6efb4df1ff686832f2bc5d91d9f6b6d1fcb6d4343298dae118c9f.jpg b/parse/train/WigDnV-_Gq/images/6426e4cd56a6efb4df1ff686832f2bc5d91d9f6b6d1fcb6d4343298dae118c9f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a09c8690069fd7dd735a90c4cf8d0002594efa3a --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/6426e4cd56a6efb4df1ff686832f2bc5d91d9f6b6d1fcb6d4343298dae118c9f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e257a5fbded7658f0ebf0ee12624a9d2c0ee85b155a9c6fce533c9b557eef1dd +size 9638 diff --git a/parse/train/WigDnV-_Gq/images/765778b1d4ff270459c69b13aef9961eb2a095d359a571028f9ab9fa76f0416a.jpg b/parse/train/WigDnV-_Gq/images/765778b1d4ff270459c69b13aef9961eb2a095d359a571028f9ab9fa76f0416a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..09d26fd271f5f89dc37ebb0d51289f57c670f5b0 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/765778b1d4ff270459c69b13aef9961eb2a095d359a571028f9ab9fa76f0416a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1dc18793a11f2112bad28a4fe3c440b19cfe83bcc57d8436be70f5d6f988176e +size 27956 diff --git a/parse/train/WigDnV-_Gq/images/79be059be4cf245f9f112beb0e01c5cd45d48d8fea28b73516bb37b15d002002.jpg b/parse/train/WigDnV-_Gq/images/79be059be4cf245f9f112beb0e01c5cd45d48d8fea28b73516bb37b15d002002.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8d989660955faa084a9cfff9db4e91896a2d6c27 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/79be059be4cf245f9f112beb0e01c5cd45d48d8fea28b73516bb37b15d002002.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ae396c5ad378dadd0821c99adf8f52cf595d8992f6b6c8c4d45b9541ff9e2ef +size 53899 diff --git a/parse/train/WigDnV-_Gq/images/999514e44500a83bd5be65c2574fdbebae72ab79b440bf043769d6fa9efaef97.jpg b/parse/train/WigDnV-_Gq/images/999514e44500a83bd5be65c2574fdbebae72ab79b440bf043769d6fa9efaef97.jpg new file mode 100644 index 0000000000000000000000000000000000000000..efa52bda850be1db4202e72be739db98ca48038a --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/999514e44500a83bd5be65c2574fdbebae72ab79b440bf043769d6fa9efaef97.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ac2d8c1ab215f3955cb81817e9277ffcee2fe5c50ccb9f15bfabb13652732ae +size 5959 diff --git a/parse/train/WigDnV-_Gq/images/9dc5140848e3d3617d3a5af27f72597b412a99a0910ba6857a332cf25b5d8f8f.jpg b/parse/train/WigDnV-_Gq/images/9dc5140848e3d3617d3a5af27f72597b412a99a0910ba6857a332cf25b5d8f8f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..86f19475f728ec7f4a233a519765d6907164ab67 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/9dc5140848e3d3617d3a5af27f72597b412a99a0910ba6857a332cf25b5d8f8f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e38e805d049164098b3520ba96f2f947e073034a199340b80c5da91cd9dbc70 +size 42462 diff --git a/parse/train/WigDnV-_Gq/images/a04bbe15dede6bfda46a24f4347c73fa94999d261b7cfd0556e6ed2a7fcce68c.jpg b/parse/train/WigDnV-_Gq/images/a04bbe15dede6bfda46a24f4347c73fa94999d261b7cfd0556e6ed2a7fcce68c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..83598c140d4103fb6a0f38ec69d6f7fa8c1adcc0 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/a04bbe15dede6bfda46a24f4347c73fa94999d261b7cfd0556e6ed2a7fcce68c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf8b1f61a525324b734ab9f371a73fa08da3910a7de6bf21be7c3cb0afef37ec +size 56067 diff --git a/parse/train/WigDnV-_Gq/images/a0ba4c43ac3275e9c568953872efbc59a96ece653711eb4b73c5eefb4ab518f1.jpg b/parse/train/WigDnV-_Gq/images/a0ba4c43ac3275e9c568953872efbc59a96ece653711eb4b73c5eefb4ab518f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b855aec78919c219d7567a4531ddf4e86148b133 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/a0ba4c43ac3275e9c568953872efbc59a96ece653711eb4b73c5eefb4ab518f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f3bed3ad6dced2976bbf61c92396c3a1cd80df162b48314b828f5b245984860 +size 96282 diff --git a/parse/train/WigDnV-_Gq/images/b0dcb357edbc74b9b1067c37e462472b96686afdfe6fe7fdb0d4e52b97d5bd83.jpg b/parse/train/WigDnV-_Gq/images/b0dcb357edbc74b9b1067c37e462472b96686afdfe6fe7fdb0d4e52b97d5bd83.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1c55c90c4734b0e3d9826428179981b41db751c2 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/b0dcb357edbc74b9b1067c37e462472b96686afdfe6fe7fdb0d4e52b97d5bd83.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:327457a4f422484924447d674888028b51f5ff0806a77bfbe42cf06cb6b4b601 +size 28712 diff --git a/parse/train/WigDnV-_Gq/images/bc3e86ae34d642196fbdc687dfd60857e436eb1f46d12dd1e07dd14d0ddfafb9.jpg b/parse/train/WigDnV-_Gq/images/bc3e86ae34d642196fbdc687dfd60857e436eb1f46d12dd1e07dd14d0ddfafb9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ad9192ee8d107dffebd0beabc1ea9692367edb30 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/bc3e86ae34d642196fbdc687dfd60857e436eb1f46d12dd1e07dd14d0ddfafb9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a0f5ccc4412f1c0cd7221fea8c091d2ffb052c6a15b3c394ffaf6f593266569d +size 7211 diff --git a/parse/train/WigDnV-_Gq/images/ce85e616d16afd0ab08a3ea37e2f184d604adbfffa7c10ba9865ece27cc01890.jpg b/parse/train/WigDnV-_Gq/images/ce85e616d16afd0ab08a3ea37e2f184d604adbfffa7c10ba9865ece27cc01890.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7eeff87706bae0a9ee85e28251eacdacf930096d --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/ce85e616d16afd0ab08a3ea37e2f184d604adbfffa7c10ba9865ece27cc01890.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd230feb2d099b8ef3444803fb3efc952a40afba36b40ec234745a926a940261 +size 35870 diff --git a/parse/train/WigDnV-_Gq/images/d51bb84d20535b7727d17b8d0c550f61bfe18a153b2b35d45b9e26d47caad0ef.jpg b/parse/train/WigDnV-_Gq/images/d51bb84d20535b7727d17b8d0c550f61bfe18a153b2b35d45b9e26d47caad0ef.jpg new file mode 100644 index 0000000000000000000000000000000000000000..21c48fa0eb55e888b2e62ca62e0e5e86d7eb000a --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/d51bb84d20535b7727d17b8d0c550f61bfe18a153b2b35d45b9e26d47caad0ef.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ec4f0d3bbe0c8d8d5cf010d7ae9d699ccb2b4951c5dc72b4e93c62f003c67e0 +size 5535 diff --git a/parse/train/WigDnV-_Gq/images/e6b3f5dec4812af694d4e40684edb2c340965c0c8c071bcfc52e4b4c749872e4.jpg b/parse/train/WigDnV-_Gq/images/e6b3f5dec4812af694d4e40684edb2c340965c0c8c071bcfc52e4b4c749872e4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..25771167e5e9813b7897dbadf9f015f95253bb85 --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/e6b3f5dec4812af694d4e40684edb2c340965c0c8c071bcfc52e4b4c749872e4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fe47a3c37c0be6b48d05159c1fced51b736aa2e0c2cbdc6e0ea4529472e583c +size 69352 diff --git a/parse/train/WigDnV-_Gq/images/f4e8724c4364b397e5b247559a6190ef11ed795d9809d061ba24aa35d2ce95a8.jpg b/parse/train/WigDnV-_Gq/images/f4e8724c4364b397e5b247559a6190ef11ed795d9809d061ba24aa35d2ce95a8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7cc55ddc1f4870d80eb7cd97840784dfd25c629b --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/f4e8724c4364b397e5b247559a6190ef11ed795d9809d061ba24aa35d2ce95a8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e85e35688fee202bee5f7d317efef1de9273a9d1c3229c1c375cb4c12e195a4 +size 10257 diff --git a/parse/train/WigDnV-_Gq/images/fd35f489ee490033710e3c8b2ca4ce8719ed7c233f214b3ccdc7624314b379f4.jpg b/parse/train/WigDnV-_Gq/images/fd35f489ee490033710e3c8b2ca4ce8719ed7c233f214b3ccdc7624314b379f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cc941a97169fcc1a875202bda6a3791e16756fff --- /dev/null +++ b/parse/train/WigDnV-_Gq/images/fd35f489ee490033710e3c8b2ca4ce8719ed7c233f214b3ccdc7624314b379f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6fee5bccdd03a540d90e356c0f7db696790666f75aab1d8f9a41e081edfbbd28 +size 5339 diff --git a/parse/train/de11dbHzAMF/images/0f581a672c00c296bc3d0de70183ba37a5835178658055bbab1c95af80a03fcf.jpg b/parse/train/de11dbHzAMF/images/0f581a672c00c296bc3d0de70183ba37a5835178658055bbab1c95af80a03fcf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..85d58d00e840f631a2b41b33197fa9292b3dc40b --- /dev/null +++ b/parse/train/de11dbHzAMF/images/0f581a672c00c296bc3d0de70183ba37a5835178658055bbab1c95af80a03fcf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a6afa7764372bccbc9d523492426ff05e39b7b3a13eddecb578f5f7dc819ef8 +size 13363 diff --git a/parse/train/de11dbHzAMF/images/1ee2271bcef4ce27ccaaa6511d831a703d071cdbd27575e4ff788e01a12977b8.jpg b/parse/train/de11dbHzAMF/images/1ee2271bcef4ce27ccaaa6511d831a703d071cdbd27575e4ff788e01a12977b8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8099d81f34c681b587dd4ac3d14c9ac33b8e0f17 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/1ee2271bcef4ce27ccaaa6511d831a703d071cdbd27575e4ff788e01a12977b8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc906af8e0b7fe8f6f9ad0b9165aa885a0849ea42764c97cb01c4c5f50b01dc9 +size 26652 diff --git a/parse/train/de11dbHzAMF/images/218453b84f43aada6eac6af231a59daa6384c3d61dad98c7c00cc8f69ac4bc2a.jpg b/parse/train/de11dbHzAMF/images/218453b84f43aada6eac6af231a59daa6384c3d61dad98c7c00cc8f69ac4bc2a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c6e521d34b8316055d5ef9f46824160d003d0612 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/218453b84f43aada6eac6af231a59daa6384c3d61dad98c7c00cc8f69ac4bc2a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef3624dcee0658630f92f6b654ddc3af691f6cf6a67a8f7bbf4b2042035f55bd +size 45590 diff --git a/parse/train/de11dbHzAMF/images/27570a00cc6c91eb9b48ac033ce502f41d9f910cdea9c085e9855462c25b535d.jpg b/parse/train/de11dbHzAMF/images/27570a00cc6c91eb9b48ac033ce502f41d9f910cdea9c085e9855462c25b535d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4400e266cefd6616aeddc80c16df38d73120d682 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/27570a00cc6c91eb9b48ac033ce502f41d9f910cdea9c085e9855462c25b535d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc4d1a9162b3e8a5246a7fada0509a00ed9a30830de0dae1a07ea6f0d99ba408 +size 49956 diff --git a/parse/train/de11dbHzAMF/images/29f9dad5e585bb0bc4752354077750eb90ed63d6d495c3045a18ce4f47033415.jpg b/parse/train/de11dbHzAMF/images/29f9dad5e585bb0bc4752354077750eb90ed63d6d495c3045a18ce4f47033415.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d5baa1bac67a5d5efa0361be0e1413d3f9d70fe9 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/29f9dad5e585bb0bc4752354077750eb90ed63d6d495c3045a18ce4f47033415.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2052b3ddfe4ce5a61bb0b083f3e7d9c8b83e627e557fa2ff241bd3affe917b23 +size 27411 diff --git a/parse/train/de11dbHzAMF/images/315b192ac7350fe31ba9cf3e0cd2422540d16310bf4018743f0239674fb09407.jpg b/parse/train/de11dbHzAMF/images/315b192ac7350fe31ba9cf3e0cd2422540d16310bf4018743f0239674fb09407.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5e64c52df31e4b5cccb5546afdebf26d0571ecf2 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/315b192ac7350fe31ba9cf3e0cd2422540d16310bf4018743f0239674fb09407.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4835cc80880d5eab9b458f4bf330a383973a9965b6ae1ad828daf9b7a609de51 +size 6003 diff --git a/parse/train/de11dbHzAMF/images/34eb77740d33c2b3598990f689875121cccead2186770f4cae1a14a288a3439e.jpg b/parse/train/de11dbHzAMF/images/34eb77740d33c2b3598990f689875121cccead2186770f4cae1a14a288a3439e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1b7a36980eb719162f05e8c1bf01e8390d8c4187 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/34eb77740d33c2b3598990f689875121cccead2186770f4cae1a14a288a3439e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64dc9d4e23cd2f2c03f363997a95f59463b9c3a47a16dc8ed6ef515ea7fe9bb7 +size 24862 diff --git a/parse/train/de11dbHzAMF/images/383f7c9a43f25a400823264ca06517a70da7d78bcb8081387461b29b043df8a1.jpg b/parse/train/de11dbHzAMF/images/383f7c9a43f25a400823264ca06517a70da7d78bcb8081387461b29b043df8a1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0d64bba4f3996b58e73ef865fe12e1d80cea3966 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/383f7c9a43f25a400823264ca06517a70da7d78bcb8081387461b29b043df8a1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da2ca21c425e64296b0333ad6c7e2a065bec059a5a6c08046a613c6e1299b33b +size 14684 diff --git a/parse/train/de11dbHzAMF/images/43d1fe4df24c55f842b492ed9785bedb5fb85c846e261d0dda666fb5c3084433.jpg b/parse/train/de11dbHzAMF/images/43d1fe4df24c55f842b492ed9785bedb5fb85c846e261d0dda666fb5c3084433.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c5597f805f3e3b65f20c74043836737ec802be9 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/43d1fe4df24c55f842b492ed9785bedb5fb85c846e261d0dda666fb5c3084433.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4333033815729b1e6406a354be747d6a57be22c7732becf4b4a0f7d806edda9 +size 72906 diff --git a/parse/train/de11dbHzAMF/images/54a5538fd45e7f02e1e3e3ae3875cf4d36765159012e9130b84cb010465917a6.jpg b/parse/train/de11dbHzAMF/images/54a5538fd45e7f02e1e3e3ae3875cf4d36765159012e9130b84cb010465917a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..557cab26cb09a754bbc9a7b49c3fc95f40744c3f --- /dev/null +++ b/parse/train/de11dbHzAMF/images/54a5538fd45e7f02e1e3e3ae3875cf4d36765159012e9130b84cb010465917a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1acbc3810ff930fa0b2015572100365865bf5c2a0732cecdcdf5233949c74560 +size 10478 diff --git a/parse/train/de11dbHzAMF/images/8b3d8035638bbcfdd27c38fce493e7ea50d5cd5fad80b2711b8239536014ac47.jpg b/parse/train/de11dbHzAMF/images/8b3d8035638bbcfdd27c38fce493e7ea50d5cd5fad80b2711b8239536014ac47.jpg new file mode 100644 index 0000000000000000000000000000000000000000..98470f6bd0e413a8c2fa1c323753da47fd409903 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/8b3d8035638bbcfdd27c38fce493e7ea50d5cd5fad80b2711b8239536014ac47.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8da06155ba886c910da6cc81033a1eaf668333d84b74d2a3db0290d278d32295 +size 10331 diff --git a/parse/train/de11dbHzAMF/images/a04307f18020ae5dc1041ec5407bbb2d85bc76aca20adc42d4177138eb6ce3ed.jpg b/parse/train/de11dbHzAMF/images/a04307f18020ae5dc1041ec5407bbb2d85bc76aca20adc42d4177138eb6ce3ed.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8398b1927f9b3056fadea5475a50b3cd550c8830 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/a04307f18020ae5dc1041ec5407bbb2d85bc76aca20adc42d4177138eb6ce3ed.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23086971d8450512c3673845031ae8f8b759fe3866ea71d2253a23a1c65be8e9 +size 51625 diff --git a/parse/train/de11dbHzAMF/images/b9bb099607e4e2107b550e94429b41f86b3e19c71b85d1243ec254f8d875585e.jpg b/parse/train/de11dbHzAMF/images/b9bb099607e4e2107b550e94429b41f86b3e19c71b85d1243ec254f8d875585e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..523d3723bc7d357d49479c84769f3497af3fbd2a --- /dev/null +++ b/parse/train/de11dbHzAMF/images/b9bb099607e4e2107b550e94429b41f86b3e19c71b85d1243ec254f8d875585e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:723766a321c101859249ab55a3d4ed9584564e5489f323622c27e561b4a539f6 +size 30683 diff --git a/parse/train/de11dbHzAMF/images/c0edcd9dbfb63940afee1cee5e485a34f2983bf79768a803581334598893f550.jpg b/parse/train/de11dbHzAMF/images/c0edcd9dbfb63940afee1cee5e485a34f2983bf79768a803581334598893f550.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fc3f25a5c1fec4f612a7af757edcb87b18f5c7e4 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/c0edcd9dbfb63940afee1cee5e485a34f2983bf79768a803581334598893f550.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:20bfb59fb7613c0cc882f3a8e852d511cf577c6b0091bcff7facbb883e1a3b52 +size 39472 diff --git a/parse/train/de11dbHzAMF/images/ec21abfc0af4768516a38bacdcfe935a0d7b0afe0997309b7dfe520b668e403b.jpg b/parse/train/de11dbHzAMF/images/ec21abfc0af4768516a38bacdcfe935a0d7b0afe0997309b7dfe520b668e403b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..57b7f3ba7ba1d145cf1ebbe64c1a0dd59f44b186 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/ec21abfc0af4768516a38bacdcfe935a0d7b0afe0997309b7dfe520b668e403b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e075f9609a369d976929abc81888a15b1ceed3df5273d02d6168907b2faff249 +size 19141 diff --git a/parse/train/de11dbHzAMF/images/eda0ab390767cc700829152d7cb3128ad5efff3115baff6c46ac7a3761e6b9a4.jpg b/parse/train/de11dbHzAMF/images/eda0ab390767cc700829152d7cb3128ad5efff3115baff6c46ac7a3761e6b9a4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7e257445336a0f11118c3ce1bc15c6c02e2d9141 --- /dev/null +++ b/parse/train/de11dbHzAMF/images/eda0ab390767cc700829152d7cb3128ad5efff3115baff6c46ac7a3761e6b9a4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7496601e268b319590c6b0a01d53f0a19471b9d6da1e38fa7e26df17ae56423f +size 14034 diff --git a/parse/train/djbC2A4uTHP/images/013be7f0e1e75e1072aef4213dc3071a40eed8f53d36cb5a52a75b32e6cdd895.jpg b/parse/train/djbC2A4uTHP/images/013be7f0e1e75e1072aef4213dc3071a40eed8f53d36cb5a52a75b32e6cdd895.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cb45d859451ff3f53be015830577d7d53c88f8d5 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/013be7f0e1e75e1072aef4213dc3071a40eed8f53d36cb5a52a75b32e6cdd895.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f23eb01a2ad38aa970a37434bbeb38adf09faee7a41c9cf9d5eaaecbbbc480c0 +size 64487 diff --git a/parse/train/djbC2A4uTHP/images/01f62b4aea9bb82dffcddb89fa9a85c9f395e99eb8cc15ec176ece76074b337b.jpg b/parse/train/djbC2A4uTHP/images/01f62b4aea9bb82dffcddb89fa9a85c9f395e99eb8cc15ec176ece76074b337b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..be22123b09358755f167df839c0635b4adcc0284 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/01f62b4aea9bb82dffcddb89fa9a85c9f395e99eb8cc15ec176ece76074b337b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:36abb9cd1ec3bc4547531eeba5a8b74a4bcf24e441731bdac8a7c81b7a2085d8 +size 119686 diff --git a/parse/train/djbC2A4uTHP/images/0f2716d399a4da7efa579f44270d8b63f236485b38e4adc90d9c678c3b3a2ee1.jpg b/parse/train/djbC2A4uTHP/images/0f2716d399a4da7efa579f44270d8b63f236485b38e4adc90d9c678c3b3a2ee1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..86cb96c91912997aac6795080bcaf409f6928b8a --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/0f2716d399a4da7efa579f44270d8b63f236485b38e4adc90d9c678c3b3a2ee1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fff87c0ab7bf6687e60d20b47c57191af6efe77c10e2a6ae67118b7c736c6c06 +size 12935 diff --git a/parse/train/djbC2A4uTHP/images/20429e2704dad1c0dc1720cb8df1c75463dee281517f045ffd9fcd6d220eec7b.jpg b/parse/train/djbC2A4uTHP/images/20429e2704dad1c0dc1720cb8df1c75463dee281517f045ffd9fcd6d220eec7b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f5a6f1267978395a1303ee8fbd4e463465c190e0 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/20429e2704dad1c0dc1720cb8df1c75463dee281517f045ffd9fcd6d220eec7b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee18222e08af582303c7356bbbe029c3c8cddde1a273f29da50a514d3e2d75e3 +size 7292 diff --git a/parse/train/djbC2A4uTHP/images/232a0cf438bd8162ffcb009760e5764148e0ddb2944864d1ba174eb053dc3ab2.jpg b/parse/train/djbC2A4uTHP/images/232a0cf438bd8162ffcb009760e5764148e0ddb2944864d1ba174eb053dc3ab2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a880b5ade588b3217e694ed12f8e021a8e0e2f8c --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/232a0cf438bd8162ffcb009760e5764148e0ddb2944864d1ba174eb053dc3ab2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c00441f3be6b8b23e1cf027b940bbf9c121ffd48a14b2598b48f0847300248e +size 6761 diff --git a/parse/train/djbC2A4uTHP/images/3415f282ed82b433ed836f811f5ab0ae836d4bcb25e77dd339ab61af4daf5a9e.jpg b/parse/train/djbC2A4uTHP/images/3415f282ed82b433ed836f811f5ab0ae836d4bcb25e77dd339ab61af4daf5a9e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d2905afed753228706d72b3bf8f4bf80e734fd19 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/3415f282ed82b433ed836f811f5ab0ae836d4bcb25e77dd339ab61af4daf5a9e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13c0018e2c37383db34a87ba5a75a59f275a0266dada020f5dd62049e1e1e87b +size 17595 diff --git a/parse/train/djbC2A4uTHP/images/49142c6f3e517ea78a7cf6503ab86315a9fab4eae97e9da18bb18134b4423853.jpg b/parse/train/djbC2A4uTHP/images/49142c6f3e517ea78a7cf6503ab86315a9fab4eae97e9da18bb18134b4423853.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7a8be39e8dc0380384f4a84e75042b0154b8d2c8 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/49142c6f3e517ea78a7cf6503ab86315a9fab4eae97e9da18bb18134b4423853.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c23b44814103347a6e56f80eba2511de5c42546a0a26d47e71f98e74888d8fe5 +size 7630 diff --git a/parse/train/djbC2A4uTHP/images/4b2324520025efa58830e21a6115ffa750dd98795319a96faee2d2d6ca2a0112.jpg b/parse/train/djbC2A4uTHP/images/4b2324520025efa58830e21a6115ffa750dd98795319a96faee2d2d6ca2a0112.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b52adda4f0bbec5e09624d166402650e859a487 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/4b2324520025efa58830e21a6115ffa750dd98795319a96faee2d2d6ca2a0112.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd2055786e919808a2715e9e382622d0e9ff075dbe2fdc6c5c46075be1456ba2 +size 18619 diff --git a/parse/train/djbC2A4uTHP/images/5d271458eb06f054cd1c9f6f42d7f9c933f577600e9497a10a7ddf4e3e3b6ade.jpg b/parse/train/djbC2A4uTHP/images/5d271458eb06f054cd1c9f6f42d7f9c933f577600e9497a10a7ddf4e3e3b6ade.jpg new file mode 100644 index 0000000000000000000000000000000000000000..75c51652ec1bbfaa8792bed80b3ccd04682d50ca --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/5d271458eb06f054cd1c9f6f42d7f9c933f577600e9497a10a7ddf4e3e3b6ade.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a524c6b1fc6caf54ab737b4f671eb0058150597508bdee216504044bdad03c37 +size 7273 diff --git a/parse/train/djbC2A4uTHP/images/6a278a274b29255ea80874edce5c640ac9b6abc9a186c108a470f5a2d42a756f.jpg b/parse/train/djbC2A4uTHP/images/6a278a274b29255ea80874edce5c640ac9b6abc9a186c108a470f5a2d42a756f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..837b7afaa90e36e9bee24bd50cb65194386f2059 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/6a278a274b29255ea80874edce5c640ac9b6abc9a186c108a470f5a2d42a756f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a86fe452f4141822d183eb2d2f1185d120380c61f4a8d1600184e2af2626125a +size 23446 diff --git a/parse/train/djbC2A4uTHP/images/7dda144bd40410a81a940ae50d5e67f85669cc267a11ca25ba90d8da35a5eb88.jpg b/parse/train/djbC2A4uTHP/images/7dda144bd40410a81a940ae50d5e67f85669cc267a11ca25ba90d8da35a5eb88.jpg new file mode 100644 index 0000000000000000000000000000000000000000..35b9e3d8fd90d9e9b3567409bb4a49dade0603d9 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/7dda144bd40410a81a940ae50d5e67f85669cc267a11ca25ba90d8da35a5eb88.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c6da4560f2af590034c2a6766692fdb7535d614a3a314221c0e021600a0604ad +size 7521 diff --git a/parse/train/djbC2A4uTHP/images/807f0b6a2f18d592a13c2f76527dd67d65704ecf0bc4f5ce156fe8502705139e.jpg b/parse/train/djbC2A4uTHP/images/807f0b6a2f18d592a13c2f76527dd67d65704ecf0bc4f5ce156fe8502705139e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a3c4f607236d769497f15cc0e23822b2066695a --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/807f0b6a2f18d592a13c2f76527dd67d65704ecf0bc4f5ce156fe8502705139e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:215f2c3c81239473b900049b5d5f7fb02f8295d357bb81084527e1e9fefad427 +size 3085 diff --git a/parse/train/djbC2A4uTHP/images/87da93feb9101d9471f5468dd3ffeb7e8e2ecfe35807360b6f865d0e977509dc.jpg b/parse/train/djbC2A4uTHP/images/87da93feb9101d9471f5468dd3ffeb7e8e2ecfe35807360b6f865d0e977509dc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..40f4eea43217eab523219d073882147326eda2bb --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/87da93feb9101d9471f5468dd3ffeb7e8e2ecfe35807360b6f865d0e977509dc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73e090eb4144eca5b4261c43698529cde114eb85b07623b9661ac1a451c7f022 +size 25607 diff --git a/parse/train/djbC2A4uTHP/images/8b9b3f68fca58fe2e42e1a8482298efe7f6b2f24634f547d5b7a3e397bf38b80.jpg b/parse/train/djbC2A4uTHP/images/8b9b3f68fca58fe2e42e1a8482298efe7f6b2f24634f547d5b7a3e397bf38b80.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9af1ad604f3f01d416b0bca9926ee18faf129528 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/8b9b3f68fca58fe2e42e1a8482298efe7f6b2f24634f547d5b7a3e397bf38b80.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89f6dd45545fa13948e8d7e202066e67f2a9ee2c77e4f5b7d26304059184e077 +size 74160 diff --git a/parse/train/djbC2A4uTHP/images/9049cac26096d5e7579e063d4751364e984a103026cb53014919ca12ce878707.jpg b/parse/train/djbC2A4uTHP/images/9049cac26096d5e7579e063d4751364e984a103026cb53014919ca12ce878707.jpg new file mode 100644 index 0000000000000000000000000000000000000000..33eca18cb6db561ce2a69a9971fd92924fc40bbc --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/9049cac26096d5e7579e063d4751364e984a103026cb53014919ca12ce878707.