aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1710.00132 | 2762736764 | For intelligent robotics applications, extending 3D mapping to 3D semantic mapping enables robots to, not only localize themselves with respect to the scene's geometrical features but also simultaneously understand the higher level meaning of the scene contexts. Most previous methods focus on geometric 3D reconstruction and scene understanding independently notwithstanding the fact that joint estimation can boost the accuracy of the semantic mapping. In this paper, a dense RGB-D semantic mapping system with a Pixel-Voxel network is proposed, which can perform dense 3D mapping while simultaneously recognizing and semantically labelling each point in the 3D map. The proposed Pixel-Voxel network obtains global context information by using PixelNet to exploit the RGB image and meanwhile, preserves accurate local shape information by using VoxelNet to exploit the corresponding 3D point cloud. Unlike the existing architecture that fuses score maps from different models with equal weights, we proposed a Softmax weighted fusion stack that adaptively learns the varying contributions of PixelNet and VoxelNet, and fuses the score maps of the two models according to their respective confidence levels. The proposed Pixel-Voxel network achieves the state-of-the-art semantic segmentation performance on the SUN RGB-D benchmark dataset. The runtime of the proposed system can be boosted to 11-12Hz, enabling near to real-time performance using an i7 8-cores PC with Titan X GPU. | All the above methods mainly focus on semantic segmentation using a single image and they only perform 3D label refinement through a recursive Bayesian update using a sequence of images. However, they do not take full advantage of the associated information provided by multiple viewpoints of a scene. Yu @cite_21 proposed a DA-RNN integrated with Kinect Fusion @cite_5 for 3D semantic mapping. DA-RNN employs a recurrent neural network to tightly combine the information contained in multiple viewpoints of an RGB-D video stream to improve the semantic segmentation performance. Ma @cite_24 proposed a multi-view consistency layer which can use multi-view context information for object-class segmentation from multiple RGB-D views. It utilizes the visual odometry trajectory from RGB-D SLAM @cite_23 to wrap semantic segmentation between two viewpoints. In addition, Armin @cite_0 proposed a network architecture for spatially and temporally coherent semantic co-segmentation and mapping of complex dynamic scenes from multiple static or moving cameras. | {
"cite_N": [
"@cite_21",
"@cite_24",
"@cite_0",
"@cite_23",
"@cite_5"
],
"mid": [
"2604365427",
"2951620021",
"",
"2069479606",
"1987648924"
],
"abstract": [
"3D scene understanding is important for robots to interact with the 3D world in a meaningful way. Most previous works on 3D scene understanding focus on recognizing geometrical or semantic properties of the scene independently. In this work, we introduce Data Associated Recurrent Neural Networks (DA-RNNs), a novel framework for joint 3D scene mapping and semantic labeling. DA-RNNs use a new recurrent neural network architecture for semantic labeling on RGB-D videos. The output of the network is integrated with mapping techniques such as KinectFusion in order to inject semantic information into the reconstructed 3D scene. Experiments conducted on a real world dataset and a synthetic dataset with RGB-D videos demonstrate the ability of our method in semantic 3D scene mapping.",
"Visual scene understanding is an important capability that enables robots to purposefully act in their environment. In this paper, we propose a novel approach to object-class segmentation from multiple RGB-D views using deep learning. We train a deep neural network to predict object-class semantics that is consistent from several view points in a semi-supervised way. At test time, the semantics predictions of our network can be fused more consistently in semantic keyframe maps than predictions of a network trained on individual views. We base our network architecture on a recent single-view deep learning approach to RGB and depth fusion for semantic object-class segmentation and enhance it with multi-scale loss minimization. We obtain the camera trajectory using RGB-D SLAM and warp the predictions of RGB-D images into ground-truth annotated frames in order to enforce multi-view consistency during training. At test time, predictions from multiple views are fused into keyframes. We propose and analyze several methods for enforcing multi-view consistency during training and testing. We evaluate the benefit of multi-view consistency training and demonstrate that pooling of deep features and fusion over multiple views outperforms single-view baselines on the NYUDv2 benchmark for semantic segmentation. Our end-to-end trained network achieves state-of-the-art performance on the NYUDv2 dataset in single-view segmentation as well as multi-view semantic fusion.",
"",
"In this paper, we present a novel mapping system that robustly generates highly accurate 3-D maps using an RGB-D camera. Our approach requires no further sensors or odometry. With the availability of low-cost and light-weight RGB-D sensors such as the Microsoft Kinect, our approach applies to small domestic robots such as vacuum cleaners, as well as flying robots such as quadrocopters. Furthermore, our system can also be used for free-hand reconstruction of detailed 3-D models. In addition to the system itself, we present a thorough experimental evaluation on a publicly available benchmark dataset. We analyze and discuss the influence of several parameters such as the choice of the feature descriptor, the number of visual features, and validation methods. The results of the experiments demonstrate that our system can robustly deal with challenging scenarios such as fast camera motions and feature-poor environments while being fast enough for online operation. Our system is fully available as open source and has already been widely adopted by the robotics community.",
"We present a system for accurate real-time mapping of complex and arbitrary indoor scenes in variable lighting conditions, using only a moving low-cost depth camera and commodity graphics hardware. We fuse all of the depth data streamed from a Kinect sensor into a single global implicit surface model of the observed scene in real-time. The current sensor pose is simultaneously obtained by tracking the live depth frame relative to the global model using a coarse-to-fine iterative closest point (ICP) algorithm, which uses all of the observed depth data available. We demonstrate the advantages of tracking against the growing full surface model compared with frame-to-frame tracking, obtaining tracking and mapping results in constant time within room sized scenes with limited drift and high accuracy. We also show both qualitative and quantitative results relating to various aspects of our tracking and mapping system. Modelling of natural scenes, in real-time with only commodity sensor and GPU hardware, promises an exciting step forward in augmented reality (AR), in particular, it allows dense surfaces to be reconstructed in real-time, with a level of detail and robustness beyond any solution yet presented using passive computer vision."
]
} |
1710.00132 | 2762736764 | For intelligent robotics applications, extending 3D mapping to 3D semantic mapping enables robots to, not only localize themselves with respect to the scene's geometrical features but also simultaneously understand the higher level meaning of the scene contexts. Most previous methods focus on geometric 3D reconstruction and scene understanding independently notwithstanding the fact that joint estimation can boost the accuracy of the semantic mapping. In this paper, a dense RGB-D semantic mapping system with a Pixel-Voxel network is proposed, which can perform dense 3D mapping while simultaneously recognizing and semantically labelling each point in the 3D map. The proposed Pixel-Voxel network obtains global context information by using PixelNet to exploit the RGB image and meanwhile, preserves accurate local shape information by using VoxelNet to exploit the corresponding 3D point cloud. Unlike the existing architecture that fuses score maps from different models with equal weights, we proposed a Softmax weighted fusion stack that adaptively learns the varying contributions of PixelNet and VoxelNet, and fuses the score maps of the two models according to their respective confidence levels. The proposed Pixel-Voxel network achieves the state-of-the-art semantic segmentation performance on the SUN RGB-D benchmark dataset. The runtime of the proposed system can be boosted to 11-12Hz, enabling near to real-time performance using an i7 8-cores PC with Titan X GPU. | According to the type of input data, semantic segmentation can be further grouped into three main sub-categories: RGB @cite_15 @cite_6 @cite_18 @cite_14 @cite_2 , RGB-D @cite_11 @cite_22 @cite_4 and 3D @cite_7 @cite_17 semantic segmentation. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_4",
"@cite_22",
"@cite_7",
"@cite_17",
"@cite_6",
"@cite_2",
"@cite_15",
"@cite_11"
],
"mid": [
"2412782625",
"2204696980",
"",
"2489780108",
"2950642167",
"2624503621",
"2963881378",
"1745334888",
"1903029394",
"2587989515"
],
"abstract": [
"In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First , we highlight convolution with upsampled filters, or ‘atrous convolution’, as a powerful tool in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second , we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third , we improve the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models. The commonly deployed combination of max-pooling and downsampling in DCNNs achieves invariance but has a toll on localization accuracy. We overcome this by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF), which is shown both qualitatively and quantitatively to improve localization performance. Our proposed “DeepLab” system sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 79.7 percent mIOU in the test set, and advances the results on three other datasets: PASCAL-Context, PASCAL-Person-Part, and Cityscapes. All of our code is made publicly available online.",
"In image labeling, local representations for image units are usually generated from their surrounding image patches, thus long-range contextual information is not effectively encoded. In this paper, we introduce recurrent neural networks (RNNs) to address this issue. Specifically, directed acyclic graph RNNs (DAG-RNNs) are proposed to process DAG-structured images, which enables the network to model long-range semantic dependencies among image units. Our DAG-RNNs are capable of tremendously enhancing the discriminative power of local representations, which significantly benefits the local classification. Meanwhile, we propose a novel class weighting function that attends to rare classes, which phenomenally boosts the recognition accuracy for non-frequent classes. Integrating with convolution and deconvolution layers, our DAG-RNNs achieve new state-of-the-art results on the challenging SiftFlow, CamVid and Barcelona benchmarks.",
"",
"Semantic labeling of RGB-D scenes is crucial to many intelligent applications including perceptual robotics. It generates pixelwise and fine-grained label maps from simultaneously sensed photometric (RGB) and depth channels. This paper addresses this problem by (i) developing a novel Long Short-Term Memorized Context Fusion (LSTM-CF) Model that captures and fuses contextual information from multiple channels of photometric and depth data, and (ii) incorporating this model into deep convolutional neural networks (CNNs) for end-to-end training. Specifically, contexts in photometric and depth channels are, respectively, captured by stacking several convolutional layers and a long short-term memory layer; the memory layer encodes both short-range and long-range spatial dependencies in an image along the vertical direction. Another long short-term memorized fusion layer is set up to integrate the contexts along the vertical direction from different channels, and perform bi-directional propagation of the fused vertical contexts along the horizontal direction to obtain true 2D global contexts. At last, the fused contextual representation is concatenated with the convolutional features extracted from the photometric channels in order to improve the accuracy of fine-scale semantic labeling. Our proposed model has set a new state of the art, i.e., ( 48.1 ) and ( 49.4 ) average class accuracy over 37 categories ( ( 2.2 ) and ( 5.4 ) improvement) on the large-scale SUNRGBD dataset and the NYUDv2 dataset, respectively.",
"Point cloud is an important type of geometric data structure. Due to its irregular format, most researchers transform such data to regular 3D voxel grids or collections of images. This, however, renders data unnecessarily voluminous and causes issues. In this paper, we design a novel type of neural network that directly consumes point clouds and well respects the permutation invariance of points in the input. Our network, named PointNet, provides a unified architecture for applications ranging from object classification, part segmentation, to scene semantic parsing. Though simple, PointNet is highly efficient and effective. Empirically, it shows strong performance on par or even better than state of the art. Theoretically, we provide analysis towards understanding of what the network has learnt and why the network is robust with respect to input perturbation and corruption.",
"Few prior works study deep learning on point sets. PointNet by is a pioneer in this direction. However, by design PointNet does not capture local structures induced by the metric space points live in, limiting its ability to recognize fine-grained patterns and generalizability to complex scenes. In this work, we introduce a hierarchical neural network that applies PointNet recursively on a nested partitioning of the input point set. By exploiting metric space distances, our network is able to learn local features with increasing contextual scales. With further observation that point sets are usually sampled with varying densities, which results in greatly decreased performance for networks trained on uniform densities, we propose novel set learning layers to adaptively combine features from multiple scales. Experiments show that our network called PointNet++ is able to learn deep point set features efficiently and robustly. In particular, results significantly better than state-of-the-art have been obtained on challenging benchmarks of 3D point clouds.",
"We present a novel and practical deep fully convolutional neural network architecture for semantic pixel-wise segmentation termed SegNet. This core trainable segmentation engine consists of an encoder network, a corresponding decoder network followed by a pixel-wise classification layer. The architecture of the encoder network is topologically identical to the 13 convolutional layers in the VGG16 network [1] . The role of the decoder network is to map the low resolution encoder feature maps to full input resolution feature maps for pixel-wise classification. The novelty of SegNet lies is in the manner in which the decoder upsamples its lower resolution input feature map(s). Specifically, the decoder uses pooling indices computed in the max-pooling step of the corresponding encoder to perform non-linear upsampling. This eliminates the need for learning to upsample. The upsampled maps are sparse and are then convolved with trainable filters to produce dense feature maps. We compare our proposed architecture with the widely adopted FCN [2] and also with the well known DeepLab-LargeFOV [3] , DeconvNet [4] architectures. This comparison reveals the memory versus accuracy trade-off involved in achieving good segmentation performance. SegNet was primarily motivated by scene understanding applications. Hence, it is designed to be efficient both in terms of memory and computational time during inference. It is also significantly smaller in the number of trainable parameters than other competing architectures and can be trained end-to-end using stochastic gradient descent. We also performed a controlled benchmark of SegNet and other architectures on both road scenes and SUN RGB-D indoor scene segmentation tasks. These quantitative assessments show that SegNet provides good performance with competitive inference time and most efficient inference memory-wise as compared to other architectures. We also provide a Caffe implementation of SegNet and a web demo at http: mi.eng.cam.ac.uk projects segnet .",
"We propose a novel semantic segmentation algorithm by learning a deep deconvolution network. We learn the network on top of the convolutional layers adopted from VGG 16-layer net. The deconvolution network is composed of deconvolution and unpooling layers, which identify pixelwise class labels and predict segmentation masks. We apply the trained network to each proposal in an input image, and construct the final semantic segmentation map by combining the results from all proposals in a simple manner. The proposed algorithm mitigates the limitations of the existing methods based on fully convolutional networks by integrating deep deconvolution network and proposal-wise prediction, our segmentation method typically identifies detailed structures and handles objects in multiple scales naturally. Our network demonstrates outstanding performance in PASCAL VOC 2012 dataset, and we achieve the best accuracy (72.5 ) among the methods trained without using Microsoft COCO dataset through ensemble with the fully convolutional network.",
"Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build “fully convolutional” networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet [20], the VGG net [31], and GoogLeNet [32]) into fully convolutional networks and transfer their learned representations by fine-tuning [3] to the segmentation task. We then define a skip architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20 relative improvement to 62.2 mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes less than one fifth of a second for a typical image.",
"In this paper we address the problem of semantic labeling of indoor scenes on RGB-D data. With the availability of RGB-D cameras, it is expected that additional depth measurement will improve the accuracy. Here we investigate a solution how to incorporate complementary depth information into a semantic segmentation framework by making use of convolutional neural networks (CNNs). Recently encoder-decoder type fully convolutional CNN architectures have achieved a great success in the field of semantic segmentation. Motivated by this observation we propose an encoder-decoder type network, where the encoder part is composed of two branches of networks that simultaneously extract features from RGB and depth images and fuse depth features into the RGB feature maps as the network goes deeper. Comprehensive experimental evaluations demonstrate that the proposed fusion-based architecture achieves competitive results with the state-of-the-art methods on the challenging SUN RGB-D benchmark obtaining 76.27 global accuracy, 48.30 average class accuracy and 37.29 average intersection-over-union score."
]
} |
1710.00132 | 2762736764 | For intelligent robotics applications, extending 3D mapping to 3D semantic mapping enables robots to, not only localize themselves with respect to the scene's geometrical features but also simultaneously understand the higher level meaning of the scene contexts. Most previous methods focus on geometric 3D reconstruction and scene understanding independently notwithstanding the fact that joint estimation can boost the accuracy of the semantic mapping. In this paper, a dense RGB-D semantic mapping system with a Pixel-Voxel network is proposed, which can perform dense 3D mapping while simultaneously recognizing and semantically labelling each point in the 3D map. The proposed Pixel-Voxel network obtains global context information by using PixelNet to exploit the RGB image and meanwhile, preserves accurate local shape information by using VoxelNet to exploit the corresponding 3D point cloud. Unlike the existing architecture that fuses score maps from different models with equal weights, we proposed a Softmax weighted fusion stack that adaptively learns the varying contributions of PixelNet and VoxelNet, and fuses the score maps of the two models according to their respective confidence levels. The proposed Pixel-Voxel network achieves the state-of-the-art semantic segmentation performance on the SUN RGB-D benchmark dataset. The runtime of the proposed system can be boosted to 11-12Hz, enabling near to real-time performance using an i7 8-cores PC with Titan X GPU. | FCN @cite_15 is the first end-to-end fashion network instead of using hand-crafted features for semantic segmentation. It replaces the fully connected layers of the classification network with the convolution layers to output the coarse map and utilizes a skip architecture to refine it. DeconvNet @cite_2 composing of deconvolution and unpooling layers, utilizes the fractionally strided convolutions to alleviate the limited resolution of labelling problem. SegNet @cite_6 proposed an encoder-decoder architecture, which records the indices of max pooling for up-sampling. DeepLab @cite_18 makes use of dilated convolutions @cite_10 to increase the receptive field without down-sampling the feature map. CRF as RNN @cite_14 reformulates the mean-field inference in dense CRF as an RNN architecture that enables it to integrate with CNN as a fully end-to-end network. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_6",
"@cite_2",
"@cite_15",
"@cite_10"
],
"mid": [
"2412782625",
"2204696980",
"2963881378",
"1745334888",
"1903029394",
"2286929393"
],
"abstract": [
"In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First , we highlight convolution with upsampled filters, or ‘atrous convolution’, as a powerful tool in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second , we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third , we improve the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models. The commonly deployed combination of max-pooling and downsampling in DCNNs achieves invariance but has a toll on localization accuracy. We overcome this by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF), which is shown both qualitatively and quantitatively to improve localization performance. Our proposed “DeepLab” system sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 79.7 percent mIOU in the test set, and advances the results on three other datasets: PASCAL-Context, PASCAL-Person-Part, and Cityscapes. All of our code is made publicly available online.",
"In image labeling, local representations for image units are usually generated from their surrounding image patches, thus long-range contextual information is not effectively encoded. In this paper, we introduce recurrent neural networks (RNNs) to address this issue. Specifically, directed acyclic graph RNNs (DAG-RNNs) are proposed to process DAG-structured images, which enables the network to model long-range semantic dependencies among image units. Our DAG-RNNs are capable of tremendously enhancing the discriminative power of local representations, which significantly benefits the local classification. Meanwhile, we propose a novel class weighting function that attends to rare classes, which phenomenally boosts the recognition accuracy for non-frequent classes. Integrating with convolution and deconvolution layers, our DAG-RNNs achieve new state-of-the-art results on the challenging SiftFlow, CamVid and Barcelona benchmarks.",
"We present a novel and practical deep fully convolutional neural network architecture for semantic pixel-wise segmentation termed SegNet. This core trainable segmentation engine consists of an encoder network, a corresponding decoder network followed by a pixel-wise classification layer. The architecture of the encoder network is topologically identical to the 13 convolutional layers in the VGG16 network [1] . The role of the decoder network is to map the low resolution encoder feature maps to full input resolution feature maps for pixel-wise classification. The novelty of SegNet lies is in the manner in which the decoder upsamples its lower resolution input feature map(s). Specifically, the decoder uses pooling indices computed in the max-pooling step of the corresponding encoder to perform non-linear upsampling. This eliminates the need for learning to upsample. The upsampled maps are sparse and are then convolved with trainable filters to produce dense feature maps. We compare our proposed architecture with the widely adopted FCN [2] and also with the well known DeepLab-LargeFOV [3] , DeconvNet [4] architectures. This comparison reveals the memory versus accuracy trade-off involved in achieving good segmentation performance. SegNet was primarily motivated by scene understanding applications. Hence, it is designed to be efficient both in terms of memory and computational time during inference. It is also significantly smaller in the number of trainable parameters than other competing architectures and can be trained end-to-end using stochastic gradient descent. We also performed a controlled benchmark of SegNet and other architectures on both road scenes and SUN RGB-D indoor scene segmentation tasks. These quantitative assessments show that SegNet provides good performance with competitive inference time and most efficient inference memory-wise as compared to other architectures. We also provide a Caffe implementation of SegNet and a web demo at http: mi.eng.cam.ac.uk projects segnet .",
"We propose a novel semantic segmentation algorithm by learning a deep deconvolution network. We learn the network on top of the convolutional layers adopted from VGG 16-layer net. The deconvolution network is composed of deconvolution and unpooling layers, which identify pixelwise class labels and predict segmentation masks. We apply the trained network to each proposal in an input image, and construct the final semantic segmentation map by combining the results from all proposals in a simple manner. The proposed algorithm mitigates the limitations of the existing methods based on fully convolutional networks by integrating deep deconvolution network and proposal-wise prediction, our segmentation method typically identifies detailed structures and handles objects in multiple scales naturally. Our network demonstrates outstanding performance in PASCAL VOC 2012 dataset, and we achieve the best accuracy (72.5 ) among the methods trained without using Microsoft COCO dataset through ensemble with the fully convolutional network.",
"Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build “fully convolutional” networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet [20], the VGG net [31], and GoogLeNet [32]) into fully convolutional networks and transfer their learned representations by fine-tuning [3] to the segmentation task. We then define a skip architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20 relative improvement to 62.2 mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes less than one fifth of a second for a typical image.",
"State-of-the-art models for semantic segmentation are based on adaptations of convolutional networks that had originally been designed for image classification. However, dense prediction and image classification are structurally different. In this work, we develop a new convolutional network module that is specifically designed for dense prediction. The presented module uses dilated convolutions to systematically aggregate multi-scale contextual information without losing resolution. The architecture is based on the fact that dilated convolutions support exponential expansion of the receptive field without loss of resolution or coverage. We show that the presented context module increases the accuracy of state-of-the-art semantic segmentation systems. In addition, we examine the adaptation of image classification networks to dense prediction and show that simplifying the adapted network can increase accuracy."
]
} |
1710.00132 | 2762736764 | For intelligent robotics applications, extending 3D mapping to 3D semantic mapping enables robots to, not only localize themselves with respect to the scene's geometrical features but also simultaneously understand the higher level meaning of the scene contexts. Most previous methods focus on geometric 3D reconstruction and scene understanding independently notwithstanding the fact that joint estimation can boost the accuracy of the semantic mapping. In this paper, a dense RGB-D semantic mapping system with a Pixel-Voxel network is proposed, which can perform dense 3D mapping while simultaneously recognizing and semantically labelling each point in the 3D map. The proposed Pixel-Voxel network obtains global context information by using PixelNet to exploit the RGB image and meanwhile, preserves accurate local shape information by using VoxelNet to exploit the corresponding 3D point cloud. Unlike the existing architecture that fuses score maps from different models with equal weights, we proposed a Softmax weighted fusion stack that adaptively learns the varying contributions of PixelNet and VoxelNet, and fuses the score maps of the two models according to their respective confidence levels. The proposed Pixel-Voxel network achieves the state-of-the-art semantic segmentation performance on the SUN RGB-D benchmark dataset. The runtime of the proposed system can be boosted to 11-12Hz, enabling near to real-time performance using an i7 8-cores PC with Titan X GPU. | FuseNet @cite_11 can fuse RGB and depth image cues in a single encoder-decoder CNN architecture for RGB-D semantic segmentation. LSTM-CF @cite_22 network fuses contextual information from multiple channels of RGB and depth image through stacking several convolution layers and a long short-term memory layer. FuseNet normalises the depth value into the interval of @math to have the same spatial range as color images, while LSTM-CF network transforms depth image to HHA image to have 3 channels as the color image. The HHA representation can improve the depth semantic segmentation, however, HHA representation requires high computational cost and hence cannot be performed in real-time. In addition, STD2P @cite_4 proposes a novel superpixel-based multi-view convolutional neural network for RGB-D semantic segmentation, which uses Spatio-temporal pooling layer to aggregate information over space and time. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_11"
],
"mid": [
"",
"2489780108",
"2587989515"
],
"abstract": [
"",
"Semantic labeling of RGB-D scenes is crucial to many intelligent applications including perceptual robotics. It generates pixelwise and fine-grained label maps from simultaneously sensed photometric (RGB) and depth channels. This paper addresses this problem by (i) developing a novel Long Short-Term Memorized Context Fusion (LSTM-CF) Model that captures and fuses contextual information from multiple channels of photometric and depth data, and (ii) incorporating this model into deep convolutional neural networks (CNNs) for end-to-end training. Specifically, contexts in photometric and depth channels are, respectively, captured by stacking several convolutional layers and a long short-term memory layer; the memory layer encodes both short-range and long-range spatial dependencies in an image along the vertical direction. Another long short-term memorized fusion layer is set up to integrate the contexts along the vertical direction from different channels, and perform bi-directional propagation of the fused vertical contexts along the horizontal direction to obtain true 2D global contexts. At last, the fused contextual representation is concatenated with the convolutional features extracted from the photometric channels in order to improve the accuracy of fine-scale semantic labeling. Our proposed model has set a new state of the art, i.e., ( 48.1 ) and ( 49.4 ) average class accuracy over 37 categories ( ( 2.2 ) and ( 5.4 ) improvement) on the large-scale SUNRGBD dataset and the NYUDv2 dataset, respectively.",
"In this paper we address the problem of semantic labeling of indoor scenes on RGB-D data. With the availability of RGB-D cameras, it is expected that additional depth measurement will improve the accuracy. Here we investigate a solution how to incorporate complementary depth information into a semantic segmentation framework by making use of convolutional neural networks (CNNs). Recently encoder-decoder type fully convolutional CNN architectures have achieved a great success in the field of semantic segmentation. Motivated by this observation we propose an encoder-decoder type network, where the encoder part is composed of two branches of networks that simultaneously extract features from RGB and depth images and fuse depth features into the RGB feature maps as the network goes deeper. Comprehensive experimental evaluations demonstrate that the proposed fusion-based architecture achieves competitive results with the state-of-the-art methods on the challenging SUN RGB-D benchmark obtaining 76.27 global accuracy, 48.30 average class accuracy and 37.29 average intersection-over-union score."
]
} |
1710.00132 | 2762736764 | For intelligent robotics applications, extending 3D mapping to 3D semantic mapping enables robots to, not only localize themselves with respect to the scene's geometrical features but also simultaneously understand the higher level meaning of the scene contexts. Most previous methods focus on geometric 3D reconstruction and scene understanding independently notwithstanding the fact that joint estimation can boost the accuracy of the semantic mapping. In this paper, a dense RGB-D semantic mapping system with a Pixel-Voxel network is proposed, which can perform dense 3D mapping while simultaneously recognizing and semantically labelling each point in the 3D map. The proposed Pixel-Voxel network obtains global context information by using PixelNet to exploit the RGB image and meanwhile, preserves accurate local shape information by using VoxelNet to exploit the corresponding 3D point cloud. Unlike the existing architecture that fuses score maps from different models with equal weights, we proposed a Softmax weighted fusion stack that adaptively learns the varying contributions of PixelNet and VoxelNet, and fuses the score maps of the two models according to their respective confidence levels. The proposed Pixel-Voxel network achieves the state-of-the-art semantic segmentation performance on the SUN RGB-D benchmark dataset. The runtime of the proposed system can be boosted to 11-12Hz, enabling near to real-time performance using an i7 8-cores PC with Titan X GPU. | The forerunner work PointNet @cite_7 provides a unified architecture for both classification and segmentation which consumes the raw unordered point clouds as input. PointNet only employs a single max-pooling to generate the global feature which describes the original input clouds, thus it does not capture the local structures induced by the 3D metric space points live in. In the improved version PointNet++ @cite_17 , it proposed a hierarchical neural network. It applies PointNet recursively on a nested partitioning of the input point set, which enables it to learn local features with increasing contextual scales. | {
"cite_N": [
"@cite_7",
"@cite_17"
],
"mid": [
"2950642167",
"2624503621"
],
"abstract": [
"Point cloud is an important type of geometric data structure. Due to its irregular format, most researchers transform such data to regular 3D voxel grids or collections of images. This, however, renders data unnecessarily voluminous and causes issues. In this paper, we design a novel type of neural network that directly consumes point clouds and well respects the permutation invariance of points in the input. Our network, named PointNet, provides a unified architecture for applications ranging from object classification, part segmentation, to scene semantic parsing. Though simple, PointNet is highly efficient and effective. Empirically, it shows strong performance on par or even better than state of the art. Theoretically, we provide analysis towards understanding of what the network has learnt and why the network is robust with respect to input perturbation and corruption.",
"Few prior works study deep learning on point sets. PointNet by is a pioneer in this direction. However, by design PointNet does not capture local structures induced by the metric space points live in, limiting its ability to recognize fine-grained patterns and generalizability to complex scenes. In this work, we introduce a hierarchical neural network that applies PointNet recursively on a nested partitioning of the input point set. By exploiting metric space distances, our network is able to learn local features with increasing contextual scales. With further observation that point sets are usually sampled with varying densities, which results in greatly decreased performance for networks trained on uniform densities, we propose novel set learning layers to adaptively combine features from multiple scales. Experiments show that our network called PointNet++ is able to learn deep point set features efficiently and robustly. In particular, results significantly better than state-of-the-art have been obtained on challenging benchmarks of 3D point clouds."
]
} |
1710.00132 | 2762736764 | For intelligent robotics applications, extending 3D mapping to 3D semantic mapping enables robots to, not only localize themselves with respect to the scene's geometrical features but also simultaneously understand the higher level meaning of the scene contexts. Most previous methods focus on geometric 3D reconstruction and scene understanding independently notwithstanding the fact that joint estimation can boost the accuracy of the semantic mapping. In this paper, a dense RGB-D semantic mapping system with a Pixel-Voxel network is proposed, which can perform dense 3D mapping while simultaneously recognizing and semantically labelling each point in the 3D map. The proposed Pixel-Voxel network obtains global context information by using PixelNet to exploit the RGB image and meanwhile, preserves accurate local shape information by using VoxelNet to exploit the corresponding 3D point cloud. Unlike the existing architecture that fuses score maps from different models with equal weights, we proposed a Softmax weighted fusion stack that adaptively learns the varying contributions of PixelNet and VoxelNet, and fuses the score maps of the two models according to their respective confidence levels. The proposed Pixel-Voxel network achieves the state-of-the-art semantic segmentation performance on the SUN RGB-D benchmark dataset. The runtime of the proposed system can be boosted to 11-12Hz, enabling near to real-time performance using an i7 8-cores PC with Titan X GPU. | In addition, the network in @cite_15 @cite_11 @cite_22 all simply fuse the score maps from different models using equal weights. Each model should have the different contributions in different situations for different categories. So in this paper, a Softmax weighted fusion stack is proposed for adoptively learning the varying contributions of each model. | {
"cite_N": [
"@cite_15",
"@cite_22",
"@cite_11"
],
"mid": [
"1903029394",
"2489780108",
"2587989515"
],
"abstract": [
"Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build “fully convolutional” networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet [20], the VGG net [31], and GoogLeNet [32]) into fully convolutional networks and transfer their learned representations by fine-tuning [3] to the segmentation task. We then define a skip architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20 relative improvement to 62.2 mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes less than one fifth of a second for a typical image.",
"Semantic labeling of RGB-D scenes is crucial to many intelligent applications including perceptual robotics. It generates pixelwise and fine-grained label maps from simultaneously sensed photometric (RGB) and depth channels. This paper addresses this problem by (i) developing a novel Long Short-Term Memorized Context Fusion (LSTM-CF) Model that captures and fuses contextual information from multiple channels of photometric and depth data, and (ii) incorporating this model into deep convolutional neural networks (CNNs) for end-to-end training. Specifically, contexts in photometric and depth channels are, respectively, captured by stacking several convolutional layers and a long short-term memory layer; the memory layer encodes both short-range and long-range spatial dependencies in an image along the vertical direction. Another long short-term memorized fusion layer is set up to integrate the contexts along the vertical direction from different channels, and perform bi-directional propagation of the fused vertical contexts along the horizontal direction to obtain true 2D global contexts. At last, the fused contextual representation is concatenated with the convolutional features extracted from the photometric channels in order to improve the accuracy of fine-scale semantic labeling. Our proposed model has set a new state of the art, i.e., ( 48.1 ) and ( 49.4 ) average class accuracy over 37 categories ( ( 2.2 ) and ( 5.4 ) improvement) on the large-scale SUNRGBD dataset and the NYUDv2 dataset, respectively.",
"In this paper we address the problem of semantic labeling of indoor scenes on RGB-D data. With the availability of RGB-D cameras, it is expected that additional depth measurement will improve the accuracy. Here we investigate a solution how to incorporate complementary depth information into a semantic segmentation framework by making use of convolutional neural networks (CNNs). Recently encoder-decoder type fully convolutional CNN architectures have achieved a great success in the field of semantic segmentation. Motivated by this observation we propose an encoder-decoder type network, where the encoder part is composed of two branches of networks that simultaneously extract features from RGB and depth images and fuse depth features into the RGB feature maps as the network goes deeper. Comprehensive experimental evaluations demonstrate that the proposed fusion-based architecture achieves competitive results with the state-of-the-art methods on the challenging SUN RGB-D benchmark obtaining 76.27 global accuracy, 48.30 average class accuracy and 37.29 average intersection-over-union score."
]
} |
1710.00126 | 2762616984 | This paper presents a novel 3DOF pedestrian trajectory prediction approach for autonomous mobile service robots. While most previously reported methods are based on learning of 2D positions in monocular camera images, our approach uses range-finder sensors to learn and predict 3DOF pose trajectories (i.e. 2D position plus 1D rotation within the world coordinate system). Our approach, T-Pose-LSTM (Temporal 3DOF-Pose Long-Short-Term Memory), is trained using long-term data from real-world robot deployments and aims to learn context-dependent (environment- and time-specific) human activities. Our approach incorporates long-term temporal information (i.e. date and time) with short-term pose observations as input. A sequence-to-sequence LSTM encoder-decoder is trained, which encodes observations into LSTM and then decodes as predictions. For deployment, it can perform on-the-fly prediction in real-time. Instead of using manually annotated data, we rely on a robust human detection, tracking and SLAM system, providing us with examples in a global coordinate system. We validate the approach using more than 15K pedestrian trajectories recorded in a care home environment over a period of three months. The experiment shows that the proposed T-Pose-LSTM model advances the state-of-the-art 2D-based method for human trajectory prediction in long-term mobile robot deployments. | Predicting pedestrian trajectories has a long history in computer vision @cite_15 @cite_14 @cite_19 @cite_0 @cite_2 @cite_17 @cite_21 . Pioneering work focuses on representing the social context with hand-crafted features @cite_15 @cite_14 @cite_19 @cite_0 , while more recent approaches have tried to learn from context information relating to the position of other pedestrians or landmarks in the environment @cite_2 @cite_17 . As the trajectory prediction can be formulated as a regression problem, Gaussian Process (GP)-based methods @cite_19 @cite_7 are very popular in pedestrian prediction. The limitation of GP-like parametric models is that the computation increases exponentially with the number of training examples. Although sparse GP methods @cite_7 can be much more efficient for large-scale datasets, this problem is unlikely be remedied. | {
"cite_N": [
"@cite_14",
"@cite_7",
"@cite_21",
"@cite_0",
"@cite_19",
"@cite_2",
"@cite_15",
"@cite_17"
],
"mid": [
"2532516272",
"2613401634",
"2519586580",
"2146183743",
"2082585576",
"2424778531",
"1970206276",
"2613814584"
],
"abstract": [
"Object tracking typically relies on a dynamic model to predict the object's location from its past trajectory. In crowded scenarios a strong dynamic model is particularly important, because more accurate predictions allow for smaller search regions, which greatly simplifies data association. Traditional dynamic models predict the location for each target solely based on its own history, without taking into account the remaining scene objects. Collisions are resolved only when they happen. Such an approach ignores important aspects of human behavior: people are driven by their future destination, take into account their environment, anticipate collisions, and adjust their trajectories at an early stage in order to avoid them. In this work, we introduce a model of dynamic social behavior, inspired by models developed for crowd simulation. The model is trained with videos recorded from birds-eye view at busy locations, and applied as a motion model for multi-people tracking from a vehicle-mounted camera. Experiments on real sequences show that accounting for social interactions and scene knowledge improves tracking performance, especially during occlusions.",
"We study the sparsity and optimality properties of crowd navigation and find that existing techniques do not satisfy both criteria simultaneously: either they achieve optimality with a prohibitive number of samples or tractability assumptions make them fragile to catastrophe. For example, if the human and robot are modeled independently, then tractability is attained but the planner is prone to overcautious or overaggressive behavior. For sampling based motion planning of joint human-robot cost functions, for @math agents and @math step lookahead, @math samples are needed for coverage of the action space. Advanced approaches statically partition the action space into free-space and then sample in those convex regions. However, if the human is into free-space, then the partition is misleading and sampling is unsafe: free space will soon be occupied. We diagnose the cause of these deficiencies---optimization happens over space---and propose a novel solution: optimize over trajectory space by using a Gaussian process (GP) basis. We exploit the \"kernel trick\" of GPs, where a continuum of trajectories are captured with a mean and covariance function. By using the mean and covariance as proxies for a trajectory family we reason about collective trajectory behavior without resorting to sampling. The GP basis is sparse and optimal with respect to collision avoidance and robot and crowd intention and flexibility. GP sparsity leans heavily on the insight that joint action space decomposes into free regions; however, the decomposition contains feasible solutions only if the partition is dynamically generated. We call our approach .",
"Humans navigate crowded spaces such as a university campus by following common sense rules based on social etiquette. In this paper, we argue that in order to enable the design of new target tracking or trajectory forecasting methods that can take full advantage of these rules, we need to have access to better data in the first place. To that end, we contribute a new large-scale dataset that collects videos of various types of targets (not just pedestrians, but also bikers, skateboarders, cars, buses, golf carts) that navigate in a real world outdoor environment such as a university campus. Moreover, we introduce a new characterization that describes the “social sensitivity” at which two targets interact. We use this characterization to define “navigation styles” and improve both forecasting models and state-of-the-art multi-target tracking–whereby the learnt forecasting models help the data association step.",
"We propose an agent-based behavioral model of pedestrians to improve tracking performance in realistic scenarios. In this model, we view pedestrians as decision-making agents who consider a plethora of personal, social, and environmental factors to decide where to go next. We formulate prediction of pedestrian behavior as an energy minimization on this model. Two of our main contributions are simple, yet effective estimates of pedestrian destination and social relationships (groups). Our final contribution is to incorporate these hidden properties into an energy formulation that results in accurate behavioral prediction. We evaluate both our estimates of destination and grouping, as well as our accuracy at prediction and tracking against state of the art behavioral model and show improvements, especially in the challenging observational situation of infrequent appearance observations–something that might occur in thousands of webcams available on the Internet.",
"In this paper, we study the safe navigation of a mobile robot through crowds of dynamic agents with uncertain trajectories. Existing algorithms suffer from the “freezing robot” problem: once the environment surpasses a certain level of complexity, the planner decides that all forward paths are unsafe, and the robot freezes in place (or performs unnecessary maneuvers) to avoid collisions. Since a feasible path typically exists, this behavior is suboptimal. Existing approaches have focused on reducing the predictive uncertainty for individual agents by employing more informed models or heuristically limiting the predictive covariance to prevent this overcautious behavior. In this work, we demonstrate that both the individual prediction and the predictive uncertainty have little to do with the frozen robot problem. Our key insight is that dynamic agents solve the frozen robot problem by engaging in “joint collision avoidance”: They cooperatively make room to create feasible trajectories. We develop IGP, a nonparametric statistical model based on dependent output Gaussian processes that can estimate crowd interaction from data. Our model naturally captures the non-Markov nature of agent trajectories, as well as their goal-driven navigation. We then show how planning in this model can be efficiently implemented using particle based inference. Lastly, we evaluate our model on a dataset of pedestrians entering and leaving a building, first comparing the model with actual pedestrians, and find that the algorithm either outperforms human pedestrians or performs very similarly to the pedestrians. We also present an experiment where a covariance reduction method results in highly overcautious behavior, while our model performs desirably.",
"Pedestrians follow different trajectories to avoid obstacles and accommodate fellow pedestrians. Any autonomous vehicle navigating such a scene should be able to foresee the future positions of pedestrians and accordingly adjust its path to avoid collisions. This problem of trajectory prediction can be viewed as a sequence generation task, where we are interested in predicting the future trajectory of people based on their past positions. Following the recent success of Recurrent Neural Network (RNN) models for sequence prediction tasks, we propose an LSTM model which can learn general human movement and predict their future trajectories. This is in contrast to traditional approaches which use hand-crafted functions such as Social forces. We demonstrate the performance of our method on several public datasets. Our model outperforms state-of-the-art methods on some of these datasets. We also analyze the trajectories predicted by our model to demonstrate the motion behaviour learned by our model.",
"We present an example-based crowd simulation technique. Most crowd simulation techniques assume that the behavior exhibited by each person in the crowd can be defined by a restricted set of rules. This assumption limits the behavioral complexity of the simulated agents. By learning from real-world examples, our autonomous agents display complex natural behaviors that are often missing in crowd simulations. Examples are created from tracked video segments of real pedestrian crowds. During a simulation, autonomous agents search for examples that closely match the situation that they are facing. Trajectories taken by real people in similar situations, are copied to the simulated agents, resulting in seemingly natural behaviors.",
"Human motion and behaviour in crowded spaces is influenced by several factors, such as the dynamics of other moving agents in the scene, as well as the static elements that might be perceived as points of attraction or obstacles. In this work, we present a new model for human trajectory prediction which is able to take advantage of both human-human and human-space interactions. The future trajectory of humans, are generated by observing their past positions and interactions with the surroundings. To this end, we propose a \"context-aware\" recurrent neural network LSTM model, which can learn and predict human motion in crowded spaces such as a sidewalk, a museum or a shopping mall. We evaluate our model on a public pedestrian datasets, and we contribute a new challenging dataset that collects videos of humans that navigate in a (real) crowded space such as a big museum. Results show that our approach can predict human trajectories better when compared to previous state-of-the-art forecasting models."
]
} |
1710.00126 | 2762616984 | This paper presents a novel 3DOF pedestrian trajectory prediction approach for autonomous mobile service robots. While most previously reported methods are based on learning of 2D positions in monocular camera images, our approach uses range-finder sensors to learn and predict 3DOF pose trajectories (i.e. 2D position plus 1D rotation within the world coordinate system). Our approach, T-Pose-LSTM (Temporal 3DOF-Pose Long-Short-Term Memory), is trained using long-term data from real-world robot deployments and aims to learn context-dependent (environment- and time-specific) human activities. Our approach incorporates long-term temporal information (i.e. date and time) with short-term pose observations as input. A sequence-to-sequence LSTM encoder-decoder is trained, which encodes observations into LSTM and then decodes as predictions. For deployment, it can perform on-the-fly prediction in real-time. Instead of using manually annotated data, we rely on a robust human detection, tracking and SLAM system, providing us with examples in a global coordinate system. We validate the approach using more than 15K pedestrian trajectories recorded in a care home environment over a period of three months. The experiment shows that the proposed T-Pose-LSTM model advances the state-of-the-art 2D-based method for human trajectory prediction in long-term mobile robot deployments. | LSTM-based data-driven approaches @cite_2 @cite_17 have achieved the state-of-the-art performance, especially with large-scale datasets @cite_21 . Later, an end-to-end trajectory prediction method was proposed by @cite_11 , which combines the CNN (Convolutional Neural Network) for local image context understanding and LSTM for consecutive position prediction. | {
"cite_N": [
"@cite_21",
"@cite_17",
"@cite_11",
"@cite_2"
],
"mid": [
"2519586580",
"2613814584",
"2617510024",
"2424778531"
],
"abstract": [
"Humans navigate crowded spaces such as a university campus by following common sense rules based on social etiquette. In this paper, we argue that in order to enable the design of new target tracking or trajectory forecasting methods that can take full advantage of these rules, we need to have access to better data in the first place. To that end, we contribute a new large-scale dataset that collects videos of various types of targets (not just pedestrians, but also bikers, skateboarders, cars, buses, golf carts) that navigate in a real world outdoor environment such as a university campus. Moreover, we introduce a new characterization that describes the “social sensitivity” at which two targets interact. We use this characterization to define “navigation styles” and improve both forecasting models and state-of-the-art multi-target tracking–whereby the learnt forecasting models help the data association step.",
"Human motion and behaviour in crowded spaces is influenced by several factors, such as the dynamics of other moving agents in the scene, as well as the static elements that might be perceived as points of attraction or obstacles. In this work, we present a new model for human trajectory prediction which is able to take advantage of both human-human and human-space interactions. The future trajectory of humans, are generated by observing their past positions and interactions with the surroundings. To this end, we propose a \"context-aware\" recurrent neural network LSTM model, which can learn and predict human motion in crowded spaces such as a sidewalk, a museum or a shopping mall. We evaluate our model on a public pedestrian datasets, and we contribute a new challenging dataset that collects videos of humans that navigate in a (real) crowded space such as a big museum. Results show that our approach can predict human trajectories better when compared to previous state-of-the-art forecasting models.",
"Trajectory Prediction of dynamic objects is a widely studied topic in the field of artificial intelligence. Thanks to a large number of applications like predicting abnormal events, navigation system for the blind, etc. there have been many approaches to attempt learning patterns of motion directly from data using a wide variety of techniques ranging from hand-crafted features to sophisticated deep learning models for unsupervised feature learning. All these approaches have been limited by problems like inefficient features in the case of hand crafted features, large error propagation across the predicted trajectory and no information of static artefacts around the dynamic moving objects. We propose an end to end deep learning model to learn the motion patterns of humans using different navigational modes directly from data using the much popular sequence to sequence model coupled with a soft attention mechanism. We also propose a novel approach to model the static artefacts in a scene and using these to predict the dynamic trajectories. The proposed method, tested on trajectories of pedestrians, consistently outperforms previously proposed state of the art approaches on a variety of large scale data sets. We also show how our architecture can be naturally extended to handle multiple modes of movement (say pedestrians, skaters, bikers and buses) simultaneously.",
"Pedestrians follow different trajectories to avoid obstacles and accommodate fellow pedestrians. Any autonomous vehicle navigating such a scene should be able to foresee the future positions of pedestrians and accordingly adjust its path to avoid collisions. This problem of trajectory prediction can be viewed as a sequence generation task, where we are interested in predicting the future trajectory of people based on their past positions. Following the recent success of Recurrent Neural Network (RNN) models for sequence prediction tasks, we propose an LSTM model which can learn general human movement and predict their future trajectories. This is in contrast to traditional approaches which use hand-crafted functions such as Social forces. We demonstrate the performance of our method on several public datasets. Our model outperforms state-of-the-art methods on some of these datasets. We also analyze the trajectories predicted by our model to demonstrate the motion behaviour learned by our model."
]
} |
1710.00126 | 2762616984 | This paper presents a novel 3DOF pedestrian trajectory prediction approach for autonomous mobile service robots. While most previously reported methods are based on learning of 2D positions in monocular camera images, our approach uses range-finder sensors to learn and predict 3DOF pose trajectories (i.e. 2D position plus 1D rotation within the world coordinate system). Our approach, T-Pose-LSTM (Temporal 3DOF-Pose Long-Short-Term Memory), is trained using long-term data from real-world robot deployments and aims to learn context-dependent (environment- and time-specific) human activities. Our approach incorporates long-term temporal information (i.e. date and time) with short-term pose observations as input. A sequence-to-sequence LSTM encoder-decoder is trained, which encodes observations into LSTM and then decodes as predictions. For deployment, it can perform on-the-fly prediction in real-time. Instead of using manually annotated data, we rely on a robust human detection, tracking and SLAM system, providing us with examples in a global coordinate system. We validate the approach using more than 15K pedestrian trajectories recorded in a care home environment over a period of three months. The experiment shows that the proposed T-Pose-LSTM model advances the state-of-the-art 2D-based method for human trajectory prediction in long-term mobile robot deployments. | Furthermore, researchers are working on understanding higher-level human behaviors from pedestrian trajectories in station surveillance videos, including work on travel time estimation @cite_13 , recognition of stationary crowd groups @cite_27 , and destination forecasting @cite_25 . Pioneering research has also considered the representation of dynamic maps from long-term datasets, including the FreMEn approach @cite_16 for signal analysis, which uses the Fourier Transform to identify underlying periodic processes in the environment and hence make predictions of human activities based on the frequency spectrum of the observations. | {
"cite_N": [
"@cite_27",
"@cite_16",
"@cite_13",
"@cite_25"
],
"mid": [
"2467483865",
"1033316025",
"2205807761",
"2134944993"
],
"abstract": [
"Pedestrian behavior modeling and analysis is important for crowd scene understanding and has various applications in video surveillance. Stationary crowd groups are a key factor influencing pedestrian walking patterns but was mostly ignored in the literature. It plays different roles for different pedestrians in a crowded scene and can change over time. In this paper, a novel model is proposed to model pedestrian behaviors by incorporating stationary crowd groups as a key component. Through inference on the interactions between stationary crowd groups and pedestrians, our model can be used to investigate pedestrian behaviors. The effectiveness of the proposed model is demonstrated through multiple applications, including walking path prediction, destination prediction, personality attribute classification, and abnormal event detection. To evaluate our model, two large pedestrian walking route datasets are built. The walking routes of around 15 000 pedestrians from two crowd surveillance videos are manually annotated. The datasets will be released to the public and benefit future research on pedestrian behavior analysis and crowd scene understanding.",
"We present a new approach to long-term mobile robot mapping in dynamic indoor environments. Unlike traditional world models that are tailored to represent static scenes, our approach explicitly models environmental dynamics. We assume that some of the hidden processes that influence the dynamic environment states are periodic and model the uncertainty of the estimated state variables by their frequency spectra. The spectral model can represent arbitrary timescales of environment dynamics with low memory requirements. Transformation of the spectral model to the time domain allows for the prediction of the future environment states, which improves the robot's long-term performance in changing environments. Experiments performed over time periods of months to years demonstrate that the approach can efficiently represent large numbers of observations and reliably predict future environment states. The experiments indicate that the model's predictive capabilities improve mobile robot localization and navigation in changing environments.",
"In this paper, we target on the problem of estimating the statistic of pedestrian travel time within a period from an entrance to a destination in a crowded scene. Such estimation is based on the global distributions of crowd densities and velocities instead of complete trajectories of pedestrians, which cannot be obtained in crowded scenes. The proposed method is motivated by our statistical investigation into the correlations between travel time and global properties of crowded scenes. Active regions are created for each source-destination pair to model the probable walking regions over the corresponding source-destination traffic flow. Two sets of scene features are specially designed for modeling moving and stationary persons inside the active regions and their influences on pedestrian travel time. The estimation of pedestrian travel time provides valuable information for both crowd scene understanding and pedestrian behavior analysis, but was not sufficiently studied in literature. The effectiveness of the proposed pedestrian travel time estimation model is demonstrated through several surveillance applications, including dynamic scene monitoring, localization of regions blocking traffics, and detection of abnormal pedestrian behaviors. Many more valuable applications based on our method are to be explored in the future.",
"In crowded spaces such as city centers or train stations, human mobility looks complex, but is often influenced only by a few causes. We propose to quantitatively study crowded environments by introducing a dataset of 42 million trajectories collected in train stations. Given this dataset, we address the problem of forecasting pedestrians' destinations, a central problem in understanding large-scale crowd mobility. We need to overcome the challenges posed by a limited number of observations (e.g. sparse cameras), and change in pedestrian appearance cues across different cameras. In addition, we often have restrictions in the way pedestrians can move in a scene, encoded as priors over origin and destination (OD) preferences. We propose a new descriptor coined as Social Affinity Maps (SAM) to link broken or unobserved trajectories of individuals in the crowd, while using the OD-prior in our framework. Our experiments show improvement in performance through the use of SAM features and OD prior. To the best of our knowledge, our work is one of the first studies that provides encouraging results towards a better understanding of crowd behavior at the scale of million pedestrians."
]
} |
1710.00126 | 2762616984 | This paper presents a novel 3DOF pedestrian trajectory prediction approach for autonomous mobile service robots. While most previously reported methods are based on learning of 2D positions in monocular camera images, our approach uses range-finder sensors to learn and predict 3DOF pose trajectories (i.e. 2D position plus 1D rotation within the world coordinate system). Our approach, T-Pose-LSTM (Temporal 3DOF-Pose Long-Short-Term Memory), is trained using long-term data from real-world robot deployments and aims to learn context-dependent (environment- and time-specific) human activities. Our approach incorporates long-term temporal information (i.e. date and time) with short-term pose observations as input. A sequence-to-sequence LSTM encoder-decoder is trained, which encodes observations into LSTM and then decodes as predictions. For deployment, it can perform on-the-fly prediction in real-time. Instead of using manually annotated data, we rely on a robust human detection, tracking and SLAM system, providing us with examples in a global coordinate system. We validate the approach using more than 15K pedestrian trajectories recorded in a care home environment over a period of three months. The experiment shows that the proposed T-Pose-LSTM model advances the state-of-the-art 2D-based method for human trajectory prediction in long-term mobile robot deployments. | Over the past 20 years of development of Simultaneous Localization and Mapping (SLAM), 2D laser-based SLAM @cite_6 and visual-SLAM @cite_5 @cite_31 have largely become off-the-shelf technologies employed in many robotic systems. With the advances in robustness and adaptability of robotic systems in real-world environments, long-term autonomy has become an important research topic, starting from museum robots interacting with visitors @cite_12 , through robots employed in office environments @cite_30 @cite_20 , to robots running for months in care homes @cite_3 . Such interactive robots provide an excellent opportunity to learn from the long-term observations of people in their vicinity and gather -- and subsequently exploit -- such experience for their own behaviour generation and planning. Data sets gathered by the long-term autonomous robotic system of the STRANDS project @cite_3 @cite_26 are also used to evaluate the approach proposed in this paper (see Sec. ). | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_6",
"@cite_3",
"@cite_5",
"@cite_31",
"@cite_20",
"@cite_12"
],
"mid": [
"2406364024",
"2594081070",
"2125420246",
"",
"1612997784",
"2069479606",
"",
"1539282749"
],
"abstract": [
"On 18 November 2014, a team of four autonomous CoBot robots reached 1,000-km of overall autonomous navigation, as a result of a 1,000-km challenge that the authors had set three years earlier. The authors are frequently asked for the lessons learned, as well as the performance results. In this article, they introduce the challenge and contribute a detailed presentation of technical insights as well as quantitative and qualitative results. They have previously presented the algorithms for the individual technical contributions, namely robot localization, symbiotic robot autonomy, and robot task scheduling. In this article, they present the data collected over the 1,000-km challenge and analyze it to evaluate the accuracy and robustness of the localization algorithms on the CoBots. Furthermore, they present technical insights into the algorithms, which they believe are responsible for the robots' continuous robust performance.",
"Adapting to users' intentions is a key requirement for autonomous robots in general, and in care settings in particular. In this paper, a comprehensive long-term study of a mobile robot providing information services to residents, visitors, and staff of a care home is presented, with a focus on adapting to the when and where the robot should be offering its services to best accommodate the users' needs. Rather than providing a fixed schedule, the presented system takes the opportunity of long-term deployment to explore the space of possibilities of interaction while concurrently exploiting the model learned to provide better services. But in order to provide effective services to users in a care home, not only then when and where are relevant, but also the way how the information is provided and accessed. Hence, also the usability of the deployed system is studied specifically, in order to provide a most comprehensive overall assessment of a robotic info-terminal implementation in a care setting. Our results back our hypotheses, (i) that learning a spatiotemporal model of users' intentions improves efficiency and usefulness of the system, and (ii) that the specific information sought after is indeed dependent on the location the info-terminal is offered.",
"Recently Rao-Blackwellized particle filters have been introduced as effective means to solve the simultaneous localization and mapping (SLAM) problem. This approach uses a particle filter in which each particle carries an individual map of the environment. Accordingly, a key question is how to reduce the number of particles. In this paper we present adaptive techniques to reduce the number of particles in a Rao-Blackwellized particle filter for learning grid maps. We propose an approach to compute an accurate proposal distribution taking into account not only the movement of the robot but also the most recent observation. This drastically decrease the uncertainty about the robot's pose in the prediction step of the filter. Furthermore, we present an approach to selectively carry out re-sampling operations which seriously reduces the problem of particle depletion. Experimental results carried out with mobile robots in large-scale indoor as well as in outdoor environments illustrate the advantages of our methods over previous approaches.",
"",
"This paper presents ORB-SLAM, a feature-based monocular simultaneous localization and mapping (SLAM) system that operates in real time, in small and large indoor and outdoor environments. The system is robust to severe motion clutter, allows wide baseline loop closing and relocalization, and includes full automatic initialization. Building on excellent algorithms of recent years, we designed from scratch a novel system that uses the same features for all SLAM tasks: tracking, mapping, relocalization, and loop closing. A survival of the fittest strategy that selects the points and keyframes of the reconstruction leads to excellent robustness and generates a compact and trackable map that only grows if the scene content changes, allowing lifelong operation. We present an exhaustive evaluation in 27 sequences from the most popular datasets. ORB-SLAM achieves unprecedented performance with respect to other state-of-the-art monocular SLAM approaches. For the benefit of the community, we make the source code public.",
"In this paper, we present a novel mapping system that robustly generates highly accurate 3-D maps using an RGB-D camera. Our approach requires no further sensors or odometry. With the availability of low-cost and light-weight RGB-D sensors such as the Microsoft Kinect, our approach applies to small domestic robots such as vacuum cleaners, as well as flying robots such as quadrocopters. Furthermore, our system can also be used for free-hand reconstruction of detailed 3-D models. In addition to the system itself, we present a thorough experimental evaluation on a publicly available benchmark dataset. We analyze and discuss the influence of several parameters such as the choice of the feature descriptor, the number of visual features, and validation methods. The results of the experiments demonstrate that our system can robustly deal with challenging scenarios such as fast camera motions and feature-poor environments while being fast enough for online operation. Our system is fully available as open source and has already been widely adopted by the robotics community.",
"",
"This paper describes an interactive tour-guide robot, which was successfully exhibited in a Smithsonian museum. During its two weeks of operation, the robot interacted with thousands of people, traversing more than 44 km at speeds of up to 163 cm sec. Our approach specifically addresses issues such as safe navigation in unmodified and dynamic environments, and short-term human-robot interaction. It uses learning pervasively at all levels of the software architecture."
]
} |
1710.00126 | 2762616984 | This paper presents a novel 3DOF pedestrian trajectory prediction approach for autonomous mobile service robots. While most previously reported methods are based on learning of 2D positions in monocular camera images, our approach uses range-finder sensors to learn and predict 3DOF pose trajectories (i.e. 2D position plus 1D rotation within the world coordinate system). Our approach, T-Pose-LSTM (Temporal 3DOF-Pose Long-Short-Term Memory), is trained using long-term data from real-world robot deployments and aims to learn context-dependent (environment- and time-specific) human activities. Our approach incorporates long-term temporal information (i.e. date and time) with short-term pose observations as input. A sequence-to-sequence LSTM encoder-decoder is trained, which encodes observations into LSTM and then decodes as predictions. For deployment, it can perform on-the-fly prediction in real-time. Instead of using manually annotated data, we rely on a robust human detection, tracking and SLAM system, providing us with examples in a global coordinate system. We validate the approach using more than 15K pedestrian trajectories recorded in a care home environment over a period of three months. The experiment shows that the proposed T-Pose-LSTM model advances the state-of-the-art 2D-based method for human trajectory prediction in long-term mobile robot deployments. | After investigation of the state-of-the-art methods for pedestrian trajectory prediction, we concluded that their limitations are twofold. Firstly, the predominantly reported methods are based on monocular cameras, see e.g. the UCY @cite_15 , ETH @cite_14 and SDD @cite_21 datasets. The drawback of monocular cameras is that the trajectories are not in real-world coordinates (meters) when the camera is not perpendicular to the ground. Also, the visual scope of monocular cameras is very narrow on mobile platforms. For mobile robot applications, alternative means of sensing need to be found. Secondly, the existing works on pedestrian trajectory prediction proposed various methods to parameterize the social context, while few of them take account of the overall spatial and temporal context. | {
"cite_N": [
"@cite_15",
"@cite_14",
"@cite_21"
],
"mid": [
"1970206276",
"2532516272",
"2519586580"
],
"abstract": [
"We present an example-based crowd simulation technique. Most crowd simulation techniques assume that the behavior exhibited by each person in the crowd can be defined by a restricted set of rules. This assumption limits the behavioral complexity of the simulated agents. By learning from real-world examples, our autonomous agents display complex natural behaviors that are often missing in crowd simulations. Examples are created from tracked video segments of real pedestrian crowds. During a simulation, autonomous agents search for examples that closely match the situation that they are facing. Trajectories taken by real people in similar situations, are copied to the simulated agents, resulting in seemingly natural behaviors.",
"Object tracking typically relies on a dynamic model to predict the object's location from its past trajectory. In crowded scenarios a strong dynamic model is particularly important, because more accurate predictions allow for smaller search regions, which greatly simplifies data association. Traditional dynamic models predict the location for each target solely based on its own history, without taking into account the remaining scene objects. Collisions are resolved only when they happen. Such an approach ignores important aspects of human behavior: people are driven by their future destination, take into account their environment, anticipate collisions, and adjust their trajectories at an early stage in order to avoid them. In this work, we introduce a model of dynamic social behavior, inspired by models developed for crowd simulation. The model is trained with videos recorded from birds-eye view at busy locations, and applied as a motion model for multi-people tracking from a vehicle-mounted camera. Experiments on real sequences show that accounting for social interactions and scene knowledge improves tracking performance, especially during occlusions.",
"Humans navigate crowded spaces such as a university campus by following common sense rules based on social etiquette. In this paper, we argue that in order to enable the design of new target tracking or trajectory forecasting methods that can take full advantage of these rules, we need to have access to better data in the first place. To that end, we contribute a new large-scale dataset that collects videos of various types of targets (not just pedestrians, but also bikers, skateboarders, cars, buses, golf carts) that navigate in a real world outdoor environment such as a university campus. Moreover, we introduce a new characterization that describes the “social sensitivity” at which two targets interact. We use this characterization to define “navigation styles” and improve both forecasting models and state-of-the-art multi-target tracking–whereby the learnt forecasting models help the data association step."
]
} |
1710.00122 | 2578818355 | Nowadays, multiprocessing is mainstream with exponentially increasing number of processors. Load balancing is, therefore, a critical operation for the efficient execution of parallel algorithms. In this paper we consider the fundamental class of tree-based algorithms that are notoriously irregular, and hard to load-balance with existing static techniques. We propose a hybrid load balancing method using the utility of statistical random sampling in estimating the tree depth and node count distributions to uniformly partition an input tree. To conduct an initial performance study, we implemented the method on an Intel Xeon Phi accelerator system. We considered the tree traversal operation on both regular and irregular unbalanced trees manifested by Fibonacci and unbalanced (biased) randomly generated trees, respectively. The results show scalable performance for up to the 60 physical processors of the accelerator, as well as an extrapolated 128 processors case. | The load-balancing problem has been addressed previously but not in the manner proposed in this work. Several load balancing algorithms are available, for example Round Robin and Randomized Algorithms, Central Manager Algorithm and Threshold Algorithm. However, these algorithms depend on static load balancing. It requires that the workload is initially known to the balancing algorithm that runs before any real computation. Dynamic algorithms, such as the Central Queue algorithm and the Local Queue algorithm @cite_3 , introduce runtime overhead, as they provide general tasking distribution mechanisms that do not exploit tree aspects. Gursoy suggested several data composition schemes @cite_4 ; however, they are concerned mainly with tree-based k-means clustering, which does not target general trees. Another related work is that of El-Mahdy and ElShishiny @cite_5 ; that method is the closest to our work in terms of the adopted hybrid static and dynamic approach; however, the method targets the statically structured objects of images, and does not access irregular data structures, such as trees. | {
"cite_N": [
"@cite_5",
"@cite_4",
"@cite_3"
],
"mid": [
"2231778486",
"1907099735",
"2118663179"
],
"abstract": [
"A method and system for load balancing the work of NP processors (NP≧3) configured to generate each image of multiple images in a display area of a display device. The process for each image includes: dividing the display area logically into NP initial segments ordered along an axis of the display area; assigning each processor to a corresponding initial segment; assigning a thickness to each initial segment; simultaneously computing an average work function per pixel for each initial segment; generating a cumulative work function from the average work function per pixel for each initial segment; partitioning a work function domain of the cumulative work function into NP sub-domains; determining NP final segments of the display area by using the cumulative work function to inversely map boundaries of the sub-domains onto the axis; assigning each processor to a final segment, and displaying and or storing the NP final segments.",
"Developing fast algorithms for clustering has been an important area of research in data mining and other fields. K-means is one of the widely used clustering algorithms. In this work, we have developed and evaluated parallelization of k-means method for low-dimensional data on message passing computers. Three different data decomposition schemes and their impact on the pruning of distance calculations in tree-based k-means algorithm have been studied. Random pattern decomposition has good load balancing but fails to prune distance calculations effectively. Compact spatial decomposition of patterns based on space filling curves outperforms random pattern decomposition even though it has load imbalance problem. In both cases, parallel tree-based k-means clustering runs significantly faster than the traditional parallel k-means.",
"Studies the performance of a randomized algorithm for balancing load across a multiprocessor executing a dynamic irregular task tree. Specifically, we show that the time taken to explore a task tree is likely to be within a small constant factor of an inherent lower bound for the tree instance. Our model permits arbitrary task times and overlap between computation and load balance, and thus extends earlier work (R.M. Karp and Y. Zhang, 1988) which assumed fixed cost tasks and used a bulk synchronous style in which the system alternated between distinct computing and load balancing steps. Our analysis is supported by experiments with application codes, demonstrating that the efficiency is high enough to make this method practical. >"
]
} |
1710.00205 | 2763190896 | Vector-space models, from word embeddings to neural network parsers, have many advantages for NLP. But how to generalise from fixed-length word vectors to a vector space for arbitrary linguistic structures is still unclear. In this paper we propose bag-of-vector embeddings of arbitrary linguistic graphs. A bag-of-vector space is the minimal nonparametric extension of a vector space, allowing the representation to grow with the size of the graph, but not tying the representation to any specific tree or graph structure. We propose efficient training and inference algorithms based on tensor factorisation for embedding arbitrary graphs in a bag-of-vector space. We demonstrate the usefulness of this representation by training bag-of-vector embeddings of dependency graphs and evaluating them on unsupervised semantic induction for the Semantic Textual Similarity and Natural Language Inference tasks. | Neural network (NN) models have been used to compute embeddings for parsed sentences. They either use a fixed-length vector for an arbitrarily long sentence (e.g. @cite_15 ), or they keep the original sentence structure and decorate it with vectors (e.g. @cite_21 @cite_3 ). Ours is the first non-parametric vector-based model that does not keep the entire structure. However, attention-based NN models could be used in this way. For example in machine translation, take a source sequence of words and encode it as a sequence of vectors, which are then used in an attention-based NN model to generate the target sequence of words for the translation. Keeping the original ordering of words is not fundamental to this method, and thus it could be interpreted as a bag-of-vector embedding method for sequences. However, it is not at all clear how to generalise such a sequence-encoding method to arbitrary graphs. Similarly, Kalman filters have been used to induce vector representations of word tokens in their sequential context @cite_11 , but it is not clear how to generalise this to arbitrary graphs. | {
"cite_N": [
"@cite_15",
"@cite_21",
"@cite_3",
"@cite_11"
],
"mid": [
"2103305545",
"2143431851",
"2133280805",
""
],
"abstract": [
"Paraphrase detection is the task of examining two sentences and determining whether they have the same meaning. In order to obtain high accuracy on this task, thorough syntactic and semantic analysis of the two statements is needed. We introduce a method for paraphrase detection based on recursive autoencoders (RAE). Our unsupervised RAEs are based on a novel unfolding objective and learn feature vectors for phrases in syntactic trees. These features are used to measure the word- and phrase-wise similarity between two sentences. Since sentences may be of arbitrary length, the resulting matrix of similarity measures is of variable size. We introduce a novel dynamic pooling layer which computes a fixed-sized representation from the variable-sized matrices. The pooled representation is then used as input to a classifier. Our method outperforms other state-of-the-art approaches on the challenging MSRP paraphrase corpus.",
"We present a neural network method for inducing representations of parse histories and using these history representations to estimate the probabilities needed by a statistical left-corner parser. The resulting statistical parser achieves performance (89.1 F-measure) on the Penn Treebank which is only 0.6 below the best current parser for this task, despite using a smaller vocabulary size and less prior linguistic knowledge. Crucial to this success is the use of structurally determined soft biases in inducing the representation of the parse history, and no use of hard independence assumptions.",
"Natural language parsing has typically been done with small sets of discrete categories such as NP and VP, but this representation does not capture the full syntactic nor semantic richness of linguistic phrases, and attempts to improve on this by lexicalizing phrases or splitting categories only partly address the problem at the cost of huge feature spaces and sparseness. Instead, we introduce a Compositional Vector Grammar (CVG), which combines PCFGs with a syntactically untied recursive neural network that learns syntactico-semantic, compositional vector representations. The CVG improves the PCFG of the Stanford Parser by 3.8 to obtain an F1 score of 90.4 . It is fast to train and implemented approximately as an efficient reranker it is about 20 faster than the current Stanford factored parser. The CVG learns a soft notion of head words and improves performance on the types of ambiguities that require semantic information such as PP attachments.",
""
]
} |
1710.00205 | 2763190896 | Vector-space models, from word embeddings to neural network parsers, have many advantages for NLP. But how to generalise from fixed-length word vectors to a vector space for arbitrary linguistic structures is still unclear. In this paper we propose bag-of-vector embeddings of arbitrary linguistic graphs. A bag-of-vector space is the minimal nonparametric extension of a vector space, allowing the representation to grow with the size of the graph, but not tying the representation to any specific tree or graph structure. We propose efficient training and inference algorithms based on tensor factorisation for embedding arbitrary graphs in a bag-of-vector space. We demonstrate the usefulness of this representation by training bag-of-vector embeddings of dependency graphs and evaluating them on unsupervised semantic induction for the Semantic Textual Similarity and Natural Language Inference tasks. | Previous work on context-sensitive word embeddings has typically treated this as a word sense disambiguation problem, with one vector per word sense (e.g. @cite_1 @cite_4 ). In contrast, for our method every different context results in a different word token vector. | {
"cite_N": [
"@cite_1",
"@cite_4"
],
"mid": [
"2949364118",
"2262907013"
],
"abstract": [
"There is rising interest in vector-space word embeddings and their use in NLP, especially given recent methods for their fast estimation at very large scale. Nearly all this work, however, assumes a single vector per word type ignoring polysemy and thus jeopardizing their usefulness for downstream tasks. We present an extension to the Skip-gram model that efficiently learns multiple embeddings per word type. It differs from recent related work by jointly performing word sense discrimination and embedding learning, by non-parametrically estimating the number of senses per word type, and by its efficiency and scalability. We present new state-of-the-art results in the word similarity in context task and demonstrate its scalability by training with one machine on a corpus of nearly 1 billion tokens in less than 6 hours.",
"Distributed word representations have a rising interest in NLP community. Most of existing models assume only one vector for each individual word, which ignores polysemy and thus degrades their effectiveness for downstream tasks. To address this problem, some recent work adopts multiprototype models to learn multiple embeddings per word type. In this paper, we distinguish the different senses of each word by their latent topics. We present a general architecture to learn the word and topic embeddings efficiently, which is an extension to the Skip-Gram model and can model the interaction between words and topics simultaneously. The experiments on the word similarity and text classification tasks show our model outperforms state-of-the-art methods."
]
} |
1710.00341 | 2761944948 | Given the constantly growing proliferation of false claims online in recent years, there has been also a growing research interest in automatically distinguishing false rumors from factually true claims. Here, we propose a general-purpose framework for fully-automatic fact checking using external sources, tapping the potential of the entire Web as a knowledge source to confirm or reject a claim. Our framework uses a deep neural network with LSTM text encoding to combine semantic kernels with task-specific embeddings that encode a claim together with pieces of potentially-relevant text fragments from the Web, taking the source reliability into account. The evaluation results show good performance on two different tasks and datasets: (i) rumor detection and (ii) fact checking of the answers to a question in community question answering forums. | For instance, studied the credibility of Twitter accounts (as opposed to tweet posts), and found that both the topical content of information sources and social network structure affect source credibility. Other work, closer to ours, aims at addressing credibility assessment of rumors on Twitter as a problem of finding false information about a newsworthy event @cite_4 . This model considers user reputation, writing style, and various time-based features, among others. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2084591134"
],
"abstract": [
"We analyze the information credibility of news propagated through Twitter, a popular microblogging service. Previous research has shown that most of the messages posted on Twitter are truthful, but the service is also used to spread misinformation and false rumors, often unintentionally. On this paper we focus on automatic methods for assessing the credibility of a given set of tweets. Specifically, we analyze microblog postings related to \"trending\" topics, and classify them as credible or not credible, based on features extracted from them. We use features from users' posting and re-posting (\"re-tweeting\") behavior, from the text of the posts, and from citations to external sources. We evaluate our methods using a significant number of human assessments about the credibility of items on a recent sample of Twitter postings. Our results shows that there are measurable differences in the way messages propagate, that can be used to classify them automatically as credible or not credible, with precision and recall in the range of 70 to 80 ."
]
} |
1710.00341 | 2761944948 | Given the constantly growing proliferation of false claims online in recent years, there has been also a growing research interest in automatically distinguishing false rumors from factually true claims. Here, we propose a general-purpose framework for fully-automatic fact checking using external sources, tapping the potential of the entire Web as a knowledge source to confirm or reject a claim. Our framework uses a deep neural network with LSTM text encoding to combine semantic kernels with task-specific embeddings that encode a claim together with pieces of potentially-relevant text fragments from the Web, taking the source reliability into account. The evaluation results show good performance on two different tasks and datasets: (i) rumor detection and (ii) fact checking of the answers to a question in community question answering forums. | Other efforts have focused on news communities. For example, several truth discovery algorithms are combined in an ensemble method for veracity estimation in the VERA system @cite_17 . They proposed a platform for end-to-end truth discovery from the Web: extracting unstructured information from multiple sources, combining information about single claims, running an ensemble of algorithms, and visualizing and explaining the results. They also explore two different real-world application scenarios for their system: fact checking for crisis situations and evaluation of trustworthiness of a rumor. However, the input to their model is structured data, while here we are interested in unstructured text as input. | {
"cite_N": [
"@cite_17"
],
"mid": [
"2338607651"
],
"abstract": [
"Social networks and the Web in general are characterized by multiple information sources often claiming conflicting data values. Data veracity is hard to estimate, especially when there is no prior knowledge about the sources or the claims and in time-dependent scenarios where initially very few observers can report first information. Despite the wide set of recently proposed truth discovery approaches, \"no-one-fits-all\" solution emerges for estimating the veracity of on-line information in open contexts. However, analyzing the space of conflicting information and disagreeing sources might be relevant, as well as ensembling multiple truth discovery methods. This demonstration presents VERA, a Web-based platform that supports information extraction from Web textual data and micro-texts from Twitter and estimates data veracity. Given a user query, VERA systematically extracts entities and relations from Web content, structures them as claims relevant to the query and gathers more conflicting corroborating information. VERA combines multiple truth discovery algorithms through ensembling returns the veracity label and score of each data value and the trustworthiness scores of the sources. VERA will be demonstrated through several real-world scenarios to show its potential value for fact-checking from Web data."
]
} |
1710.00341 | 2761944948 | Given the constantly growing proliferation of false claims online in recent years, there has been also a growing research interest in automatically distinguishing false rumors from factually true claims. Here, we propose a general-purpose framework for fully-automatic fact checking using external sources, tapping the potential of the entire Web as a knowledge source to confirm or reject a claim. Our framework uses a deep neural network with LSTM text encoding to combine semantic kernels with task-specific embeddings that encode a claim together with pieces of potentially-relevant text fragments from the Web, taking the source reliability into account. The evaluation results show good performance on two different tasks and datasets: (i) rumor detection and (ii) fact checking of the answers to a question in community question answering forums. | In the context of question answering, there has been work on assessing the credibility of an answer, e.g., based on intrinsic information @cite_26 , i.e., without any external resources. In this case, the reliability of an answer is measured by computing the divergence between language models of the question and of the answer. The spawn of community-based question answering Websites also allowed for the use of other kinds of information. Click counts, link analysis (e.g., PageRank), and user votes have been used to assess the quality of a posted answer @cite_14 @cite_25 @cite_0 . Nevertheless, these studies address the answers' credibility level just marginally. | {
"cite_N": [
"@cite_0",
"@cite_14",
"@cite_26",
"@cite_25"
],
"mid": [
"2025895610",
"2037858832",
"2090380787",
"2102956348"
],
"abstract": [
"Question-Answer portals such as Naver and Yahoo! Answers are quickly becoming rich sources of knowledge on many topics which are not well served by general web search engines. Unfortunately, the quality of the submitted answers is uneven, ranging from excellent detailed answers to snappy and insulting remarks or even advertisements for commercial content. Furthermore, user feedback for many topics is sparse, and can be insufficient to reliably identify good answers from the bad ones. Hence, estimating the authority of users is a crucial task for this emerging domain, with potential applications to answer ranking, spam detection, and incentive mechanism design. We present an analysis of the link structure of a general-purpose question answering community to discover authoritative users, and promising experimental results over a dataset of more than 3 million answers from a popular community QA site. We also describe structural differences between question topics that correlate with the success of link analysis for authority discovery.",
"The quality of user-generated content varies drastically from excellent to abuse and spam. As the availability of such content increases, the task of identifying high-quality content sites based on user contributions --social media sites -- becomes increasingly important. Social media in general exhibit a rich variety of information sources: in addition to the content itself, there is a wide array of non-content information available, such as links between items and explicit quality ratings from members of the community. In this paper we investigate methods for exploiting such community feedback to automatically identify high quality content. As a test case, we focus on Yahoo! Answers, a large community question answering portal that is particularly rich in the amount and types of content and social interactions available in it. We introduce a general classification framework for combining the evidence from different sources of information, that can be tuned automatically for a given social media type and quality definition. In particular, for the community question answering domain, we show that our system is able to separate high-quality items from the rest with an accuracy close to that of humans",
"Answer Validation is a topic of significant interest within the Question Answering community. In this paper, we propose the use of language modeling methodologies for Answer Validation, using corpus-based methods that do not require the use of external sources. Specifically, we propose a model for Answer Credibility which quantifies the reliability of a source document that contains a candidate answer and the Question's Context Model.",
"New types of document collections are being developed by various web services. The service providers keep track of non-textual features such as click counts. In this paper, we present a framework to use non-textual features to predict the quality of documents. We also show our quality measure can be successfully incorporated into the language modeling-based retrieval model. We test our approach on a collection of question and answer pairs gathered from a community based question answering service where people ask and answer questions. Experimental results using our quality measure show a significant improvement over our baseline."
]
} |
1710.00341 | 2761944948 | Given the constantly growing proliferation of false claims online in recent years, there has been also a growing research interest in automatically distinguishing false rumors from factually true claims. Here, we propose a general-purpose framework for fully-automatic fact checking using external sources, tapping the potential of the entire Web as a knowledge source to confirm or reject a claim. Our framework uses a deep neural network with LSTM text encoding to combine semantic kernels with task-specific embeddings that encode a claim together with pieces of potentially-relevant text fragments from the Web, taking the source reliability into account. The evaluation results show good performance on two different tasks and datasets: (i) rumor detection and (ii) fact checking of the answers to a question in community question answering forums. | Efforts to determine the credibility of an answer in order to assess its overall quality required the inclusion of content-based information @cite_9 , e.g., verbs and adjectives such as and , which cast doubt on the answer. Similarly, used source credibility (e.g., does the document come from a government Website?), sentiment analysis, and answer contradiction compared to other related answers. | {
"cite_N": [
"@cite_9"
],
"mid": [
"762579341"
],
"abstract": [
"In this paper, we focus on the task of identifying the best answer for a usergenerated question in Collaborative Question Answering (CQA) services. Given that most existing research on CQA has focused on non-textual features such as click-through counts which are relatively difficult to access, we examine the effectiveness of diverse content-based features for the task. Specially, we propose to explore how the information of evidentiality can contribute to the task. By the comparison of diverse textual features and their combinations, the current study provides useful insight into the issues of detecting the best answer to a given question in CQA without user features or system specific link structures."