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:688244c45e0f8393adc3eb28268593bd8df95dcb458ef2e70b8c11913abf83a1 +size 20270 diff --git a/parse/train/djbC2A4uTHP/images/ac39e3c789e3adf899bf3dc4d7aa0cb25c62b933db2b17bafbd2bb47123459fc.jpg b/parse/train/djbC2A4uTHP/images/ac39e3c789e3adf899bf3dc4d7aa0cb25c62b933db2b17bafbd2bb47123459fc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d2a6410064824c3a900b601d190d102074c5ce7d --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/ac39e3c789e3adf899bf3dc4d7aa0cb25c62b933db2b17bafbd2bb47123459fc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b59c5c91f6131cb9c5affc039b786c62b01e8f2f57eaeb8271c653b08821f439 +size 20410 diff --git a/parse/train/djbC2A4uTHP/images/c74383d34e75780205a7859f62fd761513688c87abc06ef16c17185ee13fd7a4.jpg b/parse/train/djbC2A4uTHP/images/c74383d34e75780205a7859f62fd761513688c87abc06ef16c17185ee13fd7a4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..67b534a7a40b7666fee592672abaf26017eae9b2 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/c74383d34e75780205a7859f62fd761513688c87abc06ef16c17185ee13fd7a4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8b93adcdb582f94ba226e0b0fa73059584fbab5bfcdf6ce5b05c618a01acb214 +size 5778 diff --git a/parse/train/djbC2A4uTHP/images/ca2e472d072fc4f4263a91a933a1bbd1bd818c34ad3086fbc683078ed7560075.jpg b/parse/train/djbC2A4uTHP/images/ca2e472d072fc4f4263a91a933a1bbd1bd818c34ad3086fbc683078ed7560075.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0b648c180678f18088dac2e909d54306188d3434 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/ca2e472d072fc4f4263a91a933a1bbd1bd818c34ad3086fbc683078ed7560075.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:572a2d52104823ad9beb00bda162b4ed26a40ec8ea4233618e9be4382c79c0f3 +size 42373 diff --git a/parse/train/djbC2A4uTHP/images/d590410ae5320acbe610c74382f7978aeec1c5aa980fb2c6b19f816af2109735.jpg b/parse/train/djbC2A4uTHP/images/d590410ae5320acbe610c74382f7978aeec1c5aa980fb2c6b19f816af2109735.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c3e44b3b72ab4d2dac7f3a66ef0ce04474e8ee5b --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/d590410ae5320acbe610c74382f7978aeec1c5aa980fb2c6b19f816af2109735.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d01635dc890c8b1b9f53d2949b6caaa14513a23771b8af21eb5aa38eae1530c +size 138698 diff --git a/parse/train/djbC2A4uTHP/images/f03c44e8274f6944635a3eaf220484b2dffc981faaf10c88764078d41e3aee5f.jpg b/parse/train/djbC2A4uTHP/images/f03c44e8274f6944635a3eaf220484b2dffc981faaf10c88764078d41e3aee5f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..318bfd41fed4d3ffba9523fa0d01a275f9b0eaf6 --- /dev/null +++ b/parse/train/djbC2A4uTHP/images/f03c44e8274f6944635a3eaf220484b2dffc981faaf10c88764078d41e3aee5f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4bba05c8b85d1bfaa1f23365d82d79a40fd22e4238398881baeb25630e5a8772 +size 5007 diff --git a/parse/train/fUxqIofPPi/images/0783840e719a4067baf0886192f65b93cbc8c6b8ce9cb284f4f4aa9ca3f1b6b4.jpg b/parse/train/fUxqIofPPi/images/0783840e719a4067baf0886192f65b93cbc8c6b8ce9cb284f4f4aa9ca3f1b6b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b3617398cf3a9f139ddeb201ceebb4e4fe97e8f7 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/0783840e719a4067baf0886192f65b93cbc8c6b8ce9cb284f4f4aa9ca3f1b6b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ade2edc11a5162a02b4a3a6444bd0a74668f63fe37bfe946f5e1682646153716 +size 181028 diff --git a/parse/train/fUxqIofPPi/images/262d3179ec27dff6ff9840b60352189736cabac3c2b780c63656ebff3f792930.jpg b/parse/train/fUxqIofPPi/images/262d3179ec27dff6ff9840b60352189736cabac3c2b780c63656ebff3f792930.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1afa9ee756c534e36ef69fcb0a1f7d416d2dc0f6 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/262d3179ec27dff6ff9840b60352189736cabac3c2b780c63656ebff3f792930.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0694a1c6e2ebd9873980a4ad1f53036ba2c20d9a191c2757056ac538557bc6ea +size 45399 diff --git a/parse/train/fUxqIofPPi/images/286fbe8b3c1b202601efc7433e4086776d4341d3681a0692ed6230be2e0e1844.jpg b/parse/train/fUxqIofPPi/images/286fbe8b3c1b202601efc7433e4086776d4341d3681a0692ed6230be2e0e1844.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6661ecd9af6ff9ab8663175a7ed5fb2af326f124 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/286fbe8b3c1b202601efc7433e4086776d4341d3681a0692ed6230be2e0e1844.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:14b51b37f14070f404fd5c723b6084879f44fd85601b4c2f0ddef10d6cf2d78f +size 88172 diff --git a/parse/train/fUxqIofPPi/images/49ee5915d7fd475a516ce08c9ee80d2bc56371dfdca42948cc5b77f9b446b8fb.jpg b/parse/train/fUxqIofPPi/images/49ee5915d7fd475a516ce08c9ee80d2bc56371dfdca42948cc5b77f9b446b8fb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..622b88a07df046c4a7359132ffa2a801f0348b7a --- /dev/null +++ b/parse/train/fUxqIofPPi/images/49ee5915d7fd475a516ce08c9ee80d2bc56371dfdca42948cc5b77f9b446b8fb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d68c28589c28fde281da9976df73059a91013431b3332b8bd0baa400946bbc32 +size 18955 diff --git a/parse/train/fUxqIofPPi/images/58f8014f5496a2b69ee499d788bfdffd0879232dd58fdda16a76fdeb47fbd9fd.jpg b/parse/train/fUxqIofPPi/images/58f8014f5496a2b69ee499d788bfdffd0879232dd58fdda16a76fdeb47fbd9fd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4b7b5f01ba7f2b0501906537da45fbf0f3d78ce9 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/58f8014f5496a2b69ee499d788bfdffd0879232dd58fdda16a76fdeb47fbd9fd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90958105afb058f3ff9496075ed6d37edb3451d680aaa0114aa005203db52c26 +size 109841 diff --git a/parse/train/fUxqIofPPi/images/6aeb5b35114e0ffcb27b889c8f6efda7b1e03c75bc828526b043c80699da6067.jpg b/parse/train/fUxqIofPPi/images/6aeb5b35114e0ffcb27b889c8f6efda7b1e03c75bc828526b043c80699da6067.jpg new file mode 100644 index 0000000000000000000000000000000000000000..259dfd64f67db33849b10ce2530c160017e5cd54 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/6aeb5b35114e0ffcb27b889c8f6efda7b1e03c75bc828526b043c80699da6067.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:34189b524a4017f9d39827c0088e0eadaa8313e48e4a83dd046987cf8148f192 +size 7551 diff --git a/parse/train/fUxqIofPPi/images/a0fe6abaf21c6270a3c608c0c9511d9a5e1da6365adf60b259000b133b070663.jpg b/parse/train/fUxqIofPPi/images/a0fe6abaf21c6270a3c608c0c9511d9a5e1da6365adf60b259000b133b070663.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f4a0adf8864f6c638969f834e60bc9cf49aed91c --- /dev/null +++ b/parse/train/fUxqIofPPi/images/a0fe6abaf21c6270a3c608c0c9511d9a5e1da6365adf60b259000b133b070663.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b81612149c10b4957ed1184b72cd3c2f54f39594fc84dee242ba74dab111aaa +size 10833 diff --git a/parse/train/fUxqIofPPi/images/aaa199bcd852ba250a50b0e4fa19c774e05bf5db00029dfe9e012a6cc4efec6b.jpg b/parse/train/fUxqIofPPi/images/aaa199bcd852ba250a50b0e4fa19c774e05bf5db00029dfe9e012a6cc4efec6b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b48e34069840a5b4d75e1e3dcdb2805205ebb0f --- /dev/null +++ b/parse/train/fUxqIofPPi/images/aaa199bcd852ba250a50b0e4fa19c774e05bf5db00029dfe9e012a6cc4efec6b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73dbd0c6575305573d3619eec887ea860de97cf8600f77790ba8d2d25c8fdce2 +size 13901 diff --git a/parse/train/fUxqIofPPi/images/ae0c53cd1b97844fa8180802230f7d2ece451efea963ce834374ad854fe23c5c.jpg b/parse/train/fUxqIofPPi/images/ae0c53cd1b97844fa8180802230f7d2ece451efea963ce834374ad854fe23c5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9770000951c6b679bd064bfc733d7e38f7ebcbd0 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/ae0c53cd1b97844fa8180802230f7d2ece451efea963ce834374ad854fe23c5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ad1be89dc471a1a23f61a3d1351201bd5c9a145b2ef7f39dc45c16ca41834af +size 10221 diff --git a/parse/train/fUxqIofPPi/images/b0daf437d42b8753ba9cb84d4e91753512629b3119522e23e9b1f8198557046d.jpg b/parse/train/fUxqIofPPi/images/b0daf437d42b8753ba9cb84d4e91753512629b3119522e23e9b1f8198557046d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ff5090936c9f3bc0d752002f93ebc62066fe0fa3 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/b0daf437d42b8753ba9cb84d4e91753512629b3119522e23e9b1f8198557046d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3eb51138d138294b161b2a91149a341cac2c57ee9a0bf0ecccbd8fc35a22a231 +size 55474 diff --git a/parse/train/fUxqIofPPi/images/c613d56c378d9adb5ccd64519c6837c7f6c69eb4fe98ed6515aeeb136758ad4e.jpg b/parse/train/fUxqIofPPi/images/c613d56c378d9adb5ccd64519c6837c7f6c69eb4fe98ed6515aeeb136758ad4e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cd9679998ff7fee3bc2708c3c3f5ca0be20262d4 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/c613d56c378d9adb5ccd64519c6837c7f6c69eb4fe98ed6515aeeb136758ad4e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7e48e58ee817463b894c8ffc47b73374d066ce767253f4c18758c53a81ae0d1 +size 178899 diff --git a/parse/train/fUxqIofPPi/images/c883e6a958f6034a0b565bab650b45781fb0ee5d5d9d6b3449ca0d3672c443e5.jpg b/parse/train/fUxqIofPPi/images/c883e6a958f6034a0b565bab650b45781fb0ee5d5d9d6b3449ca0d3672c443e5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..246a57abc857557c2c7b7b1eacc88884e930c0d2 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/c883e6a958f6034a0b565bab650b45781fb0ee5d5d9d6b3449ca0d3672c443e5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe3c82b79274a3ec7c91bb25e2743ac50ef2f6d1cd219ef90b8e9c4b6def81d4 +size 74307 diff --git a/parse/train/fUxqIofPPi/images/ca0b8c7f7b5eca23b1fa27490dd4df4336e34dbf7b8ec8321ed5e9c4d23557e3.jpg b/parse/train/fUxqIofPPi/images/ca0b8c7f7b5eca23b1fa27490dd4df4336e34dbf7b8ec8321ed5e9c4d23557e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d4d067e8d13c8afc967f2faa75f124ac296143e4 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/ca0b8c7f7b5eca23b1fa27490dd4df4336e34dbf7b8ec8321ed5e9c4d23557e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2caca2fd79bc48fc6b95b0dfd7b265bb6898b8ff20d97935f89cf4971d05c652 +size 10252 diff --git a/parse/train/fUxqIofPPi/images/ce1468a0184c8b54ad5d2ea3bbe252d82a8d7c854c77bcfaaf05668d7b251d24.jpg b/parse/train/fUxqIofPPi/images/ce1468a0184c8b54ad5d2ea3bbe252d82a8d7c854c77bcfaaf05668d7b251d24.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8e1a8863fde5c4a2c9281f288db0e7e5c1619e17 --- /dev/null +++ b/parse/train/fUxqIofPPi/images/ce1468a0184c8b54ad5d2ea3bbe252d82a8d7c854c77bcfaaf05668d7b251d24.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52778e39736ab0963ffcffcac723f266bc4b73588045127c960054c4099ce205 +size 6081 diff --git a/parse/train/jSz59N8NvUP/images/01b3a40f43eb0a3156bdfa79e1d96e28aaf5c27fd312154845a27f5978e4f461.jpg b/parse/train/jSz59N8NvUP/images/01b3a40f43eb0a3156bdfa79e1d96e28aaf5c27fd312154845a27f5978e4f461.jpg new file mode 100644 index 0000000000000000000000000000000000000000..072c04c22c6fee3a6125f4c9e213071d81304032 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/01b3a40f43eb0a3156bdfa79e1d96e28aaf5c27fd312154845a27f5978e4f461.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82291c59bc0cd69ba44455e2c970285679a604c73b8fb5097b0b6d865a59be7a +size 46799 diff --git a/parse/train/jSz59N8NvUP/images/07bb1558d20aacee31e83db28fe51363cced91302bb2e7a9eb05b85875526036.jpg b/parse/train/jSz59N8NvUP/images/07bb1558d20aacee31e83db28fe51363cced91302bb2e7a9eb05b85875526036.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a06ee755f8d13435105f5811bf0aec4afc30a8b5 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/07bb1558d20aacee31e83db28fe51363cced91302bb2e7a9eb05b85875526036.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2bfd1a6e5040fec6209a33d60158a7675559e8be1397fd0dfb3cee816ce23ee1 +size 4908 diff --git a/parse/train/jSz59N8NvUP/images/1c8381dc1dc248d217d7d56f44e796043ced99bd5419b0a5cc891905203c9f39.jpg b/parse/train/jSz59N8NvUP/images/1c8381dc1dc248d217d7d56f44e796043ced99bd5419b0a5cc891905203c9f39.jpg new file mode 100644 index 0000000000000000000000000000000000000000..07c6a09afb11771b5e50d20955d49cdb352b7fbd --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/1c8381dc1dc248d217d7d56f44e796043ced99bd5419b0a5cc891905203c9f39.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:622c06fcc433bd33097da93fea86d9ae30a41d6344b6672985bc40b154ed4748 +size 5333 diff --git a/parse/train/jSz59N8NvUP/images/2b8bc3895048c9c9c8db1d4c6dced3334aab1aca3500f15b23687765b6a84924.jpg b/parse/train/jSz59N8NvUP/images/2b8bc3895048c9c9c8db1d4c6dced3334aab1aca3500f15b23687765b6a84924.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fa99186c6ea32b83b5dc4066da8d0f445ebe305f --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/2b8bc3895048c9c9c8db1d4c6dced3334aab1aca3500f15b23687765b6a84924.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:310f766da8623f9d2d72c7b0bf47db73f91ae00bb23e1be605331eb4aa07e985 +size 87178 diff --git a/parse/train/jSz59N8NvUP/images/2de6497b14092f7d097b8cefb6a97821b9691c0f06529650672de2f2a0f7f7ac.jpg b/parse/train/jSz59N8NvUP/images/2de6497b14092f7d097b8cefb6a97821b9691c0f06529650672de2f2a0f7f7ac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..41e9099ca7de3cb481889814f697057d6270cef2 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/2de6497b14092f7d097b8cefb6a97821b9691c0f06529650672de2f2a0f7f7ac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f30cb24454de567bbf8e2036289316648943469c48c3360c889da6b3478925f +size 1443 diff --git a/parse/train/jSz59N8NvUP/images/36462717557b89b7bb6548f8dd3c22326ea43e9cf8bb2619d18e303e29ba37f0.jpg b/parse/train/jSz59N8NvUP/images/36462717557b89b7bb6548f8dd3c22326ea43e9cf8bb2619d18e303e29ba37f0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7171c630e0e7d1957f48d68e08fe96c5c1626a7e --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/36462717557b89b7bb6548f8dd3c22326ea43e9cf8bb2619d18e303e29ba37f0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:047eec9fbe339e405bc80c7718c506bb6f2afdf06b9f5ec6d0d84c419929628d +size 35940 diff --git a/parse/train/jSz59N8NvUP/images/40ab1abc6353dc009e6215522df2f222c2e4246b190684a7cc4f972009903ee1.jpg b/parse/train/jSz59N8NvUP/images/40ab1abc6353dc009e6215522df2f222c2e4246b190684a7cc4f972009903ee1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..73143ea4e4d3deceab57513dd599773beb3c4cf2 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/40ab1abc6353dc009e6215522df2f222c2e4246b190684a7cc4f972009903ee1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:052664d90c252316a910f1d7df7ef666e802d2f5e295aa2ab6c0feee0d1d98a5 +size 57402 diff --git a/parse/train/jSz59N8NvUP/images/43786e0eb492dae93915dedb19372bd94de34c9e31336c2fa47a38e19e95aab2.jpg b/parse/train/jSz59N8NvUP/images/43786e0eb492dae93915dedb19372bd94de34c9e31336c2fa47a38e19e95aab2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8fce8de47180fccf5a4c9c50b2380fffa6f16426 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/43786e0eb492dae93915dedb19372bd94de34c9e31336c2fa47a38e19e95aab2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0377406a819a725ba7fc8dc07cd88b51ed8565422d6fe9432ff739a014392e78 +size 7746 diff --git a/parse/train/jSz59N8NvUP/images/58ca1d47fbcb25e7433df9e3c688bffbaadce3a4dc4009f011f43c8774c66982.jpg b/parse/train/jSz59N8NvUP/images/58ca1d47fbcb25e7433df9e3c688bffbaadce3a4dc4009f011f43c8774c66982.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9059baab53a19985fc5a3ca40a187cb7b244dcb1 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/58ca1d47fbcb25e7433df9e3c688bffbaadce3a4dc4009f011f43c8774c66982.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:957a7553c3a659f7359d1a1f4c41ac7b0a371c13e80fdca0e0bcd38c4fc20d34 +size 9123 diff --git a/parse/train/jSz59N8NvUP/images/5cc90cb8da2aceaf2d07b3bee384f5c4e5bec7f2733a9bd58f550b427aba4a9c.jpg b/parse/train/jSz59N8NvUP/images/5cc90cb8da2aceaf2d07b3bee384f5c4e5bec7f2733a9bd58f550b427aba4a9c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..24a96dab01651bea51efa652cb7b497db4b1e422 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/5cc90cb8da2aceaf2d07b3bee384f5c4e5bec7f2733a9bd58f550b427aba4a9c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e2a16722edb1e8dfacc5aabea19e394b1047e259493e6dc65d52aa1f5a771184 +size 10239 diff --git a/parse/train/jSz59N8NvUP/images/8f612b06fefb8df68ce87a482eaf42ad80ea1b7a73259a95f98cd09315cd54be.jpg b/parse/train/jSz59N8NvUP/images/8f612b06fefb8df68ce87a482eaf42ad80ea1b7a73259a95f98cd09315cd54be.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dfbf0ba85e065e67df1bfe5fdf4c06e99b9a3ea6 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/8f612b06fefb8df68ce87a482eaf42ad80ea1b7a73259a95f98cd09315cd54be.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b41a0cc4c470f0abaa07d25e7165aa398b14a4a66ac756c1c171b24e75320d8 +size 25245 diff --git a/parse/train/jSz59N8NvUP/images/ab983e7608a0cd9e8df35bcb966d3f574bdd62cc254254795b5a19f77059b08d.jpg b/parse/train/jSz59N8NvUP/images/ab983e7608a0cd9e8df35bcb966d3f574bdd62cc254254795b5a19f77059b08d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38ada0a3e2705dd840598527f7748ba872de2e17 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/ab983e7608a0cd9e8df35bcb966d3f574bdd62cc254254795b5a19f77059b08d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5496a9fd98f1e604000f1230eede8a09a997fe6dd531fad379417a7737d6063b +size 21405 diff --git a/parse/train/jSz59N8NvUP/images/bc42f183690147a5522b6b07c852df22694bb03aba97adf9e552b9088e1c2c81.jpg b/parse/train/jSz59N8NvUP/images/bc42f183690147a5522b6b07c852df22694bb03aba97adf9e552b9088e1c2c81.jpg new file mode 100644 index 0000000000000000000000000000000000000000..652ba3fa11b82663ff7c547219acd03e898b2f10 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/bc42f183690147a5522b6b07c852df22694bb03aba97adf9e552b9088e1c2c81.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e3d73a76441771b0f4ca756e52ddfda11a1639fd32dfd34a4b39f25d9eeb1ae +size 14277 diff --git a/parse/train/jSz59N8NvUP/images/bd28cdf21835c75d0dba7b8e63b28e68453c2829ff7cffd08ea7edd112509f93.jpg b/parse/train/jSz59N8NvUP/images/bd28cdf21835c75d0dba7b8e63b28e68453c2829ff7cffd08ea7edd112509f93.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9069406988a514e0b8536960d6936640e61bf67b --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/bd28cdf21835c75d0dba7b8e63b28e68453c2829ff7cffd08ea7edd112509f93.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:062e8b1763a2eae5ed3460fe7e31a55b0ae28c83023c877edfae5d9ed1b7c06f +size 9976 diff --git a/parse/train/jSz59N8NvUP/images/df6d6c6ea6aa3788e8bab52caac9b19f4b0fe7d1db713bae7e2c1641cba7d385.jpg b/parse/train/jSz59N8NvUP/images/df6d6c6ea6aa3788e8bab52caac9b19f4b0fe7d1db713bae7e2c1641cba7d385.jpg new file mode 100644 index 0000000000000000000000000000000000000000..95002f58ad3b873dff081678b79b61c8bd6ffc60 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/df6d6c6ea6aa3788e8bab52caac9b19f4b0fe7d1db713bae7e2c1641cba7d385.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70a9896f28572cd1bd4a6915b5af3d55af38c35b56ed371c05eabfe947e0b95a +size 17722 diff --git a/parse/train/jSz59N8NvUP/images/e0274f1ed1c83c6486627189f4a2308b107173defdad8c8c8d2b5d68a9261267.jpg b/parse/train/jSz59N8NvUP/images/e0274f1ed1c83c6486627189f4a2308b107173defdad8c8c8d2b5d68a9261267.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4a81546dc6147d4d4f62b11e97ba6f740f4f7790 --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/e0274f1ed1c83c6486627189f4a2308b107173defdad8c8c8d2b5d68a9261267.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03a0b0a682b8be0b75ad42f80413c5eb37a522ab9f112299076cbae7560aab18 +size 3560 diff --git a/parse/train/jSz59N8NvUP/images/ee93cef6cc6cc2130375adef678bf9393f780300704deb29c0d593e51c4e4761.jpg b/parse/train/jSz59N8NvUP/images/ee93cef6cc6cc2130375adef678bf9393f780300704deb29c0d593e51c4e4761.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6569f0c9e964a20b7c44fa4f5d24ad3d6392dbdb --- /dev/null +++ b/parse/train/jSz59N8NvUP/images/ee93cef6cc6cc2130375adef678bf9393f780300704deb29c0d593e51c4e4761.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b06268f551b1e4949220864bf5a7cc74a77eda9a57ec21ba51b2e9fac77cb330 +size 8412 diff --git a/parse/train/kW_zpEmMLdP/images/1dc3d6c2fb8f3b215277ef1af14b46a4f59d5fd6e880787af9c6e7845f441bac.jpg b/parse/train/kW_zpEmMLdP/images/1dc3d6c2fb8f3b215277ef1af14b46a4f59d5fd6e880787af9c6e7845f441bac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..03124e24f6d6e4f2e0b78419bfe5ecee26b6a9cf --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/1dc3d6c2fb8f3b215277ef1af14b46a4f59d5fd6e880787af9c6e7845f441bac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f8a45a8539568b6413bbdb2b70ecb7e6463873cb53c2d6de05ef3006ae5879e +size 6689 diff --git a/parse/train/kW_zpEmMLdP/images/31de5e991dcb48e2bf6cc0e92b8e763fbf2182ef49aac36d061ef80686f0faad.jpg b/parse/train/kW_zpEmMLdP/images/31de5e991dcb48e2bf6cc0e92b8e763fbf2182ef49aac36d061ef80686f0faad.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7eb70834fa9b039d09b75a3ede3945f5d5255678 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/31de5e991dcb48e2bf6cc0e92b8e763fbf2182ef49aac36d061ef80686f0faad.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32051ebfb5d1b1262efde19749672f921a8a0662979e7607d881b0d91eb632f9 +size 7829 diff --git a/parse/train/kW_zpEmMLdP/images/439c59cde8dfe0bb27a8ee74bc6dc758e8eb735080377b30a7fc6db0f5acbe9a.jpg b/parse/train/kW_zpEmMLdP/images/439c59cde8dfe0bb27a8ee74bc6dc758e8eb735080377b30a7fc6db0f5acbe9a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d2df667977e893d4a531b31b186c8dd46079ca63 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/439c59cde8dfe0bb27a8ee74bc6dc758e8eb735080377b30a7fc6db0f5acbe9a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7319038aed7e1c884ac822621ff263f22802027710a2807be1f33b7aef25ffb6 +size 5221 diff --git a/parse/train/kW_zpEmMLdP/images/45fd467e9cb0430256bc37144176702a3e7b015e4d8d96da2d6b8f58462af568.jpg b/parse/train/kW_zpEmMLdP/images/45fd467e9cb0430256bc37144176702a3e7b015e4d8d96da2d6b8f58462af568.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ebd48ffc8649859a1134a9c2960807b691ea00ac --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/45fd467e9cb0430256bc37144176702a3e7b015e4d8d96da2d6b8f58462af568.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30815a24c3517af155990b86ef43511b47ef59dc9f7f665da8a3544a46e67be7 +size 10435 diff --git a/parse/train/kW_zpEmMLdP/images/4ec221b23762b137e1f58f81383ed732ceccacdfe1e9c1602a7dc96ab5b7260f.jpg b/parse/train/kW_zpEmMLdP/images/4ec221b23762b137e1f58f81383ed732ceccacdfe1e9c1602a7dc96ab5b7260f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..86c2e5c72b888bfb1011665f0f26b5f852578cb6 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/4ec221b23762b137e1f58f81383ed732ceccacdfe1e9c1602a7dc96ab5b7260f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67da9209c0279a3bbd6474c7eecfcbf10885f943ac43823bda3a626e783e1100 +size 5165 diff --git a/parse/train/kW_zpEmMLdP/images/50557ab1d4f71ef4e1b4650e143ca13aa43084266980a6d3349248a0bb3970b5.jpg b/parse/train/kW_zpEmMLdP/images/50557ab1d4f71ef4e1b4650e143ca13aa43084266980a6d3349248a0bb3970b5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4923e11091cfeba087877c8de66dffd60d6955f4 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/50557ab1d4f71ef4e1b4650e143ca13aa43084266980a6d3349248a0bb3970b5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00e488351822e5a3430549573c57568c1584dba98b5a7bb6b15713fa4fae6d56 +size 7957 diff --git a/parse/train/kW_zpEmMLdP/images/5123c9cb7769d896ed0233aa658a140c32ef93df2f2f62a45068e0055febdbe3.jpg b/parse/train/kW_zpEmMLdP/images/5123c9cb7769d896ed0233aa658a140c32ef93df2f2f62a45068e0055febdbe3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f3a7499acb3a5ef27c6098c0ae34dea14ae01dda --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/5123c9cb7769d896ed0233aa658a140c32ef93df2f2f62a45068e0055febdbe3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1bfd409d8e7d09953f50631c5fe5fdb3f944bb24e3fdc82458abaefb5323452 +size 9835 diff --git a/parse/train/kW_zpEmMLdP/images/5707d0258bbeed54a6692ddecd00be20874e57b32e1e131d790d881cade1f44c.jpg b/parse/train/kW_zpEmMLdP/images/5707d0258bbeed54a6692ddecd00be20874e57b32e1e131d790d881cade1f44c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..303beb534a55f947c8b60c3129ad7125b0fd61dd --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/5707d0258bbeed54a6692ddecd00be20874e57b32e1e131d790d881cade1f44c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddfb986add1e642c99871702aecfc3ffa13e979a032957f47874f398e82abba9 +size 3894 diff --git a/parse/train/kW_zpEmMLdP/images/581fbe5ab8fc9f499754572ae6d8155a908a0facfbc16233beade0569938bd0f.jpg b/parse/train/kW_zpEmMLdP/images/581fbe5ab8fc9f499754572ae6d8155a908a0facfbc16233beade0569938bd0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e3988130d4bb4b5385ad9cd2fc765734d763a563 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/581fbe5ab8fc9f499754572ae6d8155a908a0facfbc16233beade0569938bd0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbe387d48d8d5bda43355d006ef9dfc250cff81c4225f5a4bc883054a950e7c5 +size 5733 diff --git a/parse/train/kW_zpEmMLdP/images/5988ad9cb668c04010a32c82d65072808d4ad376d29f525daa76b045200ed46f.jpg b/parse/train/kW_zpEmMLdP/images/5988ad9cb668c04010a32c82d65072808d4ad376d29f525daa76b045200ed46f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6d0d8cb7a031d99364854a22e45640c4171a86b7 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/5988ad9cb668c04010a32c82d65072808d4ad376d29f525daa76b045200ed46f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19e2c156da5319c52428c72731e3ac3d780df07f1864bd698fcd15feb9814a61 +size 31776 diff --git a/parse/train/kW_zpEmMLdP/images/723379b4d392bced631053012d738d93f625d274b7cf0837fc0e762fa42d6e5e.jpg b/parse/train/kW_zpEmMLdP/images/723379b4d392bced631053012d738d93f625d274b7cf0837fc0e762fa42d6e5e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8665fe008d2134a31e430b352b259204d3dc435f --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/723379b4d392bced631053012d738d93f625d274b7cf0837fc0e762fa42d6e5e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa22f4813ed802bc10db111ea5b841372232f16ee5014880d952c6a55a794ea2 +size 71589 diff --git a/parse/train/kW_zpEmMLdP/images/78f231da1c57dbceeb5712a0f9cb646e4004963c44a3ad38cd186e549eda27fc.jpg b/parse/train/kW_zpEmMLdP/images/78f231da1c57dbceeb5712a0f9cb646e4004963c44a3ad38cd186e549eda27fc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..caf1ab46e8d70ff57a76570fc6aa60edb89b4531 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/78f231da1c57dbceeb5712a0f9cb646e4004963c44a3ad38cd186e549eda27fc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b35491d82c609098074b36724cc4b4fca0cae8ab9834ae2726c6528776ef31ef +size 109312 diff --git a/parse/train/kW_zpEmMLdP/images/79cab534e6022fe94537953ae38ca5dcf1b8a0083d49a9b4551a51cfd11e60f3.jpg b/parse/train/kW_zpEmMLdP/images/79cab534e6022fe94537953ae38ca5dcf1b8a0083d49a9b4551a51cfd11e60f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..032eb71f9214c1d9d935b8a54c80764785d77811 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/79cab534e6022fe94537953ae38ca5dcf1b8a0083d49a9b4551a51cfd11e60f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dfb114816cd9fe5d9b876310c60f874a8cdc33e65dce0e3c7b4884b1bb32f2de +size 9786 diff --git a/parse/train/kW_zpEmMLdP/images/8386c67b8db0e78fb86996acb7e4b896268a8141dc9c7ea8c47253b229a71fe2.jpg b/parse/train/kW_zpEmMLdP/images/8386c67b8db0e78fb86996acb7e4b896268a8141dc9c7ea8c47253b229a71fe2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..675e3847216384a19f026c42f10f976534db6517 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/8386c67b8db0e78fb86996acb7e4b896268a8141dc9c7ea8c47253b229a71fe2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8dbda61a7208cd8264dccb6fc71598128dd9150b9b84794993e1e49114101da +size 1984 diff --git a/parse/train/kW_zpEmMLdP/images/8cd50fe01605b942bf745a2b6fbb8f3a112ffdf0487b9028a86b729464049de8.jpg b/parse/train/kW_zpEmMLdP/images/8cd50fe01605b942bf745a2b6fbb8f3a112ffdf0487b9028a86b729464049de8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a12c1119f7696cc07e1e6c7c2070fda9a2e967f0 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/8cd50fe01605b942bf745a2b6fbb8f3a112ffdf0487b9028a86b729464049de8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f0c57c68116afa6f2e1b552c1fcdff726da72d91e34b34c178ec2b5c8b82e7d +size 7785 diff --git a/parse/train/kW_zpEmMLdP/images/95555e40d0ebdb768ca569f3ee3022eccd0013102f4766579c16be768cd28ea4.jpg b/parse/train/kW_zpEmMLdP/images/95555e40d0ebdb768ca569f3ee3022eccd0013102f4766579c16be768cd28ea4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1fd54be5d134aa2a27ee27bdb58296cc6a9a0d4a --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/95555e40d0ebdb768ca569f3ee3022eccd0013102f4766579c16be768cd28ea4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f7955c800ab0b269f9f7ee8021587db97bdfb67a9e91e91aa800f65c9f2d139 +size 5976 diff --git a/parse/train/kW_zpEmMLdP/images/9b765b5f7c28c8d8bb324a2b76644d786cf7566f23bb173ccaf48456308a14a9.jpg b/parse/train/kW_zpEmMLdP/images/9b765b5f7c28c8d8bb324a2b76644d786cf7566f23bb173ccaf48456308a14a9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ff84725c19ec34d122a45e08791e4293f14f2802 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/9b765b5f7c28c8d8bb324a2b76644d786cf7566f23bb173ccaf48456308a14a9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aae9a17664b6bc68aeb41340c90e9a894ddb58aafda7cfe09e70d7fdbbc8969d +size 15334 diff --git a/parse/train/kW_zpEmMLdP/images/ab901b01da3f4fce0e69991398aeab32fd7a469fbdfba991cbc9db87ac402724.jpg b/parse/train/kW_zpEmMLdP/images/ab901b01da3f4fce0e69991398aeab32fd7a469fbdfba991cbc9db87ac402724.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f08e3ffadc4b22ba0a4db5097c9f29ee2b0afc60 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/ab901b01da3f4fce0e69991398aeab32fd7a469fbdfba991cbc9db87ac402724.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab9fec13ff745df1bfb4d07f0e8eeeb7134f389e08ee436c70cdcaa3402267f9 +size 70281 diff --git a/parse/train/kW_zpEmMLdP/images/abfa1e619ffd89e2e673c408547dc23710893eead4ab288bc89a70be71c733b4.jpg b/parse/train/kW_zpEmMLdP/images/abfa1e619ffd89e2e673c408547dc23710893eead4ab288bc89a70be71c733b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9efa294f5c6d49c99842d618faba3930cb1f725a --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/abfa1e619ffd89e2e673c408547dc23710893eead4ab288bc89a70be71c733b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31194c03fd239088f9b70231e05bed3907f62947555433a81952d6232b8ee591 +size 4832 diff --git a/parse/train/kW_zpEmMLdP/images/af8004ec87444a6c2f5b72d0b26768a866a7f6ccfe2b0594d73fb8b4ff2cd7d4.jpg b/parse/train/kW_zpEmMLdP/images/af8004ec87444a6c2f5b72d0b26768a866a7f6ccfe2b0594d73fb8b4ff2cd7d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6d3c2ed8baa698cc0c46dc7cb3e967c36bc7694e --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/af8004ec87444a6c2f5b72d0b26768a866a7f6ccfe2b0594d73fb8b4ff2cd7d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b4598a0a7713e8784f8912ba6795df43afdd65b1b06325bd43b76d84a1c5023 +size 9262 diff --git a/parse/train/kW_zpEmMLdP/images/b40b57a3093dc26df3c32bb247ee92538f14fd2060baebc43a0f1e4be718d92f.jpg b/parse/train/kW_zpEmMLdP/images/b40b57a3093dc26df3c32bb247ee92538f14fd2060baebc43a0f1e4be718d92f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6d362ce9e0626e7744e5b6fae4c71cc1f1938851 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/b40b57a3093dc26df3c32bb247ee92538f14fd2060baebc43a0f1e4be718d92f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c99ba970aab9f474499f7ae96ffea00cd507a83fff6285c8cf3fa4407abe8602 +size 89393 diff --git a/parse/train/kW_zpEmMLdP/images/bbb5aabd1c4a24b828354c7c2936108cfbab251255c095c7f2528ce273dbf6d4.jpg b/parse/train/kW_zpEmMLdP/images/bbb5aabd1c4a24b828354c7c2936108cfbab251255c095c7f2528ce273dbf6d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..57dcb0684a10e50039fae0af28880f97cbcbe9a8 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/bbb5aabd1c4a24b828354c7c2936108cfbab251255c095c7f2528ce273dbf6d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25a3586f746d5fde8743885461939652ea532a8d3ed8292396ecdca892972102 +size 5582 diff --git a/parse/train/kW_zpEmMLdP/images/c17b4fb89729e7c3df0884807ec3b5d3a137a2fb80ea35781aca77db4f3e6cdf.jpg b/parse/train/kW_zpEmMLdP/images/c17b4fb89729e7c3df0884807ec3b5d3a137a2fb80ea35781aca77db4f3e6cdf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ae876887fbe9367990f229a03f68200bf24ab352 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/c17b4fb89729e7c3df0884807ec3b5d3a137a2fb80ea35781aca77db4f3e6cdf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7b8bcc7ccc2d9caf66f832d69454544631a05e15409743d5bace2dad5191ff1 +size 8632 diff --git a/parse/train/kW_zpEmMLdP/images/cca3f6f9c9e0cdd9f912ba4765682e7a110fcc7c5a6497503ac85095be41d02b.jpg b/parse/train/kW_zpEmMLdP/images/cca3f6f9c9e0cdd9f912ba4765682e7a110fcc7c5a6497503ac85095be41d02b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d7c564ebec22e979c4560db8765750f8cb9d5c28 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/cca3f6f9c9e0cdd9f912ba4765682e7a110fcc7c5a6497503ac85095be41d02b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c2b870f380c2a59a328d3f2398777f1225cd403a1f230a366629492ffd6814f +size 9960 diff --git a/parse/train/kW_zpEmMLdP/images/d7abe975a06a178aa5526fb411d20e67258c7b12155dab8db96e73ee1d637544.jpg b/parse/train/kW_zpEmMLdP/images/d7abe975a06a178aa5526fb411d20e67258c7b12155dab8db96e73ee1d637544.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5d1630adf3437c362cc7d9c2d8611d8a16153ddd --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/d7abe975a06a178aa5526fb411d20e67258c7b12155dab8db96e73ee1d637544.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f555a4379b5e1082cf27a9ea3271cde295581e972224529e2566a1851c362ab0 +size 26632 diff --git a/parse/train/kW_zpEmMLdP/images/dcb5c9f961ed752c4c4c97e3d5aa4ed98d1f3815b407132760e20b7035c2c79d.jpg b/parse/train/kW_zpEmMLdP/images/dcb5c9f961ed752c4c4c97e3d5aa4ed98d1f3815b407132760e20b7035c2c79d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..90b56b6339e60dde8985279debd6fc8d532cb4b8 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/dcb5c9f961ed752c4c4c97e3d5aa4ed98d1f3815b407132760e20b7035c2c79d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f3d6d475feef8b81f4fff5d0f7d41b3a908b9338fb84f833e88a79c20af3eeb +size 20552 diff --git a/parse/train/kW_zpEmMLdP/images/e52e47a88c4515e116f1e80d8788fa3081faf9f3ab8656268d23029ead0c9c82.jpg b/parse/train/kW_zpEmMLdP/images/e52e47a88c4515e116f1e80d8788fa3081faf9f3ab8656268d23029ead0c9c82.jpg new file mode 100644 index 0000000000000000000000000000000000000000..124278e89a305bb61d453d1cb25dd5060ac06648 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/e52e47a88c4515e116f1e80d8788fa3081faf9f3ab8656268d23029ead0c9c82.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa198924c2857c8c06d5d0e86df844f72687780456d2127c9e77acdd98bc912e +size 6825 diff --git a/parse/train/kW_zpEmMLdP/images/e7d56149874e2bf9449faf02fcc2b3ab53c453ca5dc9320e9be8d3f0b21de7b0.jpg b/parse/train/kW_zpEmMLdP/images/e7d56149874e2bf9449faf02fcc2b3ab53c453ca5dc9320e9be8d3f0b21de7b0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eafdad7cb0f40e087ae2e12521a6faa1d923e804 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/e7d56149874e2bf9449faf02fcc2b3ab53c453ca5dc9320e9be8d3f0b21de7b0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ec15dbc55d5c961320f8d704e831d69bb08af94b34621343479d1036508afec +size 38487 diff --git a/parse/train/kW_zpEmMLdP/images/eb9496f880a86a00296bf11c61334bada0577dab9986509a9b669b12c05858e8.jpg b/parse/train/kW_zpEmMLdP/images/eb9496f880a86a00296bf11c61334bada0577dab9986509a9b669b12c05858e8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f03659e2712c8c620d494b5fcf3a2ea5a759dba3 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/eb9496f880a86a00296bf11c61334bada0577dab9986509a9b669b12c05858e8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d738f9f8a8c4dc40d06fa3be666d2b7c2afd50baedff49a8b53bc05d8fe28385 +size 7932 diff --git a/parse/train/kW_zpEmMLdP/images/f07179b3ae0b8c27d15b0063b7e10b3f61a3a7ef91265699908a3bfd6b27374d.jpg b/parse/train/kW_zpEmMLdP/images/f07179b3ae0b8c27d15b0063b7e10b3f61a3a7ef91265699908a3bfd6b27374d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bd8804614267511e65cc23a6188c98ff921421b8 --- /dev/null +++ b/parse/train/kW_zpEmMLdP/images/f07179b3ae0b8c27d15b0063b7e10b3f61a3a7ef91265699908a3bfd6b27374d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:88b8edb34dbb82daa71c7d25bff9b9d7f84cc990a0e2d1ab812ea4cb24dd0004 +size 3935 diff --git a/parse/train/r1gBOxSFwr/images/077b92a2b85c8f67b7fb4f51840cbf117525b2b21bdcfa183d7d03e965e901c4.jpg b/parse/train/r1gBOxSFwr/images/077b92a2b85c8f67b7fb4f51840cbf117525b2b21bdcfa183d7d03e965e901c4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4d116130a703dd4ece97ce5935c6e7791b84bb6a --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/077b92a2b85c8f67b7fb4f51840cbf117525b2b21bdcfa183d7d03e965e901c4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89610afa4f10d0172bb4fe455085ed989d246422ede7215d0f97fb3ceee6e9ce +size 28044 diff --git a/parse/train/r1gBOxSFwr/images/0943bb6b2d46a873250c1ed78ffda01ebc514f54a4cdaa1abae0000e38c9ba0e.jpg b/parse/train/r1gBOxSFwr/images/0943bb6b2d46a873250c1ed78ffda01ebc514f54a4cdaa1abae0000e38c9ba0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9ae5d3a23cf8d9600844716106d80bbceb0e1d51 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/0943bb6b2d46a873250c1ed78ffda01ebc514f54a4cdaa1abae0000e38c9ba0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:301059eb4449b2aa052a8a0c6e6325e12b6719e33829f7e92e14d4c59b1b5c88 +size 64419 diff --git a/parse/train/r1gBOxSFwr/images/0f0d834e3a679d33a2d87a3dfce5c000306f282f56c483413282f036d58e7087.jpg b/parse/train/r1gBOxSFwr/images/0f0d834e3a679d33a2d87a3dfce5c000306f282f56c483413282f036d58e7087.jpg new file mode 100644 index 0000000000000000000000000000000000000000..87ec0b02951c986feaa307ad33fa831f715ba51a --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/0f0d834e3a679d33a2d87a3dfce5c000306f282f56c483413282f036d58e7087.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7712140f0d7aaad24b96f9a40ba671c2dbc827b4a2ed49339af8a2bc9feda18 +size 4122 diff --git a/parse/train/r1gBOxSFwr/images/13cf2b09a9fb2ccba4ee2c09854d072fdfe8b58ecf9f2249becf636cfc3e503b.jpg b/parse/train/r1gBOxSFwr/images/13cf2b09a9fb2ccba4ee2c09854d072fdfe8b58ecf9f2249becf636cfc3e503b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5a82b853f4e8a72226383e9ab6723a79a3983fe6 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/13cf2b09a9fb2ccba4ee2c09854d072fdfe8b58ecf9f2249becf636cfc3e503b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9fe272c36397b796ca0a63835344137c24778d01cbe3fdff1c42b00e5913075 +size 4413 diff --git a/parse/train/r1gBOxSFwr/images/1b02313bc2f642a2a9c2a4f79b77e02230fd123f895c99a4d477b9bb77fc661d.jpg b/parse/train/r1gBOxSFwr/images/1b02313bc2f642a2a9c2a4f79b77e02230fd123f895c99a4d477b9bb77fc661d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b65f2bd552d6aa59e3b9c1206ced1cd20aabd7f0 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/1b02313bc2f642a2a9c2a4f79b77e02230fd123f895c99a4d477b9bb77fc661d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b340649ff5fd86c6fa34c792b0195980e8a29834ed3f5aaa8ec50901303f231 +size 3748 diff --git a/parse/train/r1gBOxSFwr/images/252bfdc463ccf1d573b57ab7e7c0a9bfa131cfc11fd63d36c24ced957d47a7c2.jpg b/parse/train/r1gBOxSFwr/images/252bfdc463ccf1d573b57ab7e7c0a9bfa131cfc11fd63d36c24ced957d47a7c2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..de24fb602694e76061550c264ca2e61f4c210587 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/252bfdc463ccf1d573b57ab7e7c0a9bfa131cfc11fd63d36c24ced957d47a7c2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1aca779943f2950cfb10d98fd80e875b52cf8c9a2bbc9a6bdaf1ea72426e976b +size 63118 diff --git a/parse/train/r1gBOxSFwr/images/26cd1178dbddb29b5b36b53f084ed3aa8bdfcf0871dd1763cbfa1b64ef19a5d3.jpg b/parse/train/r1gBOxSFwr/images/26cd1178dbddb29b5b36b53f084ed3aa8bdfcf0871dd1763cbfa1b64ef19a5d3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9dc737f8d7dc2bcef917f6f36b148b59488c79ae --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/26cd1178dbddb29b5b36b53f084ed3aa8bdfcf0871dd1763cbfa1b64ef19a5d3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a28d2b5006970261178bd42dee1ed308efdae1aa7ef1e14d426137d312924358 +size 6257 diff --git a/parse/train/r1gBOxSFwr/images/2ed1f4c6d245787bf8be9744bc726c7b8ac31c7d278ba6db146015c5292b492d.jpg b/parse/train/r1gBOxSFwr/images/2ed1f4c6d245787bf8be9744bc726c7b8ac31c7d278ba6db146015c5292b492d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6e85ff28e646dc141632edeed16e6aa46715e9ad --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/2ed1f4c6d245787bf8be9744bc726c7b8ac31c7d278ba6db146015c5292b492d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f803f4b5d54d211bef7c37cb2b90c5a83dd5e4e64aa3944df4cc22b590c1ce15 +size 5208 diff --git a/parse/train/r1gBOxSFwr/images/2fc534ecec1bad3c6840ae491a336ee8ecfdb3685c8b7b14f7d14b846a1ec854.jpg b/parse/train/r1gBOxSFwr/images/2fc534ecec1bad3c6840ae491a336ee8ecfdb3685c8b7b14f7d14b846a1ec854.jpg new file mode 100644 index 0000000000000000000000000000000000000000..75c0a7bc602ccb0011ee99d7827a820fb6c4fb77 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/2fc534ecec1bad3c6840ae491a336ee8ecfdb3685c8b7b14f7d14b846a1ec854.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:180d7457d18fce54a1ff08fe4127092762a3269d25ea3a1ab2f7b5c27896706d +size 62565 diff --git a/parse/train/r1gBOxSFwr/images/4bf06ceff4d666da1d47102f965e8f890355a47514b801d5cd9e28d6619653c8.jpg b/parse/train/r1gBOxSFwr/images/4bf06ceff4d666da1d47102f965e8f890355a47514b801d5cd9e28d6619653c8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4cd4d70eb1d267b7fe1c97ee9883d289a4e39043 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/4bf06ceff4d666da1d47102f965e8f890355a47514b801d5cd9e28d6619653c8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b46efd8c51a0971436774ce5d395ce81c811bed1f5d9428336d5982e693dbc4 +size 62891 diff --git a/parse/train/r1gBOxSFwr/images/4c072761616d6607d9247646365766816eb1bce4969431a179dcada4442c92a9.jpg b/parse/train/r1gBOxSFwr/images/4c072761616d6607d9247646365766816eb1bce4969431a179dcada4442c92a9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4ad5c3c906c518cd04f63abab90d9bbee54df1c0 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/4c072761616d6607d9247646365766816eb1bce4969431a179dcada4442c92a9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b8f2df0dce5dca0bd3cdf499faffb5f7ad872b1753e7bd50cecdea7c6f90757c +size 31440 diff --git a/parse/train/r1gBOxSFwr/images/4fc93512528bccb502f0d178d5883958d951c278d715554c59383cfcdcbfe8a8.jpg b/parse/train/r1gBOxSFwr/images/4fc93512528bccb502f0d178d5883958d951c278d715554c59383cfcdcbfe8a8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf8e20a2655741dea5ea191733606e1928edb461 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/4fc93512528bccb502f0d178d5883958d951c278d715554c59383cfcdcbfe8a8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc706a60d5458ce0f9f2294d6aa9e071ce5f9a55bf3875cb71ab2bea1d0dac6a +size 65192 diff --git a/parse/train/r1gBOxSFwr/images/740c239929acaab0b63ff4de72a954ecceaae0479908fdaffeb22ceac1de755a.jpg b/parse/train/r1gBOxSFwr/images/740c239929acaab0b63ff4de72a954ecceaae0479908fdaffeb22ceac1de755a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..81a0e241365d08ae5c6c0c14dbb6255a967f46af --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/740c239929acaab0b63ff4de72a954ecceaae0479908fdaffeb22ceac1de755a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb15821281530502014e4c251dc38263ba81e5910fd37ecbca19bfae67ff3f77 +size 3730 diff --git a/parse/train/r1gBOxSFwr/images/7d757ee8a360fe0ba71f34905691dede9068f67aa0ef2dd1e18a47c8c7a08643.jpg b/parse/train/r1gBOxSFwr/images/7d757ee8a360fe0ba71f34905691dede9068f67aa0ef2dd1e18a47c8c7a08643.jpg new file mode 100644 index 0000000000000000000000000000000000000000..03bd658f1e493473f1b88195b30b8b9699c1a887 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/7d757ee8a360fe0ba71f34905691dede9068f67aa0ef2dd1e18a47c8c7a08643.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0cde264a4892588a86c4f6f22a0cd82fb26f8d1710f49b0e0c6b963f0c8a710 +size 63469 diff --git a/parse/train/r1gBOxSFwr/images/818c0a6ade5596a9eaf9b80b8ec5f7e389a7368321d048f6d012dac0081655ee.jpg b/parse/train/r1gBOxSFwr/images/818c0a6ade5596a9eaf9b80b8ec5f7e389a7368321d048f6d012dac0081655ee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..197baa89bd403a58fcf68be0dd7b26adbaf13a4a --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/818c0a6ade5596a9eaf9b80b8ec5f7e389a7368321d048f6d012dac0081655ee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dde8354e553509899417035b5830c54220c73ace6fd1331bcff303a76fdc9e59 +size 2747 diff --git a/parse/train/r1gBOxSFwr/images/974b17427a4f34781c6007e4a6d0845e6b491255ea2cf2951d3bcaf315913b00.jpg b/parse/train/r1gBOxSFwr/images/974b17427a4f34781c6007e4a6d0845e6b491255ea2cf2951d3bcaf315913b00.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e35e3ea9de73596d398fd6b221355f7e4dfd0eda --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/974b17427a4f34781c6007e4a6d0845e6b491255ea2cf2951d3bcaf315913b00.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7636544370288fdb067200f809fbc915687d81e09d7107015d7f5609ae6de284 +size 62388 diff --git a/parse/train/r1gBOxSFwr/images/98b593b7456b2f627a9c67015f8f431abf85b774bcabebfac9e23059bc3b664f.jpg b/parse/train/r1gBOxSFwr/images/98b593b7456b2f627a9c67015f8f431abf85b774bcabebfac9e23059bc3b664f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..54577fa12ecfc2a151b373f6ef5d814978824ce3 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/98b593b7456b2f627a9c67015f8f431abf85b774bcabebfac9e23059bc3b664f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4edf3c64323b1e8eb9c4cb0eb88b9c3f2502353109ab5a65a33374b4b08d4501 +size 63583 diff --git a/parse/train/r1gBOxSFwr/images/9bb6c2dcf1463042e321f6ecbf2ab5f08350b14908219dfea7c5ff1ba458d184.jpg b/parse/train/r1gBOxSFwr/images/9bb6c2dcf1463042e321f6ecbf2ab5f08350b14908219dfea7c5ff1ba458d184.jpg new file mode 100644 index 0000000000000000000000000000000000000000..141538ab0acf9dd7f03778a3720be366c4612366 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/9bb6c2dcf1463042e321f6ecbf2ab5f08350b14908219dfea7c5ff1ba458d184.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f191d39f504f66a35d9c7e10e56108d53b8d60a5a5c362fcf1237a1c0c6609c +size 2985 diff --git a/parse/train/r1gBOxSFwr/images/ab8ec62f2c2de938d5f87f8cc8279be1afa065c41ffb5aabf5dfa8d6fb3aad16.jpg b/parse/train/r1gBOxSFwr/images/ab8ec62f2c2de938d5f87f8cc8279be1afa065c41ffb5aabf5dfa8d6fb3aad16.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6ef348bdbbab7a0833fb979f976823acf427935a --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/ab8ec62f2c2de938d5f87f8cc8279be1afa065c41ffb5aabf5dfa8d6fb3aad16.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bbdd15da2647a7855c0b01a8d277574651188a9fbc6f08a8643564a85fda2e9 +size 87674 diff --git a/parse/train/r1gBOxSFwr/images/b71e9cdb931c255737c774aac177c35838265222f323612d0067f86353c81006.jpg b/parse/train/r1gBOxSFwr/images/b71e9cdb931c255737c774aac177c35838265222f323612d0067f86353c81006.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ba2c4403340a4b28a2278e37b67e57302b4003ee --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/b71e9cdb931c255737c774aac177c35838265222f323612d0067f86353c81006.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fad1c798fb0b5d97b94ecf3d73c50dd702fe0f1d4798b0b9d12b5bf31da12b53 +size 47639 diff --git a/parse/train/r1gBOxSFwr/images/bc5e7f098a58038130103414a7bee5968600db371744b5a32347d26fe22a5b5d.jpg b/parse/train/r1gBOxSFwr/images/bc5e7f098a58038130103414a7bee5968600db371744b5a32347d26fe22a5b5d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b41188962f0d63c80c07b2665114c9458f17425c --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/bc5e7f098a58038130103414a7bee5968600db371744b5a32347d26fe22a5b5d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76d8d3bd9bfe32716bd0db68d7c15b24d956bfccce47d20c68084608f5e3f29b +size 5363 diff --git a/parse/train/r1gBOxSFwr/images/ec44ffeca034a584f723f35766775b9b2e26ccb2b05583aa76a0bf11da4c3bf3.jpg b/parse/train/r1gBOxSFwr/images/ec44ffeca034a584f723f35766775b9b2e26ccb2b05583aa76a0bf11da4c3bf3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cc142accfbe542eaa67850fabd800c7978dddf87 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/ec44ffeca034a584f723f35766775b9b2e26ccb2b05583aa76a0bf11da4c3bf3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:048a4e7ee60eda2ebd989338a557a3f338a714cb66607d86875a28749e16aff8 +size 67091 diff --git a/parse/train/r1gBOxSFwr/images/f09eee29eb42da2e4c6723cabab2fc81f422c5ecd5b38913d5387344f6c39de0.jpg b/parse/train/r1gBOxSFwr/images/f09eee29eb42da2e4c6723cabab2fc81f422c5ecd5b38913d5387344f6c39de0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3c0a60a3457d105bc2d31ca55a45056454404ae8 --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/f09eee29eb42da2e4c6723cabab2fc81f422c5ecd5b38913d5387344f6c39de0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46becd322e305261a865b1961ca8ceae39e3fe44f88d28f5a19dcc08508f1cf3 +size 58010 diff --git a/parse/train/r1gBOxSFwr/images/f8b4b1926573d64d81f8aad65795b0493ffbaa53ebedfe7e4590efbf3159be4e.jpg b/parse/train/r1gBOxSFwr/images/f8b4b1926573d64d81f8aad65795b0493ffbaa53ebedfe7e4590efbf3159be4e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..65acc5906bbbfdba40a3ec25f0e257660e31460b --- /dev/null +++ b/parse/train/r1gBOxSFwr/images/f8b4b1926573d64d81f8aad65795b0493ffbaa53ebedfe7e4590efbf3159be4e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:68eb40378c9d187844be757e98741b6719f4f0bc7288b2d46fa5f9f9633916a1 +size 80553 diff --git a/parse/train/r1glDpNYwS/images/130ca587a82453324ed51da5a452a8791633884559a83a5803a3898b7eeaaf9b.jpg b/parse/train/r1glDpNYwS/images/130ca587a82453324ed51da5a452a8791633884559a83a5803a3898b7eeaaf9b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d84b8043db58a4be84f3508149f3b6cc152ab1f7 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/130ca587a82453324ed51da5a452a8791633884559a83a5803a3898b7eeaaf9b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:608f58ce88769597b45e65dec290e4fa9cb14e57d14f33dcfe0b1d5532dff4dc +size 106395 diff --git a/parse/train/r1glDpNYwS/images/239ee6197e5924863c38e21f735796e29db234aeb91960074102f63a3db476b4.jpg b/parse/train/r1glDpNYwS/images/239ee6197e5924863c38e21f735796e29db234aeb91960074102f63a3db476b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cef1a4bf5b99c09f0a08e6cdf23c2519b844d544 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/239ee6197e5924863c38e21f735796e29db234aeb91960074102f63a3db476b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d45e1441301d4fa86ae705380e8ec6a918889b87d7cfc15f4d4c9eefb506f52c +size 5931 diff --git a/parse/train/r1glDpNYwS/images/408de8d892d3acc28d50ae6ef93d25372c1a2e6ef6b7cc9a41e548f0726ad348.jpg b/parse/train/r1glDpNYwS/images/408de8d892d3acc28d50ae6ef93d25372c1a2e6ef6b7cc9a41e548f0726ad348.