]
} |
1710.00269 | 2762672168 | In an information-rich world, people's time and attention must be divided among rapidly changing information sources and the diverse tasks demanded of them. How people decide which of the many sources, such as scientific articles or patents, to read and use in their own work affects dissemination of scholarly knowledge and adoption of innovation. We analyze the choices people make about what information to propagate on the citation networks of Physical Review journals, US patents and legal opinions. We observe regularities in behavior consistent with human bounded rationality: rather than evaluate all available choices, people rely on simply cognitive heuristics to decide what information to attend to. We demonstrate that these heuristics bias choices, so that people preferentially propagate information that is easier to discover, often because it is newer or more popular. However, we do not find evidence that popular sources help to amplify the spread of information beyond making it more salient. Our paper provides novel evidence of the critical role that bounded rationality plays in the decisions to allocate attention in social communication. | Bounded rationality is recognized as an important factor shaping human decisions and behavior @cite_3 , including consumer behavior @cite_43 @cite_1 @cite_36 . Computer scientists have begun to use attention, a concept related to bounded rationality, to explain social behavior online @cite_20 , including peer production of content in social media @cite_29 @cite_47 . Eye tracking studies @cite_9 @cite_12 and controlled online experiments @cite_14 demonstrated the importance of cognitive heuristics and biases in online behavior. Specifically, people pay more attention to content near the top of the page than to lower content, a cognitive bias known as position bias @cite_19 . Other studies examined how news articles, Web pages, and YouTube videos, compete for the limited collective attention @cite_39 @cite_26 @cite_13 @cite_51 .Such studies often reproduced the observed macroscopic phenomena, such as the extreme inequality of item popularity, but failed to note the link to individual's cognitive heuristics and biases. | {
"cite_N": [
"@cite_14",
"@cite_26",
"@cite_36",
"@cite_29",
"@cite_9",
"@cite_1",
"@cite_3",
"@cite_39",
"@cite_43",
"@cite_19",
"@cite_51",
"@cite_47",
"@cite_13",
"@cite_12",
"@cite_20"
],
"mid": [
"2025574664",
"2964002805",
"",
"2007590415",
"",
"2165757809",
"",
"1615730030",
"2147775745",
"2118303196",
"2952051872",
"2078939860",
"2072606289",
"2404406225",
"2014790997"
],
"abstract": [
"With the advent of social media and peer production, the amount of new online content has grown dramatically. To identify interesting items in the vast stream of new content, providers must rely on peer recommendation to aggregate opinions of their many users. Due to human cognitive biases, the presentation order strongly affects how people allocate attention to the available content. Moreover, we can manipulate attention through the presentation order of items to change the way peer recommendation works. We experimentally evaluate this effect using Amazon Mechanical Turk. We find that different policies for ordering content can steer user attention so as to improve the outcomes of peer recommendation.",
"Online popularity has an enormous impact on opinions, culture, policy, and profits. We provide a quantitative, large scale, temporal analysis of the dynamics of online content popularity in two massive model systems: the Wikipedia and an entire country’s Web space. We find that the dynamics of popularity are characterized by bursts, displaying characteristic features of critical systems such as fat-tailed distributions of magnitude and interevent time. We propose a minimal model combining the classic preferential popularity increase mechanism with the occurrence of random popularity shifts due to exogenous factors. The model recovers the critical features observed in the empirical analysis of the systems analyzed here, highlighting the key factors needed in the description of popularity dynamics.",
"",
"Online peer production systems have enabled people to coactively create, share, classify, and rate content on an unprecedented scale. This paper describes strong macroscopic regularities in how people contribute to peer production systems, and shows how these regularities arise from simple dynamical rules. First, it is demonstrated that the probability a person stops contributing varies inversely with the number of contributions he has made. This rule leads to a power law distribution for the number of contributions per person in which a small number of very active users make most of the contributions. The rule also implies that the power law exponent is proportional to the effort required to contribute, as justified by the data. Second, the level of activity per topic is shown to follow a lognormal distribution generated by a stochastic reinforcement mechanism. A small number of very popular topics thus accumulate the vast majority of contributions. These trends are demonstrated to hold across hundreds of millions of contributions to four disparate peer production systems of differing scope, interface style, and purpose.",
"",
"While limited attention has been analyzed in a variety of economic and psychological settings, its impact on financial markets is not well understood. In this paper, we examine individual NYSE specialist portfolios and test whether liquidity provision is affected as specialists allocate their attention across stocks. Our results indicate that specialists allocate effort toward their most active stocks during periods of increased activity, resulting in less frequent price improvement and increased transaction costs for their remaining assigned stocks. Thus, the allocation of effort due to limited attention has a significant impact on liquidity provision in securities markets.",
"",
"In our modern society, people are daily confronted with an increasing amount of information of any kind. As a consequence, the attention capacities and processing abilities of individuals often saturate. People, therefore, have to select which elements of their environment they are focusing on, and which are ignored. Moreover, recent work shows that individuals are naturally attracted by what other people are interested in. This imitative behaviour gives rise to various herding phenomena, such as the spread of ideas or the outbreak of commercial trends, turning the understanding of collective attention an important issue of our society. In this article, we propose an individual-based model of collective attention. In a situation where a group of people is facing a steady flow of novel information, the model naturally reproduces the log-normal distribution of the attention each news item receives, in agreement with empirical observations. Furthermore, the model predicts that the popularity of a news item strongly depends on the number of concurrent news appearing at approximately the same moment. We confirmed this prediction by means of empirical data extracted from the website digg.com. This result can be interpreted from the point of view of competition between the news for the limited attention capacity of individuals. The proposed model, therefore, provides new elements to better understand the dynamics of collective attention in an information-rich world.",
"Limited consumer attention limits product market competition: prices are stochastically lower the more attention is paid. Ads compete to be the lowest price in a sector but compete for attention with ads from other sectors: equilibrium ad shares follow a CES form. When a sector gets more proÞtable, its advertising expands: others lose ad market share. The \"information hump\" shows highest ad levels for intermediate attention levels. The Information Age takes off when the number of viable sectors grows, but total ad volume reaches an upper limit. Overall, advertising is excessive, though the allocation across sectors is optimal.",
"While the statisticians are trying to knock a few tenths off the statistical error, says Mr. Payne, errors of tens of percents occur because of bad question wording. Mr. Payne's shrewd critique of the problems of asking questions reveals much about the nature of language and words, and a good deal about the public who must answer the poller's questions. For public opinion pollers, census takers, advertising copywriters, and survey makers of all kinds this book will be a tool for the achievement of more reliable results.Originally published in 1951.The Princeton Legacy Library uses the latest print-on-demand technology to again make available previously out-of-print books from the distinguished backlist of Princeton University Press. These paperback editions preserve the original texts of these important books while presenting them in durable paperback editions. The goal of the Princeton Legacy Library is to vastly increase access to the rich scholarly heritage found in the thousands of books published by Princeton University Press since its founding in 1905.",
"Micro-blogging systems such as Twitter expose digital traces of social discourse with an unprecedented degree of resolution of individual behaviors. They offer an opportunity to investigate how a large-scale social system responds to exogenous or endogenous stimuli, and to disentangle the temporal, spatial and topical aspects of users' activity. Here we focus on spikes of collective attention in Twitter, and specifically on peaks in the popularity of hashtags. Users employ hashtags as a form of social annotation, to define a shared context for a specific event, topic, or meme. We analyze a large-scale record of Twitter activity and find that the evolution of hastag popularity over time defines discrete classes of hashtags. We link these dynamical classes to the events the hashtags represent and use text mining techniques to provide a semantic characterization of the hastag classes. Moreover, we track the propagation of hashtags in the Twitter social network and find that epidemic spreading plays a minor role in hastag popularity, which is mostly driven by exogenous factors.",
"It is commonly believed that information spreads between individuals like a pathogen, with each exposure by an informed friend potentially resulting in a naive individual becoming infected. However, empirical studies of social media suggest that individual response to repeated exposure to information is far more complex. As a proxy for intervention experiments, we compare user responses to multiple exposures on two different social media sites, Twitter and Digg. We show that the position of exposing messages on the user-interface strongly affects social contagion. Accounting for this visibility significantly simplifies the dynamics of social contagion. The likelihood an individual will spread information increases monotonically with exposure, while explicit feedback about how many friends have previously spread it increases the likelihood of a response. We provide a framework for unifying information visibility, divided attention, and explicit social feedback to predict the temporal dynamics of user behavior.",
"The wide adoption of social media has increased the competition among ideas for our finite attention. We employ a parsimonious agent-based model to study whether such a competition may affect the popularity of different memes, the diversity of information we are exposed to, and the fading of our collective interests for specific topics. Agents share messages on a social network but can only pay attention to a portion of the information they receive. In the emerging dynamics of information diffusion, a few memes go viral while most do not. The predictions of our model are consistent with empirical data from Twitter, a popular microblogging platform. Surprisingly, we can explain the massive heterogeneity in the popularity and persistence of memes as deriving from a combination of the competition for our limited attention and the structure of the social network, without the need to assume different intrinsic values among ideas.",
"Microblogging environments such as Twitter present a modality for interacting with information characterized by exposure to information “streams”. In this work, we examine what information in that stream is attended to, and how that attention corresponds to other aspects of microblog consumption and participation. To do this, we measured eye gaze, memory for content, interest ratings, and intended behavior of active Twitter users as they read their tweet steams. Our analyses focus on three sets of alignments: first, whether attention corresponds to other measures of user cognition such as memory (e.g., do people even remember what they attend to?); second, whether attention corresponds to behavior (e.g., are users likely to retweet content that is given the most attention); and third, whether attention corresponds to other attributes of the content and its presentation (e.g., do links attract attention?). We show a positive but imperfect alignment between user attention and other measures of user cognition like memory and interest, and between attention and behaviors like retweeting. To the third alignment, we show that the relationship between attention and attributes of tweets, such as whether it contains a link or is from a friend versus an organization, are complicated and in some cases counterintuitive. We discuss findings in relation to large scale phenomena like information diffusion and also suggest design directions to help maximize user attention in microblog environments.",
"If the Web and the Net can be viewed as spaces in which we will increasingly live our lives, the economic laws we will live under have to be natural to this new space."
]
} |
1709.10323 | 2758049963 | Abstract A recent theoretical analysis shows the equivalence between non-negative matrix factorization (NMF) and spectral clustering based approach to subspace clustering. As NMF and many of its variants are essentially linear, we introduce a nonlinear NMF with explicit orthogonality and derive general kernel-based orthogonal multiplicative update rules to solve the subspace clustering problem. In nonlinear orthogonal NMF framework, we propose two subspace clustering algorithms, named kernel-based non-negative subspace clustering KNSC-Ncut and KNSC-Rcut and establish their connection with spectral normalized cut and ratio cut clustering. We further extend the nonlinear orthogonal NMF framework and introduce a graph regularization to obtain a factorization that respects a local geometric structure of the data after the nonlinear mapping. The proposed NMF-based approach to subspace clustering takes into account the nonlinear nature of the manifold, as well as its intrinsic local geometry, which considerably improves the clustering performance when compared to the several recently proposed state-of-the-art methods. | Furthermore, the global discriminative NMF-based NSC model introduced in @cite_0 , includes an additional nonlinear discriminative regularization term to the NMF optimization function proposed in @cite_24 . As shown in @cite_0 , the global discriminant information greatly improves the accuracy of NSC-Ncut and NSC-Rcut @cite_24 . Although in @cite_0 the nonlinear character of the manifold is taken into account through the nonlinear discriminative matrix, the NMF data fidelity terms are still defined in the input data space. | {
"cite_N": [
"@cite_0",
"@cite_24"
],
"mid": [
"2256361316",
"1989773214"
],
"abstract": [
"Based on spectral graph theory, spectral clustering is an optimal graph partition problem. It has been proven that the spectral clustering is equivalent to nonnegative matrix factorization (NMF) under certain conditions. Based on the equivalence, some spectral clustering methods are proposed, but the global discriminative information of the dataset is neglected. In this paper, based on the equivalence between spectral clustering and NMF, we simultaneously maximize the between-class scatter matrix and minimize the within-class scatter matrix to enhance the discriminating power. We integrate the geometrical structure and discriminative structure in a joint framework. With a global discriminative regularization term added into the nonnegative matrix factorization framework, we propose two novel spectral clustering methods, named global discriminative-based nonnegative and spectral clustering (GDBNSC-Ncut and GDBNSC-Rcut) These new spectral clustering algorithms can preserve both the global geometrical structure and global discriminative structure. The intrinsic geometrical information of the dataset is detected, and clustering quality is improved with enhanced discriminating power. In addition, the proposed algorithms also have very good abilities of handling out-of-sample data. Experimental results on real word data demonstrate that the proposed algorithms outperform some state-of-the-art methods with good clustering qualities. We maximize the between-class scatter matrix.Meanwhile, we minimize the within-class scatter matrix.We integrate the geometrical structure and discriminative structure in a framework.A global discriminative regularization term added into the objective functions.The new algorithms can preserve the global geometrical and discriminative structure.",
"Spectral clustering aims to partition a data set into several groups by using the Laplacian of the graph such that data points in the same group are similar while data points in different groups are dissimilar to each other. Spectral clustering is very simple to implement and has many advantages over the traditional clustering algorithms such as k-means. Non-negative matrix factorization (NMF) factorizes a non-negative data matrix into a product of two non-negative (lower rank) matrices so as to achieve dimension reduction and part-based data representation. In this work, we proved that the spectral clustering under some conditions is equivalent to NMF. Unlike the previous work, we formulate the spectral clustering as a factorization of data matrix (or scaled data matrix) rather than the symmetrical factorization of the symmetrical pairwise similarity matrix as the previous study did. Under the NMF framework, where regularization can be easily incorporated into the spectral clustering, we propose several non-negative and sparse spectral clustering algorithms. Empirical studies on real world data show much better clustering accuracy of the proposed algorithms than some state-of-the-art methods such as ratio cut and normalized cut spectral clustering and non-negative Laplacian embedding. HighlightsWe proved that the spectral clustering is equivalent to NMF.Under the NMF framework, we propose nonnegative sparse spectral clustering model.We propose several algorithms to solve the proposed models.Experiment results on real world data show much better performance."
]
} |
1709.10431 | 2740716165 | We motivate and describe a new freely available human-human dialogue dataset for interactive learning of visually grounded word meanings through ostensive definition by a tutor to a learner. The data has been collected using a novel, character-by-character variant of the DiET chat tool (, 2003; Mills and Healey, submitted) with a novel task, where a Learner needs to learn invented visual attribute words (such as " burchak " for square) from a tutor. As such, the text-based interactions closely resemble face-to-face conversation and thus contain many of the linguistic phenomena encountered in natural, spontaneous dialogue. These include self-and other-correction, mid-sentence continuations, interruptions, overlaps, fillers, and hedges. We also present a generic n-gram framework for building user (i.e. tutor) simulations from this type of incremental data, which is freely available to researchers. We show that the simulations produce outputs that are similar to the original data (e.g. 78 turn match similarity). Finally, we train and evaluate a Reinforcement Learning dialogue control agent for learning visually grounded word meanings, trained from the BURCHAK corpus. The learned policy shows comparable performance to a rule-based system built previously. | There are several existing corpora of human-human spontaneous spoken dialogue, such as SWITCHBOARD @cite_1 , and the British National Corpus, which consist of open, unrestricted telephone conversations between people, where there are no specific tasks to be achieved. These datasets contain many of the incremental dialogue phenomena that we are interested in, but there is no shared visual scene between participants, meaning we cannot use such data to explore learning of perceptually grounded language. More relevant is the MAPTASK corpus @cite_8 , where dialogue participants both have maps which are not shared. This dataset allows investigation of negotiation dialogue, where object names can be agreed, and so does support some work on language grounding. However, in the MAPTASK, grounded word meanings are not taught by ostensive definition as is the case in our new dataset. | {
"cite_N": [
"@cite_1",
"@cite_8"
],
"mid": [
"2166637769",
"2143023047"
],
"abstract": [
"SWITCHBOARD is a large multispeaker corpus of conversational speech and text which should be of interest to researchers in speaker authentication and large vocabulary speech recognition. About 2500 conversations by 500 speakers from around the US were collected automatically over T1 lines at Texas Instruments. Designed for training and testing of a variety of speech processing algorithms, especially in speaker verification, it has over an 1 h of speech from each of 50 speakers, and several minutes each from hundreds of others. A time-aligned word for word transcription accompanies each recording. >",
"The HCRC Map Task corpus has been collected and transcribed in Glasgow and Edinburgh, and recently published on CD-ROM. This effort was made possible by funding from the British Economic and Social Research Council.The corpus is composed of 128 two-person conversations in both high-quality digital audio and orthographic transcriptions, amounting to 18 hours and 150,000 words respectively.The experimental design is quite detailed and complex, allowing a number of different phonemic, syntactico-semantic and pragmatic contrasts to be explored in a controlled way.The corpus is a uniquely valuable resource for speech recognition research in particular, as we move from developing systems intended for controlled use by familiar users to systems intended for less constrained circumstances and naive or occasional users. Examples supporting this claim are given, including preliminary evidence of the phonetic consequences of second mention and the impact of different styles of referent negotiation on communicative efficacy."
]
} |
1709.10431 | 2740716165 | We motivate and describe a new freely available human-human dialogue dataset for interactive learning of visually grounded word meanings through ostensive definition by a tutor to a learner. The data has been collected using a novel, character-by-character variant of the DiET chat tool (, 2003; Mills and Healey, submitted) with a novel task, where a Learner needs to learn invented visual attribute words (such as " burchak " for square) from a tutor. As such, the text-based interactions closely resemble face-to-face conversation and thus contain many of the linguistic phenomena encountered in natural, spontaneous dialogue. These include self-and other-correction, mid-sentence continuations, interruptions, overlaps, fillers, and hedges. We also present a generic n-gram framework for building user (i.e. tutor) simulations from this type of incremental data, which is freely available to researchers. We show that the simulations produce outputs that are similar to the original data (e.g. 78 turn match similarity). Finally, we train and evaluate a Reinforcement Learning dialogue control agent for learning visually grounded word meanings, trained from the BURCHAK corpus. The learned policy shows comparable performance to a rule-based system built previously. | Training a dialogue strategy is one of the fundamental tasks of the user simulation. Approaches to user simulation can be categorised based on the level of abstraction at which the dialogue is modeled: 1) the intention-level has become the most popular user model that predicts the next possible user dialogue action according to the dialogue history and the user task goal @cite_4 @cite_0 @cite_34 @cite_6 @cite_25 @cite_18 @cite_11 ; 2) on the word utterance-level, instead of dialogue action, the user simulation can also be built for predicting the full user utterances or a sequence of words given specific information @cite_20 @cite_13 ; and 3) on the semantic-level, the whole dialogue can be modeled as a sequence of user behaviors in the semantic representation @cite_9 @cite_5 @cite_31 . | {
"cite_N": [
"@cite_18",
"@cite_4",
"@cite_9",
"@cite_6",
"@cite_0",
"@cite_5",
"@cite_31",
"@cite_34",
"@cite_13",
"@cite_25",
"@cite_20",
"@cite_11"
],
"mid": [
"",
"2106547558",
"2062175565",
"",
"2467764055",
"91852349",
"2963827272",
"2128965063",
"2126810476",
"1664253374",
"",
"160067033"
],
"abstract": [
"",
"Automatic speech dialogue systems are becoming common. In order to assess their performance, a large sample of real dialogues has to be collected and evaluated. This process is expensive, labor intensive, and prone to errors. To alleviate this situation we propose a user simulation to conduct dialogues with the system under investigation. Using stochastic modeling of real users we can both debug and evaluate a speech dialogue system while it is still in the lab, thus substantially reducing the amount of field testing with real users.",
"This paper investigates the problem of bootstrapping a statistical dialogue manager without access to training data and proposes a new probabilistic agenda-based method for simulating user behaviour. In experiments with a statistical POMDP dialogue system, the simulator was realistic enough to successfully test the prototype system and train a dialogue policy. An extensive study with human subjects showed that the learned policy was highly competitive, with task completion rates above 90 .",
"",
"User simulation is essential for generating enough data to train a statistical spoken dialogue system. Previous models for user simulation suffer from several drawbacks, such as the inability to take dialogue history into account, the need of rigid structure to ensure coherent user behaviour, heavy dependence on a specific domain, the inability to output several user intentions during one dialogue turn, or the requirement of a summarized action space for tractability. This paper introduces a data-driven user simulator based on an encoder-decoder recurrent neural network. The model takes as input a sequence of dialogue contexts and outputs a sequence of dialogue acts corresponding to user intentions. The dialogue contexts include information about the machine acts and the status of the user goal. We show on the Dialogue State Tracking Challenge 2 (DSTC2) dataset that the sequence-to-sequence model outperforms an agenda-based simulator and an n-gram simulator, according to F-score. Furthermore, we show how this model can be used on the original action space and thereby models user behaviour with finer granularity.",
"Recent work in the area of probabilistic user simulation for training statistical dialogue managers has investigated a new agenda-based user model and presented preliminary experiments with a handcrafted model parameter set. Training the model on dialogue data is an important next step, but non-trivial since the user agenda states are not observable in data and the space of possible states and state transitions is intractably large. This paper presents a summary-space mapping which greatly reduces the number of state transitions and introduces a tree-based method for representing the space of possible agenda state sequences. Treating the user agenda as a hidden variable, the forward backward algorithm can then be successfully applied to iteratively estimate the model parameters on dialogue data. © 2007 Association for Computational Linguistics.",
"",
"This paper presents a probabilistic method to simulate task-oriented human-computer dialogues at the intention level, that may be used to improve or to evaluate the performance of spoken dialogue systems. Our method uses a network of hidden Markov models (HMMs) to predict system and user intentions, where a \"language model\" predicts sequences of goals and the component HMMs predict sequences of intentions. We compare standard HMMs, input HMMs and input-output HMMs in an effort to better predict sequences of intentions. In addition, we propose a dialogue similarity measure to evaluate the realism of the simulated dialogues. We performed experiments using the DARPA communicator corpora and report results with three different metrics: dialogue length, dialogue similarity and precision-recall",
"Human-machine dialogue is heavily influenced by speech recognition and understanding errors and it is hence desirable to train and test statistical dialogue system policies under realistic noise conditions. This paper presents a novel approach to error simulation based on statistical models for word-level utterance generation, ASR confusions, and confidence score generation. While the method explicitly models the context-dependent acoustic confusability of words and allows the system specific language model and semantic decoder to be incorporated, it is computationally inexpensive and thus potentially suitable for running thousands of training simulations. Experimental evaluation results with a POMDP-based dialogue system and the Hidden Agenda User Simulator indicate a close match between the statistical properties of real and synthetic errors.",
"User simulation is frequently used to train statistical dialog managers for task-oriented domains. At present, goal-driven simulators (those that have a persistent notion of what they wish to achieve in the dialog) require some task-specific engineering, making them impossible to evaluate intrinsically. Instead, they have been evaluated extrinsically by means of the dialog managers they are intended to train, leading to circularity of argument. In this paper, we propose the first fully generative goal-driven simulator that is fully induced from data, without hand-crafting or goal annotation. Our goals are latent, and take the form of topics in a topic model, clustering together semantically equivalent and phonetically confusable strings, implicitly modelling synonymy and speech recognition noise. We evaluate on two standard dialog resources, the Communicator and Let's Go datasets, and demonstrate that our model has substantially better fit to held out data than competing approaches. We also show that features derived from our model allow significantly greater improvement over a baseline at distinguishing real from randomly permuted dialogs.",
"",
""
]
} |
1709.10494 | 2950229635 | We present a novel framework for the automatic discovery and recognition of motion primitives in videos of human activities. Given the 3D pose of a human in a video, human motion primitives are discovered by optimizing the motion flux', a quantity which captures the motion variation of a group of skelet al joints. A normalization of the primitives is proposed in order to make them invariant with respect to a subject anatomical variations and data sampling rate. The discovered primitives are unknown and unlabeled and are unsupervisedly collected into classes via a hierarchical non-parametric Bayes mixture model. Once classes are determined and labeled they are further analyzed for establishing models for recognizing discovered primitives. Each primitive model is defined by a set of learned parameters. Given new video data and given the estimated pose of the subject appearing on the video, the motion is segmented into primitives, which are recognized with a probability given according to the parameters of the learned models. Using our framework we build a publicly available dataset of human motion primitives, using sequences taken from well-known motion capture datasets. We expect that our framework, by providing an objective way for discovering and categorizing human motion, will be a useful tool in numerous research fields including video analysis, human inspired motion generation, learning by demonstration, intuitive human-robot interaction, and human behavior analysis. | Neurophysiology studies on motion primitives @cite_72 @cite_90 @cite_67 @cite_66 @cite_79 @cite_5 are based on the idea that kinetic energy and muscular activity are optimized in order to conserve energy. In these works it has been observed that curvature and velocity of joint motion are related. Earliest works such as @cite_14 proposed a relation between curvature and angular velocity. In particular, using their notation, letting @math be the curvature and @math the angular velocity, they called the equation @math the Two-Thirds Power law, valid for certain class of two-dimensional movements. Viviani and Schneider @cite_46 formulated an extension of this law, relating the radius of curvature @math at any point @math along the trajectory with the corresponding tangential velocity @math , in their notation: | {
"cite_N": [
"@cite_67",
"@cite_14",
"@cite_90",
"@cite_79",
"@cite_72",
"@cite_5",
"@cite_46",
"@cite_66"
],
"mid": [
"2067977053",
"2008886640",
"2101143083",
"2032375919",
"13017323",
"2092895305",
"2162235402",
"2294846656"
],
"abstract": [
"The seemingly simple everyday actions of moving limb and body to accomplish a motor task or interact with the environment are incredibly complex. To reach for a target we first need to sense the target's position with respect to an external coordinate system; we then need to plan a limb trajectory which is executed by issuing an appropriate series of neural commands to the muscles. These, in turn, exert appropriate forces and torques on the joints leading to the desired movement of the arm. Here we review some of the earlier work as well as more recent studies on the control of human movement, focusing on behavioral and modeling studies dealing with task space and joint-space movement planning. At the task level, we describe studies investigating trajectory planning and inverse kinematics problems during point-to-point reaching movements as well as two-dimensional (2D) and three-dimensional (3D) drawing movements. We discuss models dealing with the two-thirds power law, particularly differential geometrical approaches dealing with the relation between path geometry and movement velocity. We also discuss optimization principles such as the minimum-jerk model and the isochrony principle for point-to-point and curved movements. We next deal with joint-space movement planning and generation, discussing the inverse kinematics problem and common solutions to the problems of kinematic redundancy. We address the question of which reference frames are used by the nervous system and review studies examining the employment of kinematic constraints such as Donders' and Listing's laws. We also discuss optimization approaches based on Riemannian geometry. One principle of motor coordination during human locomotion emerging from this body of work is the intersegmental law of coordination. However, the nature of the coordinate systems underlying motion planning remains of interest as they are related to the principles governing the control of human arm movements.",
"Considerable evidence exists that the velocity of execution of handwriting and drawing movements depends on some global metric properties of the movement (size, linear extent etc.). Recent experiments have demonstrated that the instantaneous velocity also depends on the local curvature of the trajectory, that is, on the differential geometrical properties of the movement. In this paper we investigate further the role of the differential factors. Experiments are described in which drawing movements of simple geometrical forms and scribbles are performed either freely and extemporaneously, or in the presence of external constraints. It is shown that, at any time during the movement, the velocity component related to differential factors only depends on the value of the curvature of the trajectory at the same time (no dynamics). The relation can be described quantitatively as a specific Power Law and applies to all movements considered here, including those which are performed by following the edge of a template. The fact that the velocity of execution increases with the radius of curvature implies a built-in tendency of the motor control system to keep angular velocity relatively constant and qualifies the Isogony Principle proposed previously. The specific exponent of the Power Law suggests a possible interpretation of this empirical relation.",
"In recent years different lines of evidence have led to the idea that motor actions and movements in both vertebrates and invertebrates are composed of elementary building blocks. The entire motor repertoire can be spanned by applying a well-defined set of operations and transformations to these primitives and by combining them in many different ways according to well-defined syntactic rules. Motor and movement primitives and modules might exist at the neural, dynamic and kinematic levels with complicated mapping among the elementary building blocks subserving these different levels of representation. Hence, while considerable progress has been made in recent years in unravelling the nature of these primitives, new experimental, computational and conceptual approaches are needed to further advance our understanding of motor compositionality.",
"Humans interact with their environment through sensory information and motor actions. These interactions may be understood via the underlying geometry of both perception and action. While the motor space is typically considered by default to be Euclidean, persistent behavioral observations point to a different underlying geometric structure. These observed regularities include the “two-thirds power law”, which connects path curvature with velocity, and “local isochrony”, which prescribes the relation between movement time and its extent. Starting with these empirical observations, we have developed a mathematical framework based on differential geometry, Lie group theory and Cartan’s moving frame method for the analysis of human hand trajectories. We also use this method to identify possible motion primitives, i.e., elementary building blocks from which more complicated movements are constructed. We show that a natural geometric description of continuous repetitive hand trajectories is not Euclidean but equi-affine. Specifically, equi-affine velocity is piecewise constant along movement segments, and movement execution time for a given segment is proportional to its equi-affine arc-length. Using this mathematical framework, we then analyze experimentally recorded drawing movements. To examine movement segmentation and classification, the two fundamental equi-affine differential invariants—equi-affine arc-length and curvature are calculated for the recorded movements. We also discuss the possible role of conic sections, i.e., curves with constant equi-affine curvature, as motor primitives and focus in more detail on parabolas, the equi-affine geodesics. Finally, we explore possible schemes for the internal neural coding of motor commands by showing that the equi-affine framework is compatible with the common model of population coding of the hand velocity vector when combined with a simple assumption on its dynamics. We then discuss several alternative explanations for the role that the equi-affine metric may play in internal representations of motion perception and production.",
"",
"Few computational models have addressed the spatiotemporal features of unconstrained three-dimensional (3D) arm motion. Empirical observations made on hand paths, speed profiles, and arm postures during point-to-point movements led to the assumption that hand path and arm posture are independent of movement speed, suggesting that the geometric and temporal properties of movements are decoupled. In this study, we present a computational model of 3D movements for an arm with four degrees of freedom based on the assumption that optimization principles are separately applied at the geometric and temporal levels of control. Geometric properties (path and posture) are defined in terms of geodesic paths with respect to the kinetic energy metric in the Riemannian configuration space. Accordingly, a geodesic path can be generated with less muscular effort than on any other, nongeodesic path, because the sum of all configuration-speed-dependent torques vanishes. The temporal properties of the movement (speed) are determined in task space by minimizing the squared jerk along the selected end-effector path. The integration of both planning levels into a single spatiotemporal representation simplifies the control of arm dynamics along geodesic paths and results in movements with near minimal torque change and minimal peak value of kinetic energy. Thus, the application of Riemannian geometry allows for a reconciliation of computational models previously proposed for the description of arm movements. We suggest that geodesics are an emergent property of the motor system through the exploration of dynamical space. Our data validated the predictions for joint trajectories, hand paths, final postures, speed profiles, and driving torques.",
"Trajectory and kinematics of drawing movements are mutually constrained by functional relationships that reduce the degrees of freedom of the hand-arm system. Previous investigations of these relationships are extended here by considering their development in children between 5 and 12 years of age. Performances in a simple motor task—the continuous tracing of elliptic trajectories—demonstrate that both the phenomenon of isochrony (increase of the average movement velocity with the linear extent of the trajectory) and the so-called two-thirds power law (relation between tangential velocity and curvature) are qualitatively present already at the age of 5. The quantitative aspects of these regularities evolve with age, however, and steady-state adult performance is not attained even by the oldest children. The power-law formalism developed in previous reports is generalized to encompass these developmental aspects of the control of movement. Two general frameworks are currently available to conceptualize the motor-control problem. Broadly, the two frameworks differ in the answer that they give to the question \"Where do form and structure come from?\" According to the motor-program view (e.g., Schmidt, 1988), form and structure in a movement come from a central abstract representation of the intended result. Thus, making a movement entails a detailed mapping of this representation into an appropriate motor plan. The competing task-dynamic view (e.g., Kugler, 1989; Saltzman & Kelso, 1987) emphasizes instead the global morphogenetic power of the nonlinearities of the motor system. These two views need not be mutually exclusive. In fact, different classes of movements may require different conceptual frameworks for their analysis. At any rate (at least in the case of drawing movements considered here), an internal representation of the intended trajectory and kinematics must be available before execution. Consequently, much effort is being devoted to understanding how this internal model is translated into a set of motor commands. In this respect, one basic conceptual difficulty is because the geometry of the arm-forearm-hand system affords a large number of degrees of freedom. Provided that there are no constraints in the workspace, the only trajectory that matters in many manual",
""
]
} |
1709.10494 | 2950229635 | We present a novel framework for the automatic discovery and recognition of motion primitives in videos of human activities. Given the 3D pose of a human in a video, human motion primitives are discovered by optimizing the motion flux', a quantity which captures the motion variation of a group of skelet al joints. A normalization of the primitives is proposed in order to make them invariant with respect to a subject anatomical variations and data sampling rate. The discovered primitives are unknown and unlabeled and are unsupervisedly collected into classes via a hierarchical non-parametric Bayes mixture model. Once classes are determined and labeled they are further analyzed for establishing models for recognizing discovered primitives. Each primitive model is defined by a set of learned parameters. Given new video data and given the estimated pose of the subject appearing on the video, the motion is segmented into primitives, which are recognized with a probability given according to the parameters of the learned models. Using our framework we build a publicly available dataset of human motion primitives, using sequences taken from well-known motion capture datasets. We expect that our framework, by providing an objective way for discovering and categorizing human motion, will be a useful tool in numerous research fields including video analysis, human inspired motion generation, learning by demonstration, intuitive human-robot interaction, and human behavior analysis. | where the constants @math , @math and @math has a value close to @math . An equivalent Power law for trajectories in 3D space is introduced by @cite_6 and it is called the curvature-torsion power law and is defined as @math , where @math is the curvature of the trajectory, @math the torsion, @math the spatial movement speed, @math and @math are constants. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2320004782"
],
"abstract": [
"The two-thirds power law, postulating an inverse local relation between the velocity and cubed root of curvature of planar trajectories, is a long-established simplifying principle of human hand movements. In perception, the motion of a dot along a planar elliptical path appears most uniform for speed profiles closer to those predicted by the power law than to constant Euclidean speed, a kinetic-visual illusion. Mathematically, complying with this law is equivalent to moving at constant planar equi-affine speed, while unconstrained three-dimensional drawing movements generally follow constant spatial equi-affine speed. Here we test the generalization of this illusion to visual perception of spatial motion for a dot moving along five differently shaped paths, using stereoscopic projection. The movements appeared most uniform for speed profiles closer to constant spatial equi-affine speed than to constant Euclidean speed, with path torsion (i.e., local deviation from planarity) directly affecting the speed ..."
]
} |
1709.10494 | 2950229635 | We present a novel framework for the automatic discovery and recognition of motion primitives in videos of human activities. Given the 3D pose of a human in a video, human motion primitives are discovered by optimizing the motion flux', a quantity which captures the motion variation of a group of skelet al joints. A normalization of the primitives is proposed in order to make them invariant with respect to a subject anatomical variations and data sampling rate. The discovered primitives are unknown and unlabeled and are unsupervisedly collected into classes via a hierarchical non-parametric Bayes mixture model. Once classes are determined and labeled they are further analyzed for establishing models for recognizing discovered primitives. Each primitive model is defined by a set of learned parameters. Given new video data and given the estimated pose of the subject appearing on the video, the motion is segmented into primitives, which are recognized with a probability given according to the parameters of the learned models. Using our framework we build a publicly available dataset of human motion primitives, using sequences taken from well-known motion capture datasets. We expect that our framework, by providing an objective way for discovering and categorizing human motion, will be a useful tool in numerous research fields including video analysis, human inspired motion generation, learning by demonstration, intuitive human-robot interaction, and human behavior analysis. | In robotics the paradigm of transferring human motion primitives to robot movements is paramount for imitation learning and, more recently to implement human-robot collaboration @cite_1 . A good amount of research in robotics has approached primitives in terms of Dynamic Movement Primitives (DMP) @cite_1 to model elementary motor behaviors as attractor systems, representing them with differential equations. Typical applications are learning by imitation or learning from demonstration @cite_53 @cite_58 @cite_71 @cite_42 , learning task specifications @cite_27 , modeling interaction primitives @cite_18 . Motion primitives are represented either via Hidden Markov models or Gaussian Mixture Models (GMM). @cite_30 present an approach based on HMM for imitation learning of arm movements, and @cite_3 model arm motion primitives via GMM. | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_53",
"@cite_42",
"@cite_1",
"@cite_3",
"@cite_27",
"@cite_71",
"@cite_58"
],
"mid": [
"2005870832",
"2012058822",
"1894373409",
"2149192758",
"2136719407",
"2210365848",
"2182585001",
"2167117957",
"2161395589"
],
"abstract": [
"In this paper, we deal with imitation learning of arm movements in humanoid robots. Hidden Markov Models (HMM) are used to generalize movements demonstrated to a robot multiple times. They are trained with the characteristic features (key points) of each demonstration. Using the same HMM, key points that are common to all demonstrations are identified; only those are considered when reproducing a movement. We also show how HMM can be used to detect temporal dependencies between both arms in dual-arm tasks. We created a model of the human upper body to simulate the reproduction of dual-arm movements and generate natural-looking joint configurations from tracked hand paths. Results are presented and discussed.",
"To engage in cooperative activities with human partners, robots have to possess basic interactive abilities and skills. However, programming such interactive skills is a challenging task, as each interaction partner can have different timing or an alternative way of executing movements. In this paper, we propose to learn interaction skills by observing how two humans engage in a similar task. To this end, we introduce a new representation called Interaction Primitives. Interaction primitives build on the framework of dynamic motor primitives (DMPs) by maintaining a distribution over the parameters of the DMP. With this distribution, we can learn the inherent correlations of cooperative activities which allow us to infer the behavior of the partner and to participate in the cooperation. We will provide algorithms for synchronizing and adapting the behavior of humans and robots during joint physical activities.",
"In this paper we propose and evaluate a control system to (1) learn and (2) adapt robot motion for continuous non-rigid contact with the environment. We present the approach in the context of wiping surfaces with robots. Our approach is based on learning by demonstration. First an initial periodic motion, covering the essence of the wiping task, is transferred from a human to a robot. The system extracts and learns one period of motion. Once the user demonstrator is content with the motion, the robot seeks and establishes contact with a given surface, maintaining a predefined force of contact through force feedback. The shape of the surface is encoded for the complete period of motion, but the robot can adapt to a different surface, perturbations or obstacles. The novelty stems from the fact that the feedforward component is learned and encoded in a dynamic movement primitive. By using the feedforward component, the feedback component is greatly reduced if not completely canceled. Finally, if the user is not satisfied with the periodic pattern, he she can change parts of motion through predefined gestures or through physical contact in a manner of a tutor or a coach.The complete system thus allows not only a transfer of motion, but a transfer of motion with matching correspondences, i.e. wiping motion is constrained to maintain physical contact with the surface to be wiped. The interface for both learning and adaptation is simple and intuitive and allows for fast and reliable knowledge transfer to the robot.Simulated and real world results in the application domain of wiping a surface are presented on three different robotic platforms. Results of the three robotic platforms, namely a 7 degree-of-freedom Kuka LWR-4 robot, the ARMAR-IIIa humanoid platform and the Sarcos CB-i humanoid robot, depict different methods of adaptation to the environment and coaching. An intuitive and user friendly system for transferring of skills from a person to a robot.It allows online learning and adaptation of motion trajectories.It allows adaptation of trajectories through human coaching, from either force or visual feedback.It is based on the dynamic motion primitives framework.Surface wiping use-case through non-rigid contact is demonstrated and evaluated.",
"Robots in a human environment need to be compliant. This compliance requires that a preplanned movement can be adapted to an obstacle that may be moving or appearing unexpectedly. Here, we present a general framework for movement generation and mid-flight adaptation to obstacles. For robust motion generation, developed the framework of dynamic movement primitives which represent a demonstrated movement with a set of differential equations. These equations allow adding a perturbing force without sacrificing stability of the desired movement. We extend this framework such that arbitrary movements in end-effector space can be represented - which was not possible before. Furthermore, we include obstacle avoidance by adding to the equations of motion a repellent force - a gradient of a potential field centered around the obstacle. In addition, this article compares different potential fields and shows how to avoid obstacle-link collisions within this framework. We demonstrate the abilities of our approach in simulations and with an anthropomorphic robot arm.",
"Nonlinear dynamical systems have been used in many disciplines to model complex behaviors, including biological motor control, robotics, perception, economics, traffic prediction, and neuroscience. While often the unexpected emergent behavior of nonlinear systems is the focus of investigations, it is of equal importance to create goal-directed behavior e.g., stable locomotion from a system of coupled oscillators under perceptual guidance. Modeling goal-directed behavior with nonlinear systems is, however, rather difficult due to the parameter sensitivity of these systems, their complex phase transitions in response to subtle parameter changes, and the difficulty of analyzing and predicting their long-term behavior; intuition and time-consuming parameter tuning play a major role. This letter presents and reviews dynamical movement primitives, a line of research for modeling attractor behaviors of autonomous nonlinear dynamical systems with the help of statistical learning techniques. The essence of our approach is to start with a simple dynamical system, such as a set of linear differential equations, and transform those into a weakly nonlinear system with prescribed attractor dynamics by means of a learnable autonomous forcing term. Both point attractors and limit cycle attractors of almost arbitrary complexity can be generated. We explain the design principle of our approach and evaluate its properties in several example applications in motor control and robotics.",
"This paper focuses on recognition and prediction of human reaching motion in industrial manipulation tasks. Several supervised learning methods have been proposed for this purpose, but we seek a method that can build models on-the-fly and adapt to new people and new motion styles as they emerge. Thus, unlike previous work, we propose an unsupervised online learning approach to the problem, which requires no offline training or manual categorization of trajectories. Our approach consists of a two-layer library of Gaussian Mixture Models that can be used both for recognition and prediction. We do not assume that the number of motion classes is known a priori, and thus the library grows if it cannot explain a new observed trajectory. Given an observed portion of a trajectory, the framework can predict the remainder of the trajectory by first determining what GMM it belongs to, and then using Gaussian Mixture Regression to predict the remainder of the trajectory. We tested our method on motion-capture data recorded during assembly tasks. Our results suggest that the proposed framework outperforms supervised methods in terms of both recognition and prediction. We also show the benefit of using our two-layer framework over simpler approaches.",
"In this paper, we propose an approach for learning task specifications automatically, by observing human demonstrations. Using this approach allows a robot to combine representations of individual actions to achieve a high-level goal. We hypothesize that task specifications consist of variables that present a pattern of change that is invariant across demonstrations. We identify these specifications at different stages of task completion. Changes in task constraints allow us to identify transitions in the task description and to segment them into subtasks. We extract the following task-space constraints: 1) the reference frame in which to express the task variables; 2) the variable of interest at each time step, position, or force at the end effector; and 3) a factor that can modulate the contribution of force and position in a hybrid impedance controller. The approach was validated on a seven-degree-of-freedom Kuka arm, performing two different tasks: grating vegetables and extracting a battery from a charging stand.",
"Many motor skills in humanoid robotics can be learned using parametrized motor primitives as done in imitation learning. However, most interesting motor learning problems are high-dimensional reinforcement learning problems often beyond the reach of current methods. In this paper, we extend previous work on policy learning from the immediate reward case to episodic reinforcement learning. We show that this results in a general, common framework also connected to policy gradient methods and yielding a novel algorithm for policy learning that is particularly well-suited for dynamic motor primitives. The resulting algorithm is an EM-inspired algorithm applicable to complex motor learning tasks. We compare this algorithm to several well-known parametrized policy search methods and show that it outperforms them. We apply it in the context of motor learning and show that it can learn a complex Ball-in-a-Cup task using a real Barrett WAM™ robot arm.",
"We provide a general approach for learning robotic motor skills from human demonstration. To represent an observed movement, a non-linear differential equation is learned such that it reproduces this movement. Based on this representation, we build a library of movements by labeling each recorded movement according to task and context (e.g., grasping, placing, and releasing). Our differential equation is formulated such that generalization can be achieved simply by adapting a start and a goal parameter in the equation to the desired position values of a movement. For object manipulation, we present how our framework extends to the control of gripper orientation and finger position. The feasibility of our approach is demonstrated in simulation as well as on the Sarcos dextrous robot arm. The robot learned a pick-and-place operation and a water-serving task and could generalize these tasks to novel situations."