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4d0495ee275f1bcc02b83b05b5720f92ef48bf17 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/408de8d892d3acc28d50ae6ef93d25372c1a2e6ef6b7cc9a41e548f0726ad348.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd476801fc46d564935678c6c039a71d11be5b20afedb65dc22d9b1c9562342b +size 4346 diff --git a/parse/train/r1glDpNYwS/images/549416db53bf719da6cae0d5ce3460fa6fa29dd8b3f41f41d2c957b2530bf131.jpg b/parse/train/r1glDpNYwS/images/549416db53bf719da6cae0d5ce3460fa6fa29dd8b3f41f41d2c957b2530bf131.jpg new file mode 100644 index 0000000000000000000000000000000000000000..36b613ac7665da98175869e7e01309d9aa8515de --- /dev/null +++ b/parse/train/r1glDpNYwS/images/549416db53bf719da6cae0d5ce3460fa6fa29dd8b3f41f41d2c957b2530bf131.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b15f55bedd228358c3cbd7363e68112bbd9af57e433bec99c88c47fe4578082a +size 72175 diff --git a/parse/train/r1glDpNYwS/images/5b15f8f4e910472b4d11cf21cf87b2c22d63c70a811de85807b897e853634270.jpg b/parse/train/r1glDpNYwS/images/5b15f8f4e910472b4d11cf21cf87b2c22d63c70a811de85807b897e853634270.jpg new file mode 100644 index 0000000000000000000000000000000000000000..776b6ba938ba7ec247847a4bf2c4a9ae4e4a4680 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/5b15f8f4e910472b4d11cf21cf87b2c22d63c70a811de85807b897e853634270.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80480b33d91458e3bab2c0f67b17d89ed97ddae9d92bee726c2c3b15f665d10b +size 23846 diff --git a/parse/train/r1glDpNYwS/images/641ad54c33971d3bc7cf120270597da0706328bfe4d90f2dd5946013e1f160d8.jpg b/parse/train/r1glDpNYwS/images/641ad54c33971d3bc7cf120270597da0706328bfe4d90f2dd5946013e1f160d8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..395fd1cea928112da8922cbe555e0b9f6a895e90 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/641ad54c33971d3bc7cf120270597da0706328bfe4d90f2dd5946013e1f160d8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:154204a5c14dcad22b4ad0909c9be779f4f71795209342b17d4382dedc36a4f5 +size 42401 diff --git a/parse/train/r1glDpNYwS/images/68ba82957f5f95e3dcd977744f4aaa68e10413b4b341ab50c91d806ce43d34af.jpg b/parse/train/r1glDpNYwS/images/68ba82957f5f95e3dcd977744f4aaa68e10413b4b341ab50c91d806ce43d34af.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8ab5197a8d2579791755b697cb166a7860767a9b --- /dev/null +++ b/parse/train/r1glDpNYwS/images/68ba82957f5f95e3dcd977744f4aaa68e10413b4b341ab50c91d806ce43d34af.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:220e191b2277e2e383fb374cbe7b90d93813bd11abde4caad0b050fe16820dc4 +size 34978 diff --git a/parse/train/r1glDpNYwS/images/722f7ed9fcb97702dfd0d8b93124a553ccfd8eee983cd74823821e4bb8c821ab.jpg b/parse/train/r1glDpNYwS/images/722f7ed9fcb97702dfd0d8b93124a553ccfd8eee983cd74823821e4bb8c821ab.jpg new file mode 100644 index 0000000000000000000000000000000000000000..caca6ddd4fd03c15805acbfdc6bd4f74d9d4f5d3 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/722f7ed9fcb97702dfd0d8b93124a553ccfd8eee983cd74823821e4bb8c821ab.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af2da152ae8b382e2e9f5f69dc216817dbef7b1d38655777341abf5bae45aea8 +size 4775 diff --git a/parse/train/r1glDpNYwS/images/7b07a9188b688b615bcb891efb10f668c9948f1b49b2ea95f2ccd7546cf68d03.jpg b/parse/train/r1glDpNYwS/images/7b07a9188b688b615bcb891efb10f668c9948f1b49b2ea95f2ccd7546cf68d03.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c631b8582f979fc4384716539c3cb456d7aa1097 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/7b07a9188b688b615bcb891efb10f668c9948f1b49b2ea95f2ccd7546cf68d03.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e93dcfd3bbbec0e136017c7acaf0c780aba37738c7a9531b48874981cab90609 +size 37562 diff --git a/parse/train/r1glDpNYwS/images/8e2a1cacce0cf0456a3cd0e12b860e50c471b52d9a7cc871f309365bacdc287d.jpg b/parse/train/r1glDpNYwS/images/8e2a1cacce0cf0456a3cd0e12b860e50c471b52d9a7cc871f309365bacdc287d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8cc941dc3a7265a07acf5bac3ebd94aba6767367 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/8e2a1cacce0cf0456a3cd0e12b860e50c471b52d9a7cc871f309365bacdc287d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec92d510b584684afb499a53193de6dfb8527f2bdddb9b7697fe0e20b9dfc189 +size 8200 diff --git a/parse/train/r1glDpNYwS/images/9712671a42dd9bc7ad66310ebe725ff1fb9b7026e368d207f4430fbfbb04b916.jpg b/parse/train/r1glDpNYwS/images/9712671a42dd9bc7ad66310ebe725ff1fb9b7026e368d207f4430fbfbb04b916.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ba343dac8dca0309c8c21dd79fbc3c8eecabb3e5 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/9712671a42dd9bc7ad66310ebe725ff1fb9b7026e368d207f4430fbfbb04b916.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b873b14eba31c6c6c113344dd156da90fbf5f2d82c11d55a040fbbf65ce859b +size 20451 diff --git a/parse/train/r1glDpNYwS/images/9e6c05a79367119d8a3b7be272d129a1544bbc734c74747da2b43d92367ae90f.jpg b/parse/train/r1glDpNYwS/images/9e6c05a79367119d8a3b7be272d129a1544bbc734c74747da2b43d92367ae90f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9270c56e105d989104c58385102788ee9ba3e856 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/9e6c05a79367119d8a3b7be272d129a1544bbc734c74747da2b43d92367ae90f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6b049ceecb43be501506c59b3e7af016da894d631f578714f4d6979130c533c +size 28109 diff --git a/parse/train/r1glDpNYwS/images/a86925c24d1da2a9bbbd5f719bdd968d50e528b34f4526c191a2cfa41244e9f9.jpg b/parse/train/r1glDpNYwS/images/a86925c24d1da2a9bbbd5f719bdd968d50e528b34f4526c191a2cfa41244e9f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..08ef07dcfd478beb4f6127420562b3e9d48d1619 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/a86925c24d1da2a9bbbd5f719bdd968d50e528b34f4526c191a2cfa41244e9f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e15d20514cfd37c32d16d28265102e9657986d7a316855a45c988cf631d4026e +size 49887 diff --git a/parse/train/r1glDpNYwS/images/ca7df80dddb9a26126c80fe3080f88f177c7491f3cf13888cf85b195d5fa50e9.jpg b/parse/train/r1glDpNYwS/images/ca7df80dddb9a26126c80fe3080f88f177c7491f3cf13888cf85b195d5fa50e9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..881d84ee31e6ef8c7712b747249e37e52551bd2d --- /dev/null +++ b/parse/train/r1glDpNYwS/images/ca7df80dddb9a26126c80fe3080f88f177c7491f3cf13888cf85b195d5fa50e9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a12782550f4ffe62b0c81192f562a2227450e91a32f1291368845a92c5025d04 +size 10839 diff --git a/parse/train/r1glDpNYwS/images/d2769905aff41a3c95c88243e3c51a511a1ab2ea1e29406caaf09d8ab3699e69.jpg b/parse/train/r1glDpNYwS/images/d2769905aff41a3c95c88243e3c51a511a1ab2ea1e29406caaf09d8ab3699e69.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2eedd7a8801287526cdc37f2888ec011894550b6 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/d2769905aff41a3c95c88243e3c51a511a1ab2ea1e29406caaf09d8ab3699e69.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d2d701bce318b725fddd73200a24b9f9a2c1a844828cb22afe1f856710e0a8b +size 32950 diff --git a/parse/train/r1glDpNYwS/images/d27b18d28b4fd6123cce7fb3d178e723891357a0599bc37a964b48ff63dab500.jpg b/parse/train/r1glDpNYwS/images/d27b18d28b4fd6123cce7fb3d178e723891357a0599bc37a964b48ff63dab500.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e568aeac905c999fd6d982a80c4cf48dcf8a38fc --- /dev/null +++ b/parse/train/r1glDpNYwS/images/d27b18d28b4fd6123cce7fb3d178e723891357a0599bc37a964b48ff63dab500.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d07d386c7ee392f8be74f1e1e7a5c61ab1c4d367e4ccc704d056a6ac24645f +size 53568 diff --git a/parse/train/r1glDpNYwS/images/e5922eb303b62b2ded923d7c28b2a4746f6e1fb1a8e5e62c4a780a35f97110db.jpg b/parse/train/r1glDpNYwS/images/e5922eb303b62b2ded923d7c28b2a4746f6e1fb1a8e5e62c4a780a35f97110db.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a5d56098ead03e261883a2f666ee2278ef4dd85e --- /dev/null +++ b/parse/train/r1glDpNYwS/images/e5922eb303b62b2ded923d7c28b2a4746f6e1fb1a8e5e62c4a780a35f97110db.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4bcba15526493c3d32c05cddde7d29004736fa7bb36372d50f577e17c5fb4082 +size 42985 diff --git a/parse/train/r1glDpNYwS/images/e8f507b3f8da581fc1158f84ba28bf096f1f3bef694348e8e264e0fb9e3821df.jpg b/parse/train/r1glDpNYwS/images/e8f507b3f8da581fc1158f84ba28bf096f1f3bef694348e8e264e0fb9e3821df.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d01cc7794042b13b5b39fb24a640d064a0425172 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/e8f507b3f8da581fc1158f84ba28bf096f1f3bef694348e8e264e0fb9e3821df.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1eda0754de70d43d5b6f1ee4689333642cd92dec475126a394a4afc826d2a55f +size 34205 diff --git a/parse/train/r1glDpNYwS/images/eccefb207082d4ff210ccb586ffbb616b97e0347e3d512eb9ef7f7ccd18ebd5c.jpg b/parse/train/r1glDpNYwS/images/eccefb207082d4ff210ccb586ffbb616b97e0347e3d512eb9ef7f7ccd18ebd5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ef835980c553d52ce936ba5e4ff1d3376fcb6c74 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/eccefb207082d4ff210ccb586ffbb616b97e0347e3d512eb9ef7f7ccd18ebd5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c856a47e166a6ce265241ab180005b2658404765734842c156f26f0e242e0ec6 +size 70013 diff --git a/parse/train/r1glDpNYwS/images/ed9e42db6e82a9808f050163a4330bc20b03303d3bf3ddc677dfb578e01c720e.jpg b/parse/train/r1glDpNYwS/images/ed9e42db6e82a9808f050163a4330bc20b03303d3bf3ddc677dfb578e01c720e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1ce3b4af547bb0769ccfb44e57e0e0bbf9779029 --- /dev/null +++ b/parse/train/r1glDpNYwS/images/ed9e42db6e82a9808f050163a4330bc20b03303d3bf3ddc677dfb578e01c720e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30cd5196da0df3681e9ffc6f578a9ee27e69593d605dfbd0e8254fe3753eaae9 +size 28953 diff --git a/parse/train/rJXMpikCZ/images/4e1ae4efc4fa983421833163e22f476529ef6f08daa6b89ef7efc0ec43f17f47.jpg b/parse/train/rJXMpikCZ/images/4e1ae4efc4fa983421833163e22f476529ef6f08daa6b89ef7efc0ec43f17f47.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6be65bb4335f5ec1a5537664bbef1c3f7cc5b07b --- /dev/null +++ b/parse/train/rJXMpikCZ/images/4e1ae4efc4fa983421833163e22f476529ef6f08daa6b89ef7efc0ec43f17f47.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b091bcaf4cc9a708eb42f5548074a1995dcf0c3b28993c1d5e9895dec21dbaf +size 104059 diff --git a/parse/train/rJXMpikCZ/images/5bc182b8ae8e0b035f6e25f92206ba03d4860eac1f4a475b8a8a70717759ee38.jpg b/parse/train/rJXMpikCZ/images/5bc182b8ae8e0b035f6e25f92206ba03d4860eac1f4a475b8a8a70717759ee38.jpg new file mode 100644 index 0000000000000000000000000000000000000000..43943aabd80b8060effb17c15450c101003bf79b --- /dev/null +++ b/parse/train/rJXMpikCZ/images/5bc182b8ae8e0b035f6e25f92206ba03d4860eac1f4a475b8a8a70717759ee38.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f31160fa9a2c2c3d8eda8738807400935954e793c60a9dbf6bc07031c5c735f4 +size 46716 diff --git a/parse/train/rJXMpikCZ/images/674efbb5f0c9fa0204dd41bac98bacc949febada53e8c72dbb47413271e09319.jpg b/parse/train/rJXMpikCZ/images/674efbb5f0c9fa0204dd41bac98bacc949febada53e8c72dbb47413271e09319.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5aba16a5d4be8e67e208a027df8fa97aa2147a6 --- /dev/null +++ b/parse/train/rJXMpikCZ/images/674efbb5f0c9fa0204dd41bac98bacc949febada53e8c72dbb47413271e09319.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c1a0c51b92cd76f18189d6d43d7209aa4d6da81fc13e77db89a9143759d128e +size 6995 diff --git a/parse/train/rJXMpikCZ/images/6934f8356d262e1827d82ac83ae643ea2001976c4f7f8bafababb3ded84909fc.jpg b/parse/train/rJXMpikCZ/images/6934f8356d262e1827d82ac83ae643ea2001976c4f7f8bafababb3ded84909fc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d76bf7b7724718d81e5773a615abd3e50b735d26 --- /dev/null +++ b/parse/train/rJXMpikCZ/images/6934f8356d262e1827d82ac83ae643ea2001976c4f7f8bafababb3ded84909fc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:637413df5911f2b649ede4404c67faba9bd60b1f6af00fbcd82f8b277c4f180e +size 14286 diff --git a/parse/train/rJXMpikCZ/images/8a0e7b2990b0eac03323566fc8ca498b8fc96233394aafe28e02924d02a52ebf.jpg b/parse/train/rJXMpikCZ/images/8a0e7b2990b0eac03323566fc8ca498b8fc96233394aafe28e02924d02a52ebf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8cfa087c8caa484673eba402177d5baac6ee2c35 --- /dev/null +++ b/parse/train/rJXMpikCZ/images/8a0e7b2990b0eac03323566fc8ca498b8fc96233394aafe28e02924d02a52ebf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e450798df59e0fc6927a6690e853ce14ac2dcf3e2fd7762e10c524c7f35a47d5 +size 58257 diff --git a/parse/train/rJXMpikCZ/images/8fa08bdead52222b758f78595ffd6765b51151b0e7707b243fec0b6f696de82c.jpg b/parse/train/rJXMpikCZ/images/8fa08bdead52222b758f78595ffd6765b51151b0e7707b243fec0b6f696de82c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ac7ff36cc5c41f8d2b05182655a4e0138696d1f2 --- /dev/null +++ b/parse/train/rJXMpikCZ/images/8fa08bdead52222b758f78595ffd6765b51151b0e7707b243fec0b6f696de82c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed0e5cf2533cc484f89121fd78df2c47f54c318ff53bdcd45b1415c4fd405b80 +size 6715 diff --git a/parse/train/rJXMpikCZ/images/9077b7ceec008880953f013e1bcb82af80813a9eba73948be954511aebe65889.jpg b/parse/train/rJXMpikCZ/images/9077b7ceec008880953f013e1bcb82af80813a9eba73948be954511aebe65889.jpg new file mode 100644 index 0000000000000000000000000000000000000000..abb1b6671a3d5fb5f312356b83cd1cdfbcc33aa6 --- /dev/null +++ b/parse/train/rJXMpikCZ/images/9077b7ceec008880953f013e1bcb82af80813a9eba73948be954511aebe65889.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1586bc6bee79ede787a4c0cc40f5f3d4bb8f12c810e1b8b73d1d0322ed4ab614 +size 4728 diff --git a/parse/train/rJXMpikCZ/images/ac453ededadace9124fc25c7858b7bc55c1a124a40e16ff4b71dd9f8c10113ce.jpg b/parse/train/rJXMpikCZ/images/ac453ededadace9124fc25c7858b7bc55c1a124a40e16ff4b71dd9f8c10113ce.jpg new file mode 100644 index 0000000000000000000000000000000000000000..77d879247bf9f6334c1e67c841633048ebf68ddb --- /dev/null +++ b/parse/train/rJXMpikCZ/images/ac453ededadace9124fc25c7858b7bc55c1a124a40e16ff4b71dd9f8c10113ce.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a85f1df13875b44e1af8d7d4da69c282f752aadbc977d0d91c2c97c6433d27e5 +size 87005 diff --git a/parse/train/rJXMpikCZ/images/cb9bc7121143ec6fa00ac7138066b16cc3d845ea6bb4f9546e9ea5d3d7d8b0d4.jpg b/parse/train/rJXMpikCZ/images/cb9bc7121143ec6fa00ac7138066b16cc3d845ea6bb4f9546e9ea5d3d7d8b0d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a657c4bc482224abca390395ae5e1913d8f95305 --- /dev/null +++ b/parse/train/rJXMpikCZ/images/cb9bc7121143ec6fa00ac7138066b16cc3d845ea6bb4f9546e9ea5d3d7d8b0d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e929ccfe1bfcdd13179258ef4ee7e3cba183e228dc4d2f823809ccbcfc19828 +size 5864 diff --git a/parse/train/rJXMpikCZ/images/cc698b5207998b2ef3e1a12cc280894adc295767dfef74602cdacd2b755507a6.jpg b/parse/train/rJXMpikCZ/images/cc698b5207998b2ef3e1a12cc280894adc295767dfef74602cdacd2b755507a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..01e8c02978cfe9e739ff474c75f3ab545a940281 --- /dev/null +++ b/parse/train/rJXMpikCZ/images/cc698b5207998b2ef3e1a12cc280894adc295767dfef74602cdacd2b755507a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:991388444124f585585e22b69fed17728b0ed2118286b1dabc524c37c39deb0c +size 53426 diff --git a/parse/train/rJXMpikCZ/images/f386200c88e7dece8ce2bf48bbdc5757fa1ec30f02a3cf78b17c606fb2e76683.jpg b/parse/train/rJXMpikCZ/images/f386200c88e7dece8ce2bf48bbdc5757fa1ec30f02a3cf78b17c606fb2e76683.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9c48c23b9709490df24380072df5b6a477cfadd4 --- /dev/null +++ b/parse/train/rJXMpikCZ/images/f386200c88e7dece8ce2bf48bbdc5757fa1ec30f02a3cf78b17c606fb2e76683.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:463faa7acd750277221aafe0026b443dc8ceac6f17503dd001f662f7265840b5 +size 2984 diff --git a/parse/train/rJeKjwvclx/images/085444cc5b6dcb9a4ec51b52f4a781f31f790443b465e8a01b0e5d0fd4bca99f.jpg b/parse/train/rJeKjwvclx/images/085444cc5b6dcb9a4ec51b52f4a781f31f790443b465e8a01b0e5d0fd4bca99f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b8fd156ddbce60b56c71696775900cf489094bf2 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/085444cc5b6dcb9a4ec51b52f4a781f31f790443b465e8a01b0e5d0fd4bca99f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39ede5d17fd81eae0c4fd68d075e0fedfca5529c02cab928898644cca4786774 +size 54460 diff --git a/parse/train/rJeKjwvclx/images/0b9612fdcd62bbde420d3294715e2b488ff20e952a7bac9155fd001efebd85ca.jpg b/parse/train/rJeKjwvclx/images/0b9612fdcd62bbde420d3294715e2b488ff20e952a7bac9155fd001efebd85ca.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f9a750d7053f7c1694bf97886f4b416b0e2ea6c0 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/0b9612fdcd62bbde420d3294715e2b488ff20e952a7bac9155fd001efebd85ca.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ee554dc57318a3f26a845dc0c733873d91e04a5d2f0654361d4679e2a6c3a3c +size 9011 diff --git a/parse/train/rJeKjwvclx/images/1605164114228002420daf4077ddc6f0a8eeb0996db3932f94f49047687b88a7.jpg b/parse/train/rJeKjwvclx/images/1605164114228002420daf4077ddc6f0a8eeb0996db3932f94f49047687b88a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3c1eec29188ca42335271b708f9496702baa0bbd --- /dev/null +++ b/parse/train/rJeKjwvclx/images/1605164114228002420daf4077ddc6f0a8eeb0996db3932f94f49047687b88a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32f73e9bea10dac19fe66ee6ff5bcb7dc89a1f4034a0afb8ba9e7e3958fe5dc1 +size 95710 diff --git a/parse/train/rJeKjwvclx/images/34c9b1af90d072819538f6780f37239b9e8f6a72c45b555988700cec3a94e07c.jpg b/parse/train/rJeKjwvclx/images/34c9b1af90d072819538f6780f37239b9e8f6a72c45b555988700cec3a94e07c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2e0426c972ba14b71b7ce3b8dc40233d98773fd6 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/34c9b1af90d072819538f6780f37239b9e8f6a72c45b555988700cec3a94e07c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:08dec8462e383650f3829a8d12b0a3014c2ceb114f6ecc33fa5c73817caab125 +size 33390 diff --git a/parse/train/rJeKjwvclx/images/366b7e744a65a1c479359e0000da5451a5aa3e0614250d64bb4588d807625476.jpg b/parse/train/rJeKjwvclx/images/366b7e744a65a1c479359e0000da5451a5aa3e0614250d64bb4588d807625476.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bb5e8a2569e071dee3b082c23c3110034f0c2d1a --- /dev/null +++ b/parse/train/rJeKjwvclx/images/366b7e744a65a1c479359e0000da5451a5aa3e0614250d64bb4588d807625476.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4346c5dcf3a1da7ae7e622291e62eb351f307fd0888518b490ee9d56a91397cb +size 58688 diff --git a/parse/train/rJeKjwvclx/images/3bf21e545f3b01bf0d8cf59d0bf55537d163c353404750b88f7d6d4ee5175cbe.jpg b/parse/train/rJeKjwvclx/images/3bf21e545f3b01bf0d8cf59d0bf55537d163c353404750b88f7d6d4ee5175cbe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a84eaf7c20e37218c5d1f16136a70b133f137c5f --- /dev/null +++ b/parse/train/rJeKjwvclx/images/3bf21e545f3b01bf0d8cf59d0bf55537d163c353404750b88f7d6d4ee5175cbe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d84812d7c6015271df3cd7d1e34e6462803cc1e3f601c2e93bed5f624fd197e8 +size 6584 diff --git a/parse/train/rJeKjwvclx/images/587fa0e39bc3021fe61bff9f3565c68bc05181a466d0bc36dfea698f3988018c.jpg b/parse/train/rJeKjwvclx/images/587fa0e39bc3021fe61bff9f3565c68bc05181a466d0bc36dfea698f3988018c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..47b1bc6549643952772ff2265ececb3511633259 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/587fa0e39bc3021fe61bff9f3565c68bc05181a466d0bc36dfea698f3988018c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd327abf1175c23b5b17f7e9046aa991a1cca19b3615a0e525ebcf12319154ee +size 45413 diff --git a/parse/train/rJeKjwvclx/images/653f0cc5f5b4e69dc4ebc621a1c337abbcdd7c24b52621540fafc0ac3422a7ea.jpg b/parse/train/rJeKjwvclx/images/653f0cc5f5b4e69dc4ebc621a1c337abbcdd7c24b52621540fafc0ac3422a7ea.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ee3f8636e01b217780db6254224020596000e543 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/653f0cc5f5b4e69dc4ebc621a1c337abbcdd7c24b52621540fafc0ac3422a7ea.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6aaa1a0999868064307feb76dd4f46f23bbb4b88ee5dd7dd061a817f58f2b6c +size 18424 diff --git a/parse/train/rJeKjwvclx/images/659657d3664439f5aaf2ad20059c8aff9fa095ee215405eee86563f70a315687.jpg b/parse/train/rJeKjwvclx/images/659657d3664439f5aaf2ad20059c8aff9fa095ee215405eee86563f70a315687.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1ccde5c100db02c965ca0c130d858d34521c9be2 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/659657d3664439f5aaf2ad20059c8aff9fa095ee215405eee86563f70a315687.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a9ad57d872928597c2a4eeefdc4de388134d363d9c0a0120561dfe78a5f8d35e +size 63995 diff --git a/parse/train/rJeKjwvclx/images/6b17096e746bf2a05c497268ea80bc69a1959ba4479008235c346ae565e33292.jpg b/parse/train/rJeKjwvclx/images/6b17096e746bf2a05c497268ea80bc69a1959ba4479008235c346ae565e33292.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6bd2e7b9ff712ab4f94863dad89e9136ff9d19e6 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/6b17096e746bf2a05c497268ea80bc69a1959ba4479008235c346ae565e33292.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2735159d14215e58727e5d52dbb54e8999a4cc92764d03b79fe749db0dcc41d3 +size 44951 diff --git a/parse/train/rJeKjwvclx/images/99a7b0e24a018c83940362ae014b6d290ee2bbd0ebfc0b5879e6f5bb0a803486.jpg b/parse/train/rJeKjwvclx/images/99a7b0e24a018c83940362ae014b6d290ee2bbd0ebfc0b5879e6f5bb0a803486.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a32a732948b34517b154c6360f4ab3aa01d18a2b --- /dev/null +++ b/parse/train/rJeKjwvclx/images/99a7b0e24a018c83940362ae014b6d290ee2bbd0ebfc0b5879e6f5bb0a803486.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02160b17b3745ee58fd84534705bd1f7f8a517c04b1acd1c1d9e63a4af5ee861 +size 20909 diff --git a/parse/train/rJeKjwvclx/images/a845dc12bdf8267e6cba84587670144160e05035c8b5b3713d05e247bc5ea6f6.jpg b/parse/train/rJeKjwvclx/images/a845dc12bdf8267e6cba84587670144160e05035c8b5b3713d05e247bc5ea6f6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e8abcdc21e1a1f8bc9ea2dd242a8828d1eded43e --- /dev/null +++ b/parse/train/rJeKjwvclx/images/a845dc12bdf8267e6cba84587670144160e05035c8b5b3713d05e247bc5ea6f6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9e0bf229f091fcb79d95f3951e4a811ee9fb44160ed683c1b9b3e0386488e6f +size 59766 diff --git a/parse/train/rJeKjwvclx/images/b70446b0603fb23ef8d132a24326f397cb6862b4b6496cb265c16224b89e171f.jpg b/parse/train/rJeKjwvclx/images/b70446b0603fb23ef8d132a24326f397cb6862b4b6496cb265c16224b89e171f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ddc337c45c0f14b13fe7a97e31d72cf3fe041e93 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/b70446b0603fb23ef8d132a24326f397cb6862b4b6496cb265c16224b89e171f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e48ccc391d2e2a07dfbaea5512f19fdadd1005824abec6f5ce8848344ec8bef9 +size 5630 diff --git a/parse/train/rJeKjwvclx/images/c3102bd8496d3d4cf1baf2bcf5fff89efedb6e23756c3ea7ee32a876b2887df7.jpg b/parse/train/rJeKjwvclx/images/c3102bd8496d3d4cf1baf2bcf5fff89efedb6e23756c3ea7ee32a876b2887df7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cffbc3a74464ddeb65fef98fbf4dc8b9dc1a715e --- /dev/null +++ b/parse/train/rJeKjwvclx/images/c3102bd8496d3d4cf1baf2bcf5fff89efedb6e23756c3ea7ee32a876b2887df7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18be6c82fd343642233cdde933c4b851466a73c6ad0658bd32cfb0fadd9eb96f +size 50235 diff --git a/parse/train/rJeKjwvclx/images/d5fce1c08850a8d2b2699f50725885b800b1aaecba9142ff5956d772638d5843.jpg b/parse/train/rJeKjwvclx/images/d5fce1c08850a8d2b2699f50725885b800b1aaecba9142ff5956d772638d5843.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1a42dbb18a8c1047f5d91be0db3a106802922463 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/d5fce1c08850a8d2b2699f50725885b800b1aaecba9142ff5956d772638d5843.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d2b0f678d40de20148fac5a19cf09ca68225446aa0d2a085bdd0c9bf67d2e3f +size 4191 diff --git a/parse/train/rJeKjwvclx/images/d8c0c6d7af28df7f8be0568d155ae503bad0f6993110863dc8ce1aed3014439b.jpg b/parse/train/rJeKjwvclx/images/d8c0c6d7af28df7f8be0568d155ae503bad0f6993110863dc8ce1aed3014439b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..471f2a8b418dd49f5c487b2b9f5baeef8a9d8752 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/d8c0c6d7af28df7f8be0568d155ae503bad0f6993110863dc8ce1aed3014439b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8a60bac9fcc0c89d3979cde0bc0d06e6b35ac6c7756703f1b8053c2ad9d9e2b5 +size 44335 diff --git a/parse/train/rJeKjwvclx/images/deb287d5c17e8226321ff80afd968fc83ca7761f3b9ba1ba78d6395364b0c112.jpg b/parse/train/rJeKjwvclx/images/deb287d5c17e8226321ff80afd968fc83ca7761f3b9ba1ba78d6395364b0c112.jpg new file mode 100644 index 0000000000000000000000000000000000000000..97329a3aa06d56cc937f99743c6e9acc27f1ab7a --- /dev/null +++ b/parse/train/rJeKjwvclx/images/deb287d5c17e8226321ff80afd968fc83ca7761f3b9ba1ba78d6395364b0c112.