]
} |
1709.10494 | 2950229635 | We present a novel framework for the automatic discovery and recognition of motion primitives in videos of human activities. Given the 3D pose of a human in a video, human motion primitives are discovered by optimizing the motion flux', a quantity which captures the motion variation of a group of skelet al joints. A normalization of the primitives is proposed in order to make them invariant with respect to a subject anatomical variations and data sampling rate. The discovered primitives are unknown and unlabeled and are unsupervisedly collected into classes via a hierarchical non-parametric Bayes mixture model. Once classes are determined and labeled they are further analyzed for establishing models for recognizing discovered primitives. Each primitive model is defined by a set of learned parameters. Given new video data and given the estimated pose of the subject appearing on the video, the motion is segmented into primitives, which are recognized with a probability given according to the parameters of the learned models. Using our framework we build a publicly available dataset of human motion primitives, using sequences taken from well-known motion capture datasets. We expect that our framework, by providing an objective way for discovering and categorizing human motion, will be a useful tool in numerous research fields including video analysis, human inspired motion generation, learning by demonstration, intuitive human-robot interaction, and human behavior analysis. | There is a significant amount of literature on abnormality detection in surveillance videos. Only few of them, though, are concerned with dangerous behaviors. These methods can be further divided into those detecting dangerous crowd behaviors, in which the individual motion is superseded by large flows as in @cite_60 @cite_85 @cite_80 @cite_24 , and those detecting closer dangerous human behaviors. | {
"cite_N": [
"@cite_24",
"@cite_80",
"@cite_85",
"@cite_60"
],
"mid": [
"",
"1999499433",
"1931112481",
"2520489243"
],
"abstract": [
"",
"This paper presents a novel video descriptor, referred to as Histogram of Oriented Track lets, for recognizing abnormal situation in crowded scenes. Unlike standard approaches that use optical flow, which estimates motion vectors only from two successive frames, we built our descriptor over long-range motion trajectories which is called track lets in the literature. Following the standard procedure, we divided video sequences in spatio-temporal cuboids within which we collected statistics on the track lets passing through them. In particular, we quantized orientation and magnitude in a 2-dimensional histogram which encodes the motion patterns expected in each cuboid. We classify frames as normal and abnormal by using Latent Dirichlet Allocation and Support Vector Machines. We evaluated the effectiveness of the proposed descriptors on three datasets: UCSD, Violence in Crowds and UMN. The experiments demonstrated (i) very promising results in abnormality detection, (ii) setting new state-of-the-art on two of them, and (iii) outperforming former descriptors based on the optical flow, dense trajectories and the social force model.",
"This paper presents a novel video descriptor based on substantial derivative, an important concept in fluid mechanics, that captures the rate of change of a fluid property as it travels through a velocity field. Unlike standard approaches that only use temporal motion information, our descriptor exploits the spatio-temporal characteristic of substantial derivative. In particular, the spatial and temporal motion patterns are captured by respectively the convective and local accelerations. After estimating the convective and local field from the optic flow, we followed the standard bag-of-word procedure for each motion pattern separately, and we concatenated the two resulting histograms to form the final descriptor. We extensively evaluated the effectiveness of the proposed method on five benchmarks, including three standard datasets (Violence in Movies, Violence In Crowd, and BEHAVE), and two new video-survelliance sequences downloaded from Youtube. Our experiments show how the proposed approach sets the new state-of-the-art on all benchmarks and how the structural information captured by convective acceleration is essential to detect violent episodes in crowded scenarios.",
"Approaches inspired by Newtonian mechanics have been successfully applied for detecting abnormal behaviors in crowd scenarios, being the most notable example the Social Force Model (SFM). This class of approaches describes the movements and local interactions among individuals in crowds by means of repulsive and attractive forces. Despite their promising performance, recent socio-psychology studies have shown that current SFM-based methods may not be capable of explaining behaviors in complex crowd scenarios. An alternative approach consists in describing the cognitive processes that gives rise to the behavioral patterns observed in crowd using heuristics. Inspired by these studies, we propose a new hybrid framework to detect violent events in crowd videos. More specifically, (i) we define a set of simple behavioral heuristics to describe people behaviors in crowd, and (ii) we implement these heuristics into physical equations, being able to model and classify such behaviors in the videos. The resulting heuristic maps are used to extract video features to distinguish violence from normal events. Our violence detection results set the new state of the art on several standard benchmarks and demonstrate the superiority of our method compared to standard motion descriptors, previous physics-inspired models used for crowd analysis and pre-trained ConvNet for crowd behavior analysis."
]
} |
1709.10494 | 2950229635 | We present a novel framework for the automatic discovery and recognition of motion primitives in videos of human activities. Given the 3D pose of a human in a video, human motion primitives are discovered by optimizing the motion flux', a quantity which captures the motion variation of a group of skelet al joints. A normalization of the primitives is proposed in order to make them invariant with respect to a subject anatomical variations and data sampling rate. The discovered primitives are unknown and unlabeled and are unsupervisedly collected into classes via a hierarchical non-parametric Bayes mixture model. Once classes are determined and labeled they are further analyzed for establishing models for recognizing discovered primitives. Each primitive model is defined by a set of learned parameters. Given new video data and given the estimated pose of the subject appearing on the video, the motion is segmented into primitives, which are recognized with a probability given according to the parameters of the learned models. Using our framework we build a publicly available dataset of human motion primitives, using sequences taken from well-known motion capture datasets. We expect that our framework, by providing an objective way for discovering and categorizing human motion, will be a useful tool in numerous research fields including video analysis, human inspired motion generation, learning by demonstration, intuitive human-robot interaction, and human behavior analysis. | Among the latter there are methods focusing on fights @cite_75 , methods specialized on violence @cite_28 @cite_63 @cite_23 @cite_19 , on aggressive behaviors @cite_45 , and on crime @cite_29 . A review on methods for detecting abnormal behaviors, taking into account some of the above mentioned ones, and also discussing available datasets, is provided in @cite_77 . | {
"cite_N": [
"@cite_28",
"@cite_29",
"@cite_19",
"@cite_77",
"@cite_45",
"@cite_23",
"@cite_63",
"@cite_75"
],
"mid": [
"2895554091",
"",
"2054888992",
"2756233686",
"2331812431",
"1932571874",
"2286655808",
""
],
"abstract": [
"",
"",
"To detect violence in a video, a common video description method is to apply local spatio-temporal description on the query video. Then, the low-level description is further summarized onto the high-level feature based on Bag-of-Words (BoW) model. However, traditional spatio-temporal descriptors are not discriminative enough. Moreover, BoW model roughly assigns each feature vector to only one visual word, therefore inevitably causing quantization error. To tackle the constrains, this paper employs Motion SIFT (MoSIFT) algorithm to extract the low-level description of a query video. To eliminate the feature noise, Kernel Density Estimation (KDE) is exploited for feature selection on the MoSIFT descriptor. In order to obtain the highly discriminative video feature, this paper adopts sparse coding scheme to further process the selected MoSIFTs. Encouraging experimental results are obtained based on two challenging datasets which record both crowded scenes and non-crowded scenes.",
"Different levels of an intelligent video surveillance system (IVVS) are studied in this review.Existing approaches for abnormal behavior recognition relative to each level of an IVVS are extensively reviewed.Challenging datasets for IVVS evaluation are presented.Limitations of the abnormal behavior recognition area are discussed. With the increasing number of surveillance cameras in both indoor and outdoor locations, there is a grown demand for an intelligent system that detects abnormal events. Although human action recognition is a highly reached topic in computer vision, abnormal behavior detection is lately attracting more research attention. Indeed, several systems are proposed in order to ensure human safety. In this paper, we are interested in the study of the two main steps composing a video surveillance system which are the behavior representation and the behavior modeling. Techniques related to feature extraction and description for behavior representation are reviewed. Classification methods and frameworks for behavior modeling are also provided. Moreover, available datasets and metrics for performance evaluation are presented. Finally, examples of existing video surveillance systems used in real world are described.",
"A system to monitor aggression in surveillance scenes from audio and video.Person motion and proximity measured in volumetric representation of tracked people.Informative sound classes are extracted in challenging acoustic conditions.DBN fuses context and the multi-modal features into latent aggression estimate.Comparison to previous work and system parts shows benefit of combining modalities. This paper presents a smart surveillance system named CASSANDRA, aimed at detecting instances of aggressive human behavior in public environments. A distinguishing aspect of CASSANDRA is the exploitation of complementary audio and video cues to disambiguate scene activity in real-life environments. From the video side, the system uses overlapping cameras to track persons in 3D and to extract features regarding the limb motion relative to the torso. From the audio side, it classifies instances of speech, screaming, singing, and kicking-object. The audio and video cues are fused with contextual cues (interaction, auxiliary objects); a Dynamic Bayesian Network (DBN) produces an estimate of the ambient aggression level.Our prototype system is validated on a realistic set of scenarios performed by professional actors at an actual train station to ensure a realistic audio and video noise setting.",
"Whereas the action recognition problem has become a hot topic within computer vision, the detection of fights or in general aggressive behavior has been comparatively less studied. Such capability may be extremely useful in some video surveillance scenarios like in prisons, psychiatric centers or even embedded in camera phones. Recent work has considered the well-known Bag-of-Words framework often used in generic action recognition for the specific problem of fight detection. Under this framework, spatio-temporal features are extracted from the video sequences and used for classification. Despite encouraging results in which near 90 accuracy rates were achieved for this specific task, the computational cost of extracting such features is prohibitive for practical applications, particularly in surveillance and media rating systems. The task of violence detection may have, however, specific features that can be leveraged. Inspired by psychology results that suggest that kinematic features alone are discriminant for specific actions, this work proposes a novel method which uses extreme acceleration patterns as the main feature. These extreme accelerations are efficiently estimated by applying the Radon transform to the power spectrum of consecutive frames. Experiments show that accuracy improvements of up to 12 are achieved with respect to state-of-the-art generic action recognition methods. Most importantly, the proposed method is at least 15 times faster.",
"Nowadays, with so many surveillance cameras having been installed, the market demand for intelligent violence detection is continuously growing, while it is still a challenging topic in research area. Therefore, we attempt to make some improvements of existing violence detectors. The primary contributions of this paper are two-fold. Firstly, a novel feature extraction method named Oriented VIolent Flows (OViF), which takes full advantage of the motion magnitude change information in statistical motion orientations, is proposed for practical violence detection in videos. The comparison of OViF and baseline approaches on two public databases demonstrates the efficiency of the proposed method. Secondly, feature combination and multi-classifier combination strategies are adopted and excellent results are obtained. Experimental results show that using combined features with AdaBoost+Linear-SVM achieves improved performance over the state-of-the-art on the Violent-Flows benchmark. Display Omitted A novel feature named OViF for violence detection.OViF makes full use of the orientation information of optical flows compared to ViF.We use AdaBoost as a feature selector and Linear SVM as a classifier.Experiments show that ViF+OViF obtain the state-of-the-art violence detection performance when using AdaBoost with SVM.",
""
]
} |
1709.10257 | 2758952034 | Detection of engagement during a conversation is an important function of human-robot interaction. The level of user engagement can influence the dialogue strategy of the robot. Our motivation in this work is to detect several behaviors which will be used as social signal inputs for a real-time engagement recognition model. These behaviors are nodding, laughter, verbal backchannels and eye gaze. We describe models of these behaviors which have been learned from a large corpus of human-robot interactions with the android robot ERICA. Input data to the models comes from a Kinect sensor and a microphone array. Using our engagement recognition model, we can achieve reasonable performance using the inputs from automatic social signal detection, compared to using manual annotation as input. | The ability to recognize engagement in the user can influence the conversation. Previous works have found that engagement is related to turn-taking @cite_30 @cite_20 . There has also been an implementation of an agent which modifies its dialogue strategy according to the engagement level of the user @cite_7 . We intend for ERICA to be used in a number of conversational scenarios such as interviewing, so the ability to detect the engagement level of the user is crucial to maintain a positive interaction. | {
"cite_N": [
"@cite_30",
"@cite_7",
"@cite_20"
],
"mid": [
"2084624217",
"2565288664",
"2553783396"
],
"abstract": [
"Recognizing users' engagement state and intentions is a pressing task for computational agents to facilitate fluid conversations in situated interactions. We investigate how to quantitatively evaluate high-level user engagement and intentions based on low-level visual cues, and how to design engagement-aware behaviors for the conversational agents to behave in a sociable manner. Drawing on machine learning techniques, we propose two computational models to quantify users' attention saliency and engagement intentions. Their performances are validated by a close match between the predicted values and the ground truth annotation data. Next, we design a novel engagement-aware behavior model for the agent to adjust its direction of attention and manage the conversational floor based on the estimated users' engagement. In a user study, we evaluated the agent's behaviors in a multiparty dialog scenario. The results show that the agent's engagement-aware behaviors significantly improved the effectiveness of communication and positively affected users' experience.",
"",
"We address the annotation of engagement in the context of human-machine interaction. Engagement represents the level of how much a user is being interested in and willing to continue the current interaction. The conversational data used in the annotation work is a human-robot interaction corpus where a human subject talks with the android ERICA, which is remotely operated by another human subject. The annotation work was done by multiple third-party annotators, and the task was to detect the time point when the level of engagement becomes high. The annotation results indicate that there are agreements among the annotators although the numbers of annotated points are different among them. It is also found that the level of engagement is related to turn-taking behaviors. Furthermore, we conducted interviews with the annotators to reveal behaviors used to show a high level of engagement. The results suggest that laughing, backchannels and nodding are related to the level of engagement."
]
} |
1709.10257 | 2758952034 | Detection of engagement during a conversation is an important function of human-robot interaction. The level of user engagement can influence the dialogue strategy of the robot. Our motivation in this work is to detect several behaviors which will be used as social signal inputs for a real-time engagement recognition model. These behaviors are nodding, laughter, verbal backchannels and eye gaze. We describe models of these behaviors which have been learned from a large corpus of human-robot interactions with the android robot ERICA. Input data to the models comes from a Kinect sensor and a microphone array. Using our engagement recognition model, we can achieve reasonable performance using the inputs from automatic social signal detection, compared to using manual annotation as input. | We must first detect the relevant social signals which indicate engagement. There have been numerous works which detail these behaviors and apply them to agents and robots. They include facial expressions @cite_13 @cite_3 , posture @cite_22 , conversational phenomena @cite_18 @cite_30 as well as the four social signals which we investigate in this work. Typically the models to recognize social signals are trained using machine learning techniques. | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_22",
"@cite_3",
"@cite_13"
],
"mid": [
"2084624217",
"2029731463",
"2105410877",
"2887762521",
"2077106070"
],
"abstract": [
"Recognizing users' engagement state and intentions is a pressing task for computational agents to facilitate fluid conversations in situated interactions. We investigate how to quantitatively evaluate high-level user engagement and intentions based on low-level visual cues, and how to design engagement-aware behaviors for the conversational agents to behave in a sociable manner. Drawing on machine learning techniques, we propose two computational models to quantify users' attention saliency and engagement intentions. Their performances are validated by a close match between the predicted values and the ground truth annotation data. Next, we design a novel engagement-aware behavior model for the agent to adjust its direction of attention and manage the conversational floor based on the estimated users' engagement. In a user study, we evaluated the agent's behaviors in a multiparty dialog scenario. The results show that the agent's engagement-aware behaviors significantly improved the effectiveness of communication and positively affected users' experience.",
"Based on a study of the engagement process between humans, we have developed and implemented an initial computational model for recognizing engagement between a human and a humanoid robot. Our model contains recognizers for four types of connection events involving gesture and speech: directed gaze, mutual facial gaze, conversational adjacency pairs and backchannels. To facilitate integrating and experimenting with our model in a broad range of robot architectures, we have packaged it as a node in the open-source Robot Operating System (ROS) framework. We have conducted a preliminary validation of our computational model and implementation in a simple human-robot pointing game.",
"The design of an affect recognition system for socially perceptive robots relies on representative data: human-robot interaction in naturalistic settings requires an affect recognition system to be trained and validated with contextualised affective expressions, that is, expressions that emerge in the same interaction scenario of the target application. In this paper we propose an initial computational model to automatically analyse human postures and body motion to detect engagement of children playing chess with an iCat robot that acts as a game companion. Our approach is based on vision-based automatic extraction of expressive postural features from videos capturing the behaviour of the children from a lateral view. An initial evaluation, conducted by training several recognition models with contextualised affective postural expressions, suggests that patterns of postural behaviour can be used to accurately predict the engagement of the children with the robot, thus making our approach suitable for integration into an affect recognition system for a game companion in a real world scenario.",
"In complex conversation tasks, people react to their interlocutor’s state, such as uncertainty and engagement to improve conversation effectiveness Forbes-Riley and Litman (Adapting to student uncertainty improves tutoring dialogues, pp 33–40, 2009 [2]). If a conversational system reacts to a user’s state, would that lead to a better conversation experience? To test this hypothesis, we designed and implemented a dialog system that tracks and reacts to a user’s state, such as engagement, in real time. We designed and implemented a conversational job interview task based on the proposed framework. The system acts as an interviewer and reacts to user’s disengagement in real-time with positive feedback strategies designed to re-engage the user in the job interview process. Experiments suggest that users speak more while interacting with the engagement-coordinated version of the system as compared to a non-coordinated version. Users also reported the former system as being more engaging and providing a better user experience.",
"Affect sensitivity is of the utmost importance for a robot companion to be able to display socially intelligent behaviour, a key requirement for sustaining long-term interactions with humans. This paper explores a naturalistic scenario in which children play chess with the iCat, a robot companion. A person-independent, Bayesian approach to detect the user's engagement with the iCat robot is presented. Our framework models both causes and effects of engagement: features related to the user's non-verbal behaviour, the task and the companion's affective reactions are identified to predict the children's level of engagement. An experiment was carried out to train and validate our model. Results show that our approach based on multimodal integration of task and social interaction-based features outperforms those based solely on non-verbal behaviour or contextual information (94.79 vs. 93.75 and 78.13 )."
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | * Hough transform We devote this section to a brief introduction to the HT technique, while we refer to recent surveys (for instance @cite_49 @cite_20 ) for a detailed overview. | {
"cite_N": [
"@cite_20",
"@cite_49"
],
"mid": [
"2094583515",
"2029895480"
],
"abstract": [
"The generalised Hough transform (GHT) is useful for detecting or locating translated two-dimensional objects. However, a weakness of the GHT is its storage requirements and hence the increased computational complexity resulting from the four-dimensional parameter space. In this paper, we present the results of our work which involves investigation of the performance of several efficient GHT techniques including an extension of Thomas's rotation-invariant algorithm. It is shown that our extension of Thomas's algorithm has very low memory requirements and computational complexity, and produces the best results in various tests.",
"In 1962 Hough earned the patent for a method 1], popularly called Hough Transform (HT) that efficiently identifies lines in images. It is an important tool even after the golden jubilee year of existence, as evidenced by more than 2500 research papers dealing with its variants, generalizations, properties and applications in diverse fields. The current paper is a survey of HT and its variants, their limitations and the modifications made to overcome them, the implementation issues in software and hardware, and applications in various fields. Our survey, along with more than 200 references, will help the researchers and students to get a comprehensive view on HT and guide them in applying it properly to their problems of interest. HighlightsA survey of Hough Transform (HT) has been done.The main variants of HT and its formulations have been described.The hardware and software implementations of HT have been described.HT has many applications in diverse fields. Out of them some applications in some selected fields have been mentioned."
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | HT is a standard pattern recognition technique originally used to detect straight lines in images, @cite_2 @cite_53 . Since its original conception, HT has been extensively used and many generalizations have been proposed for identifying instances of arbitrary shapes over images @cite_5 , more commonly circles or ellipses. The first very popular extension concerning the detection of any parametric analytic curve is usually referred to as the Standard Hough Transform (SHT). In spite of the robustness of SHT to discontinuity or missing data on the curve, its use has been limited by a number of drawbacks, like the need of a parametric expression, or the dependence of the computation time and the memory requirements on the number of curve parameters, or even the need of finer parameters quantization for a higher accuracy of results. | {
"cite_N": [
"@cite_5",
"@cite_53",
"@cite_2"
],
"mid": [
"22745672",
"2095905764",
"2104241941"
],
"abstract": [
"Abstract The Hough transform is a method for detecting curves by exploiting the duality between points on a curve and parameters of that curve. The initial work showed how to detect both analytic curves (1,2) and non-analytic curves, (3) but these methods were restricted to binary edge images. This work was generalized to the detection of some analytic curves in grey level images, specifically lines, (4) circles (5) and parabolas. (6) The line detection case is the best known of these and has been ingeniously exploited in several applications. (7,8,9) We show how the boundaries of an arbitrary non-analytic shape can be used to construct a mapping between image space and Hough transform space. Such a mapping can be exploited to detect instances of that particular shape in an image. Furthermore, variations in the shape such as rotations, scale changes or figure ground reversals correspond to straightforward transformations of this mapping. However, the most remarkable property is that such mappings can be composed to build mappings for complex shapes from the mappings of simpler component shapes. This makes the generalized Hough transform a kind of universal transform which can be used to find arbitrarily complex shapes.",
"Hough has proposed an interesting and computationally efficient procedure for detecting lines in pictures. This paper points out that the use of angle-radius rather than slope-intercept parameters simplifies the computation further. It also shows how the method can be used for more general curve fitting, and gives alternative interpretations that explain the source of its efficiency.",
""
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | To overcome some of these limitations, other variants have been proposed, one of the most popular being the Generalized Hough Transform (GHT) by Ballard @cite_5 . Since its conception, it proved to be very useful for detecting and locating translated two-dimensional objects, without requiring a parametric analytic expression. Thus, GHT is more general than SHT, as it is able to detect a larger class of rigid objects, still retaining the robustness of SHT. Nevertheless, GHT often requires brute force to enumerate all the possible orientations and scales of the input shape, thus the number of parameters needs to be increased in its process. Further, GHT cannot adequately handle shapes that are more flexible, as in the case of different instances of the same shape, which are similar but not identical, e.g. pet als and leaves. | {
"cite_N": [
"@cite_5"
],
"mid": [
"22745672"
],
"abstract": [
"Abstract The Hough transform is a method for detecting curves by exploiting the duality between points on a curve and parameters of that curve. The initial work showed how to detect both analytic curves (1,2) and non-analytic curves, (3) but these methods were restricted to binary edge images. This work was generalized to the detection of some analytic curves in grey level images, specifically lines, (4) circles (5) and parabolas. (6) The line detection case is the best known of these and has been ingeniously exploited in several applications. (7,8,9) We show how the boundaries of an arbitrary non-analytic shape can be used to construct a mapping between image space and Hough transform space. Such a mapping can be exploited to detect instances of that particular shape in an image. Furthermore, variations in the shape such as rotations, scale changes or figure ground reversals correspond to straightforward transformations of this mapping. However, the most remarkable property is that such mappings can be composed to build mappings for complex shapes from the mappings of simpler component shapes. This makes the generalized Hough transform a kind of universal transform which can be used to find arbitrarily complex shapes."
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | Recently, thanks to algebraic geometry concepts, theoretical foundations have been laid to extend the HT technique to the detection of algebraic objects of codimension greater than one (for instance algebraic space curves) taking advantage of various families of algebraic plane curves (see @cite_8 and @cite_54 ). Being so general, such a method allows to deal with different shapes, possibly compound, and to get the most suitable approximating profile among a large vocabulary of curves. | {
"cite_N": [
"@cite_54",
"@cite_8"
],
"mid": [
"1965230789",
"2963795227"
],
"abstract": [
"The Hough transform is a standard pattern recognition technique introduced between the 1960s and the 1970s for the detection of straight lines, circles, and ellipses. Here we offer a mathematical foundation, based on algebraic-geometry arguments, of an extension of this approach to the automated recognition of rational cubic, quartic, and elliptic curves. The accuracy of this approach is tested against synthetic data and in the case of experimental observations provided by the NASA Solar Dynamics Observatory mission.",
"Abstract The main purpose of this paper is to lay the foundations of a general theory which encompasses the features of the classical Hough transform and extend them to general algebraic objects such as affine schemes. The main motivation comes from problems of detection of special shapes in medical and astronomical images. The classical Hough transform has been used mainly to detect simple curves such as lines and circles. We generalize this notion using reduced Grobner bases of flat families of affine schemes. To this end we introduce and develop the theory of Hough regularity . The theory is highly effective and we give some examples computed with CoCoA (see [1] )."
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | In 3D, other variants of HT have been introduced and used but as far as we know none of them exploits the huge variety of algebraic plane curves (for this we refer again to the surveys @cite_49 and @cite_20 ). For instance, in @cite_10 the HT has been employed to identify recurring straight line elements on the walls of buildings. In that application, the HT is applied only to planar point sets and line elements are clustered according to their angle with respect to a main wall direction; in this sense, the Hough aggregator is used to select the feature line directions (horizontal, vertical, slanting) one at a time. | {
"cite_N": [
"@cite_20",
"@cite_10",
"@cite_49"
],
"mid": [
"2094583515",
"2064723966",
"2029895480"
],
"abstract": [
"The generalised Hough transform (GHT) is useful for detecting or locating translated two-dimensional objects. However, a weakness of the GHT is its storage requirements and hence the increased computational complexity resulting from the four-dimensional parameter space. In this paper, we present the results of our work which involves investigation of the performance of several efficient GHT techniques including an extension of Thomas's rotation-invariant algorithm. It is shown that our extension of Thomas's algorithm has very low memory requirements and computational complexity, and produces the best results in various tests.",
"Abstract We present a method for automatic reconstruction of permanent structures, such as walls, floors and ceilings, given a raw point cloud of an indoor scene. The main idea behind our approach is a graph-cut formulation to solve an inside outside labeling of a space partitioning. We first partition the space in order to align the reconstructed models with permanent structures. The horizontal structures are located through analysis of the vertical point distribution, while vertical wall structures are detected through feature preserving multi-scale line fitting, followed by clustering in a Hough transform space. The final surface is extracted through a graph-cut formulation that trades faithfulness to measurement data for geometric complexity. A series of experiments show watertight surface meshes reconstructed from point clouds measured on multi-level buildings.",
"In 1962 Hough earned the patent for a method 1], popularly called Hough Transform (HT) that efficiently identifies lines in images. It is an important tool even after the golden jubilee year of existence, as evidenced by more than 2500 research papers dealing with its variants, generalizations, properties and applications in diverse fields. The current paper is a survey of HT and its variants, their limitations and the modifications made to overcome them, the implementation issues in software and hardware, and applications in various fields. Our survey, along with more than 200 references, will help the researchers and students to get a comprehensive view on HT and guide them in applying it properly to their problems of interest. HighlightsA survey of Hough Transform (HT) has been done.The main variants of HT and its formulations have been described.The hardware and software implementations of HT have been described.HT has many applications in diverse fields. Out of them some applications in some selected fields have been mentioned."
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | * Feature characterization Overviews on methods for extracting feature points are provided in @cite_77 @cite_31 . In the realm of feature characterization it is possible to distinguish between features that are view-dependent, like silhouettes, suggestive contours and principal highlights @cite_76 , mainly useful for rendering purposes @cite_35 , or those that are independent of the spatial embedding and, therefore, more suitable for feature recognition and classification processes. In the following we sketch some of the methods that are relevant for our approach, i.e., characterizations that do not depend on the spatial embedding of the surface. | {
"cite_N": [
"@cite_77",
"@cite_31",
"@cite_76",
"@cite_35"
],
"mid": [
"2036490943",
"2169231734",
"2129751996",
"2507622548"
],
"abstract": [
"This paper presents the results of a study in which artists made line drawings intended to convey specific 3D shapes. The study was designed so that drawings could be registered with rendered images of 3D models, supporting an analysis of how well the locations of the artists' lines correlate with other artists', with current computer graphics line definitions, and with the underlying differential properties of the 3D surface. Lines drawn by artists in this study largely overlapped one another (75 are within 1mm of another line), particularly along the occluding contours of the object. Most lines that do not overlap contours overlap large gradients of the image intensity, and correlate strongly with predictions made by recent line drawing algorithms in computer graphics. 14 were not well described by any of the local properties considered in this study. The result of our work is a publicly available data set of aligned drawings, an analysis of where lines appear in that data set based on local properties of 3D models, and algorithms to predict where artists will draw lines for new scenes.",
"Sharp edges, ridges, valleys, and prongs are critical for the appearance and an accurate representation of a 3D model. In this paper, we propose a novel approach that deals with the global shape of features in a robust way. Based on a remeshing algorithm which delivers an isotropic mesh in a feature-sensitive metric, features are recognized on multiple scales via integral invariants of local neighborhoods. Morphological and smoothing operations are then used for feature region extraction and classification into basic types such as ridges, valleys, and prongs. The resulting representation of feature regions is further used for feature-specific editing operations",
"Recent work has shown that sparse lines defined on 3D shapes, including occluding contours and suggestive contours, are effective at conveying shape. We introduce two new families of lines called suggestive highlights and principal highlights, based on definitions related to suggestive contours and geometric creases. We show that when these are drawn in white on a gray background they are naturally interpreted as highlight lines, and complement contours and suggestive contours. We provide object-space definitions and algorithms for extracting these lines, explore several stylization possibilities, and compare the lines to ridges and valleys of intensity in diffuse-shaded images.",
"We present novel techniques for visualizing, illustrating, analyzing, and generating carvings in surfaces. In particular, we consider the carvings in the plaster of the cloister of the Magdeburg cathedral, which dates to the 13th century. Due to aging and weathering, the carvings have flattened. Historians and restorers are highly interested in using digitalization techniques to analyze carvings in historic artifacts and monuments and to get impressions and illustrations of their original shape and appearance. Moreover, museums and churches are interested in such illustrations for presenting them to visitors. The techniques that we propose allow for detecting, selecting, and visualizing carving structures. In addition, we introduce an example-based method for generating carvings. The resulting tool, which integrates all techniques, was evaluated by three experienced restorers to assess the usefulness and applicability. Furthermore, we compared our approach with exaggerated shading and other state-of-the-art methods."
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | A popular choice to locate features is to estimate the curvature, either on meshes @cite_60 @cite_43 or point clouds @cite_13 @cite_63 ). When dealing with curvature estimation, parameters have to be tuned according to the target feature scale and the underlying noise. The Moving Least Square method @cite_50 and its variations are robust to scale variations @cite_63 @cite_32 and does not incorporate smoothness effects in the estimation. Alternative approaches to locate curvature extrema use discrete differential operators @cite_34 or probabilistic methods, such as random walks @cite_45 . For details on the comparison of methods for curvature estimation, we refer to a recent benchmark @cite_64 . | {
"cite_N": [
"@cite_64",
"@cite_60",
"@cite_32",
"@cite_43",
"@cite_45",
"@cite_50",
"@cite_63",
"@cite_34",
"@cite_13"
],
"mid": [
"2517665947",
"2022182585",
"",
"1984490467",
"2179544992",
"2149887512",
"2098552070",
"2167832640",
"206063426"
],
"abstract": [
"While it is usually not difficult to compute principal curvatures of a smooth surface of sufficient differentiability, it is a rather difficult task when only a polygonal approximation of the surface is available, because of the inherent ambiguity of such representation. A number of different approaches has been proposed in the past that tackle this problem using various techniques. Most papers tend to focus on a particular method, while an comprehensive comparison of the different approaches is usually missing. We present results of a large experiment, involving both common and recently proposed curvature estimation techniques, applied to triangle meshes of varying properties. It turns out that none of the approaches provides reliable results under all circumstances. Motivated by this observation, we investigate mesh statistics, which can be computed from vertex positions and mesh connectivity information only, and which can help in deciding which estimator will work best for a particular case. Finally, we propose a meta-estimator, which makes a choice between existing algorithms based on the value of the mesh statistics, and we demonstrate that such meta-estimator, despite its simplicity, provides considerably more robust results than any existing approach.",
"Curves on objects can convey the inherent features of the shape. This paper defines a new class of view-independent curves, denoted demarcating curves. In a nutshell, demarcating curves are the loci of the \"strongest\" inflections on the surface. Due to their appealing capabilities to extract and emphasize 3D textures, they are applied to artifact illustration in archaeology, where they can serve as a worthy alternative to the expensive, time-consuming, and biased manual depiction currently used.",
"",
"The crest lines, salient subsets of the extrema of the principal curvatures over their corresponding curvature lines, are powerful shape descriptors which are widely used for shape matching, interrogation, and visualization purposes. In this paper, we develop fast, accurate, and reliable methods for detecting the crest lines on surfaces approximated by dense triangle meshes. The methods exploit intrinsic geometric properties of the curvature extrema and provide with an inherent level-of-detail control of the detected crest lines. As an immediate application, we use of the crest lines for adaptive mesh simplification purposes.",
"Archaeological line drawing is an essential component of an archaeological report. In this paper, we propose a multi-scale approach to generating 3D line drawing on a triangular mesh model for archaeological illustration. The discrete multi-scale representation of a given model is first constructed based on random walks. Then, we propose a probabilistic method for local scale selection based on the minimum description length (MDL) principle. Finally, the ridges or valleys are detected with the selected scales. Experimental results show that the multi-scale 3D line drawings can well depict the shapes of cultural heritage objects. Compared with the traditional manual method, our method is more accurate and convenient. Furthermore, the time cost of archaeological mapping is decreased to a large extent and the detected lines can be rendered with specific scales and view directions. The computer-generated line drawing can be used to assist the archaeologists to draw archaeological line drawing.",
"We present a new technique for extracting line-type features on point-sampled geometry. Given an unstructured point cloud as input, our method first applies principal component analysis on local neighborhoods to classify points according to the likelihood that they belong to a feature. Using hysteresis thresholding, we then compute a minimum spanning graph as an initial approximation of the feature lines. To smooth out the features while maintaining a close connection to the underlying surface, we use an adaptation of active contour models. Central to our method is a multi-scale classification operator that allows feature analysis at multiple scales, using the size of the local neighborhoods as a discrete scale parameter. This significantly improves the reliability of the detection phase and makes our method more robust in the presence of noise. To illustrate the usefulness of our method, we have implemented a non-photorealistic point renderer to visualize point-sampled surfaces as line drawings of their extracted feature curves.",
"Defining sharp features in a 3D model facilitates a better understanding of the surface and aids geometric processing and graphics applications, such as reconstruction, filtering, simplification, reverse engineering, visualization, and non-photo realism. We present a robust method that identifies sharp features in a point-based model by returning a set of smooth spline curves aligned along the edges. Our feature extraction leverages the concepts of robust moving least squares to locally project points to potential features. The algorithm processes these points to construct arc-length parameterized spline curves fit using an iterative refinement method, aligning smooth and continuous curves through the feature points. We demonstrate the benefits of our method with three applications: surface segmentation, surface meshing and point-based compression.",
"Feature lines are salient surface characteristics. Their definition involves third and fourth order surface derivatives. This often yields to unpleasantly rough and squiggly feature lines since third order derivatives are highly sensitive against unwanted surface noise. The present work proposes two novel concepts for a more stable algorithm producing visually more pleasing feature lines: First, a new computation scheme based on discrete differential geometry is presented, avoiding costly computations of higher order approximating surfaces. Secondly, this scheme is augmented by a filtering method for higher order surface derivatives to improve both the stability of the extraction of feature lines and the smoothness of their appearance.",
""
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | Partial similarity and, in particular, self-similarity, is the keyword used to detect repeated features over a surface. For instance, the method @cite_74 is able to recognize repeated surface features (circles or stars) over a surface. However, being based on geometry hashing, the method is scale dependent and does not provide the exact parameters that characterize such features. | {
"cite_N": [
"@cite_74"
],
"mid": [
"2010209818"
],
"abstract": [
"This article introduces a method for partial matching of surfaces represented by triangular meshes. Our method matches surface regions that are numerically and topologically dissimilar, but approximately similar regions. We introduce novel local surface descriptors which efficiently represent the geometry of local regions of the surface. The descriptors are defined independently of the underlying triangulation, and form a compatible representation that allows matching of surfaces with different triangulations. To cope with the combinatorial complexity of partial matching of large meshes, we introduce the abstraction of salient geometric features and present a method to construct them. A salient geometric feature is a compound high-level feature of nontrivial local shapes. We show that a relatively small number of such salient geometric features characterizes the surface well for various similarity applications. Matching salient geometric features is based on indexing rotation-invariant features and a voting scheme accelerated by geometric hashing. We demonstrate the effectiveness of our method with a number of applications, such as computing self-similarity, alignments, and subparts similarity."
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | Feature curves are often identified as ridges and valleys, thus representing the extrema of principal curvatures @cite_62 @cite_43 @cite_79 or sharp features @cite_31 . Other types of lines used for feature curve representation are parabolic ones. They partition the surface into hyperbolic and elliptic regions, and zero-mean curvature curves, which classify sub-surfaces into concave and convex shapes @cite_47 . Parabolic lines correspond to the zeros of the Gaussian and mean curvature, respectively. Finally, demarcating curves are the zero-crossings of the curvature in its gradient direction @cite_60 @cite_73 . In general, all these curves, defined as the zero set of a scalar function, do not consider curve with knots, fact that an algebraic curve like the (see @cite_75 ) could arrange. | {
"cite_N": [
"@cite_31",
"@cite_62",
"@cite_60",
"@cite_43",
"@cite_79",
"@cite_47",
"@cite_73",
"@cite_75"
],
"mid": [
"2169231734",
"2122076394",
"2022182585",
"1984490467",
"2073337928",
"2067164770",
"",
"614320654"
],
"abstract": [
"Sharp edges, ridges, valleys, and prongs are critical for the appearance and an accurate representation of a 3D model. In this paper, we propose a novel approach that deals with the global shape of features in a robust way. Based on a remeshing algorithm which delivers an isotropic mesh in a feature-sensitive metric, features are recognized on multiple scales via integral invariants of local neighborhoods. Morphological and smoothing operations are then used for feature region extraction and classification into basic types such as ridges, valleys, and prongs. The resulting representation of feature regions is further used for feature-specific editing operations",
"We propose a simple and effective method for detecting view-and scale-independent ridge-valley lines defined via first- and second-order curvature derivatives on shapes approximated by dense triangle meshes. A high-quality estimation of high-order surface derivatives is achieved by combining multi-level implicit surface fitting and finite difference approximations. We demonstrate that the ridges and valleys are geometrically and perceptually salient surface features, and, therefore, can be potentially used for shape recognition, coding, and quality evaluation purposes.",
"Curves on objects can convey the inherent features of the shape. This paper defines a new class of view-independent curves, denoted demarcating curves. In a nutshell, demarcating curves are the loci of the \"strongest\" inflections on the surface. Due to their appealing capabilities to extract and emphasize 3D textures, they are applied to artifact illustration in archaeology, where they can serve as a worthy alternative to the expensive, time-consuming, and biased manual depiction currently used.",
"The crest lines, salient subsets of the extrema of the principal curvatures over their corresponding curvature lines, are powerful shape descriptors which are widely used for shape matching, interrogation, and visualization purposes. In this paper, we develop fast, accurate, and reliable methods for detecting the crest lines on surfaces approximated by dense triangle meshes. The methods exploit intrinsic geometric properties of the curvature extrema and provide with an inherent level-of-detail control of the detected crest lines. As an immediate application, we use of the crest lines for adaptive mesh simplification purposes.",
"The patch layout of 3D surfaces reveals the high-level geometric and topological structures. In this paper, we study the patch layout computation by detecting and enclosing feature loops on surfaces. We present a hybrid framework which combines several key ingredients, including feature detection, feature filtering, feature curve extension, patch subdivision and boundary smoothing. Our framework is able to compute patch layouts through concave features as previous approaches, but also able to generate nice layouts through smoothing regions. We demonstrate the effectiveness of our framework by comparing with the state-of-the-art methods. Graphical abstractDisplay Omitted HighlightsA hybrid framework that can extract patch layouts that are better than the state-of-the-art.A novel criterion to evaluate the quality of patch layouts.Improved algorithms for smoothing curves on surfaces.",
"A new theorem is discussed that relates the apparent curvature of the occluding contour of a visual shape to the intrinsic curvature of the surface and the radial curvature. This theorem allows the formulation of general laws for the apparent curvature, independent of viewing distance and regardless of the fact that the rim (the boundary between the visible and invisible parts of the object) is a general, thus twisted, space curve. Consequently convexities, concavities, or inflextions of contours in the retinal image allow the observer to draw inferences about local surface geometry with certainty. These results appear to be counterintuitive, witness to the treatment of the problem by recent authors. It is demonstrated how well-known examples, used to show how concavities and convexities of the contour have no obvious relation to solid shape, are actually good illustrations of the fact that convexities are due to local ovoid shapes, concavities to local saddle shapes.",
"",
"Introduction Plane Curves Main Definitions and Facts Some General Classes of Plane Curves Atlas of Plane Curves Structure of the Atlas Atlas of Curves Geometric Indices for Atlas of Curves Space Curves Main Defintions and Facts Some Classes of Space Curves Applications of Curves Dynamic Properties of Plane Curves Plane Curves and Mechanisms Geometric Splines Appendices Smooth Functions of One and Two Variables Vector Functions of Scalar Argument Coordinate Systems. Coordinate Transformations Search of Some Geometric Characteristics Greek Roots of Names for Some Curves English-German-French Indices Bibliography Index of Terms"
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | Besides interpolating approaches such as splines, it is possible to fit the feature curve set with some specific family of curves, for instance the @cite_40 and the @cite_26 have been proposed as a natural way to describe line drawings and silhouettes showing their suitability for shape completion and repair. However, using one family of curves at a time implies the need of defining specific solutions and algorithms for settings the curve parameters during the reconstruction phase. | {
"cite_N": [
"@cite_40",
"@cite_26"
],
"mid": [
"2074180148",
"2207430418"
],
"abstract": [
"Logarithmic spirals are ubiquitous in nature. This paper presents a novel mathematical definition of a 3D logarithmic spiral, which provides a proper description of objects found in nature. To motivate our work, we scanned spiral-shaped objects and studied their geometric properties. We consider the extent to which the existing 3D definitions capture these properties. We identify a property that is shared by the objects we investigated and is not satisfied by the existing 3D definitions. This leads us to present our definition in which both the radius of curvature and the radius of torsion change linearly along the curve. We prove that our spiral satisfies several desirable properties, including invariance to similarity transformations, smoothness, symmetry, extensibility, and roundness. Finally, we demonstrate the utility of our curves in the modeling of several animal structures.",
"Shape completion is an intriguing problem in geometry processing with applications in CAD and graphics. This paper defines a new type of 3D curve, which can be utilized for curve completion. It can be considered as the extension to three dimensions of the 2D Euler spiral. We prove several properties of this curve - properties that have been shown to be important for the appeal of curves. We illustrate its utility in two applications. The first is ''fixing'' curves detected by algorithms for edge detection on surfaces. The second is shape illustration in archaeology, where the user would like to draw curves that are missing due to the incompleteness of the input model."
]
} |
1709.10177 | 2749749166 | Abstract Feature curves are largely adopted to highlight shape features, such as sharp lines, or to divide surfaces into meaningful segments, like convex or concave regions. Extracting these curves is not sufficient to convey prominent and meaningful information about a shape. We have first to separate the curves belonging to features from those caused by noise and then to select the lines, which describe non-trivial portions of a surface. The automatic detection of such features is crucial for the identification and or annotation of relevant parts of a given shape. To do this, the Hough transform (HT) is a feature extraction technique widely used in image analysis, computer vision and digital image processing, while, for 3D shapes, the extraction of salient feature curves is still an open problem. Thanks to algebraic geometry concepts, the HT technique has been recently extended to include a vast class of algebraic curves, thus proving to be a competitive tool for yielding an explicit representation of the diverse feature lines equations. In the paper, for the first time we apply this novel extension of the HT technique to the realm of 3D shapes in order to identify and localize semantic features like patterns, decorations or anatomical details on 3D objects (both complete and fragments), even in the case of features partially damaged or incomplete. The method recognizes various features, possibly compound, and it selects the most suitable feature profiles among families of algebraic curves. | Recently, feature curve identification has been addressed with co-occurrence analysis approaches @cite_71 @cite_52 . In this case, the curve identification is done in two steps: first, a local feature characterization is performed, for instance computing the Histogram of Oriented Curvature (HOC) @cite_4 or a depth image of the 3D model @cite_71 ; second, a learning phase is applied to the feature characterization. The learning phase is interactive and requires 2-3 training examples for every type of curve to be identified and sketched; in case of multiple curves it is necessary also to specify salient nodes for each curve. Feature lines are poly-lines (i.e. connected sequences of segments) and do not have any global equation. These methods are adopted mainly for recognizing parts of buildings (such as windows, doors, etc.) and features in architectural models that are similar to strokes. Main limitations of these methods are the partial tolerance to scale variance, the need of a number of training curves for each class of curves, the non robustness to missing data and the fact that compound features can be addressed only one curve at a time @cite_71 . | {
"cite_N": [
"@cite_4",
"@cite_52",
"@cite_71"
],
"mid": [
"2019899683",
"2172471995",
"2156073117"
],
"abstract": [
"In this paper, we present a novel method for detecting partial symmetries in very large point clouds of 3D city scans. Unlike previous work, which has only been demonstrated on data sets of a few hundred megabytes maximum, our method scales to very large scenes: We map the detection problem to a nearest-neighbour problem in a low-dimensional feature space, and follow this with a cascade of tests for geometric clustering of potential matches. Our algorithm robustly handles noisy real-world scanner data, obtaining a recognition performance comparable to that of state-of-the-art methods. In practice, it scales linearly with scene size and achieves a high absolute throughput, processing half a terabyte of scanner data overnight on a dual socket commodity PC.",
"This paper addresses approximate partial symmetry detection in 3D point clouds, a classical and foundational tool for analyzing geometry. We present a novel, fully unsupervised method that detects partial symmetry under significant geometric variability, and without constraints on the number and arrangement of instances. The core idea is a matching scheme that finds consistent co-occurrence patterns in a frame-invariant way. We obtain a canonical partition of the input shape into building blocks and can handle ambiguous data by aggregating co-occurrence information across both all building block instances and the area they cover. We evaluate our method on several benchmark data sets and demonstrate its significant improvements in handling geometric variability, including scanning noise, irregular patterns, appearance variation and shape deformation.",
"Feature detection in geometric datasets is a fundamental tool for solving shape matching problems such as partial symmetry detection. Traditional techniques usually employ a priori models such as crease lines that are unspecific to the actual application. Our paper examines the idea of learning geometric features. We introduce a formal model for a class of linear feature constellations based on a Markov chain model and propose a novel, efficient algorithm for detecting a large number of features simultaneously. After a short user-guided training stage, in which one or a few example lines are sketched directly onto the input data, our algorithm automatically finds all pieces of geometry similar to the marked areas. In particular, the algorithm is able recognize larger classes of semantically similar but geometrically varying features, which is very difficult using unsupervised techniques. In a number of experiments, we apply our technique to point cloud data from 3D scanners. The algorithm is able to detect features with very low rates of false positives and negatives and to recognize broader classes of similar geometry (such as “windows” in a building scan) even from few training examples, thereby significantly improving over previous unsupervised techniques."
]
} |
1709.09820 | 2756612184 | Generative Adversarial Networks (GANs) have shown impressive performance in generating photo-realistic images. They fit generative models by minimizing certain distance measure between the real image distribution and the generated data distribution. Several distance measures have been used, such as Jensen-Shannon divergence, @math -divergence, and Wasserstein distance, and choosing an appropriate distance measure is very important for training the generative network. In this paper, we choose to use the maximum mean discrepancy (MMD) as the distance metric, which has several nice theoretical guarantees. In fact, generative moment matching network (GMMN) (Li, Swersky, and Zemel 2015) is such a generative model which contains only one generator network @math trained by directly minimizing MMD between the real and generated distributions. However, it fails to generate meaningful samples on challenging benchmark datasets, such as CIFAR-10 and LSUN. To improve on GMMN, we propose to add an extra network @math , called mapper. @math maps both real data distribution and generated data distribution from the original data space to a feature representation space @math , and it is trained to maximize MMD between the two mapped distributions in @math , while the generator @math tries to minimize the MMD. We call the new model generative adversarial mapping networks (GAMNs). We demonstrate that the adversarial mapper @math can help @math to better capture the underlying data distribution. We also show that GAMN significantly outperforms GMMN, and is also superior to or comparable with other state-of-the-art GAN based methods on MNIST, CIFAR-10 and LSUN-Bedrooms datasets. | @math GAN is proposed to minimize the variational lower bound on @math divengence between two distributions @cite_4 . EBGAN does total variation distance minimization to learn the underlying distribution @cite_1 @cite_18 . However, these GAN based models are still hard to train, which need a careful balance during the adversarial optimization. WGAN and improved WGAN are proposed to address this issue, which are trained to minimize Wasserstein divergence @cite_18 @cite_21 . | {
"cite_N": [
"@cite_1",
"@cite_18",
"@cite_21",
"@cite_4"
],
"mid": [
"2521028896",
"",
"2605135824",
"2419501139"
],
"abstract": [
"We introduce the \"Energy-based Generative Adversarial Network\" model (EBGAN) which views the discriminator as an energy function that attributes low energies to the regions near the data manifold and higher energies to other regions. Similar to the probabilistic GANs, a generator is seen as being trained to produce contrastive samples with minimal energies, while the discriminator is trained to assign high energies to these generated samples. Viewing the discriminator as an energy function allows to use a wide variety of architectures and loss functionals in addition to the usual binary classifier with logistic output. Among them, we show one instantiation of EBGAN framework as using an auto-encoder architecture, with the energy being the reconstruction error, in place of the discriminator. We show that this form of EBGAN exhibits more stable behavior than regular GANs during training. We also show that a single-scale architecture can be trained to generate high-resolution images.",
"",
"Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to undesired behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.",
"Generative neural samplers are probabilistic models that implement sampling using feedforward neural networks: they take a random input vector and produce a sample from a probability distribution defined by the network weights. These models are expressive and allow efficient computation of samples and derivatives, but cannot be used for computing likelihoods or for marginalization. The generative-adversarial training method allows to train such models through the use of an auxiliary discriminative neural network. We show that the generative-adversarial approach is a special case of an existing more general variational divergence estimation approach. We show that any f-divergence can be used for training generative neural samplers. We discuss the benefits of various choices of divergence functions on training complexity and the quality of the obtained generative models."
]
} |
1709.09820 | 2756612184 | Generative Adversarial Networks (GANs) have shown impressive performance in generating photo-realistic images. They fit generative models by minimizing certain distance measure between the real image distribution and the generated data distribution. Several distance measures have been used, such as Jensen-Shannon divergence, @math -divergence, and Wasserstein distance, and choosing an appropriate distance measure is very important for training the generative network. In this paper, we choose to use the maximum mean discrepancy (MMD) as the distance metric, which has several nice theoretical guarantees. In fact, generative moment matching network (GMMN) (Li, Swersky, and Zemel 2015) is such a generative model which contains only one generator network @math trained by directly minimizing MMD between the real and generated distributions. However, it fails to generate meaningful samples on challenging benchmark datasets, such as CIFAR-10 and LSUN. To improve on GMMN, we propose to add an extra network @math , called mapper. @math maps both real data distribution and generated data distribution from the original data space to a feature representation space @math , and it is trained to maximize MMD between the two mapped distributions in @math , while the generator @math tries to minimize the MMD. We call the new model generative adversarial mapping networks (GAMNs). We demonstrate that the adversarial mapper @math can help @math to better capture the underlying data distribution. We also show that GAMN significantly outperforms GMMN, and is also superior to or comparable with other state-of-the-art GAN based methods on MNIST, CIFAR-10 and LSUN-Bedrooms datasets. | More related to our method, generative moment matching networks GMMNs @cite_6 @cite_19 optimize MMD between distributions to learn generative models from data. The original GMMN only includes a single generator. To boost the performance of GMMN, li2015generative introduce an auto-encoder to a GMMN and name it GMMN+AE. They first use an auto-encoder to produce the code representations of data and then minimize MMD between data code distribution and generated code distribution. The functionality of the auto-encoder here is similar to that of the mapper in GAMN to some extent: both of them map the data distribution from the real data space to another code (representation) space and can reduce the dimension of high-dimensional images. However, the mapper in GAMN is entirely different from the auto-encoder in GMMN+AE. The most different part is that the mapper is a dynamic network which always changes during training process to be a good adversary for the generator, while the auto-encoder is a static network which is trained in the beginning and then keeps fixed all the time. We believe this difference makes the mapper much powerful than the auto-encoder, thus enhancing the generator in GAMN considerably. Our experiments also demonstrate this. | {
"cite_N": [
"@cite_19",
"@cite_6"
],
"mid": [
"2949995983",
"2950292946"
],
"abstract": [
"We consider training a deep neural network to generate samples from an unknown distribution given i.i.d. data. We frame learning as an optimization minimizing a two-sample test statistic---informally speaking, a good generator network produces samples that cause a two-sample test to fail to reject the null hypothesis. As our two-sample test statistic, we use an unbiased estimate of the maximum mean discrepancy, which is the centerpiece of the nonparametric kernel two-sample test proposed by (2012). We compare to the adversarial nets framework introduced by (2014), in which learning is a two-player game between a generator network and an adversarial discriminator network, both trained to outwit the other. From this perspective, the MMD statistic plays the role of the discriminator. In addition to empirical comparisons, we prove bounds on the generalization error incurred by optimizing the empirical MMD.",
"We consider the problem of learning deep generative models from data. We formulate a method that generates an independent sample via a single feedforward pass through a multilayer perceptron, as in the recently proposed generative adversarial networks (, 2014). Training a generative adversarial network, however, requires careful optimization of a difficult minimax program. Instead, we utilize a technique from statistical hypothesis testing known as maximum mean discrepancy (MMD), which leads to a simple objective that can be interpreted as matching all orders of statistics between a dataset and samples from the model, and can be trained by backpropagation. We further boost the performance of this approach by combining our generative network with an auto-encoder network, using MMD to learn to generate codes that can then be decoded to produce samples. We show that the combination of these techniques yields excellent generative models compared to baseline approaches as measured on MNIST and the Toronto Face Database."
]
} |
1709.09890 | 2756815061 | Convolutional Neural Network (CNN) image classifiers are traditionally designed to have sequential convolutional layers with a single output layer. This is based on the assumption that all target classes should be treated equally and exclusively. However, some classes can be more difficult to distinguish than others, and classes may be organized in a hierarchy of categories. At the same time, a CNN is designed to learn internal representations that abstract from the input data based on its hierarchical layered structure. So it is natural to ask if an inverse of this idea can be applied to learn a model that can predict over a classification hierarchy using multiple output layers in decreasing order of class abstraction. In this paper, we introduce a variant of the traditional CNN model named the Branch Convolutional Neural Network (B-CNN). A B-CNN model outputs multiple predictions ordered from coarse to fine along the concatenated convolutional layers corresponding to the hierarchical structure of the target classes, which can be regarded as a form of prior knowledge on the output. To learn with B-CNNs a novel training strategy, named the Branch Training strategy (BT-strategy), is introduced which balances the strictness of the prior with the freedom to adjust parameters on the output layers to minimize the loss. In this way we show that CNN based models can be forced to learn successively coarse to fine concepts in the internal layers at the output stage, and that hierarchical prior knowledge can be adopted to boost CNN models' classification performance. Our models are evaluated to show that the B-CNN extensions improve over the corresponding baseline CNN on the benchmark datasets MNIST, CIFAR-10 and CIFAR-100. | CNN based models have been successfully exploited in many computer vision tasks and the most successful one is image classification @cite_8 . There are various techniques introduced to enhance CNN's performance in the past literature. Simonyan and Zisserman show us that increasing the depth of a CNN model @cite_19 within a boundary is likely to boost the classifier's accuracy. In @cite_0 and @cite_9 , a large number of filters has been adopted. Recently, there have been a vast amount of work to enhance the components of CNN such as novel types of activation functions @cite_20 @cite_34 , the replacement of linear filter patches @cite_5 , pooling operation @cite_17 @cite_4 @cite_23 and initialization strategies @cite_1 . These attempts mainly focus on the potential weakness inside the existent CNN models and try to find a replacement or propose any strategy to improve it. Different from them, our B-CNN model is taking existent CNNs as building blocks, exploiting natural hierarchical property of CNN @cite_0 , and boosting performance with a tailored training strategy. | {
"cite_N": [
"@cite_4",
"@cite_8",
"@cite_9",
"@cite_1",
"@cite_0",
"@cite_19",
"@cite_23",
"@cite_5",
"@cite_34",
"@cite_20",
"@cite_17"
],
"mid": [
"",
"",
"1487583988",
"2178237821",
"2952186574",
"1686810756",
"",
"",
"",
"2176412452",
"1907282891"
],
"abstract": [
"",
"",
"We present an integrated framework for using Convolutional Networks for classification, localization and detection. We show how a multiscale and sliding window approach can be efficiently implemented within a ConvNet. We also introduce a novel deep learning approach to localization by learning to predict object boundaries. Bounding boxes are then accumulated rather than suppressed in order to increase detection confidence. We show that different tasks can be learned simultaneously using a single shared network. This integrated framework is the winner of the localization task of the ImageNet Large Scale Visual Recognition Challenge 2013 (ILSVRC2013) and obtained very competitive results for the detection and classifications tasks. In post-competition work, we establish a new state of the art for the detection task. Finally, we release a feature extractor from our best model called OverFeat.",
"Layer-sequential unit-variance (LSUV) initialization - a simple method for weight initialization for deep net learning - is proposed. The method consists of the two steps. First, pre-initialize weights of each convolution or inner-product layer with orthonormal matrices. Second, proceed from the first to the final layer, normalizing the variance of the output of each layer to be equal to one. Experiment with different activation functions (maxout, ReLU-family, tanh) show that the proposed initialization leads to learning of very deep nets that (i) produces networks with test accuracy better or equal to standard methods and (ii) is at least as fast as the complex schemes proposed specifically for very deep nets such as FitNets ( (2015)) and Highway ( (2015)). Performance is evaluated on GoogLeNet, CaffeNet, FitNets and Residual nets and the state-of-the-art, or very close to it, is achieved on the MNIST, CIFAR-10 100 and ImageNet datasets.",
"Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we address both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. We also perform an ablation study to discover the performance contribution from different model layers. This enables us to find model architectures that outperform Krizhevsky al on the ImageNet classification benchmark. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets.",
"In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.",
"",
"",
"",
"We introduce the \"exponential linear unit\" (ELU) which speeds up learning in deep neural networks and leads to higher classification accuracies. Like rectified linear units (ReLUs), leaky ReLUs (LReLUs) and parametrized ReLUs (PReLUs), ELUs alleviate the vanishing gradient problem via the identity for positive values. However, ELUs have improved learning characteristics compared to the units with other activation functions. In contrast to ReLUs, ELUs have negative values which allows them to push mean unit activations closer to zero like batch normalization but with lower computational complexity. Mean shifts toward zero speed up learning by bringing the normal gradient closer to the unit natural gradient because of a reduced bias shift effect. While LReLUs and PReLUs have negative values, too, they do not ensure a noise-robust deactivation state. ELUs saturate to a negative value with smaller inputs and thereby decrease the forward propagated variation and information. Therefore, ELUs code the degree of presence of particular phenomena in the input, while they do not quantitatively model the degree of their absence. In experiments, ELUs lead not only to faster learning, but also to significantly better generalization performance than ReLUs and LReLUs on networks with more than 5 layers. On CIFAR-100 ELUs networks significantly outperform ReLU networks with batch normalization while batch normalization does not improve ELU networks. ELU networks are among the top 10 reported CIFAR-10 results and yield the best published result on CIFAR-100, without resorting to multi-view evaluation or model averaging. On ImageNet, ELU networks considerably speed up learning compared to a ReLU network with the same architecture, obtaining less than 10 classification error for a single crop, single model network.",
"We introduce a simple and effective method for regularizing large convolutional neural networks. We replace the conventional deterministic pooling operations with a stochastic procedure, randomly picking the activation within each pooling region according to a multinomial distribution, given by the activities within the pooling region. The approach is hyper-parameter free and can be combined with other regularization approaches, such as dropout and data augmentation. We achieve state-of-the-art performance on four image datasets, relative to other approaches that do not utilize data augmentation."
]
} |
1709.09890 | 2756815061 | Convolutional Neural Network (CNN) image classifiers are traditionally designed to have sequential convolutional layers with a single output layer. This is based on the assumption that all target classes should be treated equally and exclusively. However, some classes can be more difficult to distinguish than others, and classes may be organized in a hierarchy of categories. At the same time, a CNN is designed to learn internal representations that abstract from the input data based on its hierarchical layered structure. So it is natural to ask if an inverse of this idea can be applied to learn a model that can predict over a classification hierarchy using multiple output layers in decreasing order of class abstraction. In this paper, we introduce a variant of the traditional CNN model named the Branch Convolutional Neural Network (B-CNN). A B-CNN model outputs multiple predictions ordered from coarse to fine along the concatenated convolutional layers corresponding to the hierarchical structure of the target classes, which can be regarded as a form of prior knowledge on the output. To learn with B-CNNs a novel training strategy, named the Branch Training strategy (BT-strategy), is introduced which balances the strictness of the prior with the freedom to adjust parameters on the output layers to minimize the loss. In this way we show that CNN based models can be forced to learn successively coarse to fine concepts in the internal layers at the output stage, and that hierarchical prior knowledge can be adopted to boost CNN models' classification performance. Our models are evaluated to show that the B-CNN extensions improve over the corresponding baseline CNN on the benchmark datasets MNIST, CIFAR-10 and CIFAR-100. | Combining tree structure prior with CNN models has drawn interest recently. Srivastava al @cite_18 propose a method which benefits from CNN and tree-based prior when the training set is very small. They have shown that label tree prior can be used to transfer knowledge between classes and boost the performance when training examples are insufficient. But their method does not exploit the hierarchical nature within the layers which remains the whole CNN model a black box. Jia al @cite_15 introduce a graph representation to capture the hierarchical and exclusive semantics between labels. They define a joint distribution of an assignment of all labels as a Conditional Random Field and use it to replace the traditional classifiers such as softmax on the top of a deep neural network. However their work still uses the network as a feature extractor and there is no attempt to exploit the hierarchical nature of CNN itself. | {
"cite_N": [
"@cite_18",
"@cite_15"
],
"mid": [
"2107215754",
"2158535772"
],
"abstract": [
"High capacity classifiers, such as deep neural networks, often struggle on classes that have very few training examples. We propose a method for improving classification performance for such classes by discovering similar classes and transferring knowledge among them. Our method learns to organize the classes into a tree hierarchy. This tree structure imposes a prior over the classifier's parameters. We show that the performance of deep neural networks can be improved by applying these priors to the weights in the last layer. Our method combines the strength of discriminatively trained deep neural networks, which typically require large amounts of training data, with tree-based priors, making deep neural networks work well on infrequent classes as well. We also propose an algorithm for learning the underlying tree structure. Starting from an initial pre-specified tree, this algorithm modifies the tree to make it more pertinent to the task being solved, for example, removing semantic relationships in favour of visual ones for an image classification task. Our method achieves state-of-the-art classification results on the CIFAR-100 image data set and the MIR Flickr image-text data set.",
"We investigated the computational properties of natural object hierarchy in the context of constellation object class models, and its utility for object class recognition. We first observed an interesting computational property of the object hierarchy: comparing the recognition rate when using models of objects at different levels, the higher more inclusive levels (e.g., closed-frame vehicles or vehicles) exhibit higher recall but lower precision when compared with the class specific level (e.g., bus). These inherent differences suggest that combining object classifiers from different hierarchical levels into a single classifier may improve classification, as it appears like these models capture different aspects of the object. We describe a method to combine these classifiers, and analyze the conditions under which improvement can be guaranteed. When given a small sample of a new object class, we describe a method to transfer knowledge across the tree hierarchy, between related objects. Finally, we describe extensive experiments using object hierarchies obtained from publicly available datasets, and show that the combined classifiers significantly improve recognition results."
]
} |
1709.09890 | 2756815061 | Convolutional Neural Network (CNN) image classifiers are traditionally designed to have sequential convolutional layers with a single output layer. This is based on the assumption that all target classes should be treated equally and exclusively. However, some classes can be more difficult to distinguish than others, and classes may be organized in a hierarchy of categories. At the same time, a CNN is designed to learn internal representations that abstract from the input data based on its hierarchical layered structure. So it is natural to ask if an inverse of this idea can be applied to learn a model that can predict over a classification hierarchy using multiple output layers in decreasing order of class abstraction. In this paper, we introduce a variant of the traditional CNN model named the Branch Convolutional Neural Network (B-CNN). A B-CNN model outputs multiple predictions ordered from coarse to fine along the concatenated convolutional layers corresponding to the hierarchical structure of the target classes, which can be regarded as a form of prior knowledge on the output. To learn with B-CNNs a novel training strategy, named the Branch Training strategy (BT-strategy), is introduced which balances the strictness of the prior with the freedom to adjust parameters on the output layers to minimize the loss. In this way we show that CNN based models can be forced to learn successively coarse to fine concepts in the internal layers at the output stage, and that hierarchical prior knowledge can be adopted to boost CNN models' classification performance. Our models are evaluated to show that the B-CNN extensions improve over the corresponding baseline CNN on the benchmark datasets MNIST, CIFAR-10 and CIFAR-100. | Zhicheng al @cite_2 have made a significant contribution in discovering the possibility of combing the label tree and the hierarchical nature of CNN. In their work, classes are grouped as coarse categories and their HD-CNN model separates easy categories using a shared coarse classifier while distinguishing difficult classes using fine category classifiers. This model confirmed that the hierarchical property in CNN can be exploited. A potential problem of HD-CNN is that it requires the coarse and fine category components to be pretrained and followed by a fine-tune procedure which is quite time-consuming. HD-CNN also only uses one coarse category which is not scalable because there may be far more than one levels of coarse categories in a hierarchical label tree. | {
"cite_N": [
"@cite_2"
],
"mid": [
"1937922215"
],
"abstract": [
"In image classification, visual separability between different object categories is highly uneven, and some categories are more difficult to distinguish than others. Such difficult categories demand more dedicated classifiers. However, existing deep convolutional neural networks (CNN) are trained as flat N-way classifiers, and few efforts have been made to leverage the hierarchical structure of categories. In this paper, we introduce hierarchical deep CNNs (HD-CNNs) by embedding deep CNNs into a category hierarchy. An HD-CNN separates easy classes using a coarse category classifier while distinguishing difficult classes using fine category classifiers. During HD-CNN training, component-wise pretraining is followed by global finetuning with a multinomial logistic loss regularized by a coarse category consistency term. In addition, conditional executions of fine category classifiers and layer parameter compression make HD-CNNs scalable for large-scale visual recognition. We achieve state-of-the-art results on both CIFAR100 and large-scale ImageNet 1000-class benchmark datasets. In our experiments, we build up three different HD-CNNs and they lower the top-1 error of the standard CNNs by 2.65 , 3.1 and 1.1 , respectively."
]
} |
1709.09890 | 2756815061 | Convolutional Neural Network (CNN) image classifiers are traditionally designed to have sequential convolutional layers with a single output layer. This is based on the assumption that all target classes should be treated equally and exclusively. However, some classes can be more difficult to distinguish than others, and classes may be organized in a hierarchy of categories. At the same time, a CNN is designed to learn internal representations that abstract from the input data based on its hierarchical layered structure. So it is natural to ask if an inverse of this idea can be applied to learn a model that can predict over a classification hierarchy using multiple output layers in decreasing order of class abstraction. In this paper, we introduce a variant of the traditional CNN model named the Branch Convolutional Neural Network (B-CNN). A B-CNN model outputs multiple predictions ordered from coarse to fine along the concatenated convolutional layers corresponding to the hierarchical structure of the target classes, which can be regarded as a form of prior knowledge on the output. To learn with B-CNNs a novel training strategy, named the Branch Training strategy (BT-strategy), is introduced which balances the strictness of the prior with the freedom to adjust parameters on the output layers to minimize the loss. In this way we show that CNN based models can be forced to learn successively coarse to fine concepts in the internal layers at the output stage, and that hierarchical prior knowledge can be adopted to boost CNN models' classification performance. Our models are evaluated to show that the B-CNN extensions improve over the corresponding baseline CNN on the benchmark datasets MNIST, CIFAR-10 and CIFAR-100. | More recently work has combined deep learning with probabilistic graphical models such as MRFs, CRFs, etc. ( @cite_29 ), but this either requires a two-stage training process, or new methods of training the combined model, which is not as straightforward as training in a B-CNN. An end-to-end approach is the Structured Prediction Energy Networks of @cite_6 but this is not based on CNNs. | {
"cite_N": [
"@cite_29",
"@cite_6"
],
"mid": [
"2952865063",
"2276303159"
],
"abstract": [
"In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First, we highlight convolution with upsampled filters, or 'atrous convolution', as a powerful tool in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second, we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third, we improve the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models. The commonly deployed combination of max-pooling and downsampling in DCNNs achieves invariance but has a toll on localization accuracy. We overcome this by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF), which is shown both qualitatively and quantitatively to improve localization performance. Our proposed \"DeepLab\" system sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 79.7 mIOU in the test set, and advances the results on three other datasets: PASCAL-Context, PASCAL-Person-Part, and Cityscapes. All of our code is made publicly available online.",
"We introduce structured prediction energy networks (SPENs), a flexible framework for structured prediction. A deep architecture is used to define an energy function of candidate labels, and then predictions are produced by using back-propagation to iteratively optimize the energy with respect to the labels. This deep architecture captures dependencies between labels that would lead to intractable graphical models, and performs structure learning by automatically learning discriminative features of the structured output. One natural application of our technique is multi-label classification, which traditionally has required strict prior assumptions about the interactions between labels to ensure tractable learning and prediction. We are able to apply SPENs to multi-label problems with substantially larger label sets than previous applications of structured prediction, while modeling high-order interactions using minimal structural assumptions. Overall, deep learning provides remarkable tools for learning features of the inputs to a prediction problem, and this work extends these techniques to learning features of structured outputs. Our experiments provide impressive performance on a variety of benchmark multi-label classification tasks, demonstrate that our technique can be used to provide interpretable structure learning, and illuminate fundamental trade-offs between feed-forward and iterative structured prediction."
]
} |
1709.09393 | 2760670598 | Data parallelism has emerged as a necessary technique to accelerate the training of deep neural networks (DNN). In a typical data parallelism approach, the local workers push the latest updates of all the parameters to the parameter server and pull all merged parameters back periodically. However, with the increasing size of DNN models and the large number of workers in practice, this typical data parallelism cannot achieve satisfactory training acceleration, since it usually suffers from the heavy communication cost due to transferring huge amount of information between workers and the parameter server. In-depth understanding on DNN has revealed that it is usually highly redundant, that deleting a considerable proportion of the parameters will not significantly decline the model performance. This redundancy property exposes a great opportunity to reduce the communication cost by only transferring the information of those significant parameters during the parallel training. However, if we only transfer information of temporally significant parameters of the latest snapshot, we may miss the parameters that are insignificant now but have potential to become significant as the training process goes on. To this end, we design an Explore-Exploit framework to dynamically choose the subset to be communicated, which is comprised of the significant parameters in the latest snapshot together with a random explored set of other parameters. We propose to measure the significance of the parameter by the combination of its magnitude and gradient. Our experimental results demonstrate that our proposed Slim-DP can achieve better training acceleration than standard data parallelism and its communication-efficient version by saving communication time without loss of accuracy. | Many works improve the parallel training of DNN by designing new local training algorithms, new model aggregation methods, and new global model update rules. For example, to improve the local training, NG-SGD @cite_24 implements an approximate and efficient algorithm for Natural Gradient SGD; large mini-batch methods @cite_10 @cite_2 increase the learning rate and the mini-batch size to accelerate the convergence. To design new model aggregation methods, EASGD @cite_15 adds an elastic force which takes the weighted combination of the local model and the global model as the new local model after the synchronization; EC-DNN @cite_12 uses the output-average instead of the parameter-average to aggregate local models in order to improve the convergence. To improve the global model update rules, BMUF @cite_16 designs block-wise model-update filtering and utilizes the momentum of the global model to improve the speedup of Plump-DP. | {
"cite_N": [
"@cite_24",
"@cite_2",
"@cite_15",
"@cite_16",
"@cite_10",
"@cite_12"
],
"mid": [
"1779452081",
"",
"1845277745",
"2405883473",
"2622263826",
"2414666004"
],
"abstract": [
"We describe the neural-network training framework used in the Kaldi speech recognition toolkit, which is geared towards training DNNs with large amounts of training data using multiple GPU-equipped or multi-core machines. In order to be as hardware-agnostic as possible, we needed a way to use multiple machines without generating excessive network traffic. Our method is to average the neural network parameters periodically (typically every minute or two), and redistribute the averaged parameters to the machines for further training. Each machine sees different data. By itself, this method does not work very well. However, we have another method, an approximate and efficient implementation of Natural Gradient for Stochastic Gradient Descent (NG-SGD), which seems to allow our periodic-averaging method to work well, as well as substantially improving the convergence of SGD on a single machine.",
"",
"We study the problem of stochastic optimization for deep learning in the parallel computing environment under communication constraints. A new algorithm is proposed in this setting where the communication and coordination of work among concurrent processes (local workers), is based on an elastic force which links the parameters they compute with a center variable stored by the parameter server (master). The algorithm enables the local workers to perform more exploration, i.e. the algorithm allows the local variables to fluctuate further from the center variable by reducing the amount of communication between local workers and the master. We empirically demonstrate that in the deep learning setting, due to the existence of many local optima, allowing more exploration can lead to the improved performance. We propose synchronous and asynchronous variants of the new algorithm. We provide the stability analysis of the asynchronous variant in the round-robin scheme and compare it with the more common parallelized method ADMM. We show that the stability of EASGD is guaranteed when a simple stability condition is satisfied, which is not the case for ADMM. We additionally propose the momentum-based version of our algorithm that can be applied in both synchronous and asynchronous settings. Asynchronous variant of the algorithm is applied to train convolutional neural networks for image classification on the CIFAR and ImageNet datasets. Experiments demonstrate that the new algorithm accelerates the training of deep architectures compared to DOWNPOUR and other common baseline approaches and furthermore is very communication efficient.",
"We present a new approach to scalable training of deep learning machines by incremental block training with intra-block parallel optimization to leverage data parallelism and blockwise model-update filtering to stabilize learning process. By using an implementation on a distributed GPU cluster with an MPI-based HPC machine learning framework to coordinate parallel job scheduling and collective communication, we have trained successfully deep bidirectional long short-term memory (LSTM) recurrent neural networks (RNNs) and fully-connected feed-forward deep neural networks (DNNs) for large vocabulary continuous speech recognition on two benchmark tasks, namely 309-hour Switchboard-I task and 1,860-hour \"Switch-board+Fisher\" task. We achieve almost linear speedup up to 16 GPU cards on LSTM task and 64 GPU cards on DNN task, with either no degradation or improved recognition accuracy in comparison with that of running a traditional mini-batch based stochastic gradient descent training on a single GPU.",
"Deep learning thrives with large neural networks and large datasets. However, larger networks and larger datasets result in longer training times that impede research and development progress. Distributed synchronous SGD offers a potential solution to this problem by dividing SGD minibatches over a pool of parallel workers. Yet to make this scheme efficient, the per-worker workload must be large, which implies nontrivial growth in the SGD minibatch size. In this paper, we empirically show that on the ImageNet dataset large minibatches cause optimization difficulties, but when these are addressed the trained networks exhibit good generalization. Specifically, we show no loss of accuracy when training with large minibatch sizes up to 8192 images. To achieve this result, we adopt a hyper-parameter-free linear scaling rule for adjusting learning rates as a function of minibatch size and develop a new warmup scheme that overcomes optimization challenges early in training. With these simple techniques, our Caffe2-based system trains ResNet-50 with a minibatch size of 8192 on 256 GPUs in one hour, while matching small minibatch accuracy. Using commodity hardware, our implementation achieves 90 scaling efficiency when moving from 8 to 256 GPUs. Our findings enable training visual recognition models on internet-scale data with high efficiency.",
"Parallelization framework has become a necessity to speed up the training of deep neural networks (DNN) recently. Such framework typically employs the Model Average approach, denoted as MA-DNN, in which parallel workers conduct respective training based on their own local data while the parameters of local models are periodically communicated and averaged to obtain a global model which serves as the new start of local models. However, since DNN is a highly non-convex model, averaging parameters cannot ensure that such global model can perform better than those local models. To tackle this problem, we introduce a new parallel training framework called Ensemble-Compression, denoted as EC-DNN. In this framework, we propose to aggregate the local models by ensemble, i.e., averaging the outputs of local models instead of the parameters. As most of prevalent loss functions are convex to the output of DNN, the performance of ensemble-based global model is guaranteed to be at least as good as the average performance of local models. However, a big challenge lies in the explosion of model size since each round of ensemble can give rise to multiple times size increment. Thus, we carry out model compression after each ensemble, specialized by a distillation based method in this paper, to reduce the size of the global model to be the same as the local ones. Our experimental results demonstrate the prominent advantage of EC-DNN over MA-DNN in terms of both accuracy and speedup."
]
} |
1709.09585 | 2758321882 | Predicting traffic conditions has been recently explored as a way to relieve traffic congestion. Several pioneering approaches have been proposed based on traffic observations of the target location as well as its adjacent regions, but they obtain somewhat limited accuracy due to lack of mining road topology. To address the effect attenuation problem, we propose to take account of the traffic of surrounding locations(wider than adjacent range). We propose an end-to-end framework called DeepTransport, in which Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) are utilized to obtain spatial-temporal traffic information within a transport network topology. In addition, attention mechanism is introduced to align spatial and temporal information. Moreover, we constructed and released a real-world large traffic condition dataset with 5-minute resolution. Our experiments on this dataset demonstrate our method captures the complex relationship in temporal and spatial domain. It significantly outperforms traditional statistical methods and a state-of-the-art deep learning method. | On other spatial-temporal tasks, several recent deep learning works attempt to capture both time and space information. DeepST @cite_1 uses convolutional neural networks to predict citywide crowd flows. Meanwhile, ST-ResNet @cite_29 uses the framework of the residual neural networks to forecast the surrounding crowds in each region through a city. These works partition a city into an @math grid map based on the longitude and latitude @cite_20 where a grid denotes a region. However, MapBJ provides the traffic networks in the form of traffic sections instead of longitude and latitude, and the road partition method should be considered the speed limit level rather than equally cut by road length. Due to the differences in data granularity, we do not follow these methods on traffic forecasting. | {
"cite_N": [
"@cite_29",
"@cite_1",
"@cite_20"
],
"mid": [
"2952740813",
"",
"2008483594"
],
"abstract": [
"Forecasting the flow of crowds is of great importance to traffic management and public safety, yet a very challenging task affected by many complex factors, such as inter-region traffic, events and weather. In this paper, we propose a deep-learning-based approach, called ST-ResNet, to collectively forecast the in-flow and out-flow of crowds in each and every region through a city. We design an end-to-end structure of ST-ResNet based on unique properties of spatio-temporal data. More specifically, we employ the framework of the residual neural networks to model the temporal closeness, period, and trend properties of the crowd traffic, respectively. For each property, we design a branch of residual convolutional units, each of which models the spatial properties of the crowd traffic. ST-ResNet learns to dynamically aggregate the output of the three residual neural networks based on data, assigning different weights to different branches and regions. The aggregation is further combined with external factors, such as weather and day of the week, to predict the final traffic of crowds in each and every region. We evaluate ST-ResNet based on two types of crowd flows in Beijing and NYC, finding that its performance exceeds six well-know methods.",
"",
"An approach to freeway travel time prediction based on recurrent neural networks is presented. Travel time prediction requires a modeling approach that is capable of dealing with complex nonlinear spatio-temporal relationships among flows, speeds, and densities. Based on the literature, feedforward neural networks are a class of mathematical models well suited for solving this problem. A drawback of the feed-forward approach is that the size and composition of the input time series are inherently design choices and thus fixed for all input. This may lead to unnecessarily large models. Moreover, for different traffic conditions, different sizes and compositions of input time series may be required, a requirement not satisfied by any feedforward data-driven method. The recurrent neural network topology presented is capable of dealing with the spatiotemporal relationships implicitly. The topology of this neural net is derived from a state-space formulation of the travel time prediction problem, which is in l..."
]
} |
1709.09749 | 2758006462 | Previous studies have demonstrated the empirical success of word embeddings in various applications. In this paper, we investigate the problem of learning distributed representations for text documents which many machine learning algorithms take as input for a number of NLP tasks. We propose a neural network model, KeyVec, which learns document representations with the goal of preserving key semantics of the input text. It enables the learned low-dimensional vectors to retain the topics and important information from the documents that will flow to downstream tasks. Our empirical evaluations show the superior quality of KeyVec representations in two different document understanding tasks. | Le proposed a model, which extends word2vec to vectorial representations for text paragraphs @cite_3 @cite_9 . It projects both words and paragraphs into a single vector space by appending paragraph-specific vectors to typical word2vec . Different from our , Paragraph Vector does not specifically model key information of a given piece of text, while capturing its sequential information. In addition, Paragraph Vector requires extra iterative inference to generate embeddings for unseen paragraphs, whereas our embeds new documents simply via a single feed-forward run. | {
"cite_N": [
"@cite_9",
"@cite_3"
],
"mid": [
"941230081",
"2949547296"
],
"abstract": [
"Paragraph Vectors has been recently proposed as an unsupervised method for learning distributed representations for pieces of texts. In their work, the authors showed that the method can learn an embedding of movie review texts which can be leveraged for sentiment analysis. That proof of concept, while encouraging, was rather narrow. Here we consider tasks other than sentiment analysis, provide a more thorough comparison of Paragraph Vectors to other document modelling algorithms such as Latent Dirichlet Allocation, and evaluate performance of the method as we vary the dimensionality of the learned representation. We benchmarked the models on two document similarity data sets, one from Wikipedia, one from arXiv. We observe that the Paragraph Vector method performs significantly better than other methods, and propose a simple improvement to enhance embedding quality. Somewhat surprisingly, we also show that much like word embeddings, vector operations on Paragraph Vectors can perform useful semantic results.",
"Many machine learning algorithms require the input to be represented as a fixed-length feature vector. When it comes to texts, one of the most common fixed-length features is bag-of-words. Despite their popularity, bag-of-words features have two major weaknesses: they lose the ordering of the words and they also ignore semantics of the words. For example, \"powerful,\" \"strong\" and \"Paris\" are equally distant. In this paper, we propose Paragraph Vector, an unsupervised algorithm that learns fixed-length feature representations from variable-length pieces of texts, such as sentences, paragraphs, and documents. Our algorithm represents each document by a dense vector which is trained to predict words in the document. Its construction gives our algorithm the potential to overcome the weaknesses of bag-of-words models. Empirical results show that Paragraph Vectors outperform bag-of-words models as well as other techniques for text representations. Finally, we achieve new state-of-the-art results on several text classification and sentiment analysis tasks."