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f327b51007ec2b1a4723f6cf40734be1f094083ee0d0dcf9932f046ede24024e +size 4411 diff --git a/parse/train/rJeKjwvclx/images/e7fc356977edae75c16da08a71131bc1fe53766fbace45f59c4cafbd9da39dec.jpg b/parse/train/rJeKjwvclx/images/e7fc356977edae75c16da08a71131bc1fe53766fbace45f59c4cafbd9da39dec.jpg new file mode 100644 index 0000000000000000000000000000000000000000..134b53ca11e5f2343deeeae349058628768aa751 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/e7fc356977edae75c16da08a71131bc1fe53766fbace45f59c4cafbd9da39dec.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60714cb3c1931f893e1c004f00dbd9c8f82026a54f3d17c88cdc51d719273c6f +size 4283 diff --git a/parse/train/rJeKjwvclx/images/eabde3ef2b7d5396db8ce04464bea87500de6a6fae67c536112caa5e175ec1a2.jpg b/parse/train/rJeKjwvclx/images/eabde3ef2b7d5396db8ce04464bea87500de6a6fae67c536112caa5e175ec1a2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c85f7a34d1bf87416aac9d8256718bbb52ebb392 --- /dev/null +++ b/parse/train/rJeKjwvclx/images/eabde3ef2b7d5396db8ce04464bea87500de6a6fae67c536112caa5e175ec1a2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd9df99975c937c0f3f85da38cf4a10fc7b4c66305b7e23834c75d6781f5ddba +size 3366 diff --git a/parse/train/rkeNr6EKwB/images/03f356c7b26e7b3fdbd799fce775fd9415fb06a1799e302d610460f80363413c.jpg b/parse/train/rkeNr6EKwB/images/03f356c7b26e7b3fdbd799fce775fd9415fb06a1799e302d610460f80363413c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a1012cb10394f63166c41dda2f7afe011c619a46 --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/03f356c7b26e7b3fdbd799fce775fd9415fb06a1799e302d610460f80363413c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10130f5b3401a1437ec67356f59f8ff098a0c6c0f3479748752032dbacbe0558 +size 12182 diff --git a/parse/train/rkeNr6EKwB/images/43845d46be9e83e276fa02177b337479092efdd1c5bc8ea3b448e6c1a503e4e3.jpg b/parse/train/rkeNr6EKwB/images/43845d46be9e83e276fa02177b337479092efdd1c5bc8ea3b448e6c1a503e4e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..235b5fe1ceb63b8f6707de4961f49d389f04b1ad --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/43845d46be9e83e276fa02177b337479092efdd1c5bc8ea3b448e6c1a503e4e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:35ce19290054f09d1482f055f232f16a3b0ee683ff46ad81014b5f57454b6085 +size 5337 diff --git a/parse/train/rkeNr6EKwB/images/654098d91d54f346d64e70076fe78a52514a9f50fc58eaa1074b32a7676ba97c.jpg b/parse/train/rkeNr6EKwB/images/654098d91d54f346d64e70076fe78a52514a9f50fc58eaa1074b32a7676ba97c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0ad7792b2c6b9eb17fad205eefa5666d9a277e2e --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/654098d91d54f346d64e70076fe78a52514a9f50fc58eaa1074b32a7676ba97c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dfd89f34020468e79fa3f6e5c1e5e7617f74815dd032d8881ddcd2ab50c524d9 +size 14912 diff --git a/parse/train/rkeNr6EKwB/images/6723862e3ce211a5ed6d237cb781859c07aece0d15080ceeecc5a75ffb7172a4.jpg b/parse/train/rkeNr6EKwB/images/6723862e3ce211a5ed6d237cb781859c07aece0d15080ceeecc5a75ffb7172a4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..31ad729b71a45a9550fc21e38611d1ed55a8b7c7 --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/6723862e3ce211a5ed6d237cb781859c07aece0d15080ceeecc5a75ffb7172a4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7fa7f51dab27ed56b6bc0ba3321d61466ca89ea86aa94dcd0189d1218d6d3b1 +size 65472 diff --git a/parse/train/rkeNr6EKwB/images/7fe952d0a25b3b48f8e077f1bb27cff09c3989daf3242a0fcbddc30c9464fb0f.jpg b/parse/train/rkeNr6EKwB/images/7fe952d0a25b3b48f8e077f1bb27cff09c3989daf3242a0fcbddc30c9464fb0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f430b69607ccfc750f187a640e34cd2d1c4d4011 --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/7fe952d0a25b3b48f8e077f1bb27cff09c3989daf3242a0fcbddc30c9464fb0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce900c54e0b946e80b9cf6501a982d413b536d156b948983890c43faa6609ec0 +size 6440 diff --git a/parse/train/rkeNr6EKwB/images/95ea4926458d79dcc3182d068d7515dda3e7aca61eb690b2a310d456a9eaca7d.jpg b/parse/train/rkeNr6EKwB/images/95ea4926458d79dcc3182d068d7515dda3e7aca61eb690b2a310d456a9eaca7d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eadeba19c68bc221932c83c935b41256ccef6a1c --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/95ea4926458d79dcc3182d068d7515dda3e7aca61eb690b2a310d456a9eaca7d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e5c26ed90c93a3aed56a259aef922f4ca4f259b72a91e56cd6961e49ebd7943 +size 63090 diff --git a/parse/train/rkeNr6EKwB/images/a17111d2429995edacb1aced7a6c94d4c93194a316255a5ab96d7ee96a898235.jpg b/parse/train/rkeNr6EKwB/images/a17111d2429995edacb1aced7a6c94d4c93194a316255a5ab96d7ee96a898235.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2fc221074984a495a04f644ade77b159de4dfc30 --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/a17111d2429995edacb1aced7a6c94d4c93194a316255a5ab96d7ee96a898235.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:367bec4647198be1dfdbb3962e058458264781a08c7489211ffbbe10022fbf16 +size 39839 diff --git a/parse/train/rkeNr6EKwB/images/ad2f309ee6a58347d2aec12f4b19e6039e83bc6805f2abed22407b7af2449013.jpg b/parse/train/rkeNr6EKwB/images/ad2f309ee6a58347d2aec12f4b19e6039e83bc6805f2abed22407b7af2449013.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2ddd558c279fde34b43e34339ded4ba9ad52f480 --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/ad2f309ee6a58347d2aec12f4b19e6039e83bc6805f2abed22407b7af2449013.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:97c0f7832826184e976e36d0d316d98a7ead026b3e6d81431595dc4b8702c382 +size 29552 diff --git a/parse/train/rkeNr6EKwB/images/ca841bef8d4a9150762267bb7e3857a7fd3635f97e983763dd4ed3f06202b444.jpg b/parse/train/rkeNr6EKwB/images/ca841bef8d4a9150762267bb7e3857a7fd3635f97e983763dd4ed3f06202b444.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e8f5b5ccfbe015d416e3cce937051a080852748a --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/ca841bef8d4a9150762267bb7e3857a7fd3635f97e983763dd4ed3f06202b444.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0220a6733f8134d98265d70522fa73cd2afbf0d0adbb5a74cfd87c8760a6f325 +size 23023 diff --git a/parse/train/rkeNr6EKwB/images/d37fe180f85d277a81fad5f24517a0162ea4b8d8509bd8fdd1a2cfce5052d905.jpg b/parse/train/rkeNr6EKwB/images/d37fe180f85d277a81fad5f24517a0162ea4b8d8509bd8fdd1a2cfce5052d905.jpg new file mode 100644 index 0000000000000000000000000000000000000000..578c4afad0c71b8059c23076566d41d5fb979432 --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/d37fe180f85d277a81fad5f24517a0162ea4b8d8509bd8fdd1a2cfce5052d905.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d0e333482247e7c6213d3be2871757c431882c97e63a748ccda7d86db90aca4 +size 37857 diff --git a/parse/train/rkeNr6EKwB/images/f296c8040a9a863d05aadc00bc2a470a737436e05829457957ccbd87a1e44506.jpg b/parse/train/rkeNr6EKwB/images/f296c8040a9a863d05aadc00bc2a470a737436e05829457957ccbd87a1e44506.jpg new file mode 100644 index 0000000000000000000000000000000000000000..834d40f9211720bc16b2ba9f83d48a663088ed97 --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/f296c8040a9a863d05aadc00bc2a470a737436e05829457957ccbd87a1e44506.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e989bb98c542bf10e025ea0bee30c1a813b2b8a3a6ebea8da23f46c535420a60 +size 4924 diff --git a/parse/train/rkeNr6EKwB/images/fe9f1c1142dbd2e51ae5a5c8e39a0ec2e049b8ba71033d4b789b5b7bf71f87ad.jpg b/parse/train/rkeNr6EKwB/images/fe9f1c1142dbd2e51ae5a5c8e39a0ec2e049b8ba71033d4b789b5b7bf71f87ad.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a31592ad2b1e8b1a1f7477ef3303f6c9e9453c38 --- /dev/null +++ b/parse/train/rkeNr6EKwB/images/fe9f1c1142dbd2e51ae5a5c8e39a0ec2e049b8ba71033d4b789b5b7bf71f87ad.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d960d5924d54521e195bd9987d674e29adcdf618759c6bae0a49fbe768496d4d +size 22662 diff --git a/parse/train/rkmu5b0a-/images/05535a2a0d80602a3905afadd979daec66b9730ff0f098476a3f2709bb208fb6.jpg b/parse/train/rkmu5b0a-/images/05535a2a0d80602a3905afadd979daec66b9730ff0f098476a3f2709bb208fb6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..66fb2118c4fa39a84dc0b2fb403026eefd4c71e6 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/05535a2a0d80602a3905afadd979daec66b9730ff0f098476a3f2709bb208fb6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27a7eb992075f2e2b7193a9cb1fa8ba345ff2f14418939ab07f7887af9a5137d +size 37929 diff --git a/parse/train/rkmu5b0a-/images/0b5220560c4b7a2faaf9c9833d712c8a3d06b9889f1cc119cb4eded6b8e5409b.jpg b/parse/train/rkmu5b0a-/images/0b5220560c4b7a2faaf9c9833d712c8a3d06b9889f1cc119cb4eded6b8e5409b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1f5b3bc5f1cf9ea2efd4f3f38c415e5ae65cb8a1 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/0b5220560c4b7a2faaf9c9833d712c8a3d06b9889f1cc119cb4eded6b8e5409b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:257d0301f159d75739afc1c319d7d6df64a8e893efaaceab07688e67fa52e633 +size 153198 diff --git a/parse/train/rkmu5b0a-/images/0d88dae6b95f4d5fb0da1002ed38ae8f6a6b573cfc71d6e0cb22cbbf4bf2778e.jpg b/parse/train/rkmu5b0a-/images/0d88dae6b95f4d5fb0da1002ed38ae8f6a6b573cfc71d6e0cb22cbbf4bf2778e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b52e064c7d5329833ed9fc774ec31fda798f734 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/0d88dae6b95f4d5fb0da1002ed38ae8f6a6b573cfc71d6e0cb22cbbf4bf2778e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e2d05f22c93bca6901ecb4302d850cee9be4115680466902ba5433d32536a17 +size 4971 diff --git a/parse/train/rkmu5b0a-/images/135b0de476c01a8945a13ece936d42f94c6828b22922898b337ca4875eb7b91b.jpg b/parse/train/rkmu5b0a-/images/135b0de476c01a8945a13ece936d42f94c6828b22922898b337ca4875eb7b91b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c9ea0c65aa5a0410fdd42432100458f1b2ba135b --- /dev/null +++ b/parse/train/rkmu5b0a-/images/135b0de476c01a8945a13ece936d42f94c6828b22922898b337ca4875eb7b91b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c7cf82ab7c162edcff5b63eb015e79f1b5e883db5dd66924d193a625998a1787 +size 95095 diff --git a/parse/train/rkmu5b0a-/images/1a3e49269c9cf25fc180cbe2dee6813024986959838ae2f14739090abbbbe603.jpg b/parse/train/rkmu5b0a-/images/1a3e49269c9cf25fc180cbe2dee6813024986959838ae2f14739090abbbbe603.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4095e6c2413be6e26bb662fa8325dbcaeccb0da9 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/1a3e49269c9cf25fc180cbe2dee6813024986959838ae2f14739090abbbbe603.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a74ee9f2e93322c06cb8e351f004b929ebfabfc0e527e90c67faf699d52201ad +size 82580 diff --git a/parse/train/rkmu5b0a-/images/1fbf4332ca5e5e554e8de5e3a9f4bcca72886e552f01251a01d8b87407091e3b.jpg b/parse/train/rkmu5b0a-/images/1fbf4332ca5e5e554e8de5e3a9f4bcca72886e552f01251a01d8b87407091e3b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e9254581b82287375de02f7b26ad58685bf26351 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/1fbf4332ca5e5e554e8de5e3a9f4bcca72886e552f01251a01d8b87407091e3b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6efc2c663a0bd959fcf02f2d338f21b57f70fa62d98c634d33131b106ef3cf8a +size 21742 diff --git a/parse/train/rkmu5b0a-/images/3a29b11b408ca616603d319a97647b1ca629d9e63d041dc81fd41e1d5bd1e824.jpg b/parse/train/rkmu5b0a-/images/3a29b11b408ca616603d319a97647b1ca629d9e63d041dc81fd41e1d5bd1e824.jpg new file mode 100644 index 0000000000000000000000000000000000000000..167a8ee6edf80301bba996c3e067b68049096d1f --- /dev/null +++ b/parse/train/rkmu5b0a-/images/3a29b11b408ca616603d319a97647b1ca629d9e63d041dc81fd41e1d5bd1e824.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b59e6918635d66c4f2a598b5ae2213892880517319865fbb3cdeca0649d1d8b6 +size 9650 diff --git a/parse/train/rkmu5b0a-/images/3b7fa4fe0374550c5b76f7ac0e6824bd2941dffbca2e50e253c91e4f3b6bf8eb.jpg b/parse/train/rkmu5b0a-/images/3b7fa4fe0374550c5b76f7ac0e6824bd2941dffbca2e50e253c91e4f3b6bf8eb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7733f27016dd42e5ca7697957b97082461925686 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/3b7fa4fe0374550c5b76f7ac0e6824bd2941dffbca2e50e253c91e4f3b6bf8eb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa455a0f8c2f39d7f45019ee80861cb6d410156b14bdaca3609336b0dda32c27 +size 19438 diff --git a/parse/train/rkmu5b0a-/images/3d7c59e9cec1cd75ffb2bbf56067c54426c71dbfc25b747cd49fe763b28e8377.jpg b/parse/train/rkmu5b0a-/images/3d7c59e9cec1cd75ffb2bbf56067c54426c71dbfc25b747cd49fe763b28e8377.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a0a8c37c9acfab0162e7119fe3e823fdd4f681a1 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/3d7c59e9cec1cd75ffb2bbf56067c54426c71dbfc25b747cd49fe763b28e8377.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72f12174beb4322958518f2a297f816fd96498ea7df60a48558796f940ac64d0 +size 7992 diff --git a/parse/train/rkmu5b0a-/images/4500f03984e7ade89617efcf2c2114cf3f4cb22cfc6c4ba66b7cde4fab9104e0.jpg b/parse/train/rkmu5b0a-/images/4500f03984e7ade89617efcf2c2114cf3f4cb22cfc6c4ba66b7cde4fab9104e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4406c4c6b221ea52c013ecbf87570e9fdcb82cc4 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/4500f03984e7ade89617efcf2c2114cf3f4cb22cfc6c4ba66b7cde4fab9104e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f36550648cbe84b3f51bae389478b04a351d83acd63f07e8a604fa6cc0d1216e +size 237263 diff --git a/parse/train/rkmu5b0a-/images/4a8cd986f006a429cb6207fb26babdb1211cd864f8315978bbfbb812e859607b.jpg b/parse/train/rkmu5b0a-/images/4a8cd986f006a429cb6207fb26babdb1211cd864f8315978bbfbb812e859607b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f3d0cddd2ac703566cf67035efe6699fb462575 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/4a8cd986f006a429cb6207fb26babdb1211cd864f8315978bbfbb812e859607b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76edd803114944fa8583cc0ed1a2450cf342ad618ebcf0759e8ca0f36757479a +size 9722 diff --git a/parse/train/rkmu5b0a-/images/4adf59968159c77ff04a635d2f8a3a131e58d5b11b4a5f3ba29f6da1088cd0fa.jpg b/parse/train/rkmu5b0a-/images/4adf59968159c77ff04a635d2f8a3a131e58d5b11b4a5f3ba29f6da1088cd0fa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..627b4cb8cfb1a965bc8fc19c084a4389d7e757b1 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/4adf59968159c77ff04a635d2f8a3a131e58d5b11b4a5f3ba29f6da1088cd0fa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59da54736e286f66bb566c034018fba63c0bd60cd2e51a62f832fdbfd63b0105 +size 14498 diff --git a/parse/train/rkmu5b0a-/images/4f875bda45105cdd55d0b7a3d3ac604e7f4d7f68dd769810768f6176d809ad40.jpg b/parse/train/rkmu5b0a-/images/4f875bda45105cdd55d0b7a3d3ac604e7f4d7f68dd769810768f6176d809ad40.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89481fba5396bc7efd4abfd53789b558d1465395 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/4f875bda45105cdd55d0b7a3d3ac604e7f4d7f68dd769810768f6176d809ad40.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f138675109ea0729e402354e5d109063ceacd433a4d5122cdecc4020b0aea3dc +size 5637 diff --git a/parse/train/rkmu5b0a-/images/534b4451cde315612499beb7c2c41b022ebe019336c1336b237283ef4966bc5f.jpg b/parse/train/rkmu5b0a-/images/534b4451cde315612499beb7c2c41b022ebe019336c1336b237283ef4966bc5f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..78723cfda702c3abe61c1b9d8da86b7ab1cf0def --- /dev/null +++ b/parse/train/rkmu5b0a-/images/534b4451cde315612499beb7c2c41b022ebe019336c1336b237283ef4966bc5f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e78e9b6421f6ecea145ee65c9afe3e0c4fdf1bb1b5ba19c9def2b227214bf199 +size 27255 diff --git a/parse/train/rkmu5b0a-/images/538a33cc4833f2f801e49c7ee30ab98b5c692db326775b99b2078321bf2cf5d9.jpg b/parse/train/rkmu5b0a-/images/538a33cc4833f2f801e49c7ee30ab98b5c692db326775b99b2078321bf2cf5d9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..36288892378c5b7d7b498120e03ecd9709cc2941 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/538a33cc4833f2f801e49c7ee30ab98b5c692db326775b99b2078321bf2cf5d9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2dc1fe690a5f93a03cb4da8b74dfb83d22d4370c53c9003913a14d2dd3c0938 +size 186499 diff --git a/parse/train/rkmu5b0a-/images/6734c25d8064a1216ffbf124d784be9db1191db74bdd33160fb30e730eae372c.jpg b/parse/train/rkmu5b0a-/images/6734c25d8064a1216ffbf124d784be9db1191db74bdd33160fb30e730eae372c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7e48f7be1ee54a68e29bb168f9ac9521cbcc72dc --- /dev/null +++ b/parse/train/rkmu5b0a-/images/6734c25d8064a1216ffbf124d784be9db1191db74bdd33160fb30e730eae372c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb678e83908d1fa4e6948e142761f39e7eb6cc8b29bc528139834f4257a457c2 +size 71680 diff --git a/parse/train/rkmu5b0a-/images/69bc30153a07f27dbaf69e238bd7239bea809fde0fc815d34b39cc2c1f3e67d7.jpg b/parse/train/rkmu5b0a-/images/69bc30153a07f27dbaf69e238bd7239bea809fde0fc815d34b39cc2c1f3e67d7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9d709c2919e693ae1f00f71f9c2ffb642ab70c3c --- /dev/null +++ b/parse/train/rkmu5b0a-/images/69bc30153a07f27dbaf69e238bd7239bea809fde0fc815d34b39cc2c1f3e67d7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec5a26c0ffa26e8c4b0d18451da8f3ab5a6c271aa261c22fd7928f6f863ad2ab +size 177006 diff --git a/parse/train/rkmu5b0a-/images/6c851492e9d061f1f070c98f1cecc193a21a1a4ca03c999a8c960bdbc4b0c969.jpg b/parse/train/rkmu5b0a-/images/6c851492e9d061f1f070c98f1cecc193a21a1a4ca03c999a8c960bdbc4b0c969.jpg new file mode 100644 index 0000000000000000000000000000000000000000..da4ec9a9080a1b0eb65b4d1b5ce0759f8fd773fe --- /dev/null +++ b/parse/train/rkmu5b0a-/images/6c851492e9d061f1f070c98f1cecc193a21a1a4ca03c999a8c960bdbc4b0c969.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e14b6586b8838649224c72c3f625f18052c4b10b64213c0374e8174b7fd2d895 +size 111773 diff --git a/parse/train/rkmu5b0a-/images/7241f7d8e4cdc18a8fac05c33f602f76d9b546d16e3db048b245b8f4f381dee3.jpg b/parse/train/rkmu5b0a-/images/7241f7d8e4cdc18a8fac05c33f602f76d9b546d16e3db048b245b8f4f381dee3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d1907b8656987a9d6624d7dc3ddce9dc25672e8 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/7241f7d8e4cdc18a8fac05c33f602f76d9b546d16e3db048b245b8f4f381dee3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8f780b5752cc08773b44900bc83f37ebbf132a8796fb363327bb264809f3de9 +size 79078 diff --git a/parse/train/rkmu5b0a-/images/789ec7c661ff7cc829ac9a697d0c2ad0cd625d4f08f02b26856719f27d385807.jpg b/parse/train/rkmu5b0a-/images/789ec7c661ff7cc829ac9a697d0c2ad0cd625d4f08f02b26856719f27d385807.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2d2658b5219fbe93e102fae0f7cfc1ab0d98eb2b --- /dev/null +++ b/parse/train/rkmu5b0a-/images/789ec7c661ff7cc829ac9a697d0c2ad0cd625d4f08f02b26856719f27d385807.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b416453f91a8b7b3822785b1fdb1f2ced5b923f424ff2086221ba880b0eac0ec +size 8231 diff --git a/parse/train/rkmu5b0a-/images/7a7ed6cbc25244e68d95732adeaf42c0abfcab905a530b159c8851c9f4c9cc6d.jpg b/parse/train/rkmu5b0a-/images/7a7ed6cbc25244e68d95732adeaf42c0abfcab905a530b159c8851c9f4c9cc6d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c1b18e2a4d970eb756acd057354db2af5209fe41 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/7a7ed6cbc25244e68d95732adeaf42c0abfcab905a530b159c8851c9f4c9cc6d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d53e12555d884f7642f7e1a0fa90ce917d104e2a97e6a0ca30dfb4acf2f9aa7 +size 9278 diff --git a/parse/train/rkmu5b0a-/images/7f869020ed993d5a86f68aa848e6d49de144492ebcc1255ad625ad023ec10819.jpg b/parse/train/rkmu5b0a-/images/7f869020ed993d5a86f68aa848e6d49de144492ebcc1255ad625ad023ec10819.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b5785b2302fbedd3c0d5a82b02c9008e873bf7d5 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/7f869020ed993d5a86f68aa848e6d49de144492ebcc1255ad625ad023ec10819.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16b6f3ec510527760a4da105e87db0efed17b59f0f46f67e20999106de994da0 +size 27757 diff --git a/parse/train/rkmu5b0a-/images/806a4c261a0ac438515998b79ef72c38e202dcbae24e9b0c7834de8a46e049a3.jpg b/parse/train/rkmu5b0a-/images/806a4c261a0ac438515998b79ef72c38e202dcbae24e9b0c7834de8a46e049a3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..47dd6bca3ce5b787f3199e5b3ac1fbacfbeb0412 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/806a4c261a0ac438515998b79ef72c38e202dcbae24e9b0c7834de8a46e049a3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a2f9a606d195a5c0f2718ddcb64b5605d6c26fe65f649e3362d5314da3d3be8 +size 13412 diff --git a/parse/train/rkmu5b0a-/images/860d64fd0f9cc6c984f2ad5f36f65cc2ca5d594ca1e4c66b08fa49bd3680a62e.jpg b/parse/train/rkmu5b0a-/images/860d64fd0f9cc6c984f2ad5f36f65cc2ca5d594ca1e4c66b08fa49bd3680a62e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7dafd1259ad0025e0568fe00551fb3de37e57129 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/860d64fd0f9cc6c984f2ad5f36f65cc2ca5d594ca1e4c66b08fa49bd3680a62e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a98ed8280bd30fdc4fb205d431e266dfb99fe8e32c3f7ec4906ae588c3d92ae2 +size 8524 diff --git a/parse/train/rkmu5b0a-/images/8fd9f33a0872e0a1ad404bd7549e037c161cbb1011d266be798d6f672ee9b54c.jpg b/parse/train/rkmu5b0a-/images/8fd9f33a0872e0a1ad404bd7549e037c161cbb1011d266be798d6f672ee9b54c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3bc1eac5519abb3c1b30ccce10b9b6c80b4998c7 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/8fd9f33a0872e0a1ad404bd7549e037c161cbb1011d266be798d6f672ee9b54c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5bfc6b32f5a9bc88f060b6b47f1a6dba9ac4269c1c00653a5ed2e7cb1d43a46 +size 9105 diff --git a/parse/train/rkmu5b0a-/images/97b7e29cc7e8145e9981b4e98188ccd079499c52734929a6f7383000fefb158f.jpg b/parse/train/rkmu5b0a-/images/97b7e29cc7e8145e9981b4e98188ccd079499c52734929a6f7383000fefb158f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2ccee9f101146f52678ff0a7166e7d31efbef8d1 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/97b7e29cc7e8145e9981b4e98188ccd079499c52734929a6f7383000fefb158f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0632bcb42b74b0e2a60f3089730a2223647564e38cf37ee03248dccc4623191d +size 129021 diff --git a/parse/train/rkmu5b0a-/images/9e83731144e4ec40b4e1844c16b69b0bf7ed17445078047a54dfdf6157a1c52f.jpg b/parse/train/rkmu5b0a-/images/9e83731144e4ec40b4e1844c16b69b0bf7ed17445078047a54dfdf6157a1c52f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e2ea6511d639d6038a818ce856780528b4358a51 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/9e83731144e4ec40b4e1844c16b69b0bf7ed17445078047a54dfdf6157a1c52f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b5210e7494fd8f3fdc3f05252fa52d3ee4e58d301390f83d0f1809f70458fae +size 4439 diff --git a/parse/train/rkmu5b0a-/images/a34fd1d96eb6def363c341153f36959e5874bb6bfd5c0771c1cdb54bffe02243.jpg b/parse/train/rkmu5b0a-/images/a34fd1d96eb6def363c341153f36959e5874bb6bfd5c0771c1cdb54bffe02243.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b9b9f4dfe6491ec6799b2bd1f3441405cdc800c2 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/a34fd1d96eb6def363c341153f36959e5874bb6bfd5c0771c1cdb54bffe02243.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e94cd9c8714cfb17ba625ea513f38be2d3b4e4581238b329b10793c66b95e597 +size 169749 diff --git a/parse/train/rkmu5b0a-/images/a58462a2f6751a3765dc313052c392f87d864a5e04baf3f3a5b8dd3dd4920fe0.jpg b/parse/train/rkmu5b0a-/images/a58462a2f6751a3765dc313052c392f87d864a5e04baf3f3a5b8dd3dd4920fe0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..631984a067be2ca95243f86c630b17d9707cf52c --- /dev/null +++ b/parse/train/rkmu5b0a-/images/a58462a2f6751a3765dc313052c392f87d864a5e04baf3f3a5b8dd3dd4920fe0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82ee928fd8d49362a4ed4e99f968fe3551639398b8b5bbc47a9595c08d050ba8 +size 79797 diff --git a/parse/train/rkmu5b0a-/images/a64a5177f5edcab14520f68e2e40e024cd88d0aae6aed2b59751cfd87e12e224.jpg b/parse/train/rkmu5b0a-/images/a64a5177f5edcab14520f68e2e40e024cd88d0aae6aed2b59751cfd87e12e224.jpg new file mode 100644 index 0000000000000000000000000000000000000000..641dd7ff50e85d12890051f56d387d8b8157c9b0 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/a64a5177f5edcab14520f68e2e40e024cd88d0aae6aed2b59751cfd87e12e224.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d19686b37d8c47c7cfa3785182c7a826562108fcae3715aeed6ae7788fb349ec +size 10048 diff --git a/parse/train/rkmu5b0a-/images/a8fae315a8f4350c797aa21330560e4f4c69580dce7136bcd9ae20c093921e93.jpg b/parse/train/rkmu5b0a-/images/a8fae315a8f4350c797aa21330560e4f4c69580dce7136bcd9ae20c093921e93.