]
} |
1709.09749 | 2758006462 | Previous studies have demonstrated the empirical success of word embeddings in various applications. In this paper, we investigate the problem of learning distributed representations for text documents which many machine learning algorithms take as input for a number of NLP tasks. We propose a neural network model, KeyVec, which learns document representations with the goal of preserving key semantics of the input text. It enables the learned low-dimensional vectors to retain the topics and important information from the documents that will flow to downstream tasks. Our empirical evaluations show the superior quality of KeyVec representations in two different document understanding tasks. | In another recent work @cite_12 , Djuric introduced a Hierarchical Document Vector (HDV) model to learn representations from a document stream. Our differs from HDV in that we do not assume the existence of a document stream and HDV does not model sentences. | {
"cite_N": [
"@cite_12"
],
"mid": [
"2250460709"
],
"abstract": [
"We consider the problem of learning distributed representations for documents in data streams. The documents are represented as low-dimensional vectors and are jointly learned with distributed vector representations of word tokens using a hierarchical framework with two embedded neural language models. In particular, we exploit the context of documents in streams and use one of the language models to model the document sequences, and the other to model word sequences within them. The models learn continuous vector representations for both word tokens and documents such that semantically similar documents and words are close in a common vector space. We discuss extensions to our model, which can be applied to personalized recommendation and social relationship mining by adding further user layers to the hierarchy, thus learning user-specific vectors to represent individual preferences. We validated the learned representations on a public movie rating data set from MovieLens, as well as on a large-scale Yahoo News data comprising three months of user activity logs collected on Yahoo servers. The results indicate that the proposed model can learn useful representations of both documents and word tokens, outperforming the current state-of-the-art by a large margin."
]
} |
1709.09287 | 2757267823 | With the proliferation of mobile devices and location-based services, continuous generation of massive volume of streaming spatial objects (i.e., geo-tagged data) opens up new opportunities to address real-world problems by analyzing them. In this paper, we present a novel continuous bursty region detection problem that aims to continuously detect a bursty region of a given size in a specified geographical area from a stream of spatial objects. Specifically, a bursty region shows maximum spike in the number of spatial objects in a given time window. The problem is useful in addressing several real-world challenges such as surge pricing problem in online transportation and disease outbreak detection. To solve the problem, we propose an exact solution and two approximate solutions, and the approximation ratio is @math in terms of the burst score, where @math is a parameter to control the burst score. We further extend these solutions to support detection of top- @math bursty regions. Extensive experiments with real-world data are conducted to demonstrate the efficiency and effectiveness of our solutions. | Data stream management . Our work is also related to data stream management. There has been a long stream of work on various aspects of data streams since the last decade. Some examples are stream clustering @cite_11 @cite_17 , stream join processing @cite_30 , and stream summarization @cite_16 . Since most of these studies focus on general data streams, we only review the work that involves spatial information. Given a stream of spatial-textual objects, @cite_2 aims to estimate the cardinality of a spatial keyword query on objects seen so far. A host of work has also been done to study content-based publish subscribe systems @cite_35 @cite_24 @cite_4 @cite_5 @cite_3 @cite_36 over spatial object streams. In these systems, streaming published items are delivered to the users with matching interests. However, none of these studies consider the problem of detecting bursty regions. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_11",
"@cite_4",
"@cite_36",
"@cite_3",
"@cite_24",
"@cite_2",
"@cite_5",
"@cite_16",
"@cite_17"
],
"mid": [
"2118924829",
"2301494863",
"2170936641",
"",
"",
"",
"",
"975351372",
"",
"2080234606",
""
],
"abstract": [
"We consider the problem of approximating sliding window joins over data streams in a data stream processing system with limited resources. In our model, we deal with resource constraints by shedding load in the form of dropping tuples from the data streams. We first discuss alternate architectural models for data stream join processing, and we survey suitable measures for the quality of an approximation of a set-valued query result. We then consider the number of generated result tuples as the quality measure, and we give optimal offline and fast online algorithms for it. In a thorough experimental study with synthetic and real data we show the efficacy of our solutions. For applications with demand for exact results we introduce a new Archive-metric which captures the amount of work needed to complete the join in case the streams are archived for later processing.",
"As the prevalence of social media and GPS-enabled devices, a massive amount of geo-textual data has been generated in a stream fashion, leading to a variety of applications such as location-based recommendation and information dissemination. In this paper, we investigate a novel real-time top-k monitoring problem over sliding window of streaming data; that is, we continuously maintain the top-k most relevant geo-textual messages (e.g., geo-tagged tweets) for a large number of spatial-keyword subscriptions (e.g., registered users interested in local events) simultaneously. To provide the most recent information under controllable memory cost, sliding window model is employed on the streaming geo-textual data. To the best of our knowledge, this is the first work to study top-k spatial-keyword publish subscribe over sliding window. A novel system, called Skype (Top-k Spatial-keyword Publish Subscribe), is proposed in this paper. In Skype, to continuously maintain top-k results for massive subscriptions, we devise a novel indexing structure upon subscriptions such that each incoming message can be immediately delivered on its arrival. Moreover, to reduce the expensive top-k re-evaluation cost triggered by message expiration, we develop a novel cost-based k-skyband technique to reduce the number of re-evaluations in a cost-effective way. Extensive experiments verify the great efficiency and effectiveness of our proposed techniques.",
"The clustering problem is a difficult problem for the data stream domain. This is because the large volumes of data arriving in a stream renders most traditional algorithms too inefficient. In recent years, a few one-pass clustering algorithms have been developed for the data stream problem. Although such methods address the scalability issues of the clustering problem, they are generally blind to the evolution of the data and do not address the following issues: (1) The quality of the clusters is poor when the data evolves considerably over time. (2) A data stream clustering algorithm requires much greater functionality in discovering and exploring clusters over different portions of the stream. The widely used practice of viewing data stream clustering algorithms as a class of one-pass clustering algorithms is not very useful from an application point of view. For example, a simple one-pass clustering algorithm over an entire data stream of a few years is dominated by the outdated history of the stream. The exploration of the stream over different time windows can provide the users with a much deeper understanding of the evolving behavior of the clusters. At the same time, it is not possible to simultaneously perform dynamic clustering over all possible time horizons for a data stream of even moderately large volume. This paper discusses a fundamentally different philosophy for data stream clustering which is guided by application-centered requirements. The idea is divide the clustering process into an online component which periodically stores detailed summary statistics and an offine component which uses only this summary statistics. The offine component is utilized by the analyst who can use a wide variety of inputs (such as time horizon or number of clusters) in order to provide a quick understanding of the broad clusters in the data stream. The problems of efficient choice, storage, and use of this statistical data for a fast data stream turns out to be quite tricky. For this purpose, we use the concepts of a pyramidal time frame in conjunction with a microclustering approach. Our performance experiments over a number of real and synthetic data sets illustrate the effectiveness, efficiency, and insights provided by our approach.",
"",
"",
"",
"",
"In this paper, we investigate the selectivity estimation problem for streaming spatio-textual data, which arises in many social network and geo-location applications. Specifically, given a set of continuously and rapidly arriving spatio-textual objects, each of which is described by a geo-location and a short text, we aim to accurately estimate the cardinality of a spatial keyword query on objects seen so far, where a spatial keyword query consists of a search region and a set of query keywords. To the best of our knowledge, this is the first work to address this important problem. We first extend two existing techniques to solve this problem, and show their limitations. Inspired by two key observations on the \"locality\" of the correlations among query keywords, we propose a local correlation based method by utilizing an augmented adaptive space partition tree (A2SP-tree for short) to approximately learn a local Bayesian network on-the-fly for a given query and estimate its selectivity. A novel local boosting approach is presented to further enhance the learning accuracy of local Bayesian networks. Our comprehensive experiments on real-life datasets demonstrate the superior performance of the local correlation based algorithm in terms of estimation accuracy compared to other competitors.",
"",
"We introduce a new sublinear space data structure--the count-min sketch--for summarizing data streams. Our sketch allows fundamental queries in data stream summarization such as point, range, and inner product queries to be approximately answered very quickly; in addition, it can be applied to solve several important problems in data streams such as finding quantiles, frequent items, etc. The time and space bounds we show for using the CM sketch to solve these problems significantly improve those previously known--typically from 1 e2 to 1 e in factor.",
""
]
} |
1709.09384 | 2756493928 | Estimating the 6-DoF pose of a camera from a single image relative to a pre-computed 3D point-set is an important task for many computer vision applications. Perspective-n-Point (PnP) solvers are routinely used for camera pose estimation, provided that a good quality set of 2D-3D feature correspondences are known beforehand. However, finding optimal correspondences between 2D key-points and a 3D point-set is non-trivial, especially when only geometric (position) information is known. Existing approaches to the simultaneous pose and correspondence problem use local optimisation, and are therefore unlikely to find the optimal solution without a good pose initialisation, or introduce restrictive assumptions. Since a large proportion of outliers are common for this problem, we instead propose a globally-optimal inlier set cardinality maximisation approach which jointly estimates optimal camera pose and optimal correspondences. Our approach employs branch-and-bound to search the 6D space of camera poses, guaranteeing global optimality without requiring a pose prior. The geometry of SE(3) is used to find novel upper and lower bounds for the number of inliers and local optimisation is integrated to accelerate convergence. The evaluation empirically supports the optimality proof and shows that the method performs much more robustly than existing approaches, including on a large-scale outdoor data-set. | When correspondences are unknown, the problem becomes more challenging. For the 2D--2D case, problems such as correspondence-free rigid registration @cite_4 @cite_11 , SfM @cite_18 @cite_21 @cite_17 and relative camera pose @cite_49 have been addressed. For the 2D--3D case, solution have been proposed for registering a collection of images @cite_32 or multiple cameras @cite_13 to a 3D point-set. The more general problem, however, is pose estimation from a single image. David al @cite_31 proposed the SoftPOSIT algorithm, which alternates correspondence assignment with an iterative pose update algorithm. Moreno-Noguer al @cite_9 proposed the BlindPnP algorithm, which represents the pose prior as a Gaussian mixture model from which a Kalman filter is initialised for matching. It outperformed SoftPOSIT when large amounts of clutter, occlusions and repetitive patterns were present. However, both are susceptible to local optima, require a pose prior and cannot guarantee global optimality. | {
"cite_N": [
"@cite_18",
"@cite_11",
"@cite_4",
"@cite_9",
"@cite_21",
"@cite_32",
"@cite_49",
"@cite_31",
"@cite_13",
"@cite_17"
],
"mid": [
"2158054358",
"2055743756",
"2049981393",
"1752823667",
"",
"2215550549",
"2460525765",
"2165919662",
"1946683413",
"2086613398"
],
"abstract": [
"A method is presented to recover 3D scene structure and camera motion from multiple images without the need for correspondence information. The problem is framed as finding the maximum likelihood structure and motion given only the 2D measurements, integrating over all possible assignments of 3D features to 2D measurements. This goal is achieved by means of an algorithm which iteratively refines a probability distribution over the set of all correspondence assignments. At each iteration a new structure from motion problem is solved, using as input a set of 'virtual measurements' derived from this probability distribution. The distribution needed can be efficiently obtained by Markov Chain Monte Carlo sampling. The approach is cast within the framework of Expectation-Maximization, which guarantees convergence to a local maximizer of the likelihood. The algorithm works well in practice, as will be demonstrated using results on several real image sequences.",
"Algorithms for geometric matching and feature extraction that work by recursively subdividing transformation space and bounding the quality of match have been proposed in a number of different contexts and become increasingly popular over the last few years. This paper describes matchlist-based branch-and-bound techniques and presents a number of new applications of branch-and-bound methods, among them, a method for globally optimal partial line segment matching under bounded or Gaussian error, point matching under a Gaussian error model with subpixel accuracy and precise orientation models, and a simple and robust technique for finding multiple distinct objcct instances. It also contains extensive reference information for the implementation of such matching methods under a wide variety of error bounds and transformations. In addition, the paper contains a number of benchmarks and evaluations that provide new information about the runtime behavior of branch-and-bound matching algorithms in general, and that help choose among different implementation strategies, such as the use of point location data structures and space time tradeoffs involving depth-first search.",
"The authors describe a general-purpose, representation-independent method for the accurate and computationally efficient registration of 3-D shapes including free-form curves and surfaces. The method handles the full six degrees of freedom and is based on the iterative closest point (ICP) algorithm, which requires only a procedure to find the closest point on a geometric entity to a given point. The ICP algorithm always converges monotonically to the nearest local minimum of a mean-square distance metric, and the rate of convergence is rapid during the first few iterations. Therefore, given an adequate set of initial rotations and translations for a particular class of objects with a certain level of 'shape complexity', one can globally minimize the mean-square distance metric over all six degrees of freedom by testing each initial registration. One important application of this method is to register sensed data from unfixtured rigid objects with an ideal geometric model, prior to shape inspection. Experimental results show the capabilities of the registration algorithm on point sets, curves, and surfaces. >",
"Estimating a camera pose given a set of 3D-object and 2D-image feature points is a well understood problem when correspondences are given. However, when such correspondences cannot be established a priori, one must simultaneously compute them along with the pose. Most current approaches to solving this problem are too computationally intensive to be practical. An interesting exception is the SoftPosit algorithm, that looks for the solution as the minimum of a suitable objective function. It is arguably one of the best algorithms but its iterative nature means it can fail in the presence of clutter, occlusions, or repetitive patterns. In this paper, we propose an approach that overcomes this limitation by taking advantage of the fact that, in practice, some prior on the camera pose is often available. We model it as a Gaussian Mixture Model that we progressively refine by hypothesizing new correspondences. This rapidly reduces the number of potential matches for each 3D point and lets us explore the pose space more thoroughly than SoftPosit at a similar computational cost. We will demonstrate the superior performance of our approach on both synthetic and real data.",
"",
"This paper deals with the problem of registering a known structured 3D scene and its metric Structure-from-Motion (SfM) counterpart. The proposed work relies on a prior plane segmentation of the 3D scene and aligns the data obtained from both modalities by solving the point-to-plane assignment problem. An inliers-maximization approach within a Branch-and-Bound (BnB) search scheme is adopted. For the first time in this paper, a Sum-of-Squares optimization theory framework is employed for identifying point-to-plane mismatches (i.e. outliers) with certainty. This allows us to iteratively build potential inliers sets and converge to the solution satisfied by the largest number of point-to-plane assignments. Furthermore, our approach is boosted by new plane visibility conditions which are also introduced in this paper. Using this framework, we solve the registration problem in two cases: (i) a set of putative point-to-plane correspondences (with possibly overwhelmingly many outliers) is given as input and (ii) no initial correspondences are given. In both cases, our approach yields outstanding results in terms of robustness and optimality.",
"Previous work on estimating the epipolar geometry of two views relies on being able to reliably match feature points based on appearance. In this paper, we go one step further and show that it is feasible to compute both the epipolar geometry and the correspondences at the same time based on geometry only. We do this in a globally optimal manner. Our approach is based on an efficient branch and bound technique in combination with bipartite matching to solve the correspondence problem. We rely on several recent works to obtain good bounding functions to battle the combinatorial explosion of possible matchings. It is experimentally demonstrated that more difficult cases can be handled and that more inlier correspondences can be obtained by being less restrictive in the matching phase.",
"The problem of pose estimation arises in many areas of computer vision, including object recognition, object tracking, site inspection and updating, and autonomous navigation when scene models are available. We present a new algorithm, called SoftPOSIT, for determining the pose of a 3D object from a single 2D image when correspondences between object points and image points are not known. The algorithm combines the iterative softassign algorithm (Gold and Rangarajan, 1996; , 1998) for computing correspondences and the iterative POSIT algorithm (DeMenthon and Davis, 1995) for computing object pose under a full-perspective camera model. Our algorithm, unlike most previous algorithms for pose determination, does not have to hypothesize small sets of matches and then verify the remaining image points. Instead, all possible matches are treated identically throughout the search for an optimal pose. The performance of the algorithm is extensively evaluated in Monte Carlo simulations on synthetic data under a variety of levels of clutter, occlusion, and image noise. These tests show that the algorithm performs well in a variety of difficult scenarios, and empirical evidence suggests that the algorithm has an asymptotic run-time complexity that is better than previous methods by a factor of the number of image points. The algorithm is being applied to a number of practical autonomous vehicle navigation problems including the registration of 3D architectural models of a city to images, and the docking of small robots onto larger robots.",
"This paper investigates the problem of registering a scanned scene, represented by 3D Euclidean point coordinates, and two or more uncalibrated cameras. An unknown subset of the scanned points have their image projections detected and matched across images. The proposed approach assumes the cameras only known in some arbitrary projective frame and no calibration or autocalibration is required. The devised solution is based on a Linear Matrix Inequality (LMI) framework that allows simultaneously estimating the projective transformation relating the cameras to the scene and establishing 2D-3D correspondences without triangulating image points. The proposed LMI framework allows both deriving triangulation-free LMI cheirality conditions and establishing putative correspondences between 3D volumes (boxes) and 2D pixel coordinates. Two registration algorithms, one exploiting the scene's structure and the other concerned with robustness, are presented. Both algorithms employ the Branch-and-Prune paradigm and guarantee convergence to a global solution under mild initial bound conditions. The results of our experiments are presented and compared against other approaches.",
"Traditionally, the camera pose recovery problem has been formulated as one of estimating the optimal camera pose given a set of point correspondences. This critically depends on the accuracy of the point correspondences and would have problems in dealing with ambiguous features such as edge contours and high visual clutter. Joint estimation of camera pose and correspondence attempts to improve performance by explicitly acknowledging the chicken and egg nature of the pose and correspondence problem. However, such joint approaches for the two-view problem are still few and even then, they face problems when scenes contain largely edge cues with few corners, due to the fact that epipolar geometry only provides a \"soft\" point to line constraint. Viewed from the perspective of point set registration, the point matching process can be regarded as the registration of points while preserving their relative positions (i.e. preserving scene coherence). By demanding that the point set should be transformed coherently across views, this framework leverages on higher level perceptual information such as the shape of the contour. While thus potentially allowing registration of non-unique edge points, the registration framework in its traditional form is subject to substantial point localization error and is thus not suitable for estimating camera pose. In this paper, we introduce an algorithm which jointly estimates camera pose and correspondence within a point set registration framework based on motion coherence, with the camera pose helping to localize the edge registration, while the \"ambiguous\" edge information helps to guide camera pose computation. The algorithm can compute camera pose over large displacements and by utilizing the non-unique edge points can recover camera pose from what were previously regarded as feature-impoverished SfM scenes. Our algorithm is also sufficiently flexible to incorporate high dimensional feature descriptors and works well on traditional SfM scenes with adequate numbers of unique corners."
]
} |
1709.09121 | 2758852207 | This paper addresses the problem of joint detection and recounting of abnormal events in videos. Recounting of abnormal events, i.e., explaining why they are judged to be abnormal, is an unexplored but critical task in video surveillance, because it helps human observers quickly judge if they are false alarms or not. To describe the events in the human-understandable form for event recounting, learning generic knowledge about visual concepts (e.g., object and action) is crucial. Although convolutional neural networks (CNNs) have achieved promising results in learning such concepts, it remains an open question as to how to effectively use CNNs for abnormal event detection, mainly due to the environment-dependent nature of the anomaly detection. In this paper, we tackle this problem by integrating a generic CNN model and environment-dependent anomaly detectors. Our approach first learns CNN with multiple visual tasks to exploit semantic information that is useful for detecting and recounting abnormal events. By appropriately plugging the model into anomaly detectors, we can detect and recount abnormal events while taking advantage of the discriminative power of CNNs. Our approach outperforms the state-of-the-art on Avenue and UCSD Ped2 benchmarks for abnormal event detection and also produces promising results of abnormal event recounting. | The approach of anomaly detection first involves modeling normal behavior and then detecting samples that deviate from it. Modeling normal patterns of object trajectories is one standard approach @cite_4 @cite_15 @cite_19 @cite_14 to anomaly detection in videos. While it can capture long-term object-level semantics, tracking fails in crowded or cluttered scenes. An alternative approach is modeling appearance and activity patterns using low-level features extracted from local regions, which is a current standard approach, especially in crowded scenes. This approach can be divided into two stages: local anomaly detection assigning anomaly score to each local region independently, and globally consistent inference integrating local anomaly scores into a globally consistent anomaly map with statistical inferences. | {
"cite_N": [
"@cite_19",
"@cite_15",
"@cite_14",
"@cite_4"
],
"mid": [
"1498983721",
"2130349088",
"2129520225",
"2142412278"
],
"abstract": [
"This paper shows how the output of a number of detection and tracking algorithms can be fused to achieve robust tracking of people in an indoor environment. The new tracking system contains three co-operating parts: i) an Active Shape Tracker using a PCA-generated model of pedestrian outline shapes, ii) a Region Tracker, featuring region splitting and merging for multiple hypothesis matching, and iii) a Head Detector to aid in the initialisation of tracks. Data from the three parts are fused together to select the best tracking hypotheses.The new method is validated using sequences from surveillance cameras in a underground station. It is demonstrated that robust realtime tracking of people can be achieved with the new tracking system using standard PC hardware.",
"Compared to other anomalous video event detection approaches that analyze object trajectories only, we propose a context-aware method to detect anomalies. By tracking all moving objects in the video, three different levels of spatiotemporal contexts are considered, i.e., point anomaly of a video object, sequential anomaly of an object trajectory, and co-occurrence anomaly of multiple video objects. A hierarchical data mining approach is proposed. At each level, frequency-based analysis is performed to automatically discover regular rules of normal events. Events deviating from these rules are identified as anomalies. The proposed method is computationally efficient and can infer complex rules. Experiments on real traffic video validate that the detected video anomalies are hazardous or illegal according to traffic regulations.",
"Activity analysis is a basic task in video surveillance and has become an active research area. However, due to the diversity of moving objects category and their motion patterns, developing robust semantic scene models for activity analysis remains a challenging problem in traffic scenarios. This paper proposes a novel framework to learn semantic scene models. In this framework, the detected moving objects are first classified as pedestrians or vehicles via a co-trained classifier which takes advantage of the multiview information of objects. As a result, the framework can automatically learn motion patterns respectively for pedestrians and vehicles. Then, a graph is proposed to learn and cluster the motion patterns. To this end, trajectory is parameterized and the image is cut into multiple blocks which are taken as the nodes in the graph. Based on the parameters of trajectories, the primary motion patterns in each node (block) are extracted via Gaussian mixture model (GMM), and supplied to this graph. The graph cut algorithm is finally employed to group the motion patterns together, and trajectories are clustered to learn semantic scene models. Experimental results and applications to real world scenes show the validity of our proposed method.",
"We present a novel framework for learning patterns of motion and sizes of objects in static camera surveillance. The proposed method provides a new higher-level layer to the traditional surveillance pipeline for anomalous event detection and scene model feedback. Pixel level probability density functions (pdfs) of appearance have been used for background modelling in the past, but modelling pixel level pdfs of object speed and size from the tracks is novel. Each pdf is modelled as a multivariate Gaussian mixture model (GMM) of the motion (destination location & transition time) and the size (width & height) parameters of the objects at that location. Output of the tracking module is used to perform unsupervised EM-based learning of every GMM. We have successfully used the proposed scene model to detect local as well as global anomalies in object tracks. We also show the use of this scene model to improve object detection through pixel-level parameter feedback of the minimum object size and background learning rate. Most object path modelling approaches first cluster the tracks into major paths in the scene, which can be a source of error. We avoid this by building local pdfs that capture a variety of tracks which are passing through them. Qualitative and quantitative analysis of actual surveillance videos proved the effectiveness of the proposed approach."
]
} |
1709.09121 | 2758852207 | This paper addresses the problem of joint detection and recounting of abnormal events in videos. Recounting of abnormal events, i.e., explaining why they are judged to be abnormal, is an unexplored but critical task in video surveillance, because it helps human observers quickly judge if they are false alarms or not. To describe the events in the human-understandable form for event recounting, learning generic knowledge about visual concepts (e.g., object and action) is crucial. Although convolutional neural networks (CNNs) have achieved promising results in learning such concepts, it remains an open question as to how to effectively use CNNs for abnormal event detection, mainly due to the environment-dependent nature of the anomaly detection. In this paper, we tackle this problem by integrating a generic CNN model and environment-dependent anomaly detectors. Our approach first learns CNN with multiple visual tasks to exploit semantic information that is useful for detecting and recounting abnormal events. By appropriately plugging the model into anomaly detectors, we can detect and recount abnormal events while taking advantage of the discriminative power of CNNs. Our approach outperforms the state-of-the-art on Avenue and UCSD Ped2 benchmarks for abnormal event detection and also produces promising results of abnormal event recounting. | Local anomaly detection can be seen as a simple novelty detection problem @cite_42 : a model of normality is inferred using a set of normal features @math as training data, and used to assign novelty scores (anomaly scores) @math to test sample @math . Novelty detectors used in video anomaly detection include distance-based @cite_36 , reconstruction-based (e.g., autoencoders @cite_1 @cite_29 , sparse coding @cite_33 @cite_26 @cite_28 ), domain-based (one-class SVM @cite_41 ), and probabilistic methods (e.g., mixture of dynamic texture @cite_16 , mixture of probabilistic PCA @cite_3 ), following the categories in the review by @cite_42 . These models are generally built on the low-level features (e.g., HOG and HOF) extracted from densely sampled local patches. Several recent approaches have investigated the learning-based features using autoencoders @cite_29 @cite_41 , which minimize reconstruction errors without using any labeled data for training. Antic and Ommer @cite_34 detected object hypotheses by video parsing instead of dense sampling, although they relied on background subtraction that was not robust to illumination changes. | {
"cite_N": [
"@cite_26",
"@cite_33",
"@cite_36",
"@cite_28",
"@cite_29",
"@cite_42",
"@cite_1",
"@cite_41",
"@cite_3",
"@cite_34",
"@cite_16"
],
"mid": [
"2163612318",
"2012931101",
"2097363716",
"2021659075",
"",
"2115627867",
"2341058432",
"2194550927",
"2138092272",
"",
"2122361470"
],
"abstract": [
"Speedy abnormal event detection meets the growing demand to process an enormous number of surveillance videos. Based on inherent redundancy of video structures, we propose an efficient sparse combination learning framework. It achieves decent performance in the detection phase without compromising result quality. The short running time is guaranteed because the new method effectively turns the original complicated problem to one in which only a few costless small-scale least square optimization steps are involved. Our method reaches high detection rates on benchmark datasets at a speed of 140-150 frames per second on average when computing on an ordinary desktop PC using MATLAB.",
"We propose to detect abnormal events via a sparse reconstruction over the normal bases. Given an over-complete normal basis set (e.g., an image sequence or a collection of local spatio-temporal patches), we introduce the sparse reconstruction cost (SRC) over the normal dictionary to measure the normalness of the testing sample. To condense the size of the dictionary, a novel dictionary selection method is designed with sparsity consistency constraint. By introducing the prior weight of each basis during sparse reconstruction, the proposed SRC is more robust compared to other outlier detection criteria. Our method provides a unified solution to detect both local abnormal events (LAE) and global abnormal events (GAE). We further extend it to support online abnormal event detection by updating the dictionary incrementally. Experiments on three benchmark datasets and the comparison to the state-of-the-art methods validate the advantages of our algorithm.",
"Anomalies in many video surveillance applications have local spatio-temporal signatures, namely, they occur over a small time window or a small spatial region. The distinguishing feature of these scenarios is that outside this spatio-temporal anomalous region, activities appear normal. We develop a probabilistic framework to account for such local spatio-temporal anomalies. We show that our framework admits elegant characterization of optimal decision rules. A key insight of the paper is that if anomalies are local optimal decision rules are local even when the nominal behavior exhibits global spatial and temporal statistical dependencies. This insight helps collapse the large ambient data dimension for detecting local anomalies. Consequently, consistent data-driven local empirical rules with provable performance can be derived with limited training data. Our empirical rules are based on scores functions derived from local nearest neighbor distances. These rules aggregate statistics across spatio-temporal locations & scales, and produce a single composite score for video segments. We demonstrate the efficacy of our scheme on several video surveillance datasets and compare with existing work.",
"Real-time unusual event detection in video stream has been a difficult challenge due to the lack of sufficient training information, volatility of the definitions for both normality and abnormality, time constraints, and statistical limitation of the fitness of any parametric models. We propose a fully unsupervised dynamic sparse coding approach for detecting unusual events in videos based on online sparse re-constructibility of query signals from an atomically learned event dictionary, which forms a sparse coding bases. Based on an intuition that usual events in a video are more likely to be reconstructible from an event dictionary, whereas unusual events are not, our algorithm employs a principled convex optimization formulation that allows both a sparse reconstruction code, and an online dictionary to be jointly inferred and updated. Our algorithm is completely un-supervised, making no prior assumptions of what unusual events may look like and the settings of the cameras. The fact that the bases dictionary is updated in an online fashion as the algorithm observes more data, avoids any issues with concept drift. Experimental results on hours of real world surveillance video and several Youtube videos show that the proposed algorithm could reliably locate the unusual events in the video sequence, outperforming the current state-of-the-art methods.",
"",
"Novelty detection is the task of classifying test data that differ in some respect from the data that are available during training. This may be seen as ''one-class classification'', in which a model is constructed to describe ''normal'' training data. The novelty detection approach is typically used when the quantity of available ''abnormal'' data is insufficient to construct explicit models for non-normal classes. Application includes inference in datasets from critical systems, where the quantity of available normal data is very large, such that ''normality'' may be accurately modelled. In this review we aim to provide an updated and structured investigation of novelty detection research papers that have appeared in the machine learning literature during the last decade.",
"Perceiving meaningful activities in a long video sequence is a challenging problem due to ambiguous definition of 'meaningfulness' as well as clutters in the scene. We approach this problem by learning a generative model for regular motion patterns (termed as regularity) using multiple sources with very limited supervision. Specifically, we propose two methods that are built upon the autoencoders for their ability to work with little to no supervision. We first leverage the conventional handcrafted spatio-temporal local features and learn a fully connected autoencoder on them. Second, we build a fully convolutional feed-forward autoencoder to learn both the local features and the classifiers as an end-to-end learning framework. Our model can capture the regularities from multiple datasets. We evaluate our methods in both qualitative and quantitative ways - showing the learned regularity of videos in various aspects and demonstrating competitive performance on anomaly detection datasets as an application.",
"We present a novel unsupervised deep learning framework for anomalous event detection in complex video scenes. While most existing works merely use hand-crafted appearance and motion features, we propose Appearance and Motion DeepNet (AMDN) which utilizes deep neural networks to automatically learn feature representations. To exploit the complementary information of both appearance and motion patterns, we introduce a novel double fusion framework, combining both the benefits of traditional early fusion and late fusion strategies. Specifically, stacked denoising autoencoders are proposed to separately learn both appearance and motion features as well as a joint representation (early fusion). Based on the learned representations, multiple one-class SVM models are used to predict the anomaly scores of each input, which are then integrated with a late fusion strategy for final anomaly detection. We evaluate the proposed method on two publicly available video surveillance datasets, showing competitive performance with respect to state of the art approaches.",
"We propose a space-time Markov random field (MRF) model to detect abnormal activities in video. The nodes in the MRF graph correspond to a grid of local regions in the video frames, and neighboring nodes in both space and time are associated with links. To learn normal patterns of activity at each local node, we capture the distribution of its typical optical flow with a mixture of probabilistic principal component analyzers. For any new optical flow patterns detected in incoming video clips, we use the learned model and MRF graph to compute a maximum a posteriori estimate of the degree of normality at each local node. Further, we show how to incrementally update the current model parameters as new video observations stream in, so that the model can efficiently adapt to visual context changes over a long period of time. Experimental results on surveillance videos show that our space-time MRF model robustly detects abnormal activities both in a local and global sense: not only does it accurately localize the atomic abnormal activities in a crowded video, but at the same time it captures the global-level abnormalities caused by irregular interactions between local activities.",
"",
"A novel framework for anomaly detection in crowded scenes is presented. Three properties are identified as important for the design of a localized video representation suitable for anomaly detection in such scenes: 1) joint modeling of appearance and dynamics of the scene, and the abilities to detect 2) temporal, and 3) spatial abnormalities. The model for normal crowd behavior is based on mixtures of dynamic textures and outliers under this model are labeled as anomalies. Temporal anomalies are equated to events of low-probability, while spatial anomalies are handled using discriminant saliency. An experimental evaluation is conducted with a new dataset of crowded scenes, composed of 100 video sequences and five well defined abnormality categories. The proposed representation is shown to outperform various state of the art anomaly detection techniques."
]
} |
1709.09121 | 2758852207 | This paper addresses the problem of joint detection and recounting of abnormal events in videos. Recounting of abnormal events, i.e., explaining why they are judged to be abnormal, is an unexplored but critical task in video surveillance, because it helps human observers quickly judge if they are false alarms or not. To describe the events in the human-understandable form for event recounting, learning generic knowledge about visual concepts (e.g., object and action) is crucial. Although convolutional neural networks (CNNs) have achieved promising results in learning such concepts, it remains an open question as to how to effectively use CNNs for abnormal event detection, mainly due to the environment-dependent nature of the anomaly detection. In this paper, we tackle this problem by integrating a generic CNN model and environment-dependent anomaly detectors. Our approach first learns CNN with multiple visual tasks to exploit semantic information that is useful for detecting and recounting abnormal events. By appropriately plugging the model into anomaly detectors, we can detect and recount abnormal events while taking advantage of the discriminative power of CNNs. Our approach outperforms the state-of-the-art on Avenue and UCSD Ped2 benchmarks for abnormal event detection and also produces promising results of abnormal event recounting. | Globally consistent inference was introduced in several approaches to guarantee the consistency of local anomaly scores. @cite_39 enforce temporal consistency by modeling temporal sequences with hidden Markov models (HMM). The spatial consistency is also introduced in several studies using Markov random field (MRF) @cite_12 @cite_3 @cite_24 to capture spatial interdependencies between local regions. | {
"cite_N": [
"@cite_24",
"@cite_3",
"@cite_12",
"@cite_39"
],
"mid": [
"1967456674",
"2138092272",
"1983103633",
"2125105611"
],
"abstract": [
"The detection and localization of anomalous behaviors in crowded scenes is considered, and a joint detector of temporal and spatial anomalies is proposed. The proposed detector is based on a video representation that accounts for both appearance and dynamics, using a set of mixture of dynamic textures models. These models are used to implement 1) a center-surround discriminant saliency detector that produces spatial saliency scores, and 2) a model of normal behavior that is learned from training data and produces temporal saliency scores. Spatial and temporal anomaly maps are then defined at multiple spatial scales, by considering the scores of these operators at progressively larger regions of support. The multiscale scores act as potentials of a conditional random field that guarantees global consistency of the anomaly judgments. A data set of densely crowded pedestrian walkways is introduced and used to evaluate the proposed anomaly detector. Experiments on this and other data sets show that the latter achieves state-of-the-art anomaly detection results.",
"We propose a space-time Markov random field (MRF) model to detect abnormal activities in video. The nodes in the MRF graph correspond to a grid of local regions in the video frames, and neighboring nodes in both space and time are associated with links. To learn normal patterns of activity at each local node, we capture the distribution of its typical optical flow with a mixture of probabilistic principal component analyzers. For any new optical flow patterns detected in incoming video clips, we use the learned model and MRF graph to compute a maximum a posteriori estimate of the degree of normality at each local node. Further, we show how to incrementally update the current model parameters as new video observations stream in, so that the model can efficiently adapt to visual context changes over a long period of time. Experimental results on surveillance videos show that our space-time MRF model robustly detects abnormal activities both in a local and global sense: not only does it accurately localize the atomic abnormal activities in a crowded video, but at the same time it captures the global-level abnormalities caused by irregular interactions between local activities.",
"We explore a location based approach for behavior modeling and abnormality detection. In contrast to the conventional object based approach where an object may first be tagged, identified, classified, and tracked, we proceed directly with event characterization and behavior modeling at the pixel(s) level based on motion labels obtained from background subtraction. Since events are temporally and spatially dependent, this calls for techniques that account for statistics of spatiotemporal events. Based on motion labels, we learn co-occurrence statistics for normal events across space-time. For one (or many) key pixel(s), we estimate a co-occurrence matrix that accounts for any two active labels which co-occur simultaneously within the same spatiotemporal volume. This co-occurrence matrix is then used as a potential function in a Markov random field (MRF) model to describe the probability of observations within the same spatiotemporal volume. The MRF distribution implicitly accounts for speed, direction, as well as the average size of the objects passing in front of each key pixel. Furthermore, when the spatiotemporal volume is large enough, the co-occurrence distribution contains the average normal path followed by moving objects. The learned normal co-occurrence distribution can be used for abnormal detection. Our method has been tested on various outdoor videos representing various challenges.",
"Extremely crowded scenes present unique challenges to video analysis that cannot be addressed with conventional approaches. We present a novel statistical framework for modeling the local spatio-temporal motion pattern behavior of extremely crowded scenes. Our key insight is to exploit the dense activity of the crowded scene by modeling the rich motion patterns in local areas, effectively capturing the underlying intrinsic structure they form in the video. In other words, we model the motion variation of local space-time volumes and their spatial-temporal statistical behaviors to characterize the overall behavior of the scene. We demonstrate that by capturing the steady-state motion behavior with these spatio-temporal motion pattern models, we can naturally detect unusual activity as statistical deviations. Our experiments show that local spatio-temporal motion pattern modeling offers promising results in real-world scenes with complex activities that are hard for even human observers to analyze."
]
} |
1709.09121 | 2758852207 | This paper addresses the problem of joint detection and recounting of abnormal events in videos. Recounting of abnormal events, i.e., explaining why they are judged to be abnormal, is an unexplored but critical task in video surveillance, because it helps human observers quickly judge if they are false alarms or not. To describe the events in the human-understandable form for event recounting, learning generic knowledge about visual concepts (e.g., object and action) is crucial. Although convolutional neural networks (CNNs) have achieved promising results in learning such concepts, it remains an open question as to how to effectively use CNNs for abnormal event detection, mainly due to the environment-dependent nature of the anomaly detection. In this paper, we tackle this problem by integrating a generic CNN model and environment-dependent anomaly detectors. Our approach first learns CNN with multiple visual tasks to exploit semantic information that is useful for detecting and recounting abnormal events. By appropriately plugging the model into anomaly detectors, we can detect and recount abnormal events while taking advantage of the discriminative power of CNNs. Our approach outperforms the state-of-the-art on Avenue and UCSD Ped2 benchmarks for abnormal event detection and also produces promising results of abnormal event recounting. | While recent approaches have placed emphasis on global consistency @cite_17 @cite_2 @cite_24 , it is defined on top of the local anomaly scores as explained above. Besides, several critical issues remain in local anomaly detection. Despite the great success of CNN approaches in many visual tasks, the application of CNN to abnormal event detection is yet to be explored. Normally CNN requires rich supervised information (positive negative, ranking, etc.) and abundant training data. However, supervised information is unavailable for anomaly detection by definition. @cite_1 learned temporal regularity in videos with a convolutional autoencoder and used it for anomaly detection. However, we consider that autoencoders that are only learned with unlabeled data do not fully leverage the expressive power of CNN. Besides, recounting of abnormal events is yet to be considered, while several approaches have been proposed for multimedia event recounting by localizing key evidence @cite_37 @cite_8 or summarizing the evidence of detected events by text @cite_21 . | {
"cite_N": [
"@cite_37",
"@cite_8",
"@cite_21",
"@cite_1",
"@cite_24",
"@cite_2",
"@cite_17"
],
"mid": [
"1900913856",
"243985932",
"2043727559",
"2341058432",
"1967456674",
"2519730330",
"1957718552"
],
"abstract": [
"In this paper, we focus on complex event detection in internet videos while also providing the key evidences of the detection results. Convolutional Neural Networks (CNNs) have achieved promising performance in image classification and action recognition tasks. However, it remains an open problem how to use CNNs for video event detection and recounting, mainly due to the complexity and diversity of video events. In this work, we propose a flexible deep CNN infrastructure, namely Deep Event Network (DevNet), that simultaneously detects pre-defined events and provides key spatial-temporal evidences. Taking key frames of videos as input, we first detect the event of interest at the video level by aggregating the CNN features of the key frames. The pieces of evidences which recount the detection results, are also automatically localized, both temporally and spatially. The challenge is that we only have video level labels, while the key evidences usually take place at the frame levels. Based on the intrinsic property of CNNs, we first generate a spatial-temporal saliency map by back passing through DevNet, which then can be used to find the key frames which are most indicative to the event, as well as to localize the specific spatial position, usually an object, in the frame of the highly indicative area. Experiments on the large scale TRECVID 2014 MEDTest dataset demonstrate the promising performance of our method, both for event detection and evidence recounting.",
"Complex events consist of various human interactions with different objects in diverse environments. The evidences needed to recognize events may occur in short time periods with variable lengths and can happen anywhere in a video. This fact prevents conventional machine learning algorithms from effectively recognizing the events. In this paper, we propose a novel method that can automatically identify the key evidences in videos for detecting complex events. Both static instances (objects) and dynamic instances (actions) are considered by sampling frames and temporal segments respectively. To compare the characteristic power of heterogeneous instances, we embed static and dynamic instances into a multiple instance learning framework via instance similarity measures, and cast the problem as an Evidence Selective Ranking (ESR) process. We impose l1 norm to select key evidences while using the Infinite Push Loss Function to enforce positive videos to have higher detection scores than negative videos. The Alternating Direction Method of Multipliers (ADMM) algorithm is used to solve the optimization problem. Experiments on large-scale video datasets show that our method can improve the detection accuracy while providing the unique capability in discovering key evidences of each complex event.",
"Multimedia event detection has drawn a lot of attention in recent years. Given a recognized event, in this paper, we conduct a pilot study of the multimedia event recounting problem, which answers the question why this video is recognized as this event, i.e. what evidences this decision is made on. In order to provide a semantic recounting of the multimedia event, we adopt a concept-based event representation for learning a discriminative event model. Then, we present a recounting approach that exactly recovers the contribution of semantic evidence to the event classification decision. This approach can be applied on any additive discriminative classifiers. The promising result is shown on the MED11 dataset that contains 15 events in thousands of YouTube like videos.",
"Perceiving meaningful activities in a long video sequence is a challenging problem due to ambiguous definition of 'meaningfulness' as well as clutters in the scene. We approach this problem by learning a generative model for regular motion patterns (termed as regularity) using multiple sources with very limited supervision. Specifically, we propose two methods that are built upon the autoencoders for their ability to work with little to no supervision. We first leverage the conventional handcrafted spatio-temporal local features and learn a fully connected autoencoder on them. Second, we build a fully convolutional feed-forward autoencoder to learn both the local features and the classifiers as an end-to-end learning framework. Our model can capture the regularities from multiple datasets. We evaluate our methods in both qualitative and quantitative ways - showing the learned regularity of videos in various aspects and demonstrating competitive performance on anomaly detection datasets as an application.",
"The detection and localization of anomalous behaviors in crowded scenes is considered, and a joint detector of temporal and spatial anomalies is proposed. The proposed detector is based on a video representation that accounts for both appearance and dynamics, using a set of mixture of dynamic textures models. These models are used to implement 1) a center-surround discriminant saliency detector that produces spatial saliency scores, and 2) a model of normal behavior that is learned from training data and produces temporal saliency scores. Spatial and temporal anomaly maps are then defined at multiple spatial scales, by considering the scores of these operators at progressively larger regions of support. The multiscale scores act as potentials of a conditional random field that guarantees global consistency of the anomaly judgments. A data set of densely crowded pedestrian walkways is introduced and used to evaluate the proposed anomaly detector. Experiments on this and other data sets show that the latter achieves state-of-the-art anomaly detection results.",
"We address an anomaly detection setting in which training sequences are unavailable and anomalies are scored independently of temporal ordering. Current algorithms in anomaly detection are based on the classical density estimation approach of learning high-dimensional models and finding low-probability events. These algorithms are sensitive to the order in which anomalies appear and require either training data or early context assumptions that do not hold for longer, more complex videos. By defining anomalies as examples that can be distinguished from other examples in the same video, our definition inspires a shift in approaches from classical density estimation to simple discriminative learning. Our contributions include a novel framework for anomaly detection that is (1) independent of temporal ordering of anomalies, and (2) unsupervised, requiring no separate training sequences. We show that our algorithm can achieve state-of-the-art results even when we adjust the setting by removing training sequences from standard datasets.",
"This paper presents a hierarchical framework for detecting local and global anomalies via hierarchical feature representation and Gaussian process regression. While local anomaly is typically detected as a 3D pattern matching problem, we are more interested in global anomaly that involves multiple normal events interacting in an unusual manner such as car accident. To simultaneously detect local and global anomalies, we formulate the extraction of normal interactions from training video as the problem of efficiently finding the frequent geometric relations of the nearby sparse spatio-temporal interest points. A codebook of interaction templates is then constructed and modeled using Gaussian process regression. A novel inference method for computing the likelihood of an observed interaction is also proposed. As such, our model is robust to slight topological deformations and can handle the noise and data unbalance problems in the training data. Simulations show that our system outperforms the main state-of-the-art methods on this topic and achieves at least 80 detection rates based on three challenging datasets."