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9b76da0b8c0be8a1b566c5420c247df5a5425caf --- /dev/null +++ b/parse/train/rkmu5b0a-/images/a8fae315a8f4350c797aa21330560e4f4c69580dce7136bcd9ae20c093921e93.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d23e969f2bf11ca213c6702d8fe81f422757f1aab1ab0ae7c1ed273fad76f54 +size 44375 diff --git a/parse/train/rkmu5b0a-/images/a9bca4b4c8b135b0c3b822f2d3d760accc4bcd7ca8410a293b4aceb18c7564ac.jpg b/parse/train/rkmu5b0a-/images/a9bca4b4c8b135b0c3b822f2d3d760accc4bcd7ca8410a293b4aceb18c7564ac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9b4f70b80cb03b19a184c5f65e7b3aa8465a93eb --- /dev/null +++ b/parse/train/rkmu5b0a-/images/a9bca4b4c8b135b0c3b822f2d3d760accc4bcd7ca8410a293b4aceb18c7564ac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f9b96ce72370a95f01ccafd6c1696207ae68da7801a1529411f9f5720246ea9 +size 11612 diff --git a/parse/train/rkmu5b0a-/images/ad8a0ea175b73783e019e8dd90101f7b8bd443cafb375e640ac8dbb8baa54e50.jpg b/parse/train/rkmu5b0a-/images/ad8a0ea175b73783e019e8dd90101f7b8bd443cafb375e640ac8dbb8baa54e50.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0638ef0a01422aede22320f40c08acda85a7375c --- /dev/null +++ b/parse/train/rkmu5b0a-/images/ad8a0ea175b73783e019e8dd90101f7b8bd443cafb375e640ac8dbb8baa54e50.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d58df1e8819719545ff4ded3f51ec92a5813bc9099e0eddd39996f9bdfe74b81 +size 89647 diff --git a/parse/train/rkmu5b0a-/images/bf05cab90dfe6a1767bfb1575d7249154a6785f44dfbcf6e50cbdef571b4dcc7.jpg b/parse/train/rkmu5b0a-/images/bf05cab90dfe6a1767bfb1575d7249154a6785f44dfbcf6e50cbdef571b4dcc7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1bdcf1a842948b1b5d11c20199207d4d9dc8b56f --- /dev/null +++ b/parse/train/rkmu5b0a-/images/bf05cab90dfe6a1767bfb1575d7249154a6785f44dfbcf6e50cbdef571b4dcc7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85f2aea4b1f90cb9702a5e30842298f5874b34bb1d302cd33f0dda7c87379dcb +size 30446 diff --git a/parse/train/rkmu5b0a-/images/c206ea4a753a67cf5f53a2988076cd19cd7b2be0dfe03d8b4561f135ccd28bd8.jpg b/parse/train/rkmu5b0a-/images/c206ea4a753a67cf5f53a2988076cd19cd7b2be0dfe03d8b4561f135ccd28bd8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..14b3a4892555285d4d401f6a7db5ed158244d282 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/c206ea4a753a67cf5f53a2988076cd19cd7b2be0dfe03d8b4561f135ccd28bd8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e68fb934cf5573a7bebb6b7fd0ab06b7a17389416336f786f49add79fe8c453 +size 7564 diff --git a/parse/train/rkmu5b0a-/images/d269516e55ce114a9521205117db9331c17617149711f2cef4947413692db45b.jpg b/parse/train/rkmu5b0a-/images/d269516e55ce114a9521205117db9331c17617149711f2cef4947413692db45b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..805aad9cfd900d3641e7ffa428082ba87689c344 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/d269516e55ce114a9521205117db9331c17617149711f2cef4947413692db45b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87e22616126e9b084a5755368c7126252fbf55c3fd231b2691a3914d4d807c42 +size 10879 diff --git a/parse/train/rkmu5b0a-/images/d513e6508f82e24c8eb6b324fc309689a644c79c6fae895e525672068dc2d6e8.jpg b/parse/train/rkmu5b0a-/images/d513e6508f82e24c8eb6b324fc309689a644c79c6fae895e525672068dc2d6e8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..36b9e184939d0b2c7be63615e1030a509a71029a --- /dev/null +++ b/parse/train/rkmu5b0a-/images/d513e6508f82e24c8eb6b324fc309689a644c79c6fae895e525672068dc2d6e8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e79d7535d2c322a668060bfbe7d33996e0731f7e31e0d43d76feb2fd956f2c9 +size 102148 diff --git a/parse/train/rkmu5b0a-/images/d7d41c10dc13ab36b9ef700ebd062199cf5f6f1e1c5075066259648d2926eb8b.jpg b/parse/train/rkmu5b0a-/images/d7d41c10dc13ab36b9ef700ebd062199cf5f6f1e1c5075066259648d2926eb8b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..01e8623020edad188bce26802147a84711f00488 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/d7d41c10dc13ab36b9ef700ebd062199cf5f6f1e1c5075066259648d2926eb8b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:760168cdeb1056b0cc47b6b1c76449d28393fe5aa08967c9bc97b360fcb23ffc +size 26522 diff --git a/parse/train/rkmu5b0a-/images/dce53468de8960bc80f563b161a4392246d9e743ca4c228fd99e15d82fe286fa.jpg b/parse/train/rkmu5b0a-/images/dce53468de8960bc80f563b161a4392246d9e743ca4c228fd99e15d82fe286fa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6d3e0c115071bbf44475ba80ec4d2f0f21a5b71c --- /dev/null +++ b/parse/train/rkmu5b0a-/images/dce53468de8960bc80f563b161a4392246d9e743ca4c228fd99e15d82fe286fa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73add6d76c539bd3851ceb1ca44ddce5dfa04233fc96dd651d99b373c205687f +size 38097 diff --git a/parse/train/rkmu5b0a-/images/ee85b454b03e4addbad652cfcd1f40cb7e24af849d9c9bc3d6194281a28d2507.jpg b/parse/train/rkmu5b0a-/images/ee85b454b03e4addbad652cfcd1f40cb7e24af849d9c9bc3d6194281a28d2507.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d7aa0f856277ac1d84dc6a2b38aeea9c5793a631 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/ee85b454b03e4addbad652cfcd1f40cb7e24af849d9c9bc3d6194281a28d2507.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52d36167b97c23543f63ae604f27a3959d6dce03a39b1d8474bd43d60b21f55f +size 12209 diff --git a/parse/train/rkmu5b0a-/images/f9e44d74804e21226a0248f09e91c1012e0f91ae5ec34cb28fa46e22492d842b.jpg b/parse/train/rkmu5b0a-/images/f9e44d74804e21226a0248f09e91c1012e0f91ae5ec34cb28fa46e22492d842b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b72648fb4811fa2de788eadca973361c4115c024 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/f9e44d74804e21226a0248f09e91c1012e0f91ae5ec34cb28fa46e22492d842b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:36df71fadef8c039a67e2ce837e4ec5e554ed7fe3a105e63c16fd6bfc64d2fa8 +size 118221 diff --git a/parse/train/rkmu5b0a-/images/fc0d7cc9266841e39bede11278b6db77df1a74dea08d57f39e855373e6b0c1bd.jpg b/parse/train/rkmu5b0a-/images/fc0d7cc9266841e39bede11278b6db77df1a74dea08d57f39e855373e6b0c1bd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..35970f143968ac76a4ea504b9ce1151f42f89eb7 --- /dev/null +++ b/parse/train/rkmu5b0a-/images/fc0d7cc9266841e39bede11278b6db77df1a74dea08d57f39e855373e6b0c1bd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5fcb6aa8525607af204eef5589a38dd43cac46b5bfd598590052b97390ca908 +size 15839 diff --git a/parse/train/rye9LT8cee/images/07908d4c4a60884d71d85236e034944e7950ea9ea85135dfa6fdaa35629b4993.jpg b/parse/train/rye9LT8cee/images/07908d4c4a60884d71d85236e034944e7950ea9ea85135dfa6fdaa35629b4993.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a2798c65222e2b0c3e208eaea5b9ea3f76d47b8f --- /dev/null +++ b/parse/train/rye9LT8cee/images/07908d4c4a60884d71d85236e034944e7950ea9ea85135dfa6fdaa35629b4993.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5febdc87b50624a1266e04a720f0a3d88427b41d9bfca4a2447d66c9ab5919d +size 6069 diff --git a/parse/train/rye9LT8cee/images/1b9438f50f988c7e29031a72c95caf8b007169f3797e97f81005385e8193ba5c.jpg b/parse/train/rye9LT8cee/images/1b9438f50f988c7e29031a72c95caf8b007169f3797e97f81005385e8193ba5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cd8664e445d64bb1835f2f3d18c38868c91e1f5f --- /dev/null +++ b/parse/train/rye9LT8cee/images/1b9438f50f988c7e29031a72c95caf8b007169f3797e97f81005385e8193ba5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fdea1599e152fc520366627b706ee9b6918d08bb5ddd236f17cc3a2fbe2b8488 +size 75449 diff --git a/parse/train/rye9LT8cee/images/1ba787682c335d140846ebcc0f8ed66ade96dd5bcd22628cc025a7a5d50ca07d.jpg b/parse/train/rye9LT8cee/images/1ba787682c335d140846ebcc0f8ed66ade96dd5bcd22628cc025a7a5d50ca07d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..08961b56e630bd53d414efd5cf2aea61014f9006 --- /dev/null +++ b/parse/train/rye9LT8cee/images/1ba787682c335d140846ebcc0f8ed66ade96dd5bcd22628cc025a7a5d50ca07d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9bb0b450f3d7a57e57587d5b75a61c85805d88fb08a68882edd593bb95e1cec +size 38301 diff --git a/parse/train/rye9LT8cee/images/21f9cacca86800cbe60898dcaef5285412a212068ffcd3bef83ca6937190c828.jpg b/parse/train/rye9LT8cee/images/21f9cacca86800cbe60898dcaef5285412a212068ffcd3bef83ca6937190c828.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ba11a14a8402c288ddba10ec5f009215b95bb94e --- /dev/null +++ b/parse/train/rye9LT8cee/images/21f9cacca86800cbe60898dcaef5285412a212068ffcd3bef83ca6937190c828.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a3b562cd85c9f34c5432a812fdcb26da47c3d1cf586c58b668488bfc4f660c01 +size 5908 diff --git a/parse/train/rye9LT8cee/images/25ab9dbb2a25b1c9d6aca31b30fbbebcf03132dd108f4af2bd13f85ab347fb8e.jpg b/parse/train/rye9LT8cee/images/25ab9dbb2a25b1c9d6aca31b30fbbebcf03132dd108f4af2bd13f85ab347fb8e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1e371857389817d65525a95e9bda3ef0ab6766fb --- /dev/null +++ b/parse/train/rye9LT8cee/images/25ab9dbb2a25b1c9d6aca31b30fbbebcf03132dd108f4af2bd13f85ab347fb8e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37ddbf502ee3991cca2ec4f411f09fca0e87b45d96c9f01d64b5dd09c0cc87e1 +size 79626 diff --git a/parse/train/rye9LT8cee/images/2bc4bc16d787a8d97de96a28f30e9a7e08c8951f22f4cf442c2eef3101e8f0de.jpg b/parse/train/rye9LT8cee/images/2bc4bc16d787a8d97de96a28f30e9a7e08c8951f22f4cf442c2eef3101e8f0de.jpg new file mode 100644 index 0000000000000000000000000000000000000000..972948d739ad4f504d490d0723cc50b8c113f728 --- /dev/null +++ b/parse/train/rye9LT8cee/images/2bc4bc16d787a8d97de96a28f30e9a7e08c8951f22f4cf442c2eef3101e8f0de.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cdd6bd8c675ea71edc32beb229eb0aa9bc18f14a746e4b0d0743b4d9d029730 +size 7679 diff --git a/parse/train/rye9LT8cee/images/31d26842a43fae156aa79d8775b1563533b9769cd361745620104c39de42f4c7.jpg b/parse/train/rye9LT8cee/images/31d26842a43fae156aa79d8775b1563533b9769cd361745620104c39de42f4c7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..428cb68d7fd719fb4de11b056426b0ac703ff5f8 --- /dev/null +++ b/parse/train/rye9LT8cee/images/31d26842a43fae156aa79d8775b1563533b9769cd361745620104c39de42f4c7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5693141b219b64e317e76892afbbd285b8a187445fe251f7572c9550d827daeb +size 75014 diff --git a/parse/train/rye9LT8cee/images/3448f640aa7551f59a97ac98695a1cf433b2947af09721a82eab2cfc46047dec.jpg b/parse/train/rye9LT8cee/images/3448f640aa7551f59a97ac98695a1cf433b2947af09721a82eab2cfc46047dec.jpg new file mode 100644 index 0000000000000000000000000000000000000000..554110e5c87858260bc643d949e0a21e56ced474 --- /dev/null +++ b/parse/train/rye9LT8cee/images/3448f640aa7551f59a97ac98695a1cf433b2947af09721a82eab2cfc46047dec.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e371d309d038193b98c9fb796cd999afdd9962f4fea3f684758b784bac2d4105 +size 75565 diff --git a/parse/train/rye9LT8cee/images/359c7f7559b4d019f0ea90445a44be77fa58f8f7a810570ac94b2353104dd1e6.jpg b/parse/train/rye9LT8cee/images/359c7f7559b4d019f0ea90445a44be77fa58f8f7a810570ac94b2353104dd1e6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b39293029c864ca70de6fc599b77c875d161c201 --- /dev/null +++ b/parse/train/rye9LT8cee/images/359c7f7559b4d019f0ea90445a44be77fa58f8f7a810570ac94b2353104dd1e6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:206c8732ddd77fe40d15636c1fc5bec61495a2c8b15421accbb86205b51471cd +size 4397 diff --git a/parse/train/rye9LT8cee/images/39c3da3cf16fcdf41f4166139ec913ffa7cbb1514306eded77d51f37d1bb9819.jpg b/parse/train/rye9LT8cee/images/39c3da3cf16fcdf41f4166139ec913ffa7cbb1514306eded77d51f37d1bb9819.jpg new file mode 100644 index 0000000000000000000000000000000000000000..309072fbfa7878cb175489251ebe4c8e09a82c8b --- /dev/null +++ b/parse/train/rye9LT8cee/images/39c3da3cf16fcdf41f4166139ec913ffa7cbb1514306eded77d51f37d1bb9819.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64cdc8cf62cd8182b4092f5c42a57b6c88391016e0ce0aba82b35973a17de6cc +size 73921 diff --git a/parse/train/rye9LT8cee/images/3a4cae534512fc5a1a958e384bc5369500fcdfe2ce736c1f38fad47e90491f8e.jpg b/parse/train/rye9LT8cee/images/3a4cae534512fc5a1a958e384bc5369500fcdfe2ce736c1f38fad47e90491f8e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1bc99b0858d4280315ff504c274e97bee7a4a1cc --- /dev/null +++ b/parse/train/rye9LT8cee/images/3a4cae534512fc5a1a958e384bc5369500fcdfe2ce736c1f38fad47e90491f8e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8fe009c77a4a93324a58e6e9c52056f34b67e43220fa78a5b75f8356041529df +size 80284 diff --git a/parse/train/rye9LT8cee/images/3d677a5011f35eda94339f4c5314b25bae1949f329f9b661850e8ca0f75a2cc0.jpg b/parse/train/rye9LT8cee/images/3d677a5011f35eda94339f4c5314b25bae1949f329f9b661850e8ca0f75a2cc0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4e1d78441612fe2b70950b46ed6338138d4a0648 --- /dev/null +++ b/parse/train/rye9LT8cee/images/3d677a5011f35eda94339f4c5314b25bae1949f329f9b661850e8ca0f75a2cc0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c736fadd302caa3fd97cf05230b0ba8151a2643e5f40548e606a9892176ee915 +size 5548 diff --git a/parse/train/rye9LT8cee/images/4f18f4754b1828b1c31a7abf82c8a7919726ca71e9aec0a2ec68a084227d8098.jpg b/parse/train/rye9LT8cee/images/4f18f4754b1828b1c31a7abf82c8a7919726ca71e9aec0a2ec68a084227d8098.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3e27f851296fea775cb5a77257f499a4844987c2 --- /dev/null +++ b/parse/train/rye9LT8cee/images/4f18f4754b1828b1c31a7abf82c8a7919726ca71e9aec0a2ec68a084227d8098.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72ad0dd9bcfb065c4a5df4fbc0d2dbc2c6709068bd3e4ed59a1e2a7653629814 +size 102249 diff --git a/parse/train/rye9LT8cee/images/5504b155f41bfa56a3183c75a80e888be52edeee8a7acc85fbdda2ec5e985803.jpg b/parse/train/rye9LT8cee/images/5504b155f41bfa56a3183c75a80e888be52edeee8a7acc85fbdda2ec5e985803.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3b9722f53ea15f5621e130493ee6aa3c0b2f28c0 --- /dev/null +++ b/parse/train/rye9LT8cee/images/5504b155f41bfa56a3183c75a80e888be52edeee8a7acc85fbdda2ec5e985803.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1aed34582a1c88fb04e0b44bb29c6e4123d02e14277f5ed0292761c4c03971a3 +size 74780 diff --git a/parse/train/rye9LT8cee/images/56929b2fa22a486762a6b51ba622a7d64cb72b793873559465c9b20c633cc673.jpg b/parse/train/rye9LT8cee/images/56929b2fa22a486762a6b51ba622a7d64cb72b793873559465c9b20c633cc673.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7a33b17835afe4d005a5218619b60bc0d95cc31f --- /dev/null +++ b/parse/train/rye9LT8cee/images/56929b2fa22a486762a6b51ba622a7d64cb72b793873559465c9b20c633cc673.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:940d6bc5e5e097cc36b32dfe553fe8e5e0e22a9c72c1cb6b370d098230469111 +size 80345 diff --git a/parse/train/rye9LT8cee/images/5fe6573f0142c2f06af3fde8a41eea493da5531fbec69a97c7c6e453a2956241.jpg b/parse/train/rye9LT8cee/images/5fe6573f0142c2f06af3fde8a41eea493da5531fbec69a97c7c6e453a2956241.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6c6c3957976c44d3ad5c35fcd71984511b08ba7a --- /dev/null +++ b/parse/train/rye9LT8cee/images/5fe6573f0142c2f06af3fde8a41eea493da5531fbec69a97c7c6e453a2956241.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae80750b807e4dbc7c7c9b2a8754f6b55b8b2c66cf6655dbba7720bc397d91c7 +size 6578 diff --git a/parse/train/rye9LT8cee/images/63eb1e482c0c87f354a1ef3ddc10efb5af5339cbc979c890b8337f3c4ba68d23.jpg b/parse/train/rye9LT8cee/images/63eb1e482c0c87f354a1ef3ddc10efb5af5339cbc979c890b8337f3c4ba68d23.jpg new file mode 100644 index 0000000000000000000000000000000000000000..341004282fd4a9fc464e84afdca0e8eae80f7b68 --- /dev/null +++ b/parse/train/rye9LT8cee/images/63eb1e482c0c87f354a1ef3ddc10efb5af5339cbc979c890b8337f3c4ba68d23.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:275cc6a3bf7a664ecc1d2be94c6fad76e63e041202c26f77c4caa130a5acaf60 +size 7292 diff --git a/parse/train/rye9LT8cee/images/75685b49a615394f0777a1c02a8a7f34f0dfbb09552ec62000ca141a3a1c07dd.jpg b/parse/train/rye9LT8cee/images/75685b49a615394f0777a1c02a8a7f34f0dfbb09552ec62000ca141a3a1c07dd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..050c6c940c32b6d296b3eced75e6ef24e0137c00 --- /dev/null +++ b/parse/train/rye9LT8cee/images/75685b49a615394f0777a1c02a8a7f34f0dfbb09552ec62000ca141a3a1c07dd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d1890b127573a7b07a74b837186f4421d6f2e573ec6edae43162062fbfe13b38 +size 6356 diff --git a/parse/train/rye9LT8cee/images/867f9ddcd768ab95ab9fce7f437e19f1004701b1babdf357f4eabef295bb20a9.jpg b/parse/train/rye9LT8cee/images/867f9ddcd768ab95ab9fce7f437e19f1004701b1babdf357f4eabef295bb20a9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..231ec82f9662075044035b245ef9eab3c023051a --- /dev/null +++ b/parse/train/rye9LT8cee/images/867f9ddcd768ab95ab9fce7f437e19f1004701b1babdf357f4eabef295bb20a9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4b4422e230c7be2f0e716f16237cc9f6aa068e73a309b4fdc57f51a9cab3e8d +size 57581 diff --git a/parse/train/rye9LT8cee/images/97cfed648a6631a39aff6c9a119cd2b4def556f3e9ec9a52c5ebe87cff4b0f42.jpg b/parse/train/rye9LT8cee/images/97cfed648a6631a39aff6c9a119cd2b4def556f3e9ec9a52c5ebe87cff4b0f42.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1d100e6ab1441a03a70c741e03b755edd5490ec3 --- /dev/null +++ b/parse/train/rye9LT8cee/images/97cfed648a6631a39aff6c9a119cd2b4def556f3e9ec9a52c5ebe87cff4b0f42.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82dea80e90dae8b1487ac76e5e8e8e525961cb5af6cb3490a5db0d62e4664215 +size 12722 diff --git a/parse/train/rye9LT8cee/images/9ed02b9c9d557bf0efb56728387a6547f9e527798ce6862b97ed703311816ae8.jpg b/parse/train/rye9LT8cee/images/9ed02b9c9d557bf0efb56728387a6547f9e527798ce6862b97ed703311816ae8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7078b6a9294ea5271d87532b8ccdaa2b67bc86f3 --- /dev/null +++ b/parse/train/rye9LT8cee/images/9ed02b9c9d557bf0efb56728387a6547f9e527798ce6862b97ed703311816ae8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06c912a3abd3aafa85de3d7243b19f0c5f51c29969a1817e547f19b24092db49 +size 228367 diff --git a/parse/train/rye9LT8cee/images/a03ddf522a5069fc118d7904abe060d88ea9bc80ce3380abf2a32b9339444de5.jpg b/parse/train/rye9LT8cee/images/a03ddf522a5069fc118d7904abe060d88ea9bc80ce3380abf2a32b9339444de5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..258ee12cc427bca783ff392e152965c30798608b --- /dev/null +++ b/parse/train/rye9LT8cee/images/a03ddf522a5069fc118d7904abe060d88ea9bc80ce3380abf2a32b9339444de5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a77eca0f733bb439512d58d5240b2f4e49a2378a4e8b8fd8dac699da52f76792 +size 102522 diff --git a/parse/train/rye9LT8cee/images/a8c88b3ef4b68cd1367cca23ae53e2dcd8e4c6b47db36f5eb73def1a7c344f79.jpg b/parse/train/rye9LT8cee/images/a8c88b3ef4b68cd1367cca23ae53e2dcd8e4c6b47db36f5eb73def1a7c344f79.jpg new file mode 100644 index 0000000000000000000000000000000000000000..634926a9351cef9bfe4cbb1e7e05ef7d4c475e2f --- /dev/null +++ b/parse/train/rye9LT8cee/images/a8c88b3ef4b68cd1367cca23ae53e2dcd8e4c6b47db36f5eb73def1a7c344f79.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f19c3d2ccc14b6d2100bfb9e59c6b107f5b6488032dfabdb55e040c458143067 +size 102544 diff --git a/parse/train/rye9LT8cee/images/b32efdc0ae824d3e05e381f027eb6c9bf72dc3f49b79d6690084482685f14275.jpg b/parse/train/rye9LT8cee/images/b32efdc0ae824d3e05e381f027eb6c9bf72dc3f49b79d6690084482685f14275.jpg new file mode 100644 index 0000000000000000000000000000000000000000..743bcc403b7278bb07768b7729682773304fb07a --- /dev/null +++ b/parse/train/rye9LT8cee/images/b32efdc0ae824d3e05e381f027eb6c9bf72dc3f49b79d6690084482685f14275.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5338b0ce84cc2aaf71b62efe2314ffb1684bc67e406fa823e8f8dd4d5f7a950c +size 73774 diff --git a/parse/train/rye9LT8cee/images/bf5fdccded03354f1dbe17a11bf96998ee74bbb6f223eec2a26211ce9ec327cc.jpg b/parse/train/rye9LT8cee/images/bf5fdccded03354f1dbe17a11bf96998ee74bbb6f223eec2a26211ce9ec327cc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e345bbf200c1cd51fe078a555e7d267987790e04 --- /dev/null +++ b/parse/train/rye9LT8cee/images/bf5fdccded03354f1dbe17a11bf96998ee74bbb6f223eec2a26211ce9ec327cc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:132b371ea7558fcfd6b4250407587627bf20a0241837bd3fd45693d6348f6f1b +size 73835 diff --git a/parse/train/rye9LT8cee/images/d416f7464677a14320d372f279d809565e4634459f2b89dac93a37b65c882c23.jpg b/parse/train/rye9LT8cee/images/d416f7464677a14320d372f279d809565e4634459f2b89dac93a37b65c882c23.jpg new file mode 100644 index 0000000000000000000000000000000000000000..611a77dc6bf8a2786575e494a219b8febd2d14a4 --- /dev/null +++ b/parse/train/rye9LT8cee/images/d416f7464677a14320d372f279d809565e4634459f2b89dac93a37b65c882c23.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4c1baa5c92687097222515ed907528537c01b864906854231e98b08a3fff859 +size 75452 diff --git a/parse/train/rye9LT8cee/images/e0200c564e66094bf55d7b59b624bbc24bcc5d24e1579a5ccfb8eba5c530adae.jpg b/parse/train/rye9LT8cee/images/e0200c564e66094bf55d7b59b624bbc24bcc5d24e1579a5ccfb8eba5c530adae.jpg new file mode 100644 index 0000000000000000000000000000000000000000..984517f7d74d6a6d4e477b621cc03b0f85a1ade3 --- /dev/null +++ b/parse/train/rye9LT8cee/images/e0200c564e66094bf55d7b59b624bbc24bcc5d24e1579a5ccfb8eba5c530adae.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:928fdbaa465df1cb499e6fd2860e3ca88e14b2f22910d5cf24e71a12ca868aa2 +size 5496 diff --git a/parse/train/rye9LT8cee/images/ea610e6112aed77c520bb9ecd04aa3db50abb38156cda624a9a87ec2316b3d11.jpg b/parse/train/rye9LT8cee/images/ea610e6112aed77c520bb9ecd04aa3db50abb38156cda624a9a87ec2316b3d11.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8196718e5f4f4e8e00ccf56cc7707a836d24dd3f --- /dev/null +++ b/parse/train/rye9LT8cee/images/ea610e6112aed77c520bb9ecd04aa3db50abb38156cda624a9a87ec2316b3d11.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec6fec82e622167e763fda9c233a074a2af9c1a8b7c8123d60b5f7824cd93ed3 +size 35005 diff --git a/parse/train/rye9LT8cee/images/efb0084a1cbee23e0f34ce9a0c864c5dc5d59dbaf52f30405c8b81a54d3ffd50.jpg b/parse/train/rye9LT8cee/images/efb0084a1cbee23e0f34ce9a0c864c5dc5d59dbaf52f30405c8b81a54d3ffd50.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d0b130988e32152efa6eb8b8129abec0ba133e9 --- /dev/null +++ b/parse/train/rye9LT8cee/images/efb0084a1cbee23e0f34ce9a0c864c5dc5d59dbaf52f30405c8b81a54d3ffd50.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ffac29449ddfa927a9731dcd0305b8dbc40a08f2d3be8e8b8c782082ead8b23c +size 74951 diff --git a/parse/train/rye9LT8cee/images/f23e8a1026e6f5b5a4bd7ded66c617cae644b4053feb3d3043b547196cd8ebd6.jpg b/parse/train/rye9LT8cee/images/f23e8a1026e6f5b5a4bd7ded66c617cae644b4053feb3d3043b547196cd8ebd6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1c83a2c741eeb387608e51d93e44739ff72f9bc8 --- /dev/null +++ b/parse/train/rye9LT8cee/images/f23e8a1026e6f5b5a4bd7ded66c617cae644b4053feb3d3043b547196cd8ebd6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fcdf6084fe4a284e301cb9cd316dae1c14c442d8c465c97ab7ebfe60689b249c +size 42423 diff --git a/parse/train/ryghZJBKPS/images/021581459787c3b3df4ed9065c4342a96367bf511bc9af442888620fe9489f66.jpg b/parse/train/ryghZJBKPS/images/021581459787c3b3df4ed9065c4342a96367bf511bc9af442888620fe9489f66.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e36253f6a68f998174128513416b8e13c9c038d2 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/021581459787c3b3df4ed9065c4342a96367bf511bc9af442888620fe9489f66.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3acf3c12c5f26eaa0b4b96073a5ed3ee196aadb78ce6f5552b919a465a6d597 +size 152589 diff --git a/parse/train/ryghZJBKPS/images/06e0a6591dd6c4b5362a74a440c88b7e0461b3c1f9819bca861c53a4a12938ab.jpg b/parse/train/ryghZJBKPS/images/06e0a6591dd6c4b5362a74a440c88b7e0461b3c1f9819bca861c53a4a12938ab.jpg new file mode 100644 index 0000000000000000000000000000000000000000..25296ecea4165dcf8f813235b27f2811882791c0 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/06e0a6591dd6c4b5362a74a440c88b7e0461b3c1f9819bca861c53a4a12938ab.