]
} |
1709.09323 | 2950346959 | In recent years, dynamic vision sensors (DVS), also known as event-based cameras or neuromorphic sensors, have seen increased use due to various advantages over traditional frame-based cameras. Its high temporal resolution overcomes motion blurring, its high dynamic range overcomes extreme illumination conditions and its low power consumption makes it suitable for platforms such as drones and self-driving cars. While frame-based computer vision is mature due to the large amounts of data and ground truth available, event-based computer vision is still a work in progress as data sets are not as aplenty and ground truth is scarce for tasks such as object detection. In this paper, we suggest a way to overcome the lack of labeled ground truth by introducing a simple method to generate pseudo-labels for dynamic vision sensor data, assuming that the corresponding frame-based data is also available. These pseudo-labels can be treated as ground truth when training on supervised learning algorithms, and we show, for the first time, event-based car detection under ego-motion in a realistic environment at 100 frames per second with an average precision of 41.7 relative to pseudo-ground truth. | Pseudo-labeling was introduced by Lee @cite_13 for semi-supervised learning on frame-based data, where during each weight update, the unlabeled data picks up the class which has the maximum predicted probability and treats it as the ground truth. Chen al @cite_23 proposed a method to incrementally select reliable unlabeled data to give pseudo-labels to. Saito al @cite_0 proposed using three classifiers that regulate each other, to achieve domain adaptation from pseudo-labels. Pathak al @cite_8 used automatically generated masks (pseudo-labels in their context) from unsupervised motion segmentation on videos, and then trained a CNN to predict these masks from static images. The trained CNN learnt feature representations and was able to perform image classification, semantic segmentation and object detection. | {
"cite_N": [
"@cite_0",
"@cite_13",
"@cite_23",
"@cite_8"
],
"mid": [
"2949788122",
"",
"2621015177",
"2950341389"
],
"abstract": [
"Deep-layered models trained on a large number of labeled samples boost the accuracy of many tasks. It is important to apply such models to different domains because collecting many labeled samples in various domains is expensive. In unsupervised domain adaptation, one needs to train a classifier that works well on a target domain when provided with labeled source samples and unlabeled target samples. Although many methods aim to match the distributions of source and target samples, simply matching the distribution cannot ensure accuracy on the target domain. To learn discriminative representations for the target domain, we assume that artificially labeling target samples can result in a good representation. Tri-training leverages three classifiers equally to give pseudo-labels to unlabeled samples, but the method does not assume labeling samples generated from a different this http URL this paper, we propose an asymmetric tri-training method for unsupervised domain adaptation, where we assign pseudo-labels to unlabeled samples and train neural networks as if they are true labels. In our work, we use three networks asymmetrically. By asymmetric, we mean that two networks are used to label unlabeled target samples and one network is trained by the samples to obtain target-discriminative representations. We evaluate our method on digit recognition and sentiment analysis datasets. Our proposed method achieves state-of-the-art performance on the benchmark digit recognition datasets of domain adaptation.",
"",
"Aiming at improving the performance of visual classification in a cost-effective manner, this paper proposes an incremental semi-supervised learning paradigm called deep co-space (DCS). Unlike many conventional semi-supervised learning methods usually performed within a fixed feature space, our DCS gradually propagates information from labeled samples to unlabeled ones along with deep feature learning. We regard deep feature learning as a series of steps pursuing feature transformation, i.e., projecting the samples from a previous space into a new one, which tends to select the reliable unlabeled samples with respect to this setting. Specifically, for each unlabeled image instance, we measure its reliability by calculating the category variations of feature transformation from two different neighborhood variation perspectives and merged them into a unified sample mining criterion deriving from Hellinger distance. Then, those samples keeping stable correlation to their neighboring samples (i.e., having small category variation in distribution) across the successive feature space transformation are automatically received labels and incorporated into the model for incrementally training in terms of classification. Our extensive experiments on standard image classification benchmarks (e.g., Caltech-256 and SUN-397) demonstrate that the proposed framework is capable of effectively mining from large-scale unlabeled images, which boosts image classification performance and achieves promising results compared with other semi-supervised learning methods.",
"This paper presents a novel yet intuitive approach to unsupervised feature learning. Inspired by the human visual system, we explore whether low-level motion-based grouping cues can be used to learn an effective visual representation. Specifically, we use unsupervised motion-based segmentation on videos to obtain segments, which we use as 'pseudo ground truth' to train a convolutional network to segment objects from a single frame. Given the extensive evidence that motion plays a key role in the development of the human visual system, we hope that this straightforward approach to unsupervised learning will be more effective than cleverly designed 'pretext' tasks studied in the literature. Indeed, our extensive experiments show that this is the case. When used for transfer learning on object detection, our representation significantly outperforms previous unsupervised approaches across multiple settings, especially when training data for the target task is scarce."
]
} |
1709.09323 | 2950346959 | In recent years, dynamic vision sensors (DVS), also known as event-based cameras or neuromorphic sensors, have seen increased use due to various advantages over traditional frame-based cameras. Its high temporal resolution overcomes motion blurring, its high dynamic range overcomes extreme illumination conditions and its low power consumption makes it suitable for platforms such as drones and self-driving cars. While frame-based computer vision is mature due to the large amounts of data and ground truth available, event-based computer vision is still a work in progress as data sets are not as aplenty and ground truth is scarce for tasks such as object detection. In this paper, we suggest a way to overcome the lack of labeled ground truth by introducing a simple method to generate pseudo-labels for dynamic vision sensor data, assuming that the corresponding frame-based data is also available. These pseudo-labels can be treated as ground truth when training on supervised learning algorithms, and we show, for the first time, event-based car detection under ego-motion in a realistic environment at 100 frames per second with an average precision of 41.7 relative to pseudo-ground truth. | For data sets with paired modalities (e.g. RGB-D data contains RGB data of a scene synchronized with depth data of the same scene), cross modal distillation @cite_11 is a scheme that transfers knowledge from one modality, which has a lot of labels, to another modality, which has very few labels. In @cite_11 , mid-level representations of a CNN trained on RGB images were used to supervise training for another CNN to perform object detection and segmentation on depth images. @cite_17 , the visual modality of videos was used to generate pseudo-labels from CNNs and used to train a separate 1-D CNN to classify scenes from sound inputs. Our work is inspired by these cross modal methods, and we leverage on the fact that the DDD17 is a large data set with synchronized DVS and APS modalities. | {
"cite_N": [
"@cite_17",
"@cite_11"
],
"mid": [
"2544224704",
"2951874610"
],
"abstract": [
"We learn rich natural sound representations by capitalizing on large amounts of unlabeled sound data collected in the wild. We leverage the natural synchronization between vision and sound to learn an acoustic representation using two-million unlabeled videos. Unlabeled video has the advantage that it can be economically acquired at massive scales, yet contains useful signals about natural sound. We propose a student-teacher training procedure which transfers discriminative visual knowledge from well established visual recognition models into the sound modality using unlabeled video as a bridge. Our sound representation yields significant performance improvements over the state-of-the-art results on standard benchmarks for acoustic scene object classification. Visualizations suggest some high-level semantics automatically emerge in the sound network, even though it is trained without ground truth labels.",
"In this work we propose a technique that transfers supervision between images from different modalities. We use learned representations from a large labeled modality as a supervisory signal for training representations for a new unlabeled paired modality. Our method enables learning of rich representations for unlabeled modalities and can be used as a pre-training procedure for new modalities with limited labeled data. We show experimental results where we transfer supervision from labeled RGB images to unlabeled depth and optical flow images and demonstrate large improvements for both these cross modal supervision transfers. Code, data and pre-trained models are available at this https URL"
]
} |
1709.09323 | 2950346959 | In recent years, dynamic vision sensors (DVS), also known as event-based cameras or neuromorphic sensors, have seen increased use due to various advantages over traditional frame-based cameras. Its high temporal resolution overcomes motion blurring, its high dynamic range overcomes extreme illumination conditions and its low power consumption makes it suitable for platforms such as drones and self-driving cars. While frame-based computer vision is mature due to the large amounts of data and ground truth available, event-based computer vision is still a work in progress as data sets are not as aplenty and ground truth is scarce for tasks such as object detection. In this paper, we suggest a way to overcome the lack of labeled ground truth by introducing a simple method to generate pseudo-labels for dynamic vision sensor data, assuming that the corresponding frame-based data is also available. These pseudo-labels can be treated as ground truth when training on supervised learning algorithms, and we show, for the first time, event-based car detection under ego-motion in a realistic environment at 100 frames per second with an average precision of 41.7 relative to pseudo-ground truth. | Object detection on dynamic vision sensor data is relatively new since labeled event-based data sets are scarce. Liu al @cite_21 performed object detection on the predator-prey data set @cite_16 . They used dynamic vision sensor data as an attention mechanism for a frame-based CNN, and compared it to using a CNN to perform detection on the entire grayscale image. Including particle filter for both methods to aid tracking, the former method is 70X faster than the latter, with an accuracy of 90 | {
"cite_N": [
"@cite_21",
"@cite_16"
],
"mid": [
"2512653618",
"2473225499"
],
"abstract": [
"This paper reports an object tracking algorithm for a moving platform using the dynamic and active-pixel vision sensor (DAVIS). It takes advantage of both the active pixel sensor (APS) frame and dynamic vision sensor (DVS) event outputs from the DAVIS. The tracking is performed in a three step-manner: regions of interest (ROIs) are generated by a cluster-based tracking using the DVS output, likely target locations are detected by using a convolutional neural network (CNN) on the APS output to classify the ROIs as foreground and background, and finally a particle filter infers the target location from the ROIs. Doing convolution only in the ROIs boosts the speed by a factor of 70 compared with full-frame convolutions for the 240×180 frame input from the DAVIS. The tracking accuracy on a predator and prey robot database reaches 90 with a cost of less than 20ms frame in Matlab on a normal PC without using a GPU.",
"This paper describes the application of a Convolutional Neural Network (CNN) in the context of a predator prey scenario. The CNN is trained and run on data from a Dynamic and Active Pixel Sensor (DAVIS) mounted on a Summit XL robot (the predator), which follows another one (the prey). The CNN is driven by both conventional image frames and dynamic vision sensor \"frames\" that consist of a constant number of DAVIS ON and OFF events. The network is thus \"data driven\" at a sample rate proportional to the scene activity, so the effective sample rate varies from 15 Hz to 240 Hz depending on the robot speeds. The network generates four outputs: steer right, left, center and non-visible. After off-line training on labeled data, the network is imported on the on-board Summit XL robot which runs jAER and receives steering directions in real time. Successful results on closed-loop trials, with accuracies up to 87 or 92 (depending on evaluation criteria) are reported. Although the proposed approach discards the precise DAVIS event timing, it offers the significant advantage of compatibility with conventional deep learning technology without giving up the advantage of data-driven computing."
]
} |
1709.09254 | 2756725603 | We describe a data-driven approach for automatically explaining new, non-standard English expressions in a given sentence, building on a large dataset that includes 15 years of crowdsourced examples from UrbanDictionary.com. Unlike prior studies that focus on matching keywords from a slang dictionary, we investigate the possibility of learning a neural sequence-to-sequence model that generates explanations of unseen non-standard English expressions given context. We propose a dual encoder approach---a word-level encoder learns the representation of context, and a second character-level encoder to learn the hidden representation of the target non-standard expression. Our model can produce reasonable definitions of new non-standard English expressions given their context with certain confidence. | The study of non-standard language is of interests to many researchers in the social media and NLP communities. For example, propose a latent variable model to study the lexical variation of the language on Twitter, where many regional slang words are discovered. Zappavigna identifies the Internet slang as an important component in the discourse of Twitter and social media. provide a contextual analysis of how social media users shorten their messages. Notably, a study on Tweet normalization @cite_11 finds that, even when using a small slang dictionary of 5K words, Slang makes up 12 the NLP community, slang dictionary is widely used in many tasks and applications @cite_10 @cite_2 @cite_4 . However, we argue that using a small, fixed-size dictionary approach may be suboptimal: it suffers from the low coverage problem, and to keep the dictionary up to date, maintaining such dictionary is also expensive and time-consuming. To the best of our knowledge, our work is the first to build a general purpose machine learning model for explaining non-standard English terms, using a large crowdsourced dataset. | {
"cite_N": [
"@cite_2",
"@cite_10",
"@cite_4",
"@cite_11"
],
"mid": [
"",
"2132210327",
"2152460337",
"2146867136"
],
"abstract": [
"",
"We introduce the novel task of determining whether a newswire article is \"true\" or satirical. We experiment with SVMs, feature scaling, and a number of lexical and semantic feature types, and achieve promising results over the task.",
"We investigate whether wording, stylistic choices, and online behavior can be used to predict the age category of blog authors. Our hypothesis is that significant changes in writing style distinguish pre-social media bloggers from post-social media bloggers. Through experimentation with a range of years, we found that the birth dates of students in college at the time when social media such as AIM, SMS text messaging, MySpace and Facebook first became popular, enable accurate age prediction. We also show that internet writing characteristics are important features for age prediction, but that lexical content is also needed to produce significantly more accurate results. Our best results allow for 81.57 accuracy.",
"Twitter provides access to large volumes of data in real time, but is notoriously noisy, hampering its utility for NLP. In this paper, we target out-of-vocabulary words in short text messages and propose a method for identifying and normalising ill-formed words. Our method uses a classifier to detect ill-formed words, and generates correction candidates based on morphophonemic similarity. Both word similarity and context are then exploited to select the most probable correction candidate for the word. The proposed method doesn't require any annotations, and achieves state-of-the-art performance over an SMS corpus and a novel dataset based on Twitter."
]
} |
1709.09283 | 2963968420 | In recent years, various shadow detection methods from a single image have been proposed and used in vision systems; however, most of them are not appropriate for the robotic applications due to the expensive time complexity. This paper introduces a fast shadow detection method using a deep learning framework, with a time cost that is appropriate for robotic applications. In our solution, we first obtain a shadow prior map with the help of multi-class support vector machine using statistical features. Then, we use a semantic-aware patch-level Convolutional Neural Network that efficiently trains on shadow examples by combining the original image and the shadow prior map. Experiments on benchmark datasets demonstrate the proposed method significantly decreases the time complexity of shadow detection, by one or two orders of magnitude compared with state-of-the-art methods, without losing accuracy. | In this paper, we propose a novel method based on deep learning with a shadow prior. Like @cite_28 , our method can detect shadow from a single image, but at a much lower time complexity than all existing methods based on deep learning. The key insight of our method is that it performs shadow detection on a per super-pixel basis. Initial result from such a detection method would produce boundary effects between super-pixels. We overcome such artifacts with a post-processing step using edge refinement. | {
"cite_N": [
"@cite_28"
],
"mid": [
"1906669240"
],
"abstract": [
"Local structures of shadow boundaries as well as complex interactions of image regions remain largely unexploited by previous shadow detection approaches. In this paper, we present a novel learning-based framework for shadow region recovery from a single image. We exploit local structures of shadow edges by using a structured CNN learning framework. We show that using structured label information in classification can improve local consistency over pixel labels and avoid spurious labelling. We further propose and formulate shadow bright measure to model complex interactions among image regions. The shadow and bright measures of each patch are computed from the shadow edges detected by the proposed CNN. Using the global interaction constraints on patches, we formulate a least-square optimization problem for shadow recovery that can be solved efficiently. Our shadow recovery method achieves state-of-the-art results on major shadow benchmark databases collected under various conditions."
]
} |
1709.09106 | 2756627069 | Region-based image retrieval (RBIR) technique is revisited. In early attempts at RBIR in the late 90s, researchers found many ways to specify region-based queries and spatial relationships; however, the way to characterize the regions, such as by using color histograms, were very poor at that time. Here, we revisit RBIR by incorporating semantic specification of objects and intuitive specification of spatial relationships. Our contributions are the following. First, to support multiple aspects of semantic object specification (category, instance, and attribute), we propose a multitask CNN feature that allows us to use deep learning technique and to jointly handle multi-aspect object specification. Second, to help users specify spatial relationships among objects in an intuitive way, we propose recommendation techniques of spatial relationships. In particular, by mining the search results, a system can recommend feasible spatial relationships among the objects. The system also can recommend likely spatial relationships by assigned object category names based on language prior. Moreover, object-level inverted indexing supports very fast shortlist generation, and re-ranking based on spatial constraints provides users with instant RBIR experiences. | Region-based image retrieval. RBIR was proposed as an extension of global feature-based CBIR that had received much attention in the late 90s and early 00s @cite_23 @cite_34 . Unlike global feature-based approaches, RBIR extracts features from multiple sub-regions, which correspond to objects in the image. RBIR provided the way to focus on the object details, which have much benefit such as that 1) the accuracy of queried object was improved by using region-based matching, and 2) spatial localization of queried object is easy. | {
"cite_N": [
"@cite_34",
"@cite_23"
],
"mid": [
"2109868644",
"1524408959"
],
"abstract": [
"We present here SIMPLIcity (semantics-sensitive integrated matching for picture libraries), an image retrieval system, which uses semantics classification methods, a wavelet-based approach for feature extraction, and integrated region matching based upon image segmentation. An image is represented by a set of regions, roughly corresponding to objects, which are characterized by color, texture, shape, and location. The system classifies images into semantic categories. Potentially, the categorization enhances retrieval by permitting semantically-adaptive searching methods and narrowing down the searching range in a database. A measure for the overall similarity between images is developed using a region-matching scheme that integrates properties of all the regions in the images. The application of SIMPLIcity to several databases has demonstrated that our system performs significantly better and faster than existing ones. The system is fairly robust to image alterations.",
"Blobworld is a system for image retrieval based on finding coherent image regions which roughly correspond to objects. Each image is automatically segmented into regions (\"blobs\") with associated color and texture descriptors. Queryingi s based on the attributes of one or two regions of interest, rather than a description of the entire image. In order to make large-scale retrieval feasible, we index the blob descriptions usinga tree. Because indexing in the high-dimensional feature space is computationally prohibitive, we use a lower-rank approximation to the high-dimensional distance. Experiments show encouraging results for both queryinga nd indexing."
]
} |
1709.09106 | 2756627069 | Region-based image retrieval (RBIR) technique is revisited. In early attempts at RBIR in the late 90s, researchers found many ways to specify region-based queries and spatial relationships; however, the way to characterize the regions, such as by using color histograms, were very poor at that time. Here, we revisit RBIR by incorporating semantic specification of objects and intuitive specification of spatial relationships. Our contributions are the following. First, to support multiple aspects of semantic object specification (category, instance, and attribute), we propose a multitask CNN feature that allows us to use deep learning technique and to jointly handle multi-aspect object specification. Second, to help users specify spatial relationships among objects in an intuitive way, we propose recommendation techniques of spatial relationships. In particular, by mining the search results, a system can recommend feasible spatial relationships among the objects. The system also can recommend likely spatial relationships by assigned object category names based on language prior. Moreover, object-level inverted indexing supports very fast shortlist generation, and re-ranking based on spatial constraints provides users with instant RBIR experiences. | Spatial query and relationship. Several early CBIR studies @cite_9 @cite_21 investigated interactive specifications of spatial queries by using sketches, etc. VisualSEEk @cite_9 searched for images by using spatial queries such as those indicating the absolute and relative locations of objects, which were specified by diagramming spatial arrangements of regions. VideoQ @cite_21 used sketch-based interactive queries to specify spatiotemporal constraints on objects, such as motion in content-based video retrieval. More recent studies @cite_36 @cite_3 @cite_40 can handle the query composed of semantic concepts (i.e., object categories) and their spatial layout. Other studies @cite_28 @cite_18 demonstrated the effectiveness of spatial position-based interactive filtering for video search in the competitions @cite_30 . Although interactive systems make it easy for the user to specify and refine spatial queries, they have trouble specifying complex relationships among objects. | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_36",
"@cite_28",
"@cite_9",
"@cite_21",
"@cite_3",
"@cite_40"
],
"mid": [
"2063729789",
"1020536998",
"",
"",
"2091503252",
"1981723314",
"2003635569",
"2747635822"
],
"abstract": [
"The Video Browser Showdown is an international competition in the field of interactive video search and retrieval. It is held annually as a special session at the International Conference on Multimedia Modeling (MMM). The Video Browser Showdown evaluates the performance of exploratory tools for interactive content search in videos in direct competition and in front of an audience. Its goal is to push research on user-centric video search tools including video navigation, content browsing, content interaction, and video content visualization. This article summarizes the first three VBS competitions (2012-2014).",
"The success of our Signature-Based Video Browser presented last year at Video Browser Showdown 2014 (now renamed to Video Search Showcase) was mainly based on effective filtering using position-color feature signatures, while browsing in the results comprising matched keyframes was based just on a simple sequential search approach. Since the results can consist of highly similar keyframes (e.g., news studio scenes) making the browsing more difficult, we have enhanced our tool with more advanced browsing techniques considering also homogeneous result sets obtained after filtering phase. Furthermore, we have utilized improved search models based on feature signatures to make the filtering phase more effective.",
"",
"",
"",
"The rapidity with which digitat information, particularly video, is being generated, has necessitated the development of tools for efficient search of these media. Content based visual queries have been primarily focussed on still image retrieval. In this papel; we propose a novel, real-time, interactive system on the Web, based on the visual paradigm, with spatio-temporal attributesplaying a key role in video retrieval. We have developed algorithms for automated video object segmentation and tracking and use real-time video editing techniques while responding to user queries. The resulting system pe$orms well, with the user being able to retrieve complex video clips such as those of skiers, baseball players, with ease.",
"In this paper, we present a novel image search system, image search by concept map. This system enables users to indicate not only what semantic concepts are expected to appear but also how these concepts are spatially distributed in the desired images. To this end, we propose a new image search interface to enable users to formulate a query, called concept map, by intuitively typing textual queries in a blank canvas to indicate the desired spatial positions of the concepts. In the ranking process, by interpreting each textual concept as a set of representative visual instances, the concept map query is translated into a visual instance map, which is then used to evaluate the relevance of the image in the database. Experimental results demonstrate the effectiveness of the proposed system.",
"The performance of image retrieval has been improved tremendously in recent years through the use of deep feature representations. Most existing methods, however, aim to retrieve images that are visually similar or semantically relevant to the query, irrespective of spatial configuration. In this paper, we develop a spatial-semantic image search technology that enables users to search for images with both semantic and spatial constraints by manipulating concept text-boxes on a 2D query canvas. We train a convolutional neural network to synthesize appropriate visual features that captures the spatial-semantic constraints from the user canvas query. We directly optimize the retrieval performance of the visual features when training our deep neural network. These visual features then are used to retrieve images that are both spatially and semantically relevant to the user query. The experiments on large-scale datasets such as MS-COCO and Visual Genome show that our method outperforms other baseline and state-of-the-art methods in spatial-semantic image search."
]
} |
1709.09106 | 2756627069 | Region-based image retrieval (RBIR) technique is revisited. In early attempts at RBIR in the late 90s, researchers found many ways to specify region-based queries and spatial relationships; however, the way to characterize the regions, such as by using color histograms, were very poor at that time. Here, we revisit RBIR by incorporating semantic specification of objects and intuitive specification of spatial relationships. Our contributions are the following. First, to support multiple aspects of semantic object specification (category, instance, and attribute), we propose a multitask CNN feature that allows us to use deep learning technique and to jointly handle multi-aspect object specification. Second, to help users specify spatial relationships among objects in an intuitive way, we propose recommendation techniques of spatial relationships. In particular, by mining the search results, a system can recommend feasible spatial relationships among the objects. The system also can recommend likely spatial relationships by assigned object category names based on language prior. Moreover, object-level inverted indexing supports very fast shortlist generation, and re-ranking based on spatial constraints provides users with instant RBIR experiences. | Other approaches @cite_37 @cite_43 @cite_44 @cite_16 @cite_1 @cite_13 encode relationships between objects within a graph or a descriptor, where images with similar contexts are retrieved by matching them. The query is represented by graph in @cite_37 @cite_43 that includes object class information as well as spatial contexts among objects. @cite_47 encoded the spatial relationship into the descriptor; their work shows the effectiveness in describing complex spatial relationship between two objects. @cite_44 encode high-order contextual information among objects into the descriptor to measure the strength of interaction, where the objects with similar context can be retrieved by the value. While these approaches can handle complex relationships, query specification and refinement are generally difficult for the user and not suitable for interactive end-user applications. | {
"cite_N": [
"@cite_37",
"@cite_1",
"@cite_44",
"@cite_43",
"@cite_47",
"@cite_16",
"@cite_13"
],
"mid": [
"2077069816",
"5598288",
"",
"2953141364",
"2953050665",
"2140435402",
""
],
"abstract": [
"This paper develops a novel framework for semantic image retrieval based on the notion of a scene graph. Our scene graphs represent objects (“man”, “boat”), attributes of objects (“boat is white”) and relationships between objects (“man standing on boat”). We use these scene graphs as queries to retrieve semantically related images. To this end, we design a conditional random field model that reasons about possible groundings of scene graphs to test images. The likelihoods of these groundings are used as ranking scores for retrieval. We introduce a novel dataset of 5,000 human-generated scene graphs grounded to images and use this dataset to evaluate our method for image retrieval. In particular, we evaluate retrieval using full scene graphs and small scene subgraphs, and show that our method outperforms retrieval methods that use only objects or low-level image features. In addition, we show that our full model can be used to improve object localization compared to baseline methods.",
"We consider image retrieval with structured object queries --- queries that specify the objects that should be present in the scene, and their spatial relations. An example of such queries is \"car on the road\". Existing image retrieval systems typically consider queries consisting of object classes (i.e. keywords). They train a separate classifier for each object class and combine the output heuristically. In contrast, we develop a learning framework to jointly consider object classes and their relations. Our method considers not only the objects in the query (\"car\" and \"road\" in the above example), but also related object categories can be useful for retrieval. Since we do not have ground-truth labeling of object bounding boxes on the test image, we represent them as latent variables in our model. Our learning method is an extension of the ranking SVM with latent variables, which we call latent ranking SVM. We demonstrate image retrieval and ranking results on a dataset with more than a hundred of object classes.",
"",
"We propose a novel image representation, termed Attribute-Graph, to rank images by their semantic similarity to a given query image. An Attribute-Graph is an undirected fully connected graph, incorporating both local and global image characteristics. The graph nodes characterise objects as well as the overall scene context using mid-level semantic attributes, while the edges capture the object topology. We demonstrate the effectiveness of Attribute-Graphs by applying them to the problem of image ranking. We benchmark the performance of our algorithm on the 'rPascal' and 'rImageNet' datasets, which we have created in order to evaluate the ranking performance on complex queries containing multiple objects. Our experimental evaluation shows that modelling images as Attribute-Graphs results in improved ranking performance over existing techniques.",
"As humans, we regularly interpret images based on the relations between image regions. For example, a person riding object X, or a plank bridging two objects. Current methods provide limited support to search for images based on such relations. We present RAID, a relation-augmented image descriptor that supports queries based on inter-region relations. The key idea of our descriptor is to capture the spatial distribution of simple point-to-region relationships to describe more complex relationships between two image regions. We evaluate the proposed descriptor by querying into a large subset of the Microsoft COCO database and successfully extract nontrivial images demonstrating complex inter-region relations, which are easily missed or erroneously classified by existing methods.",
"The use of context is critical for scene understanding in computer vision, where the recognition of an object is driven by both local appearance and the object's relationship to other elements of the scene (context). Most current approaches rely on modeling the relationships between object categories as a source of context. In this paper we seek to move beyond categories to provide a richer appearance-based model of context. We present an exemplar-based model of objects and their relationships, the Visual Memex, that encodes both local appearance and 2D spatial context between object instances. We evaluate our model on Torralba's proposed Context Challenge against a baseline category-based system. Our experiments suggest that moving beyond categories for context modeling appears to be quite beneficial, and may be the critical missing ingredient in scene understanding systems.",
""
]
} |
1709.09304 | 2757131217 | Multi-index fusion has demonstrated impressive performances in retrieval task by integrating different visual representations in a unified framework. However, previous works mainly consider propagating similarities via neighbor structure, ignoring the high order information among different visual representations. In this paper, we propose a new multi-index fusion scheme for image retrieval. By formulating this procedure as a multilinear based optimization problem, the complementary information hidden in different indexes can be explored more thoroughly. Specially, we first build our multiple indexes from various visual representations. Then a so-called index-specific functional matrix, which aims to propagate similarities, is introduced for updating the original index. The functional matrices are then optimized in a unified tensor space to achieve a refinement, such that the relevant images can be pushed more closer. The optimization problem can be efficiently solved by the augmented Lagrangian method with theoretical convergence guarantee. Unlike the traditional multi-index fusion scheme, our approach embeds the multi-index subspace structure into the new indexes with sparse constraint, thus it has little additional memory consumption in online query stage. Experimental evaluation on three benchmark datasets reveals that the proposed approach achieves the state-of-the-art performance, i.e., N-score 3.94 on UKBench, mAP 94.1 on Holiday and 62.39 on Market-1501. | Meanwhile, a lot of works aim to produce the compact image representation @cite_14 @cite_36 @cite_24 @cite_51 , which is benefit for computational efficiency and memory cost. Furthermore, several recent proposed methods attempt to extract features from the pre-trained deep convolutional networks via compact encoding. By using the compact codes, Babenko discover that the features from the fully-connected layers of CNN (fully-connected feature) provide high-level descriptors of the visual content @cite_40 , yielding competitive results. But more recently, the research attention has moved to the activations of CNN filters (convolutional feature) @cite_38 . Convolutional features have a natural interpretation as descriptors of local image regions, which not only share the same benefits with the local features, but also hold high-level semantic information @cite_29 . Empirically, they gain even better results than the local features. Generally speaking, both of these methods hold distinct merits, resulting in different retrieval results. This may cause us to consider whether we should only focus on one type visual feature ( e.g., abandon these hand-crafted features), or combine different visual representations for retrieval. | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_36",
"@cite_29",
"@cite_24",
"@cite_40",
"@cite_51"
],
"mid": [
"2949266290",
"2045143396",
"2390601522",
"2204975001",
"1984309565",
"",
"2147238549"
],
"abstract": [
"Deep convolutional neural networks have been successfully applied to image classification tasks. When these same networks have been applied to image retrieval, the assumption has been made that the last layers would give the best performance, as they do in classification. We show that for instance-level image retrieval, lower layers often perform better than the last layers in convolutional neural networks. We present an approach for extracting convolutional features from different layers of the networks, and adopt VLAD encoding to encode features into a single vector for each image. We investigate the effect of different layers and scales of input images on the performance of convolutional features using the recent deep networks OxfordNet and GoogLeNet. Experiments demonstrate that intermediate layers or higher layers with finer scales produce better results for image retrieval, compared to the last layer. When using compressed 128-D VLAD descriptors, our method obtains state-of-the-art results and outperforms other VLAD and CNN based approaches on two out of three test datasets. Our work provides guidance for transferring deep networks trained on image classification to image retrieval tasks.",
"We consider the design of a single vector representation for an image that embeds and aggregates a set of local patch descriptors such as SIFT. More specifically we aim to construct a dense representation, like the Fisher Vector or VLAD, though of small or intermediate size. We make two contributions, both aimed at regularizing the individual contributions of the local descriptors in the final representation. The first is a novel embedding method that avoids the dependency on absolute distances by encoding directions. The second contribution is a \"democratization\" strategy that further limits the interaction of unrelated descriptors in the aggregation stage. These methods are complementary and give a substantial performance boost over the state of the art in image search with short or mid-size vectors, as demonstrated by our experiments on standard public image retrieval benchmarks.",
"Content-based image retrieval is an important research topic in the multimedia field. In large-scale image search using local features, image features are encoded and aggregated into a compact vector to avoid indexing each feature individually. In the aggregation step, sum-aggregation is wildly used in many existing works and demonstrates promising performance. However, it is based on a strong and implicit assumption that the local descriptors of an image are identically and independently distributed in descriptor space and image plane. To address this problem, we propose a new aggregation method named democratic diffusion aggregation (DDA) with weak spatial context embedded. The main idea of our aggregation method is to re-weight the embedded vectors before sum-aggregation by considering the relevance among local descriptors. Different from previous work, by conducting a diffusion process on the improved kernel matrix, we calculate the weighting coefficients more efficiently without any iterative optimization. Besides considering the relevance of local descriptors from different images, we also discuss an efficient query fusion strategy which uses the initial top-ranked image vectors to enhance the retrieval performance. Experimental results show that our aggregation method exhibits much higher efficiency (about @math faster) and better retrieval accuracy compared with previous methods, and the query fusion strategy consistently improves the retrieval quality.",
"Several recent works have shown that image descriptors produced by deep convolutional neural networks provide state-of-the-art performance for image classification and retrieval problems. It also has been shown that the activations from the convolutional layers can be interpreted as local features describing particular image regions. These local features can be aggregated using aggregating methods developed for local features (e.g. Fisher vectors), thus providing new powerful global descriptor. In this paper we investigate possible ways to aggregate local deep features to produce compact descriptors for image retrieval. First, we show that deep features and traditional hand-engineered features have quite different distributions of pairwise similarities, hence existing aggregation methods have to be carefully re-evaluated. Such re-evaluation reveals that in contrast to shallow features, the simple aggregation method based on sum pooling provides the best performance for deep convolutional features. This method is efficient, has few parameters, and bears little risk of overfitting when e.g. learning the PCA matrix. In addition, we suggest a simple yet efficient query expansion scheme suitable for the proposed aggregation method. Overall, the new compact global descriptor improves the state-of-the-art on four common benchmarks considerably.",
"This paper addresses the problem of large-scale image search. Three constraints have to be taken into account: search accuracy, efficiency, and memory usage. We first present and evaluate different ways of aggregating local image descriptors into a vector and show that the Fisher kernel achieves better performance than the reference bag-of-visual words approach for any given vector dimension. We then jointly optimize dimensionality reduction and indexing in order to obtain a precise vector comparison as well as a compact representation. The evaluation shows that the image representation can be reduced to a few dozen bytes while preserving high accuracy. Searching a 100 million image data set takes about 250 ms on one processor core.",
"",
"Within the field of pattern classification, the Fisher kernel is a powerful framework which combines the strengths of generative and discriminative approaches. The idea is to characterize a signal with a gradient vector derived from a generative probability model and to subsequently feed this representation to a discriminative classifier. We propose to apply this framework to image categorization where the input signals are images and where the underlying generative model is a visual vocabulary: a Gaussian mixture model which approximates the distribution of low-level features in images. We show that Fisher kernels can actually be understood as an extension of the popular bag-of-visterms. Our approach demonstrates excellent performance on two challenging databases: an in-house database of 19 object scene categories and the recently released VOC 2006 database. It is also very practical: it has low computational needs both at training and test time and vocabularies trained on one set of categories can be applied to another set without any significant loss in performance."
]
} |
1709.09304 | 2757131217 | Multi-index fusion has demonstrated impressive performances in retrieval task by integrating different visual representations in a unified framework. However, previous works mainly consider propagating similarities via neighbor structure, ignoring the high order information among different visual representations. In this paper, we propose a new multi-index fusion scheme for image retrieval. By formulating this procedure as a multilinear based optimization problem, the complementary information hidden in different indexes can be explored more thoroughly. Specially, we first build our multiple indexes from various visual representations. Then a so-called index-specific functional matrix, which aims to propagate similarities, is introduced for updating the original index. The functional matrices are then optimized in a unified tensor space to achieve a refinement, such that the relevant images can be pushed more closer. The optimization problem can be efficiently solved by the augmented Lagrangian method with theoretical convergence guarantee. Unlike the traditional multi-index fusion scheme, our approach embeds the multi-index subspace structure into the new indexes with sparse constraint, thus it has little additional memory consumption in online query stage. Experimental evaluation on three benchmark datasets reveals that the proposed approach achieves the state-of-the-art performance, i.e., N-score 3.94 on UKBench, mAP 94.1 on Holiday and 62.39 on Market-1501. | To overcome these drawbacks, some works focus on fusing visual features on index level. A common assumption shared in these methods is that: two images, which are nearest neighbors to each other under one type of visual representation, are probably to be true related. By pushing them closer in other visual feature spaces, the search accuracy can be greatly promoted. Under the guidance of this principle, the proposed collaborative index embedding method @cite_6 , which is most relevant to our work, utilize an alternating index update scheme to fuse feature. By enriching the corresponding feature, it refine the neighborhood structures to improve the retrieval accuracy. Chen @cite_32 extend this model for the multi-index fusion problem. However, both of these methods neglect the distance information of original feature space. More importantly, high order information is more or less ignored. | {
"cite_N": [
"@cite_32",
"@cite_6"
],
"mid": [
"2741640202",
"2591614014"
],
"abstract": [
"Different kinds of features hold some distinct merits, making them complementary to each other. Inspired by this idea an index level multiple feature fusion scheme via similarity matrix pooling is proposed in this paper. We first compute the similarity matrix of each index, and then a novel scheme is used to pool on these similarity matrices for updating the original indices. Compared with the existing fusion schemes, the proposed scheme performs feature fusion at index level to save memory and reduce computational complexity. On the other hand, the proposed scheme treats different kinds of features adaptively based on its importance, thus improves retrieval accuracy. The performance of the proposed approach is evaluated using two public datasets, which significantly outperforms the baseline methods in retrieval accuracy with low memory consumption and computational complexity.",
"In content-based image retrieval, SIFT feature and the feature from deep convolutional neural network (CNN) have demonstrated promising performance. To fully explore both visual features in a unified framework for effective and efficient retrieval, we propose a collaborative index embedding method to implicitly integrate the index matrices of them. We formulate the index embedding as an optimization problem from the perspective of neighborhood sharing and solve it with an alternating index update scheme. After the iterative embedding, only the embedded CNN index is kept for on-line query, which demonstrates significant gain in retrieval accuracy, with very economical memory cost. Extensive experiments have been conducted on the public datasets with million-scale distractor images. The experimental results reveal that, compared with the recent state-of-the-art retrieval algorithms, our approach achieves competitive accuracy performance with less memory overhead and efficient query computation."