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bbc7b182641c8f9234406db8ed42f07f37102cbd30a8b7f1c9002e0880d980e +size 127998 diff --git a/parse/train/ryghZJBKPS/images/0dc915ff317f4bdacdb7591b36b371084cd4cbb9f1f0d4f07c652c08e046f8dd.jpg b/parse/train/ryghZJBKPS/images/0dc915ff317f4bdacdb7591b36b371084cd4cbb9f1f0d4f07c652c08e046f8dd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4e414b9bfe34d9c2e210ed9dd525fb32da86a9f1 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/0dc915ff317f4bdacdb7591b36b371084cd4cbb9f1f0d4f07c652c08e046f8dd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0afefa8556d90e40c6c3567b159e57b1b352f031777745606bb811318c151aa +size 58830 diff --git a/parse/train/ryghZJBKPS/images/0f700512626eabdd7ba390eaa418d36818628e85542d87e27533dd250077d27a.jpg b/parse/train/ryghZJBKPS/images/0f700512626eabdd7ba390eaa418d36818628e85542d87e27533dd250077d27a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e138fe83769a62d00f7fc06211f9ebd246fccff9 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/0f700512626eabdd7ba390eaa418d36818628e85542d87e27533dd250077d27a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6189c53f94ffd697868e8868f830829aa3c516e1a8cb54cc13a7110bcf389ed +size 61117 diff --git a/parse/train/ryghZJBKPS/images/1500a84f46227f17a6b03c0a03417a976729932c9d4bd0b0846c341096d3bce2.jpg b/parse/train/ryghZJBKPS/images/1500a84f46227f17a6b03c0a03417a976729932c9d4bd0b0846c341096d3bce2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a6cef462b6823b44d3dd341741945418b7d60b74 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/1500a84f46227f17a6b03c0a03417a976729932c9d4bd0b0846c341096d3bce2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0555df4bc4e2b8774f05b3ac19742cfb1895bcd56ada5e35105384b2f04a5502 +size 156920 diff --git a/parse/train/ryghZJBKPS/images/186ef9f07fafdec057c5d4f7fa9b5bc24cf0fd602b6e46330071731b7cdd09b1.jpg b/parse/train/ryghZJBKPS/images/186ef9f07fafdec057c5d4f7fa9b5bc24cf0fd602b6e46330071731b7cdd09b1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f0e59568175c50b356ae7673c9123cbeb17c35eb --- /dev/null +++ b/parse/train/ryghZJBKPS/images/186ef9f07fafdec057c5d4f7fa9b5bc24cf0fd602b6e46330071731b7cdd09b1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86b9bfc4598da17f96a063aa4b4f11a58b38f6ed9d73b9dfac270fa2be18b94e +size 10661 diff --git a/parse/train/ryghZJBKPS/images/1c591f31c495cb4e2c3d089926a4a9dcc80732c6e0a95a3492a7a4b106da884d.jpg b/parse/train/ryghZJBKPS/images/1c591f31c495cb4e2c3d089926a4a9dcc80732c6e0a95a3492a7a4b106da884d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a08e89f9627199f8c854201f4d529e4906fd092a --- /dev/null +++ b/parse/train/ryghZJBKPS/images/1c591f31c495cb4e2c3d089926a4a9dcc80732c6e0a95a3492a7a4b106da884d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4caf643100ad438ce80e6f5ebaf27058fe1d10bf414e1a4a079c55ae5022f31c +size 56826 diff --git a/parse/train/ryghZJBKPS/images/1e8686a7c69e02d8ee3ed0ecbbd8213345b8e48ed6000061e556cbde8724f18f.jpg b/parse/train/ryghZJBKPS/images/1e8686a7c69e02d8ee3ed0ecbbd8213345b8e48ed6000061e556cbde8724f18f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1106293c0ecb44c58233cebb59218c4c4507e0d2 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/1e8686a7c69e02d8ee3ed0ecbbd8213345b8e48ed6000061e556cbde8724f18f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13e9c3bec8aec7c83d2fee5a450349a87ba1ca672d4f6400de4e5e566c415183 +size 56543 diff --git a/parse/train/ryghZJBKPS/images/1f627c49dffa6687b15a575a630b10f4186558a62c8cbad6574f787671935456.jpg b/parse/train/ryghZJBKPS/images/1f627c49dffa6687b15a575a630b10f4186558a62c8cbad6574f787671935456.jpg new file mode 100644 index 0000000000000000000000000000000000000000..33a49bddeabe8a09888108a3b52073d6be266a53 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/1f627c49dffa6687b15a575a630b10f4186558a62c8cbad6574f787671935456.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c75c8dfada3d1bde1e409c84a9ac585602b2ce83a710c9ce581c5f4d3464eecd +size 6077 diff --git a/parse/train/ryghZJBKPS/images/32eeed035433770dcd2fc45f0a5eb44ee367692d933f0fc9411bc45e38f7149f.jpg b/parse/train/ryghZJBKPS/images/32eeed035433770dcd2fc45f0a5eb44ee367692d933f0fc9411bc45e38f7149f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..413215315d2a0f190f74dbf4a33feb85274dfe48 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/32eeed035433770dcd2fc45f0a5eb44ee367692d933f0fc9411bc45e38f7149f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ed13966344f96b9d48ce0ebd80c5486e920c0a5156b6f8df28e71246c347e83 +size 7383 diff --git a/parse/train/ryghZJBKPS/images/3398b065aad57ae31c9b307f82c2049a16bcf7ee25b9677c8a6caa71d142daf3.jpg b/parse/train/ryghZJBKPS/images/3398b065aad57ae31c9b307f82c2049a16bcf7ee25b9677c8a6caa71d142daf3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e6734a30aa253205d3c5da044d54139882156f87 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/3398b065aad57ae31c9b307f82c2049a16bcf7ee25b9677c8a6caa71d142daf3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd141649538d1cfcda6b1bb3538e76694abb3638ee468f45f76652e66a818d78 +size 79835 diff --git a/parse/train/ryghZJBKPS/images/3fe9bade8f08ff05364a6d8e2a7baee936a987903dffc50b0cac6374c5710223.jpg b/parse/train/ryghZJBKPS/images/3fe9bade8f08ff05364a6d8e2a7baee936a987903dffc50b0cac6374c5710223.jpg new file mode 100644 index 0000000000000000000000000000000000000000..463ea69e676e2127edb73f955d0461ae84eac79b --- /dev/null +++ b/parse/train/ryghZJBKPS/images/3fe9bade8f08ff05364a6d8e2a7baee936a987903dffc50b0cac6374c5710223.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3db88ddf44d40873470f54742cb039aeb71e98ae7c3eed007999f8bf576a294a +size 47536 diff --git a/parse/train/ryghZJBKPS/images/432ede9d567f05ff30cc84e24734b9684b3267b38f6ff82e344349767b470a38.jpg b/parse/train/ryghZJBKPS/images/432ede9d567f05ff30cc84e24734b9684b3267b38f6ff82e344349767b470a38.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4bab58d579959823b2024828e30c53cecbfaecae --- /dev/null +++ b/parse/train/ryghZJBKPS/images/432ede9d567f05ff30cc84e24734b9684b3267b38f6ff82e344349767b470a38.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:abf9c2dc3ceaebdaccab3b757a0f0f08f4596c749bba24592d0b8c8a3f9c2daf +size 86135 diff --git a/parse/train/ryghZJBKPS/images/5d40d22c75721a059cbb0b806209efb76c254e0e8846d044ac3890ff3d1626ee.jpg b/parse/train/ryghZJBKPS/images/5d40d22c75721a059cbb0b806209efb76c254e0e8846d044ac3890ff3d1626ee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..36f5b1cf1f5c6e9db64b9fe395064b4a3f26b0c3 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/5d40d22c75721a059cbb0b806209efb76c254e0e8846d044ac3890ff3d1626ee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4baa8f4280eae49f4f88426c9a79e4708fd6765f378baf0434b2f344bec7282d +size 42379 diff --git a/parse/train/ryghZJBKPS/images/68897bb6eb4bbcd2af8922a41897bb08fb0a8da3dee3dc4ea852ae1ac708751f.jpg b/parse/train/ryghZJBKPS/images/68897bb6eb4bbcd2af8922a41897bb08fb0a8da3dee3dc4ea852ae1ac708751f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8020e2697a5327f7df242fefaa1311c645fb60b2 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/68897bb6eb4bbcd2af8922a41897bb08fb0a8da3dee3dc4ea852ae1ac708751f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87c2640018b4f4701b20e36398744133051b0b4a9f7047f22385ab2577811712 +size 43764 diff --git a/parse/train/ryghZJBKPS/images/761d32e1fc0ea3a16b3cc392c02bb4a4bc88fb94d123a527dc896c44b63172ea.jpg b/parse/train/ryghZJBKPS/images/761d32e1fc0ea3a16b3cc392c02bb4a4bc88fb94d123a527dc896c44b63172ea.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d13e6800f60d0ac01ec42a50126b8f094d63b09 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/761d32e1fc0ea3a16b3cc392c02bb4a4bc88fb94d123a527dc896c44b63172ea.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d4eaba42773379e4427523dc81a45ca699e3a5784fa688f1fbde98cdf6ad7cc +size 6806 diff --git a/parse/train/ryghZJBKPS/images/76e18f8566dc277eb165ad2a542cf0aeb87cdf8690bfcf233cdf2a0caac590d4.jpg b/parse/train/ryghZJBKPS/images/76e18f8566dc277eb165ad2a542cf0aeb87cdf8690bfcf233cdf2a0caac590d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..96b73d5dd048fe40f908ccc81baec4b26dd5d391 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/76e18f8566dc277eb165ad2a542cf0aeb87cdf8690bfcf233cdf2a0caac590d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46ba31ace3be96160450295d5c324df650cbbafdf6ba530f7b503332536532a6 +size 41264 diff --git a/parse/train/ryghZJBKPS/images/7dd5a955d88848c7fffd4e19963044529b8b5dd217cfd79f1fe6e676bf7c481e.jpg b/parse/train/ryghZJBKPS/images/7dd5a955d88848c7fffd4e19963044529b8b5dd217cfd79f1fe6e676bf7c481e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4926bb00ad86348027f575ba9f2425f3985bd21c --- /dev/null +++ b/parse/train/ryghZJBKPS/images/7dd5a955d88848c7fffd4e19963044529b8b5dd217cfd79f1fe6e676bf7c481e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eca07c458f6842f60268ad4cb505d8867eced0b0c8e7a36ec00f96cf44c60543 +size 58563 diff --git a/parse/train/ryghZJBKPS/images/862b919b13024e0ae4ccf8ea1350510ffaca7bdbca2409f37db62423dba2128a.jpg b/parse/train/ryghZJBKPS/images/862b919b13024e0ae4ccf8ea1350510ffaca7bdbca2409f37db62423dba2128a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..57f9b7472b2e3180b4ce104aa8baf084dc48cf25 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/862b919b13024e0ae4ccf8ea1350510ffaca7bdbca2409f37db62423dba2128a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5b81ead93f714bb7f5cb1de842a83f2943033f35a8ada740ac1bfca6b2d884f +size 61469 diff --git a/parse/train/ryghZJBKPS/images/8aea56706be71d726b40aab86de0168f6950ac1a8c1af6affb837d9b574a0b9b.jpg b/parse/train/ryghZJBKPS/images/8aea56706be71d726b40aab86de0168f6950ac1a8c1af6affb837d9b574a0b9b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..23491b3a4abab5fcf44bf6ac290c0935b2011eeb --- /dev/null +++ b/parse/train/ryghZJBKPS/images/8aea56706be71d726b40aab86de0168f6950ac1a8c1af6affb837d9b574a0b9b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d64911c42de0f1e93da2a66640961501d1d9883fd7b70585957b13d0c188e052 +size 30769 diff --git a/parse/train/ryghZJBKPS/images/8ec656b0d27579e17e5d426494b01e24c71753019e1725e0565a8cb78ab69c17.jpg b/parse/train/ryghZJBKPS/images/8ec656b0d27579e17e5d426494b01e24c71753019e1725e0565a8cb78ab69c17.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9b000b8299732df92f99c9296b43e55db1b08eba --- /dev/null +++ b/parse/train/ryghZJBKPS/images/8ec656b0d27579e17e5d426494b01e24c71753019e1725e0565a8cb78ab69c17.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47939af176581afc5b3538409f9cb2d18ccaba03ed898bbf0e0c1706551bcdb9 +size 78025 diff --git a/parse/train/ryghZJBKPS/images/8fe1677b8a9ebbca89d82595804c40b53899f667292aace8887172757fda3d48.jpg b/parse/train/ryghZJBKPS/images/8fe1677b8a9ebbca89d82595804c40b53899f667292aace8887172757fda3d48.jpg new file mode 100644 index 0000000000000000000000000000000000000000..da20e2b9b397752eb5f44748b289baae98cb3955 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/8fe1677b8a9ebbca89d82595804c40b53899f667292aace8887172757fda3d48.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab96aed62f0d77ed21f0e40f6f55f417b763f3d4a194687e1476708cf773c110 +size 44750 diff --git a/parse/train/ryghZJBKPS/images/9441310347a519c3bf0c4e0278a502b2a72f0f5cf89c8efe0a5342d05f3ee882.jpg b/parse/train/ryghZJBKPS/images/9441310347a519c3bf0c4e0278a502b2a72f0f5cf89c8efe0a5342d05f3ee882.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d5209478226e30251c66a847b0f044a53306e055 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/9441310347a519c3bf0c4e0278a502b2a72f0f5cf89c8efe0a5342d05f3ee882.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbc59237cbd3a70fc0288325bded9dfb166770b9bf1ae5d9e47aaab0be1add92 +size 7464 diff --git a/parse/train/ryghZJBKPS/images/990f157de5f16287b5cc2739568242f27ae18014df01472d1d1c4978f90a5d0a.jpg b/parse/train/ryghZJBKPS/images/990f157de5f16287b5cc2739568242f27ae18014df01472d1d1c4978f90a5d0a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b6350baae98befce50375ebfecaf85aecd758b7d --- /dev/null +++ b/parse/train/ryghZJBKPS/images/990f157de5f16287b5cc2739568242f27ae18014df01472d1d1c4978f90a5d0a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b45ed350f26a3c53dc59e87af0728ca06c95836752c77194e12e3ac1aa9c7373 +size 54376 diff --git a/parse/train/ryghZJBKPS/images/a4a03ba238975dad7d4d9706c2fc6f8bf6ace30f7ad25c97337a95b53ba674be.jpg b/parse/train/ryghZJBKPS/images/a4a03ba238975dad7d4d9706c2fc6f8bf6ace30f7ad25c97337a95b53ba674be.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d780284d39384f74d37a2c527e0255aca053a496 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/a4a03ba238975dad7d4d9706c2fc6f8bf6ace30f7ad25c97337a95b53ba674be.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:108f1c7db729ca3e68a8cec7d98c75734bd4e6d85280734ef0fff991e9a3a824 +size 152540 diff --git a/parse/train/ryghZJBKPS/images/b1f42c320b083c071d89711fbe35a86eb10b8d1f0c2d817a99adf3ecf93c99a1.jpg b/parse/train/ryghZJBKPS/images/b1f42c320b083c071d89711fbe35a86eb10b8d1f0c2d817a99adf3ecf93c99a1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7138ce4fb584c164d51fba170e683ce68fca4e9e --- /dev/null +++ b/parse/train/ryghZJBKPS/images/b1f42c320b083c071d89711fbe35a86eb10b8d1f0c2d817a99adf3ecf93c99a1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0ae1b6bf56365399b9cd5e8292aa5909a19208b68ed82f50b2dcb5bb18d54cc +size 44306 diff --git a/parse/train/ryghZJBKPS/images/b6198f1f5a9453b9546972ece2536fc119869120e54d54025af727432c2cd297.jpg b/parse/train/ryghZJBKPS/images/b6198f1f5a9453b9546972ece2536fc119869120e54d54025af727432c2cd297.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5c806fcccb7464affbf7f69d35fc6c91a7b8ab5e --- /dev/null +++ b/parse/train/ryghZJBKPS/images/b6198f1f5a9453b9546972ece2536fc119869120e54d54025af727432c2cd297.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99a82e7230ef617e0252743f7882663966c6de58565bfb71f2a88048f05d5c6e +size 48181 diff --git a/parse/train/ryghZJBKPS/images/bcdf7607b2dfbb32a55c7026c517c34f609fe133885d1d155bef016d2d58a66c.jpg b/parse/train/ryghZJBKPS/images/bcdf7607b2dfbb32a55c7026c517c34f609fe133885d1d155bef016d2d58a66c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1496149476acbfca23bcd2737bb6e581ea5be125 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/bcdf7607b2dfbb32a55c7026c517c34f609fe133885d1d155bef016d2d58a66c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:568d4af97ef1177b4a675fa64fdf68101478396a04c14ea4c4db990361f0e7bf +size 79223 diff --git a/parse/train/ryghZJBKPS/images/c0a77f9cb45b53e1dbc20841a41ef19cb516cd593b6a7208b585f88a23eee329.jpg b/parse/train/ryghZJBKPS/images/c0a77f9cb45b53e1dbc20841a41ef19cb516cd593b6a7208b585f88a23eee329.jpg new file mode 100644 index 0000000000000000000000000000000000000000..42c7813902d0aaaa0272d9f938318b69e24cf68d --- /dev/null +++ b/parse/train/ryghZJBKPS/images/c0a77f9cb45b53e1dbc20841a41ef19cb516cd593b6a7208b585f88a23eee329.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee423d89f908e7dc8355a45d529944428d7ce12e580a96ca9524ac91f2ddda2b +size 8284 diff --git a/parse/train/ryghZJBKPS/images/d654ac52f0e8e7b4de997ddc93ec4e0f4ce9c1df3e103a48670fd2e5bf5edd73.jpg b/parse/train/ryghZJBKPS/images/d654ac52f0e8e7b4de997ddc93ec4e0f4ce9c1df3e103a48670fd2e5bf5edd73.jpg new file mode 100644 index 0000000000000000000000000000000000000000..44abe074369fbc09d6944f6ff13ab74a7cbbd081 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/d654ac52f0e8e7b4de997ddc93ec4e0f4ce9c1df3e103a48670fd2e5bf5edd73.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84f01ad2974262f32f88f92c30bdf870372d41b2005770677e42da38ae161de3 +size 62015 diff --git a/parse/train/ryghZJBKPS/images/d9030f407149623b5e1e3c6e2181cca01dc0f054d2cc7ee5b9e65e37b788430f.jpg b/parse/train/ryghZJBKPS/images/d9030f407149623b5e1e3c6e2181cca01dc0f054d2cc7ee5b9e65e37b788430f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..acf68d2d879df311fb23c205c207c5cadc2dadf9 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/d9030f407149623b5e1e3c6e2181cca01dc0f054d2cc7ee5b9e65e37b788430f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11651ddd81ca6e634f4d0261a0ce2adc97a83e18d2229683a9fe9a9a70b9c276 +size 55126 diff --git a/parse/train/ryghZJBKPS/images/e12959b2bd715346b53c680fd61135d51d59b955ba6bea4935589b7e90cd23d7.jpg b/parse/train/ryghZJBKPS/images/e12959b2bd715346b53c680fd61135d51d59b955ba6bea4935589b7e90cd23d7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6bb860bc70601a220722200534b08cf94c503acb --- /dev/null +++ b/parse/train/ryghZJBKPS/images/e12959b2bd715346b53c680fd61135d51d59b955ba6bea4935589b7e90cd23d7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d08caecd60c8f04f08990049df0b528bfb03863c06f0f3bcbe5b616a49270f4c +size 4366 diff --git a/parse/train/ryghZJBKPS/images/e8fe0e3b347610236751383630edb45029111c672cb94ecfd283ffc78dca523d.jpg b/parse/train/ryghZJBKPS/images/e8fe0e3b347610236751383630edb45029111c672cb94ecfd283ffc78dca523d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7c49dd9f98c910fa3ba8f0f7ce7ee2006996b196 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/e8fe0e3b347610236751383630edb45029111c672cb94ecfd283ffc78dca523d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b59202b103e67021d89e00fa55c1a3c51e3d287ce272039dd391df0b09bbde9 +size 52189 diff --git a/parse/train/ryghZJBKPS/images/e9edd9bc4085eb347e0355361c980da855565eadfb3c1c4c32469cc726e95c60.jpg b/parse/train/ryghZJBKPS/images/e9edd9bc4085eb347e0355361c980da855565eadfb3c1c4c32469cc726e95c60.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e1ef77b429aa3ee65898b212ff7809ad79e40802 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/e9edd9bc4085eb347e0355361c980da855565eadfb3c1c4c32469cc726e95c60.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00583df6bca5b7684157ac70268014bb3809bb223831ae391fdb65b33d80460b +size 67665 diff --git a/parse/train/ryghZJBKPS/images/eb67dcecb5e582087261eb4ed09d95f125d1a62a5adfb0b8d0f0f0be6bba3e0a.jpg b/parse/train/ryghZJBKPS/images/eb67dcecb5e582087261eb4ed09d95f125d1a62a5adfb0b8d0f0f0be6bba3e0a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c6b2dc5cdeca82e16ae3831eaefdea10933f37ae --- /dev/null +++ b/parse/train/ryghZJBKPS/images/eb67dcecb5e582087261eb4ed09d95f125d1a62a5adfb0b8d0f0f0be6bba3e0a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99ada428b755aa8b10020d56772500d89f9687f1110bfd8b0d79e7806829e847 +size 56853 diff --git a/parse/train/ryghZJBKPS/images/f0c42a0c3e2d768cd833fa7d50e3800e2873b503e576998c7a448383e9630f34.jpg b/parse/train/ryghZJBKPS/images/f0c42a0c3e2d768cd833fa7d50e3800e2873b503e576998c7a448383e9630f34.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f3dc67559e1ae720bf806dcc5efdff3da6241783 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/f0c42a0c3e2d768cd833fa7d50e3800e2873b503e576998c7a448383e9630f34.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:777568f173e89b3ccb1fdf680b62a0f2aa8ac44a1c34af44297e0e21d14a62cf +size 157032 diff --git a/parse/train/ryghZJBKPS/images/f2afa1c87f4feaa6e3796baea1d6a22218fb040453bcdb82a250d55b4d0a9738.jpg b/parse/train/ryghZJBKPS/images/f2afa1c87f4feaa6e3796baea1d6a22218fb040453bcdb82a250d55b4d0a9738.jpg new file mode 100644 index 0000000000000000000000000000000000000000..30946e4ee531ed18022a87dc31c45d9c7397814a --- /dev/null +++ b/parse/train/ryghZJBKPS/images/f2afa1c87f4feaa6e3796baea1d6a22218fb040453bcdb82a250d55b4d0a9738.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a148b3e741927d66fc5323b458636c6a0c5ef59c12a942d3a508cba2618038aa +size 57181 diff --git a/parse/train/ryghZJBKPS/images/fbce56c22162e29e6d225aa0e5d67f5788bf75bd60797bed3ade2cebd28df0bb.jpg b/parse/train/ryghZJBKPS/images/fbce56c22162e29e6d225aa0e5d67f5788bf75bd60797bed3ade2cebd28df0bb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..efa4de02da18452b4c7f48d3cb33fcf80c1847c0 --- /dev/null +++ b/parse/train/ryghZJBKPS/images/fbce56c22162e29e6d225aa0e5d67f5788bf75bd60797bed3ade2cebd28df0bb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d03e627cd760b63bd48b177a50e62ee701437887c682e383a4355703d9d4c5d7 +size 56738 diff --git a/parse/train/rylKB3A9Fm/images/17180044125dc55c34949f4e6b9829cd33664d97e574ca6a9da1af179a20d5c5.jpg b/parse/train/rylKB3A9Fm/images/17180044125dc55c34949f4e6b9829cd33664d97e574ca6a9da1af179a20d5c5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db1fe6d0c3d8f895b4f830c9bc9201aa306e3a6a --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/17180044125dc55c34949f4e6b9829cd33664d97e574ca6a9da1af179a20d5c5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:793c418a31bdacae77829f0527dfe988b3ce7085d867ae01ef0166ac403ffc83 +size 6992 diff --git a/parse/train/rylKB3A9Fm/images/210c85e05fe6eaba16bc613c4655e0c793a2f1fb59dc8cf95188409ed5c21e5c.jpg b/parse/train/rylKB3A9Fm/images/210c85e05fe6eaba16bc613c4655e0c793a2f1fb59dc8cf95188409ed5c21e5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3958e6e967b9329f78576b16ce5b1ebaea8500df --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/210c85e05fe6eaba16bc613c4655e0c793a2f1fb59dc8cf95188409ed5c21e5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86da741a4f9026486f7f676b6f3b9d23bca46b139a3405b9a8ad0092035f05a1 +size 74960 diff --git a/parse/train/rylKB3A9Fm/images/21e803b52ef48a8451ccccdb160cc80a8f7bc0f4036267729d02f7a683c5f9f4.jpg b/parse/train/rylKB3A9Fm/images/21e803b52ef48a8451ccccdb160cc80a8f7bc0f4036267729d02f7a683c5f9f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..41d939f1221f2658c15d40744247eabe3e94aa81 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/21e803b52ef48a8451ccccdb160cc80a8f7bc0f4036267729d02f7a683c5f9f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c4d404541175de25e23202ab701ef9cb4aca13a7f3885aa46cccf9642e3367a +size 81069 diff --git a/parse/train/rylKB3A9Fm/images/32d8be26ba11ab10f2a1e69f863f32fcfe0f4f8ce39c5313541b952e689c758e.jpg b/parse/train/rylKB3A9Fm/images/32d8be26ba11ab10f2a1e69f863f32fcfe0f4f8ce39c5313541b952e689c758e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1c4fbe7200a406d7d867e5cbc0fae2fbfaed870a --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/32d8be26ba11ab10f2a1e69f863f32fcfe0f4f8ce39c5313541b952e689c758e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63edffd7b93f3c8c668f22ed2f4e06b873f68a86d3fd24388a1ff44296d57a9c +size 35282 diff --git a/parse/train/rylKB3A9Fm/images/36e59b5be411ce825c4ce9c6002be23d24bb871e3d203dc80efc1f117e6c64f9.jpg b/parse/train/rylKB3A9Fm/images/36e59b5be411ce825c4ce9c6002be23d24bb871e3d203dc80efc1f117e6c64f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b6b23a7deb7d3b6a0a5489e69109fdfb15d67ddb --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/36e59b5be411ce825c4ce9c6002be23d24bb871e3d203dc80efc1f117e6c64f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90a40e7b471662a85a91c9cc9a87471bd9079decfc9bb510da59acee972c0b88 +size 82491 diff --git a/parse/train/rylKB3A9Fm/images/3a151edbbd19584736b246733607cc249d862a8a4c37636d77a3be809a0ed327.jpg b/parse/train/rylKB3A9Fm/images/3a151edbbd19584736b246733607cc249d862a8a4c37636d77a3be809a0ed327.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f5ecb4c254612d9fe8d6dd612ddf25bd7aeb0a6f --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/3a151edbbd19584736b246733607cc249d862a8a4c37636d77a3be809a0ed327.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f01e85bdbd7fd8f3c960440d4eb043e09b89333ff34439b2825b0ff0ebf886ec +size 33776 diff --git a/parse/train/rylKB3A9Fm/images/437909a2b2123a3197e73bc9d6d39f69a6f1a3e1da8b04dd3d611ced90e5b0cc.jpg b/parse/train/rylKB3A9Fm/images/437909a2b2123a3197e73bc9d6d39f69a6f1a3e1da8b04dd3d611ced90e5b0cc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..99777b9871927f186d59701ea74b950e0f05f60d --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/437909a2b2123a3197e73bc9d6d39f69a6f1a3e1da8b04dd3d611ced90e5b0cc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a5fdc7ecc3ad7d460bb11db30cfc606035f1e27b3fd5f3d64d6d92bc3b41b08 +size 66247 diff --git a/parse/train/rylKB3A9Fm/images/6ce41ad84af195666df6da446e72efeadc2a4526954774ebcd06faceabde362f.jpg b/parse/train/rylKB3A9Fm/images/6ce41ad84af195666df6da446e72efeadc2a4526954774ebcd06faceabde362f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..42d06cba78caeea6ebbda6214c7f3f1b97dc38f7 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/6ce41ad84af195666df6da446e72efeadc2a4526954774ebcd06faceabde362f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:814c39633247001d21d83f9ecbeaeb2f3b157f989df3d60d171ce61339ad6163 +size 123484 diff --git a/parse/train/rylKB3A9Fm/images/7f3ff340813c73e770668f03c240c43095ac6dcf5c8d05728f14fa131b849471.jpg b/parse/train/rylKB3A9Fm/images/7f3ff340813c73e770668f03c240c43095ac6dcf5c8d05728f14fa131b849471.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a18f29988d67968bee86f01c728ef528d6afda03 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/7f3ff340813c73e770668f03c240c43095ac6dcf5c8d05728f14fa131b849471.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89ee4710d165900d2cc45184273fb661bf5c8ed2e34e64624e41eb0c055238d4 +size 130521 diff --git a/parse/train/rylKB3A9Fm/images/9609859ebe7143f35034b1c299228a3bb1330bf8cbc18850cf9e024ce675ce3b.jpg b/parse/train/rylKB3A9Fm/images/9609859ebe7143f35034b1c299228a3bb1330bf8cbc18850cf9e024ce675ce3b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e04c5ed3fa48239f3aab79a23870170bf43e4970 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/9609859ebe7143f35034b1c299228a3bb1330bf8cbc18850cf9e024ce675ce3b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afa1d4d6195f2a999e3eda99b1688e20f07fda0eb8174ee874fb240f5df10c3b +size 85163 diff --git a/parse/train/rylKB3A9Fm/images/9b7a72b0516d9a399de4fa55e2734d0a146d127372b622ed5657290b4c74f781.jpg b/parse/train/rylKB3A9Fm/images/9b7a72b0516d9a399de4fa55e2734d0a146d127372b622ed5657290b4c74f781.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eec821b963f1b6495e6f43215b76bbb75a3face6 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/9b7a72b0516d9a399de4fa55e2734d0a146d127372b622ed5657290b4c74f781.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:481ea6024e200ccf95ab330ac6f9a3de015e405e5cf49a8485f7d92f039fd2e3 +size 67684 diff --git a/parse/train/rylKB3A9Fm/images/ade4172e500217cb3c9196945360bd9f9e5616d975f9bf5240e024b27ecf7e1b.jpg b/parse/train/rylKB3A9Fm/images/ade4172e500217cb3c9196945360bd9f9e5616d975f9bf5240e024b27ecf7e1b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5c61e7adaa9a7750cffb82c2008a699db7fbb87b --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/ade4172e500217cb3c9196945360bd9f9e5616d975f9bf5240e024b27ecf7e1b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d5a9c94c1b15fbbd15c3553de54637e2788dadcf9144247acaa9c9558d09f01 +size 85328 diff --git a/parse/train/rylKB3A9Fm/images/d6a7e2d42ae94b5912a386f2a96ffea3505d15df09ead4d3f9a857d98dacd36b.jpg b/parse/train/rylKB3A9Fm/images/d6a7e2d42ae94b5912a386f2a96ffea3505d15df09ead4d3f9a857d98dacd36b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8190b69155dc633388abf7f37ff45d8f15de5bf7 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/d6a7e2d42ae94b5912a386f2a96ffea3505d15df09ead4d3f9a857d98dacd36b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3abba7dfdacc74df3382546c7e177ad562d7e425afc5d7661e3b7f17b4554c34 +size 32489 diff --git a/parse/train/rylKB3A9Fm/images/d79ded067aeebe0f25ab2e294e64d585f2ae0c816854d98ab92b0139159b3ca9.jpg b/parse/train/rylKB3A9Fm/images/d79ded067aeebe0f25ab2e294e64d585f2ae0c816854d98ab92b0139159b3ca9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8176830614ca8cbe392451a2e9d374626e73a192 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/d79ded067aeebe0f25ab2e294e64d585f2ae0c816854d98ab92b0139159b3ca9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27dbf0ba98c0cd282e76f4b451272d07b88af3bec3ad23968e7a78614f1efa0b +size 88999 diff --git a/parse/train/rylKB3A9Fm/images/ee8866b278925a384fc07b5300a9f50b53dc9bf57f51c882a5916a754f24420d.jpg b/parse/train/rylKB3A9Fm/images/ee8866b278925a384fc07b5300a9f50b53dc9bf57f51c882a5916a754f24420d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9b9e1d895223216050e6729d268abb07b2f28b23 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/ee8866b278925a384fc07b5300a9f50b53dc9bf57f51c882a5916a754f24420d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a614d6968122b85ef38c05c1ba5a2caff11a8ad13a56d3ed767725dd87ceb173 +size 81920 diff --git a/parse/train/rylKB3A9Fm/images/f3ed7773ca245d7fd9e63b9069d511938ca80bec185f15fcdb1157f17852d1f0.jpg b/parse/train/rylKB3A9Fm/images/f3ed7773ca245d7fd9e63b9069d511938ca80bec185f15fcdb1157f17852d1f0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bc9e2796ab48a97cb00464ee1eff657bb852cde7 --- /dev/null +++ b/parse/train/rylKB3A9Fm/images/f3ed7773ca245d7fd9e63b9069d511938ca80bec185f15fcdb1157f17852d1f0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4fe27ce20eabab3f1755e51bd34de23234b865d2ca83020b3047d62ddd2f9be7 +size 7269 diff --git a/parse/train/rytstxWAW/images/168baf68e963671312e2dd7ef874b58754e3b1f657650d690bdc0dee5fe10252.jpg b/parse/train/rytstxWAW/images/168baf68e963671312e2dd7ef874b58754e3b1f657650d690bdc0dee5fe10252.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c699dfa6c5128fc5fae1def50c5e86d8934fd25f --- /dev/null +++ b/parse/train/rytstxWAW/images/168baf68e963671312e2dd7ef874b58754e3b1f657650d690bdc0dee5fe10252.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc1caf3aba45aed62e185c82ffea1d50a9f6898dd2efa7d6485cfcaae4845471 +size 32488 diff --git a/parse/train/rytstxWAW/images/2b5516ae51f2040394c7a65ceb34b5024929322f1e47249874dc61080572ba52.jpg b/parse/train/rytstxWAW/images/2b5516ae51f2040394c7a65ceb34b5024929322f1e47249874dc61080572ba52.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2422251ad44bf99b3cdb8fc26bcecc96359dbee0 --- /dev/null +++ b/parse/train/rytstxWAW/images/2b5516ae51f2040394c7a65ceb34b5024929322f1e47249874dc61080572ba52.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f97ff6bb2fab3ba890961aafe540774fa9eadd23a9e1ec6c8cc561b95d0463d +size 11973 diff --git a/parse/train/rytstxWAW/images/4c22dc9d6f088e64a28a6322f913c74b32d28c4dda617bba7c3685ab52c805e0.jpg b/parse/train/rytstxWAW/images/4c22dc9d6f088e64a28a6322f913c74b32d28c4dda617bba7c3685ab52c805e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f8275b5bf00b9b282e7b4678ccc4e7c57a7d1434 --- /dev/null +++ b/parse/train/rytstxWAW/images/4c22dc9d6f088e64a28a6322f913c74b32d28c4dda617bba7c3685ab52c805e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76adf279c82825c6be812871e191c93703cc97e7430b3979da5631451987d62a +size 7084 diff --git a/parse/train/rytstxWAW/images/8aa5c430eb60a66f066a0a641e4e904d711aa491a6697b94f9f203653b4055e3.jpg b/parse/train/rytstxWAW/images/8aa5c430eb60a66f066a0a641e4e904d711aa491a6697b94f9f203653b4055e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..479c6a72b02125d78ce2ae898bfb1fb2ec6b3a99 --- /dev/null +++ b/parse/train/rytstxWAW/images/8aa5c430eb60a66f066a0a641e4e904d711aa491a6697b94f9f203653b4055e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de07da52092e480916308f5d5ee84bc10bb7f4757c75982a5b49accf23f119b2 +size 5996 diff --git a/parse/train/rytstxWAW/images/8afbd81877ff338c8f417d92d120b64a5acc99e64c611c7338749644db57408f.jpg b/parse/train/rytstxWAW/images/8afbd81877ff338c8f417d92d120b64a5acc99e64c611c7338749644db57408f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4f251658f419823c73bf62a3dd26a95e469582a3 --- /dev/null +++ b/parse/train/rytstxWAW/images/8afbd81877ff338c8f417d92d120b64a5acc99e64c611c7338749644db57408f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f6d03153b682187a73183bc89b5e9d60c70fef6db2143ea0bc03e4c36ba47be +size 9592 diff --git a/parse/train/rytstxWAW/images/a642c0f916609a1979359abadb4fcbfef3bf65d3b9b149ccd5a918a7627521a6.jpg b/parse/train/rytstxWAW/images/a642c0f916609a1979359abadb4fcbfef3bf65d3b9b149ccd5a918a7627521a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dc59f32bf4f394534a62c398e65028c769accd7f --- /dev/null +++ b/parse/train/rytstxWAW/images/a642c0f916609a1979359abadb4fcbfef3bf65d3b9b149ccd5a918a7627521a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e10147a83a71ff2c9913764cfe5584d4c88da09a8aa0143bc8196bd0a089b76f +size 21539 diff --git a/parse/train/rytstxWAW/images/cc08b7a622b64096f7a33fc2b3a0b330c2215aef55ab8ccee3aba551c75e2925.jpg b/parse/train/rytstxWAW/images/cc08b7a622b64096f7a33fc2b3a0b330c2215aef55ab8ccee3aba551c75e2925.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6ddb2e7a82ec740ed8a7d56067e72d83f164dc95 --- /dev/null +++ b/parse/train/rytstxWAW/images/cc08b7a622b64096f7a33fc2b3a0b330c2215aef55ab8ccee3aba551c75e2925.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d891ca9ec4ad90424b02c48fe4a62db0937050c41126ba49e41eb66862a97512 +size 7113 diff --git a/parse/train/rytstxWAW/images/d5e307a35ff927c88fbf8dca81a4529e3da7e6b8fc128d005f01e76b8d41dc03.jpg b/parse/train/rytstxWAW/images/d5e307a35ff927c88fbf8dca81a4529e3da7e6b8fc128d005f01e76b8d41dc03.jpg new file mode 100644 index 0000000000000000000000000000000000000000..54e3023786f609035e3a1635d38ed6dff8a3ca52 --- /dev/null +++ b/parse/train/rytstxWAW/images/d5e307a35ff927c88fbf8dca81a4529e3da7e6b8fc128d005f01e76b8d41dc03.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c168065bbc6a6b90b8ccabd90b9c69775947b768433fb76ddcb9f7d720b59d2 +size 31228 diff --git a/parse/train/rytstxWAW/images/de0530a1d4d07e8004d972616b211612a44f60f41db600db4240e9ac0f277e6c.jpg b/parse/train/rytstxWAW/images/de0530a1d4d07e8004d972616b211612a44f60f41db600db4240e9ac0f277e6c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e666b9ccc6d874d54c8c0a6614f13b3d7e9e4340 --- /dev/null +++ b/parse/train/rytstxWAW/images/de0530a1d4d07e8004d972616b211612a44f60f41db600db4240e9ac0f277e6c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39e2b0446df7befefd238d7c62f4d5c97b63a5aed277a2f79cdd2eec0cc753e9 +size 5038 diff --git a/parse/train/rytstxWAW/images/e910886eb9c996ba3f2c2b160c0262d01583ef6b108415c9c68485a245db573e.jpg b/parse/train/rytstxWAW/images/e910886eb9c996ba3f2c2b160c0262d01583ef6b108415c9c68485a245db573e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5d32a80545403127acf8ca77a5004623cd9b7839 --- /dev/null +++ b/parse/train/rytstxWAW/images/e910886eb9c996ba3f2c2b160c0262d01583ef6b108415c9c68485a245db573e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c8f8bf3fdda271fec207fd126a24281c71aca6458331ec09e071ca0adfda84b +size 14092 diff --git a/parse/train/rytstxWAW/images/f3bd5c9d7a11c629a0e67b3fc0bfd299e4f1666d5164c811520cd18ee4155dfb.jpg b/parse/train/rytstxWAW/images/f3bd5c9d7a11c629a0e67b3fc0bfd299e4f1666d5164c811520cd18ee4155dfb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fb24018e7727c3fef47fb7e585119fa10f94b1e2 --- /dev/null +++ b/parse/train/rytstxWAW/images/f3bd5c9d7a11c629a0e67b3fc0bfd299e4f1666d5164c811520cd18ee4155dfb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f235d8d0de1278edf0bd71543e86bbcac0bce0ab8c9427a3390aea4305d7de8b +size 6549 diff --git a/parse/train/rytstxWAW/images/f8673b9ec4ecd6622e9df2e0f521922bcd3dc8ae8ece989c4964f8c18d1ef29b.jpg b/parse/train/rytstxWAW/images/f8673b9ec4ecd6622e9df2e0f521922bcd3dc8ae8ece989c4964f8c18d1ef29b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9b3b5e44833b82fd1a1f1f5e36e320ec3a1b3dc6 --- /dev/null +++ b/parse/train/rytstxWAW/images/f8673b9ec4ecd6622e9df2e0f521922bcd3dc8ae8ece989c4964f8c18d1ef29b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00df9a0ba6f7b382b949e4b059e0fbca59cc7c3d27a72b6e772ec13bf808a788 +size 4494 diff --git a/parse/train/tUeeRzMXJZ/images/00276619a26958d1d719dd9340160f5de593498236ee87d2b14a46630e799ca8.jpg b/parse/train/tUeeRzMXJZ/images/00276619a26958d1d719dd9340160f5de593498236ee87d2b14a46630e799ca8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..961bc8904b2283b9ba86b92d3fdf7e76859b6cbe --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/00276619a26958d1d719dd9340160f5de593498236ee87d2b14a46630e799ca8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d7472473797176232cf44f91c6235c4f72eccf89a6d8bba4753e00f3c228698 +size 10922 diff --git a/parse/train/tUeeRzMXJZ/images/059836c7da018a534eff53eda7a1e45a3a02d20ce1b65fc5e20428c702fbe592.jpg b/parse/train/tUeeRzMXJZ/images/059836c7da018a534eff53eda7a1e45a3a02d20ce1b65fc5e20428c702fbe592.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6cf51cd75358935620bc9fcffad28ec90b5eba83 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/059836c7da018a534eff53eda7a1e45a3a02d20ce1b65fc5e20428c702fbe592.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8c1d8f672cd0e9117ae0c1b05b460cc53b577757af57b9cae32b2c3263a60ec +size 15419 diff --git a/parse/train/tUeeRzMXJZ/images/0f122a8ae933bd73b5713d4ddbf31f0c8f735cb2caf0fc2e02ecdfd7fb2642a7.jpg b/parse/train/tUeeRzMXJZ/images/0f122a8ae933bd73b5713d4ddbf31f0c8f735cb2caf0fc2e02ecdfd7fb2642a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6b2459af43e7b18207a33403e9c9390b53b28151 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/0f122a8ae933bd73b5713d4ddbf31f0c8f735cb2caf0fc2e02ecdfd7fb2642a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:299a8df955fcf64a1820cc34adf44916d5e6cb0d83223a57ccc45ce6a3307405 +size 17370 diff --git a/parse/train/tUeeRzMXJZ/images/189c807a993e11ca4039e04c060fabb9f9e57a0d42c41a2df64a2d68e2152e21.jpg b/parse/train/tUeeRzMXJZ/images/189c807a993e11ca4039e04c060fabb9f9e57a0d42c41a2df64a2d68e2152e21.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fa29cf5e9ec907135da2c7fa8bde16f7c0f26ee8 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/189c807a993e11ca4039e04c060fabb9f9e57a0d42c41a2df64a2d68e2152e21.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18a4d74585dd6c6524bfb95d4d27745ab89e757d4242931091be52d934eefe18 +size 7463 diff --git a/parse/train/tUeeRzMXJZ/images/1e1463ab472f1fe08bcfd47b5a6d82ee6a1a4f9621ea4be0058f53a1b36a86a8.jpg b/parse/train/tUeeRzMXJZ/images/1e1463ab472f1fe08bcfd47b5a6d82ee6a1a4f9621ea4be0058f53a1b36a86a8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ffac60b7321e509eae23124b79ede9b11113864c --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/1e1463ab472f1fe08bcfd47b5a6d82ee6a1a4f9621ea4be0058f53a1b36a86a8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6328935b6ebc1c9bab3c41e49f7888ea552535362df0ea72deb48fe00d45633e +size 10408 diff --git a/parse/train/tUeeRzMXJZ/images/2e525db580e7bf8a223928427fe6d5be53ee99784e348c11e0e79203219864a0.jpg b/parse/train/tUeeRzMXJZ/images/2e525db580e7bf8a223928427fe6d5be53ee99784e348c11e0e79203219864a0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..80b8fda7e3bfd8a0e511ecc13d7a5eda3c6d02ca --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/2e525db580e7bf8a223928427fe6d5be53ee99784e348c11e0e79203219864a0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8ed7140b4eb9821eef4a1c0a679cf5536d7a439da7c799386c058241e00d3dd +size 11373 diff --git a/parse/train/tUeeRzMXJZ/images/3a81ff5cfa75945277e9efb13e4bdd0228b923eba96644fb716833337fffa5a7.jpg b/parse/train/tUeeRzMXJZ/images/3a81ff5cfa75945277e9efb13e4bdd0228b923eba96644fb716833337fffa5a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..266e90d513d1461210f8347b799a68e5f8407fc8 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/3a81ff5cfa75945277e9efb13e4bdd0228b923eba96644fb716833337fffa5a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b35a011b30612261085c1b4f9e9bc5669bb47fee28b540a64a4acd065dde97b4 +size 9857 diff --git a/parse/train/tUeeRzMXJZ/images/3d4bf1b5494b558533d2eb411d055d8b9606d8bf0c2f9c2090f940d59e9107c9.jpg b/parse/train/tUeeRzMXJZ/images/3d4bf1b5494b558533d2eb411d055d8b9606d8bf0c2f9c2090f940d59e9107c9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c24b94ca7d69a72641edf04e8fff3b167919791c --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/3d4bf1b5494b558533d2eb411d055d8b9606d8bf0c2f9c2090f940d59e9107c9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e6b6120ade5b9a9b3324f5788baab6e7a7ac438dfd4a60a4f518dbb0746772c +size 8793 diff --git a/parse/train/tUeeRzMXJZ/images/3d71bb01073b4ccc29a099e99cdb27b1c6751c6d42a1a8d6983d67e07de043b1.jpg b/parse/train/tUeeRzMXJZ/images/3d71bb01073b4ccc29a099e99cdb27b1c6751c6d42a1a8d6983d67e07de043b1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..35efbb889fa99acf3c7a3cc174d92667a00d88fd --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/3d71bb01073b4ccc29a099e99cdb27b1c6751c6d42a1a8d6983d67e07de043b1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2bbe0053b8f5f4d61e8a4938f393db203f01e3c176d80aab6bee5fd259a9872 +size 8611 diff --git a/parse/train/tUeeRzMXJZ/images/41e2c695bfb18eb02bb514dfc61496934865e2e52e5f44e1b92dc538cb2c0a0d.jpg b/parse/train/tUeeRzMXJZ/images/41e2c695bfb18eb02bb514dfc61496934865e2e52e5f44e1b92dc538cb2c0a0d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0452066eb693118b8b8018cee6088de948a83b4c --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/41e2c695bfb18eb02bb514dfc61496934865e2e52e5f44e1b92dc538cb2c0a0d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:560b7d8f2835fba271fdf79ab35e730bd58d37cf3f153314122037ec1257da20 +size 9026 diff --git a/parse/train/tUeeRzMXJZ/images/481eebe3f789e03295bf20f2b0dcb79112ae4b052985e502458bb7aec2c7f22b.jpg b/parse/train/tUeeRzMXJZ/images/481eebe3f789e03295bf20f2b0dcb79112ae4b052985e502458bb7aec2c7f22b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..22600f17e24f134c3812bfac22cffd2a50d2d9b3 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/481eebe3f789e03295bf20f2b0dcb79112ae4b052985e502458bb7aec2c7f22b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc5866140187bf5995cfa1de521975513f03dc0880111104a00ee3bc1dc8f48c +size 10648 diff --git a/parse/train/tUeeRzMXJZ/images/49129cab931398ea35a2ba4607da169e4ea87a2936eba77511ba5631d40fc304.jpg b/parse/train/tUeeRzMXJZ/images/49129cab931398ea35a2ba4607da169e4ea87a2936eba77511ba5631d40fc304.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5c1c429ede7667108985f1b31ec4138bc2f799b0 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/49129cab931398ea35a2ba4607da169e4ea87a2936eba77511ba5631d40fc304.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:becbf9fbac26782c8be9c0beb03e0b71f7566a562f26005a766b1fcd37a9a0ae +size 9701 diff --git a/parse/train/tUeeRzMXJZ/images/51177b65c88dd97314c6231a832f5ffcf96cba9ffa6b4d70beaefce6af3ae24a.jpg b/parse/train/tUeeRzMXJZ/images/51177b65c88dd97314c6231a832f5ffcf96cba9ffa6b4d70beaefce6af3ae24a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..218ab4294e7beebb9cdf91796236952dda4845be --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/51177b65c88dd97314c6231a832f5ffcf96cba9ffa6b4d70beaefce6af3ae24a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9859faad10c3f71a332ff84b899608b6a42b249f96bde133f029e6b3f1e3f33 +size 8432 diff --git a/parse/train/tUeeRzMXJZ/images/6be6f15e2613eb07b18ddbbd97bf9d67d9e9dd4e0760cd868cd22d59987fd94e.jpg b/parse/train/tUeeRzMXJZ/images/6be6f15e2613eb07b18ddbbd97bf9d67d9e9dd4e0760cd868cd22d59987fd94e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..92aabc3e6d8ac2572b1887e7d384506a679d77cc --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/6be6f15e2613eb07b18ddbbd97bf9d67d9e9dd4e0760cd868cd22d59987fd94e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eee7b51c8f938bb6640d28ef553ef30936f7acccd237b4fbb3f994564a66cd5a +size 10045 diff --git a/parse/train/tUeeRzMXJZ/images/83e5ded32ea6e94f617dbe297da602a8cb66de0afead66b3b87f0c3621dab677.jpg b/parse/train/tUeeRzMXJZ/images/83e5ded32ea6e94f617dbe297da602a8cb66de0afead66b3b87f0c3621dab677.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8de3cae3d4ddae52922819531858f39c9589faf9 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/83e5ded32ea6e94f617dbe297da602a8cb66de0afead66b3b87f0c3621dab677.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:924680436b14d5525c0d990a3a39ac91698fe9bb36ae9635865348c31a6f99ed +size 10274 diff --git a/parse/train/tUeeRzMXJZ/images/8462f13d3c79f88b8c48d3e995a4fc53894bbdd733d1629a3a796c0480e32225.jpg b/parse/train/tUeeRzMXJZ/images/8462f13d3c79f88b8c48d3e995a4fc53894bbdd733d1629a3a796c0480e32225.jpg new file mode 100644 index 0000000000000000000000000000000000000000..28271cb9df66c9f80501400048ec2c26fd2a422a --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/8462f13d3c79f88b8c48d3e995a4fc53894bbdd733d1629a3a796c0480e32225.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74ec5f3f34c88b45bb4de4d244e950b31d98c57b2ee7fe9b2281e7a46ac708f9 +size 14506 diff --git a/parse/train/tUeeRzMXJZ/images/994e4219bdef564248594be4ae72dbec81da8c0f7ebf8889fbf98279250036de.jpg b/parse/train/tUeeRzMXJZ/images/994e4219bdef564248594be4ae72dbec81da8c0f7ebf8889fbf98279250036de.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aab1250d3c779a1363cacda52c40fcd35444ce1b --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/994e4219bdef564248594be4ae72dbec81da8c0f7ebf8889fbf98279250036de.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37a054ee32dfdf372eb65cd7d51542d509ea5b99dc6f78a273d1f07818b4725a +size 8865 diff --git a/parse/train/tUeeRzMXJZ/images/a4aa6b31c3f9d9796703f72bb0af44d1e877087ea43a4d60a5f1ad96bbed5380.jpg b/parse/train/tUeeRzMXJZ/images/a4aa6b31c3f9d9796703f72bb0af44d1e877087ea43a4d60a5f1ad96bbed5380.jpg new file mode 100644 index 0000000000000000000000000000000000000000..42b449aec06ac722a5a4a4855f4fcfb39ae4d013 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/a4aa6b31c3f9d9796703f72bb0af44d1e877087ea43a4d60a5f1ad96bbed5380.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9fe1821d3e8f115a680d48f208b0f07f93c07d0e9efe2327caa539cc64a80515 +size 8491 diff --git a/parse/train/tUeeRzMXJZ/images/aa8c03ce0fab0192ad8c18b39a01daaa6ff0d1849fb37f5f490992090ccff2f4.jpg b/parse/train/tUeeRzMXJZ/images/aa8c03ce0fab0192ad8c18b39a01daaa6ff0d1849fb37f5f490992090ccff2f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f795ad5abb945a1a8539b4223e5bc2de16c9bafa --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/aa8c03ce0fab0192ad8c18b39a01daaa6ff0d1849fb37f5f490992090ccff2f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e308e068424280f25f1757fea5b68633a856c010617ad7364b722738321942f +size 12693 diff --git a/parse/train/tUeeRzMXJZ/images/b82b627fde2c397c2b10446ea970c2a3725e9791eacad3503fd29412bf5a587e.jpg b/parse/train/tUeeRzMXJZ/images/b82b627fde2c397c2b10446ea970c2a3725e9791eacad3503fd29412bf5a587e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..546c55e253adc8f6888426620715ace984ddcb92 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/b82b627fde2c397c2b10446ea970c2a3725e9791eacad3503fd29412bf5a587e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3665152a8a6e5128b8fc7adea40b646aa326d46e0c76d7bd6400a49eb9767aaf +size 7770 diff --git a/parse/train/tUeeRzMXJZ/images/b8e32641d96953b892d012f75bdef243b53a934eff77e59cf9ee6214a8027913.jpg b/parse/train/tUeeRzMXJZ/images/b8e32641d96953b892d012f75bdef243b53a934eff77e59cf9ee6214a8027913.jpg new file mode 100644 index 0000000000000000000000000000000000000000..27091a65c3839b450d3ec6a5ba5590ecd04b3785 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/b8e32641d96953b892d012f75bdef243b53a934eff77e59cf9ee6214a8027913.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddc7340ec2ab8d50d24f74e182975b1cbcf8d62ca4ce800dd03b3dc0175f0616 +size 14996 diff --git a/parse/train/tUeeRzMXJZ/images/b9aaf7bae6d9cc9f0c16f9eec91864874bb1c150456627caa2d7d753e6be4181.jpg b/parse/train/tUeeRzMXJZ/images/b9aaf7bae6d9cc9f0c16f9eec91864874bb1c150456627caa2d7d753e6be4181.jpg new file mode 100644 index 0000000000000000000000000000000000000000..20d69b925db10d4b5bb1d0271a34b567781d6ee5 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/b9aaf7bae6d9cc9f0c16f9eec91864874bb1c150456627caa2d7d753e6be4181.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38d8c17c9797352b604d42a632b9fafddf416d18ecab601416cb09aea8613a69 +size 9731 diff --git a/parse/train/tUeeRzMXJZ/images/c66ff3bd2d8f5aabe69542823d651c069fe4f25a7d63286a666ce9b54c20f57a.jpg b/parse/train/tUeeRzMXJZ/images/c66ff3bd2d8f5aabe69542823d651c069fe4f25a7d63286a666ce9b54c20f57a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..01da100cf03ada0481f9619afcdac1e3d2c1cddc --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/c66ff3bd2d8f5aabe69542823d651c069fe4f25a7d63286a666ce9b54c20f57a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ee8f171d02d42c0197a7c8e89baf3765093441c8ffc137964e05e355f47849a +size 9773 diff --git a/parse/train/tUeeRzMXJZ/images/d246d52e6cd291b5d9112b4a63cc7980f6324f3d35cc2cc9e7d00a5121e878be.jpg b/parse/train/tUeeRzMXJZ/images/d246d52e6cd291b5d9112b4a63cc7980f6324f3d35cc2cc9e7d00a5121e878be.jpg new file mode 100644 index 0000000000000000000000000000000000000000..19f667a54d3c9ebe91545a77d4e4e778ff73ede9 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/d246d52e6cd291b5d9112b4a63cc7980f6324f3d35cc2cc9e7d00a5121e878be.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e036e1112a2b41b4b71e26a38fd2eb8597b6dc06697ce97092e645216c49207c +size 7763 diff --git a/parse/train/tUeeRzMXJZ/images/d65c9f481f5cb9b2a12c0df6b5b8a9420c2214d472f0b5ede89e07d85852a305.jpg b/parse/train/tUeeRzMXJZ/images/d65c9f481f5cb9b2a12c0df6b5b8a9420c2214d472f0b5ede89e07d85852a305.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4aaa9676e4cfb99f2412a9738ffa0d4b5713e57e --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/d65c9f481f5cb9b2a12c0df6b5b8a9420c2214d472f0b5ede89e07d85852a305.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32e185764b95ede454ffaa44cdeb344e2525f1825a3d3da5a1b2b26e1901a9f2 +size 12190 diff --git a/parse/train/tUeeRzMXJZ/images/e870ea1008b81d480b706dee5a5ae15005be28db51268a0c533e4a1db300eda3.jpg b/parse/train/tUeeRzMXJZ/images/e870ea1008b81d480b706dee5a5ae15005be28db51268a0c533e4a1db300eda3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1a789bcc8e65e1106cd1be20bc4375d3245879a3 --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/e870ea1008b81d480b706dee5a5ae15005be28db51268a0c533e4a1db300eda3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:708d2f2b560ce0344536e9bf380192ff7e937f4db44d4fae8bceffd9fb8774c7 +size 19474 diff --git a/parse/train/tUeeRzMXJZ/images/f68bfebcc9a1a30f964126981d1b7792b1239c770379f8cedc8935cb4aa9df30.jpg b/parse/train/tUeeRzMXJZ/images/f68bfebcc9a1a30f964126981d1b7792b1239c770379f8cedc8935cb4aa9df30.jpg new file mode 100644 index 0000000000000000000000000000000000000000..29d520d4ca2479d753efa5beffd5c78a1394388a --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/f68bfebcc9a1a30f964126981d1b7792b1239c770379f8cedc8935cb4aa9df30.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b9f2054938d6ed6931d82171dff4bc4aac07ae0443122cdc5031ee2e9ff1965 +size 17410 diff --git a/parse/train/tUeeRzMXJZ/images/fa812e4d1a72af3a441491ece0e503dfa5d5d4fc3fb3f61e57797a91312491fe.jpg b/parse/train/tUeeRzMXJZ/images/fa812e4d1a72af3a441491ece0e503dfa5d5d4fc3fb3f61e57797a91312491fe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f75ff2bd689cf2cfaa38b73702bf3a898cdcc80d --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/fa812e4d1a72af3a441491ece0e503dfa5d5d4fc3fb3f61e57797a91312491fe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:edc2b7f75033c12d5ff184db3a4394553fa88664f63d5a17f5238b46266ce181 +size 9563 diff --git a/parse/train/tUeeRzMXJZ/images/fc242512d131e0246c4a3f1a2122f7150bde2eca97f93e0ff5f87914970bb452.jpg b/parse/train/tUeeRzMXJZ/images/fc242512d131e0246c4a3f1a2122f7150bde2eca97f93e0ff5f87914970bb452.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4b192952b7b3621efbd746dcdf32c37c01d468bc --- /dev/null +++ b/parse/train/tUeeRzMXJZ/images/fc242512d131e0246c4a3f1a2122f7150bde2eca97f93e0ff5f87914970bb452.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7209103e3d5e238b36c8a1a9fa6b2a9eff8d56f836f6b15b9fb7a96f31bafa47 +size 21894