]
} |
1709.09304 | 2757131217 | Multi-index fusion has demonstrated impressive performances in retrieval task by integrating different visual representations in a unified framework. However, previous works mainly consider propagating similarities via neighbor structure, ignoring the high order information among different visual representations. In this paper, we propose a new multi-index fusion scheme for image retrieval. By formulating this procedure as a multilinear based optimization problem, the complementary information hidden in different indexes can be explored more thoroughly. Specially, we first build our multiple indexes from various visual representations. Then a so-called index-specific functional matrix, which aims to propagate similarities, is introduced for updating the original index. The functional matrices are then optimized in a unified tensor space to achieve a refinement, such that the relevant images can be pushed more closer. The optimization problem can be efficiently solved by the augmented Lagrangian method with theoretical convergence guarantee. Unlike the traditional multi-index fusion scheme, our approach embeds the multi-index subspace structure into the new indexes with sparse constraint, thus it has little additional memory consumption in online query stage. Experimental evaluation on three benchmark datasets reveals that the proposed approach achieves the state-of-the-art performance, i.e., N-score 3.94 on UKBench, mAP 94.1 on Holiday and 62.39 on Market-1501. | Our work is also related with the multi-view subspace learning methods, especially the subspace clustering methods. Sparse subspace clustering @cite_26 and low-rank representation @cite_47 are most popular subspace clustering methods, which explore the relationships between samples via self-representation matrix. Zhang @cite_28 extend the low-rank representation to the multi-view setting via imposing a unfolding high-order norm to the subspace coefficient tensor. However, this tensor constraint can not explore the complementary information thoroughly, due to the fact that the low-rank norm penalize each view equally. By using a new tensor construction method, Xie @cite_10 replace the unfolding tensor norm with a recently proposed t-SVD based norm @cite_35 @cite_52 , which is based on a new tensor computational framework @cite_27 . This framework provides a closed multiplication operation between tensors @cite_19 , where the familiar tools used in matrix case can be directly extended to tensor case. Hence, it has good theoretical properties for handling complicated relationship among different views. For more detail information, we refer readers to read the section . | {
"cite_N": [
"@cite_35",
"@cite_26",
"@cite_28",
"@cite_52",
"@cite_19",
"@cite_27",
"@cite_47",
"@cite_10"
],
"mid": [
"2953204310",
"1993962865",
"2197707282",
"2080843093",
"",
"1992426838",
"1997201895",
"2949447882"
],
"abstract": [
"In this paper we propose novel methods for completion (from limited samples) and de-noising of multilinear (tensor) data and as an application consider 3-D and 4- D (color) video data completion and de-noising. We exploit the recently proposed tensor-Singular Value Decomposition (t-SVD)[11]. Based on t-SVD, the notion of multilinear rank and a related tensor nuclear norm was proposed in [11] to characterize informational and structural complexity of multilinear data. We first show that videos with linear camera motion can be represented more efficiently using t-SVD compared to the approaches based on vectorizing or flattening of the tensors. Since efficiency in representation implies efficiency in recovery, we outline a tensor nuclear norm penalized algorithm for video completion from missing entries. Application of the proposed algorithm for video recovery from missing entries is shown to yield a superior performance over existing methods. We also consider the problem of tensor robust Principal Component Analysis (PCA) for de-noising 3-D video data from sparse random corruptions. We show superior performance of our method compared to the matrix robust PCA adapted to this setting as proposed in [4].",
"Many real-world problems deal with collections of high-dimensional data, such as images, videos, text, and web documents, DNA microarray data, and more. Often, such high-dimensional data lie close to low-dimensional structures corresponding to several classes or categories to which the data belong. In this paper, we propose and study an algorithm, called sparse subspace clustering, to cluster data points that lie in a union of low-dimensional subspaces. The key idea is that, among the infinitely many possible representations of a data point in terms of other points, a sparse representation corresponds to selecting a few points from the same subspace. This motivates solving a sparse optimization program whose solution is used in a spectral clustering framework to infer the clustering of the data into subspaces. Since solving the sparse optimization program is in general NP-hard, we consider a convex relaxation and show that, under appropriate conditions on the arrangement of the subspaces and the distribution of the data, the proposed minimization program succeeds in recovering the desired sparse representations. The proposed algorithm is efficient and can handle data points near the intersections of subspaces. Another key advantage of the proposed algorithm with respect to the state of the art is that it can deal directly with data nuisances, such as noise, sparse outlying entries, and missing entries, by incorporating the model of the data into the sparse optimization program. We demonstrate the effectiveness of the proposed algorithm through experiments on synthetic data as well as the two real-world problems of motion segmentation and face clustering.",
"In this paper, we explore the problem of multiview subspace clustering. We introduce a low-rank tensor constraint to explore the complementary information from multiple views and, accordingly, establish a novel method called Low-rank Tensor constrained Multiview Subspace Clustering (LT-MSC). Our method regards the subspace representation matrices of different views as a tensor, which captures dexterously the high order correlations underlying multiview data. Then the tensor is equipped with a low-rank constraint, which models elegantly the cross information among different views, reduces effectually the redundancy of the learned subspace representations, and improves the accuracy of clustering as well. The inference process of the affinity matrix for clustering is formulated as a tensor nuclear norm minimization problem, constrained with an additional L2,1-norm regularizer and some linear equalities. The minimization problem is convex and thus can be solved efficiently by an Augmented Lagrangian Alternating Direction Minimization (AL-ADM) method. Extensive experimental results on four benchmark datasets show the effectiveness of our proposed LT-MSC method.",
"The development of energy selective, photon counting X-ray detectors allows for a wide range of new possibilities in the area of computed tomographic image formation. Under the assumption of perfect energy resolution, here we propose a tensor-based iterative algorithm that simultaneously reconstructs the X-ray attenuation distribution for each energy. We use a multilinear image model rather than a more standard stacked vector representation in order to develop novel tensor-based regularizers. In particular, we model the multispectral unknown as a three-way tensor where the first two dimensions are space and the third dimension is energy. This approach allows for the design of tensor nuclear norm regularizers, which like its 2D counterpart, is a convex function of the multispectral unknown. The solution to the resulting convex optimization problem is obtained using an alternating direction method of multipliers approach. Simulation results show that the generalized tensor nuclear norm can be used as a standalone regularization technique for the energy selective (spectral) computed tomography problem and when combined with total variation regularization it enhances the regularization capabilities especially at low energy images where the effects of noise are most prominent.",
"",
"Recent work by Kilmer and Martin [Linear Algebra Appl., 435 (2011), pp. 641--658] and Braman [Linear Algebra Appl., 433 (2010), pp. 1241--1253] provides a setting in which the familiar tools of linear algebra can be extended to better understand third-order tensors. Continuing along this vein, this paper investigates further implications including (1) a bilinear operator on the matrices which is nearly an inner product and which leads to definitions for length of matrices, angle between two matrices, and orthogonality of matrices, and (2) the use of t-linear combinations to characterize the range and kernel of a mapping defined by a third-order tensor and the t-product and the quantification of the dimensions of those sets. These theoretical results lead to the study of orthogonal projections as well as an effective Gram--Schmidt process for producing an orthogonal basis of matrices. The theoretical framework also leads us to consider the notion of tensor polynomials and their relation to tensor eigentupl...",
"In this paper, we address the subspace clustering problem. Given a set of data samples (vectors) approximately drawn from a union of multiple subspaces, our goal is to cluster the samples into their respective subspaces and remove possible outliers as well. To this end, we propose a novel objective function named Low-Rank Representation (LRR), which seeks the lowest rank representation among all the candidates that can represent the data samples as linear combinations of the bases in a given dictionary. It is shown that the convex program associated with LRR solves the subspace clustering problem in the following sense: When the data is clean, we prove that LRR exactly recovers the true subspace structures; when the data are contaminated by outliers, we prove that under certain conditions LRR can exactly recover the row space of the original data and detect the outlier as well; for data corrupted by arbitrary sparse errors, LRR can also approximately recover the row space with theoretical guarantees. Since the subspace membership is provably determined by the row space, these further imply that LRR can perform robust subspace clustering and error correction in an efficient and effective way.",
"In this paper, we address the multi-view subspace clustering problem. Our method utilizes the circulant algebra for tensor, which is constructed by stacking the subspace representation matrices of different views and then rotating, to capture the low rank tensor subspace so that the refinement of the view-specific subspaces can be achieved, as well as the high order correlations underlying multi-view data can be explored. By introducing a recently proposed tensor factorization, namely tensor-Singular Value Decomposition (t-SVD) kilmer13 , we can impose a new type of low-rank tensor constraint on the rotated tensor to capture the complementary information from multiple views. Different from traditional unfolding based tensor norm, this low-rank tensor constraint has optimality properties similar to that of matrix rank derived from SVD, so the complementary information among views can be explored more efficiently and thoroughly. The established model, called t-SVD based Multi-view Subspace Clustering (t-SVD-MSC), falls into the applicable scope of augmented Lagrangian method, and its minimization problem can be efficiently solved with theoretical convergence guarantee and relatively low computational complexity. Extensive experimental testing on eight challenging image dataset shows that the proposed method has achieved highly competent objective performance compared to several state-of-the-art multi-view clustering methods."
]
} |
1709.08915 | 2758592735 | We consider the fundamental problem of inferring the causal direction between two univariate numeric random variables X and Y from observational data. The two-variable case is especially difficult to solve since it is not possible to use standard conditional independence tests between the variables. To tackle this problem, we follow an information theoretic approach based on Kolmogorov complexity and use the Minimum Description Length (MDL) principle to provide a practical solution. In particular, we propose a compression scheme to encode local and global functional relations using MDL-based regression. We infer X causes Y in case it is shorter to describe Y as a function of X than the inverse direction. In addition, we introduce Slope, an efficient linear-time algorithm that through thorough empirical evaluation on both synthetic and real world data we show outperforms the state of the art by a wide margin. | A well studied framework to infer the causal direction for the two-variable case relies on the additive noise assumption @cite_19 . Simply put, it makes the strong assumption that @math is generated as a function of @math plus additive noise, @math , with @math . It can then be shown that while such a function is admissible in the causal direction, this is not possible in the anti-causal direction. There exist many approaches based on this framework that try to exploit linear @cite_19 or non-linear functions @cite_18 and can be applied to real valued @cite_19 @cite_18 @cite_35 @cite_31 as well as discrete data @cite_1 . Recently, @cite_17 reviewed several ANM-based approaches from which ANM-pHSIC, a method employing the Hilbert-Schmidt Independence Criterion (HSIC) to test for independence, performed best. For ANMs the confidence value is often expressed as the negative logarithm of the p-value from the used independence test @cite_17 . P-values are, however, quite sensitive to the data size @cite_32 , which leads to a less reliable confidence value. As we will show in the experiments, our score is robust and nearly unaffected by the data size. | {
"cite_N": [
"@cite_35",
"@cite_18",
"@cite_1",
"@cite_32",
"@cite_19",
"@cite_31",
"@cite_17"
],
"mid": [
"2146531590",
"2165582599",
"1497574396",
"1991445071",
"2151226328",
"2132547334",
"1850984366"
],
"abstract": [
"By taking into account the nonlinear effect of the cause, the inner noise effect, and the measurement distortion effect in the observed variables, the post-nonlinear (PNL) causal model has demonstrated its excellent performance in distinguishing the cause from effect. However, its identifiability has not been properly addressed, and how to apply it in the case of more than two variables is also a problem. In this paper, we conduct a systematic investigation on its identifiability in the two-variable case. We show that this model is identifiable in most cases; by enumerating all possible situations in which the model is not identifiable, we provide sufficient conditions for its identifiability. Simulations are given to support the theoretical results. Moreover, in the case of more than two variables, we show that the whole causal structure can be found by applying the PNL causal model to each structure in the Markov equivalent class and testing if the disturbance is independent of the direct causes for each variable. In this way the exhaustive search over all possible causal structures is avoided.",
"The discovery of causal relationships between a set of observed variables is a fundamental problem in science. For continuous-valued data linear acyclic causal models with additive noise are often used because these models are well understood and there are well-known methods to fit them to data. In reality, of course, many causal relationships are more or less nonlinear, raising some doubts as to the applicability and usefulness of purely linear methods. In this contribution we show that the basic linear framework can be generalized to nonlinear models. In this extended framework, nonlinearities in the data-generating process are in fact a blessing rather than a curse, as they typically provide information on the underlying causal system and allow more aspects of the true data-generating mechanisms to be identified. In addition to theoretical results we show simulations and some simple real data experiments illustrating the identification power provided by nonlinearities.",
"Inferring the causal structure of a set of random variables from a finite sample of the joint distribution is an important problem in science. Recently, methods using additive noise models have been suggested to approach the case of continuous variables. In many situations, however, the variables of interest are discrete or even have only finitely many states. In this work we extend the notion of additive noise models to these cases. Whenever the joint distribution P ) admits such a model in one direction, e.g. Y = f(X) + N, N ⊥ X, it does not admit the reversed model X = g(Y ) + N , N ⊥ Y as long as the model is chosen in a generic way. Based on these deliberations we propose an efficient new algorithm that is able to distinguish between cause and effect for a finite sample of discrete variables. We show that this algorithm works both on synthetic and real data sets.",
"This paper presents a review and critique of statistical null hypothesis testing in ecological studies in general, and wildlife studies in particular, and describes an alternative. Our review of Ecology and the Journal of Wildlife Management found the use of null hypothesis testing to be pervasive. The estimated number of P-values appearing within articles of Ecology exceeded 8,000 in 1991 and has exceeded 3,000 in each year since 1984, whereas the estimated number of P-values in the Journal of Wildlife Management exceeded 8,000 in 1997 and has exceeded 3,000 in each year since 1994. We estimated that 47 (SE = 3.9 ) of the P-values in the Journal of Wildlife Management lacked estimates of means or effect sizes or even the sign of the difference in means or other parameters. We find that null hypothesis testing is uninformative when no estimates of means or effect size and their precision are given. Contrary to common dogma, tests of statistical null hypotheses have relatively little utility in science and are not a fundamental aspect of the scientific method. We recommend their use be reduced in favor of more informative approaches. Towards this objective, we describe a relatively new paradigm of data analysis based on Kullback-Leibler information. This paradigm is an extension of likelihood theory and, when used correctly, avoids many of the fundamental limitations and common misuses of null hypothesis testing. Information-theoretic methods focus on providing a strength of evidence for an a priori set of alternative hypotheses, rather than a statistical test of a null hypothesis. This paradigm allows the following types of evidence for the alternative hypotheses: the rank of each hypothesis, expressed as a model; an estimate of the formal likelihood of each model, given the data; a measure of precision that incorporates model selection uncertainty; and simple methods to allow the use of the set of alternative models in making, formal inference. We provide an example of the information-theoretic approach using data on the effect of lead on survival in spectacled eider ducks (Somateria fischeri). Regardless of the analysis paradigm used, we strongly recommend inferences based on a priori considerations be clearly separated from those resulting from some form of data dredging.",
"In recent years, several methods have been proposed for the discovery of causal structure from non-experimental data. Such methods make various assumptions on the data generating process to facilitate its identification from purely observational data. Continuing this line of research, we show how to discover the complete causal structure of continuous-valued data, under the assumptions that (a) the data generating process is linear, (b) there are no unobserved confounders, and (c) disturbance variables have non-Gaussian distributions of non-zero variances. The solution relies on the use of the statistical method known as independent component analysis, and does not require any pre-specified time-ordering of the variables. We provide a complete Matlab package for performing this LiNGAM analysis (short for Linear Non-Gaussian Acyclic Model), and demonstrate the effectiveness of the method using artificially generated data and real-world data.",
"We consider the problem of learning causal directed acyclic graphs from an observational joint distribution. One can use these graphs to predict the outcome of interventional experiments, from which data are often not available. We show that if the observational distribution follows a structural equation model with an additive noise structure, the directed acyclic graph becomes identifiable from the distribution under mild conditions. This constitutes an interesting alternative to traditional methods that assume faithfulness and identify only the Markov equivalence class of the graph, thus leaving some edges undirected. We provide practical algorithms for finitely many samples, RESIT (regression with subsequent independence test) and two methods based on an independence score. We prove that RESIT is correct in the population setting and provide an empirical evaluation.",
"The discovery of causal relationships from purely observational data is a fundamental problem in science. The most elementary form of such a causal discovery problem is to decide whether X causes Y or, alternatively, Y causes X, given joint observations of two variables X,Y. An example is to decide whether altitude causes temperature, or vice versa, given only joint measurements of both variables. Even under the simplifying assumptions of no confounding, no feedback loops, and no selection bias, such bivariate causal discovery problems are challenging. Nevertheless, several approaches for addressing those problems have been proposed in recent years. We review two families of such methods: methods based on Additive Noise Models (ANMs) and Information Geometric Causal Inference (IGCI). We present the benchmark CAUSEEFFECTPAIRS that consists of data for 100 different causee ffect pairs selected from 37 data sets from various domains (e.g., meteorology, biology, medicine, engineering, economy, etc.) and motivate our decisions regarding the \"ground truth\" causal directions of all pairs. We evaluate the performance of several bivariate causal discovery methods on these real-world benchmark data and in addition on artificially simulated data. Our empirical results on real-world data indicate that certain methods are indeed able to distinguish cause from effect using only purely observational data, although more benchmark data would be needed to obtain statistically significant conclusions. One of the best performing methods overall is the method based on Additive Noise Models that has originally been proposed by (2009), which obtains an accuracy of 63 ± 10 and an AUC of 0.74 ± 0.05 on the real-world benchmark. As the main theoretical contribution of this work we prove the consistency of that method."
]
} |
1709.08915 | 2758592735 | We consider the fundamental problem of inferring the causal direction between two univariate numeric random variables X and Y from observational data. The two-variable case is especially difficult to solve since it is not possible to use standard conditional independence tests between the variables. To tackle this problem, we follow an information theoretic approach based on Kolmogorov complexity and use the Minimum Description Length (MDL) principle to provide a practical solution. In particular, we propose a compression scheme to encode local and global functional relations using MDL-based regression. We infer X causes Y in case it is shorter to describe Y as a function of X than the inverse direction. In addition, we introduce Slope, an efficient linear-time algorithm that through thorough empirical evaluation on both synthetic and real world data we show outperforms the state of the art by a wide margin. | Recently, Janzing and Sch " o lkopf postulated that if @math , the complexity of the description of the joint distribution in terms of Kolmogorov complexity, @math , will be shorter when first describing the distribution of the cause @math and than describing the distribution of the effect given the cause @math than vice versa @cite_26 @cite_9 . To the best of our knowledge, @cite_28 were the first to propose a practical instantiation of this framework based on the Minimum Message Length principle (MML) @cite_13 using Bayesian priors. Vreeken @cite_4 proposed to approximate the Kolmogorov complexity for numeric data using the cumulative residual entropy, and gave an instantiation for multivariate continuous-valued data. Perhaps most related to is @cite_14 , which uses MDL to infer causal direction on binary data, whereas we focus on univariate numeric data. | {
"cite_N": [
"@cite_26",
"@cite_4",
"@cite_14",
"@cite_28",
"@cite_9",
"@cite_13"
],
"mid": [
"2083689856",
"811835876",
"2584044794",
"2164001983",
"",
"2110381504"
],
"abstract": [
"Inferring the causal structure that links n observables is usually based upon detecting statistical dependences and choosing simple graphs that make the joint measure Markovian. Here we argue why causal inference is also possible when the sample size is one. We develop a theory how to generate causal graphs explaining similarities between single objects. To this end, we replace the notion of conditional stochastic independence in the causal Markov condition with the vanishing of conditional algorithmic mutual information and describe the corresponding causal inference rules. We explain why a consistent reformulation of causal inference in terms of algorithmic complexity implies a new inference principle that takes into account also the complexity of conditional probability densities, making it possible to select among Markov equivalent causal graphs. This insight provides a theoretical foundation of a heuristic principle proposed in earlier work. We also sketch some ideas on how to replace Kolmogorov complexity with decidable complexity criteria. This can be seen as an algorithmic analog of replacing the empirically undecidable question of statistical independence with practical independence tests that are based on implicit or explicit assumptions on the underlying distribution.",
"We focus on data-driven causal inference. In particular, we propose a new principle for causal inference based on algorithmic information theory, i.e. Kolmogorov complexity. In a nutshell, we determine how much information one data object gives about the other, and vice versa, and identify the most likely causal direction by the strongest direction of information. To apply this principle in practice, we propose ERGO, an efficient instantiation for inferring the causal direction between multivariate real-valued data pairs. ERGO is based on cumulative and Shannon entropy. Therewith, we do not have to assume distributions, nor have to restrict the type of correlation. Extensive empirical evaluation on synthetic, benchmark, and real-world data shows that ERGO is robust against both noise and dimensionality, efficient, and outperforms the state of the art by a wide margin.",
"Causal inference is one of the fundamental problems in science. In recent years, several methods have been proposed for discovering causal structure from observational data. These methods, however, focus specifically on numeric data, and are not applicable on nominal or binary data. In this work, we focus on causal inference for binary data. Simply put, we propose causal inference by compression. To this end we propose an inference framework based on solid information theoretic foundations, i.e. Kolmogorov complexity. However, Kolmogorov complexity is not computable, and hence we propose a practical and computable instantiation based on the Minimum Description Length (MDL) principle. To apply the framework in practice, we propose ORIGO, an efficient method for inferring the causal direction from binary data. ORIGO employs the lossless PACK compressor, works directly on the data and does not require assumptions about neither distributions nor the type of causal relations. Extensive evaluation on synthetic, benchmark, and real-world data shows that ORIGO discovers meaningful causal relations, and outperforms state-of-the-art methods by a wide margin.",
"We propose a novel method for inferring whether X causes Y or vice versa from joint observations of X and Y. The basic idea is to model the observed data using probabilistic latent variable models, which incorporate the effects of unobserved noise. To this end, we consider the hypothetical effect variable to be a function of the hypothetical cause variable and an independent noise term (not necessarily additive). An important novel aspect of our work is that we do not restrict the model class, but instead put general non-parametric priors on this function and on the distribution of the cause. The causal direction can then be inferred by using standard Bayesian model selection. We evaluate our approach on synthetic data and real-world data and report encouraging results.",
"",
"1. The class to which each thing belongs. 2. The average properties of each class. 3. The deviations of each thing from the average properties of its parent class. If the things are found to be concentrated in a small area of the region of each class in the measurement space then the deviations will be small, and with reference to the average class properties most of the information about a thing is given by naming the class to which it belongs. In this case the information may be recorded much more briefly than if a classification had not been used. We suggest that the best classification is that which results in the briefest recording of all the attribute information. In this context, we will regard the measurements of each thing as being a message about that thing. Shannon (1948) showed that where messages may be regarded as each nominating the occurrence of a particular event among a universe of possible events, the information needed to record a series of such messages is minimised if the messages are encoded so that the length of each message is proportional to minus the logarithm of the relative frequency of occurrence of the event which it nominates. The information required is greatest when all frequencies are equal. The messages here nominate the positions in measurement space of the 5 1 points representing the attributes of the things. If the expected density of points in the measurement space is everywhere uniform, the positions of the points cannot be encoded more briefly than by a simple list of the measured values. However, if the expected density is markedly non-uniform, application"
]
} |
1709.08605 | 2626348203 | The CRAYFIS experiment proposes to use privately owned mobile phones as a ground detector array for Ultra High Energy Cosmic Rays. Upon interacting with Earth's atmosphere, these events produce extensive particle showers which can be detected by cameras on mobile phones. A typical shower contains minimally-ionizing particles such as muons. As these particles interact with CMOS image sensors, they may leave tracks of faintly-activated pixels that are sometimes hard to distinguish from random detector noise. Triggers that rely on the presence of very bright pixels within an image frame are not efficient in this case. We present a trigger algorithm based on Convolutional Neural Networks which selects images containing such tracks and are evaluated in a lazy manner: the response of each successive layer is computed only if activation of the current layer satisfies a continuation criterion. Usage of neural networks increases the sensitivity considerably comparable with image thresholding, while the lazy evaluation allows for execution of the trigger under the limited computational power of mobile phones. | Minimally-ionizing particles are characterized by the pattern of activated pixels they leave over a small region of an exposure. Hence the problem of minimally-ionizing particle detection can be transformed to the problem of pattern detection over the image. Several attempts were performed to solve pattern detection problem in different setups. The solution proposed in works @cite_6 and @cite_2 utilizes Convolutional Neural Networks (CNNs). Certain properties of CNNs such as translation invariance, locality, and correspondingly few weights to be learned, make them particularly well suited to extracting features from images. The performance of even simple CNNs on image classification tasks have been shown to be very good compared to other methods @cite_4 . However, the training and evaluation of CNNs requires relatively intense computation to which smartphones are not particularly well suited. Viola and Jones in @cite_7 introduced the idea of a simple feature-based cascading classifier. They proposed using small binary filters as feature detectors and to increase computational power from cascade to cascade. | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_6",
"@cite_2"
],
"mid": [
"1536929369",
"",
"1934410531",
"2613718673"
],
"abstract": [
"From the Publisher: Dramatically updating and extending the first edition, published in 1995, the second edition of The Handbook of Brain Theory and Neural Networks presents the enormous progress made in recent years in the many subfields related to the two great questions: How does the brain work? and, How can we build intelligent machines? Once again, the heart of the book is a set of almost 300 articles covering the whole spectrum of topics in brain theory and neural networks. The first two parts of the book, prepared by Michael Arbib, are designed to help readers orient themselves in this wealth of material. Part I provides general background on brain modeling and on both biological and artificial neural networks. Part II consists of \"Road Maps\" to help readers steer through articles in part III on specific topics of interest. The articles in part III are written so as to be accessible to readers of diverse backgrounds. They are cross-referenced and provide lists of pointers to Road Maps, background material, and related reading. The second edition greatly increases the coverage of models of fundamental neurobiology, cognitive neuroscience, and neural network approaches to language. It contains 287 articles, compared to the 266 in the first edition. Articles on topics from the first edition have been updated by the original authors or written anew by new authors, and there are 106 articles on new topics.",
"",
"In real-world face detection, large visual variations, such as those due to pose, expression, and lighting, demand an advanced discriminative model to accurately differentiate faces from the backgrounds. Consequently, effective models for the problem tend to be computationally prohibitive. To address these two conflicting challenges, we propose a cascade architecture built on convolutional neural networks (CNNs) with very powerful discriminative capability, while maintaining high performance. The proposed CNN cascade operates at multiple resolutions, quickly rejects the background regions in the fast low resolution stages, and carefully evaluates a small number of challenging candidates in the last high resolution stage. To improve localization effectiveness, and reduce the number of candidates at later stages, we introduce a CNN-based calibration stage after each of the detection stages in the cascade. The output of each calibration stage is used to adjust the detection window position for input to the subsequent stage. The proposed method runs at 14 FPS on a single CPU core for VGA-resolution images and 100 FPS using a GPU, and achieves state-of-the-art detection performance on two public face detection benchmarks.",
"State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet [7] and Fast R-CNN [5] have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully-convolutional network that simultaneously predicts object bounds and objectness scores at each position. RPNs are trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. With a simple alternating optimization, RPN and Fast R-CNN can be trained to share convolutional features. For the very deep VGG-16 model [19], our detection system has a frame rate of 5fps (including all steps) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007 (73.2 mAP) and 2012 (70.4 mAP) using 300 proposals per image. Code is available at https: github.com ShaoqingRen faster_rcnn."
]
} |
1709.08810 | 2757631670 | Place recognition is an essential component of Simultaneous Localization And Mapping (SLAM). Under severe appearance change, reliable place recognition is a difficult perception task since the same place is perceptually very different in the morning, at night, or over different seasons. This work addresses place recognition as a domain translation task. Using a pair of coupled Generative Adversarial Networks (GANs), we show that it is possible to generate the appearance of one domain (such as summer) from another (such as winter) without requiring image-to-image correspondences across the domains. Mapping between domains is learned from sets of images in each domain without knowing the instance-to-instance correspondence by enforcing a cyclic consistency constraint. In the process, meaningful feature spaces are learned for each domain, the distances in which can be used for the task of place recognition. Experiments show that learned features correspond to visual similarity and can be effectively used for place recognition across seasons. | Place recognition with cameras is a well studied problem and a good overview can be found in @cite_20 . Traditionally, the problem of visual loop closure detection has been addressed using feature-based approaches, where, the extracted features serve as a description of the structure being observed by the robot @cite_21 @cite_1 . Such methods operate very well within the short time scale with limited appearance or structural changes. They would, however, fail in situations where the change in the environment is greater than the invariances (such a rotation, scale, illumination etc.) provided by the underlying feature descriptor. | {
"cite_N": [
"@cite_21",
"@cite_1",
"@cite_20"
],
"mid": [
"2144824356",
"1989484209",
"2284029970"
],
"abstract": [
"This paper describes a probabilistic approach to the problem of recognizing places based on their appearance. The system we present is not limited to localization, but can determine that a new observation comes from a previously unseen place, and so augment its map. Effectively this is a SLAM system in the space of appearance. Our probabilistic approach allows us to explicitly account for perceptual aliasing in the environment—identical but indistinctive observations receive a low probability of having come from the same place. We achieve this by learning a generative model of place appearance. By partitioning the learning problem into two parts, new place models can be learned online from only a single observation of a place. The algorithm complexity is linear in the number of places in the map, and is particularly suitable for online loop closure detection in mobile robotics.",
"We propose a novel method for visual place recognition using bag of words obtained from accelerated segment test (FAST)+BRIEF features. For the first time, we build a vocabulary tree that discretizes a binary descriptor space and use the tree to speed up correspondences for geometrical verification. We present competitive results with no false positives in very different datasets, using exactly the same vocabulary and settings. The whole technique, including feature extraction, requires 22 ms frame in a sequence with 26 300 images that is one order of magnitude faster than previous approaches.",
"Visual place recognition is a challenging problem due to the vast range of ways in which the appearance of real-world places can vary. In recent years, improvements in visual sensing capabilities, an ever-increasing focus on long-term mobile robot autonomy, and the ability to draw on state-of-the-art research in other disciplines—particularly recognition in computer vision and animal navigation in neuroscience—have all contributed to significant advances in visual place recognition systems. This paper presents a survey of the visual place recognition research landscape. We start by introducing the concepts behind place recognition—the role of place recognition in the animal kingdom, how a “place” is defined in a robotics context, and the major components of a place recognition system. Long-term robot operations have revealed that changing appearance can be a significant factor in visual place recognition failure; therefore, we discuss how place recognition solutions can implicitly or explicitly account for appearance change within the environment. Finally, we close with a discussion on the future of visual place recognition, in particular with respect to the rapid advances being made in the related fields of deep learning, semantic scene understanding, and video description."
]
} |
1709.08810 | 2757631670 | Place recognition is an essential component of Simultaneous Localization And Mapping (SLAM). Under severe appearance change, reliable place recognition is a difficult perception task since the same place is perceptually very different in the morning, at night, or over different seasons. This work addresses place recognition as a domain translation task. Using a pair of coupled Generative Adversarial Networks (GANs), we show that it is possible to generate the appearance of one domain (such as summer) from another (such as winter) without requiring image-to-image correspondences across the domains. Mapping between domains is learned from sets of images in each domain without knowing the instance-to-instance correspondence by enforcing a cyclic consistency constraint. In the process, meaningful feature spaces are learned for each domain, the distances in which can be used for the task of place recognition. Experiments show that learned features correspond to visual similarity and can be effectively used for place recognition across seasons. | Visual place recognition is the task of localizing a robot in an environment using information from a single camera. It is a well studied problem and can be split into two dominant approaches; feature based, that represent a particular image as a set of key features extracted from interesting location in the image, and image-based, that instead of extracting features, tries to reason using all the image information @cite_20 . Each approach has it own limitations: feature-based methods may fail under illumination changes while image-based methods are sensitive to view-point changes. | {
"cite_N": [
"@cite_20"
],
"mid": [
"2284029970"
],
"abstract": [
"Visual place recognition is a challenging problem due to the vast range of ways in which the appearance of real-world places can vary. In recent years, improvements in visual sensing capabilities, an ever-increasing focus on long-term mobile robot autonomy, and the ability to draw on state-of-the-art research in other disciplines—particularly recognition in computer vision and animal navigation in neuroscience—have all contributed to significant advances in visual place recognition systems. This paper presents a survey of the visual place recognition research landscape. We start by introducing the concepts behind place recognition—the role of place recognition in the animal kingdom, how a “place” is defined in a robotics context, and the major components of a place recognition system. Long-term robot operations have revealed that changing appearance can be a significant factor in visual place recognition failure; therefore, we discuss how place recognition solutions can implicitly or explicitly account for appearance change within the environment. Finally, we close with a discussion on the future of visual place recognition, in particular with respect to the rapid advances being made in the related fields of deep learning, semantic scene understanding, and video description."
]
} |
1709.08810 | 2757631670 | Place recognition is an essential component of Simultaneous Localization And Mapping (SLAM). Under severe appearance change, reliable place recognition is a difficult perception task since the same place is perceptually very different in the morning, at night, or over different seasons. This work addresses place recognition as a domain translation task. Using a pair of coupled Generative Adversarial Networks (GANs), we show that it is possible to generate the appearance of one domain (such as summer) from another (such as winter) without requiring image-to-image correspondences across the domains. Mapping between domains is learned from sets of images in each domain without knowing the instance-to-instance correspondence by enforcing a cyclic consistency constraint. In the process, meaningful feature spaces are learned for each domain, the distances in which can be used for the task of place recognition. Experiments show that learned features correspond to visual similarity and can be effectively used for place recognition across seasons. | Instead of whole images or features, several structure based methods have also been proposed that use edges in the images are the illumination invariant description @cite_8 @cite_2 . Techniques such as shadow removal @cite_6 has shown remarkable improvement by removing unwanted artifacts from images. Additionally, features from Convolutional Neural Networks (CNNs) are also successfully used for the task of place recognition due to their invariance to illumination and viewpoint changes @cite_7 . | {
"cite_N": [
"@cite_2",
"@cite_7",
"@cite_6",
"@cite_8"
],
"mid": [
"2092249759",
"2951399172",
"2038390631",
"2121308169"
],
"abstract": [
"Visual localization systems that are practical for autonomous vehicles in outdoor industrial applications must perform reliably in a wide range of conditions. Changing outdoor conditions cause difficulty by drastically altering the information available in the camera images. To confront the problem, we have developed a visual localization system that uses a surveyed three-dimensional (3D)-edge map of permanent structures in the environment. The map has the invariant properties necessary to achieve long-term robust operation. Previous 3D-edge map localization systems usually maintain a single pose hypothesis, making it difficult to initialize without an accurate prior pose estimate and also making them susceptible to misalignment with unmapped edges detected in the camera image. A multihypothesis particle filter is employed here to perform the initialization procedure with significant uncertainty in the vehicle's initial pose. A novel observation function for the particle filter is developed and evaluated against two existing functions. The new function is shown to further improve the abilities of the particle filter to converge given a very coarse estimate of the vehicle's initial pose. An intelligent exposure control algorithm is also developed that improves the quality of the pertinent information in the image. Results gathered over an entire sunny day and also during rainy weather illustrate that the localization system can operate in a wide range of outdoor conditions. The conclusion is that an invariant map, a robust multihypothesis localization algorithm, and an intelligent exposure control algorithm all combine to enable reliable visual localization through challenging outdoor conditions. © 2009 Wiley Periodicals, Inc.",
"After the incredible success of deep learning in the computer vision domain, there has been much interest in applying Convolutional Network (ConvNet) features in robotic fields such as visual navigation and SLAM. Unfortunately, there are fundamental differences and challenges involved. Computer vision datasets are very different in character to robotic camera data, real-time performance is essential, and performance priorities can be different. This paper comprehensively evaluates and compares the utility of three state-of-the-art ConvNets on the problems of particular relevance to navigation for robots; viewpoint-invariance and condition-invariance, and for the first time enables real-time place recognition performance using ConvNets with large maps by integrating a variety of existing (locality-sensitive hashing) and novel (semantic search space partitioning) optimization techniques. We present extensive experiments on four real world datasets cultivated to evaluate each of the specific challenges in place recognition. The results demonstrate that speed-ups of two orders of magnitude can be achieved with minimal accuracy degradation, enabling real-time performance. We confirm that networks trained for semantic place categorization also perform better at (specific) place recognition when faced with severe appearance changes and provide a reference for which networks and layers are optimal for different aspects of the place recognition problem.",
"In outdoor environments shadows are common. These typically strong visual features cause considerable change in the appearance of a place, and therefore confound vision-based localisation approaches. In this paper we describe how to convert a colour image of the scene to a greyscale invariant image where pixel values are a function of underlying material property not lighting. We summarise the theory of shadow invariant images and discuss the modelling and calibration issues which are important for non-ideal off-the-shelf colour cameras. We evaluate the technique with a commonly used robotic camera and an autonomous car operating in an outdoor environment, and show that it can outperform the use of ordinary greyscale images for the task of visual localisation.",
"While many visual simultaneous localization and mapping (SLAM) systems use point features as landmarks, few take advantage of the edge information in images. Those SLAM systems that do observe edge features do not consider edges with all degrees of freedom. Edges are difficult to use in vision SLAM because of selection, observation, initialization and data association challenges. A map that includes edge features, however, contains higher-order geometric information useful both during and after SLAM. We define a well-localized edge landmark and present an efficient algorithm for selecting such landmarks. Further, we describe how to initialize new landmarks, observe mapped landmarks in subsequent images, and address the data association challenges of edges. Our methods, implemented in a particle-filter SLAM system, operate at frame rate on live video sequences."
]
} |
1709.08810 | 2757631670 | Place recognition is an essential component of Simultaneous Localization And Mapping (SLAM). Under severe appearance change, reliable place recognition is a difficult perception task since the same place is perceptually very different in the morning, at night, or over different seasons. This work addresses place recognition as a domain translation task. Using a pair of coupled Generative Adversarial Networks (GANs), we show that it is possible to generate the appearance of one domain (such as summer) from another (such as winter) without requiring image-to-image correspondences across the domains. Mapping between domains is learned from sets of images in each domain without knowing the instance-to-instance correspondence by enforcing a cyclic consistency constraint. In the process, meaningful feature spaces are learned for each domain, the distances in which can be used for the task of place recognition. Experiments show that learned features correspond to visual similarity and can be effectively used for place recognition across seasons. | There is significant of literature regarding GANs and their various applications, however, we are particularly interested in the types of networks that perform transfer. One such work is DiscoGAN @cite_9 , which discovers mappings between two image domains using unpaired examples from both domains. They show that a pair of coupled GANs can discover relationships between styles of handbags and shoes and discover common attributes (such as orientation) between images of cars and human faces, etc. Along similar lines, CycleGAN @cite_3 enforces cyclic-consistency to learn one-to-one mapping between domains and show translation from satellite images to maps. If the two domain have paired information, pix2pix @cite_4 , can be used to learn the translation from one domain to another. | {
"cite_N": [
"@cite_9",
"@cite_4",
"@cite_3"
],
"mid": [
"2951939904",
"",
"2605287558"
],
"abstract": [
"While humans easily recognize relations between data from different domains without any supervision, learning to automatically discover them is in general very challenging and needs many ground-truth pairs that illustrate the relations. To avoid costly pairing, we address the task of discovering cross-domain relations given unpaired data. We propose a method based on generative adversarial networks that learns to discover relations between different domains (DiscoGAN). Using the discovered relations, our proposed network successfully transfers style from one domain to another while preserving key attributes such as orientation and face identity. Source code for official implementation is publicly available this https URL",
"",
"Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain @math to a target domain @math in the absence of paired examples. Our goal is to learn a mapping @math such that the distribution of images from @math is indistinguishable from the distribution @math using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping @math and introduce a cycle consistency loss to push @math (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach."
]
} |
1709.08767 | 2757372899 | We present a novel computational framework that connects Blue Waters, the NSF-supported, leadership-class supercomputer operated by NCSA, to the Laser Interferometer Gravitational-Wave Observatory (LIGO) Data Grid via Open Science Grid technology. To enable this computational infrastructure, we configured, for the first time, a LIGO Data Grid Tier-1 Center that can submit heterogeneous LIGO workflows using Open Science Grid facilities. In order to enable a seamless connection between the LIGO Data Grid and Blue Waters via Open Science Grid, we utilize Shifter to containerize LIGO’s workflow software. This work represents the first time Open Science Grid, Shifter, and Blue Waters are unified to tackle a scientific problem and, in particular, it is the first time a framework of this nature is used in the context of large scale gravitational wave data analysis. This new framework has been used in the last several weeks of LIGO’s second discovery campaign to run the most computationally demanding gravitational wave search workflows on Blue Waters, and accelerate discovery in the emergent field of gravitational wave astrophysics. We discuss the implications of this novel framework for a wider ecosystem of Higher Performance Computing users. | Other projects have aimed at using HPC clusters to handle what is essentially a HTC-type workflow. Within the LHC, the ATLAS experiment is making use of leadership HPC resources using PaNDA and has shown this works well to backfill into empty compute nodes on Titan @cite_15 , with hardware similar to that of Blue Waters, making an estimated 300M core hours per year available to the ATLAS experiment. Additional work is currently under way to use Shifter and OSG on Blue Waters for the LHC, and this began just after our own project, building on what we have done. Other projects have used workflow managers without Shifter on Blue Waters, for example the ArcticDEM project @cite_20 uses Swift @cite_29 , and the Southern California Earthquake Center's CyberShake project @cite_34 uses Pegasus @cite_17 . Finally some of the authors of this paper were involved in porting the OSG LIGO software stack to XSEDE HPC resources in the past @cite_6 . | {
"cite_N": [
"@cite_29",
"@cite_6",
"@cite_15",
"@cite_34",
"@cite_20",
"@cite_17"
],
"mid": [
"2077579791",
"2963938112",
"2790539989",
"2038014757",
"",
"2129666400"
],
"abstract": [
"Scientists, engineers, and statisticians must execute domain-specific application programs many times on large collections of file-based data. This activity requires complex orchestration and data management as data is passed to, from, and among application invocations. Distributed and parallel computing resources can accelerate such processing, but their use further increases programming complexity. The Swift parallel scripting language reduces these complexities by making file system structures accessible via language constructs and by allowing ordinary application programs to be composed into powerful parallel scripts that can efficiently utilize parallel and distributed resources. We present Swift's implicitly parallel and deterministic programming model, which applies external applications to file collections using a functional style that abstracts and simplifies distributed parallel execution.",
"During 2015 and 2016, the Laser Interferometer Gravitational-Wave Observatory (LIGO) conducted a three-month observing campaign. These observations delivered the first direct detection of gravitational waves from binary black hole mergers. To search for these signals, the LIGO Scientific Collaboration uses the PyCBC search pipeline. To deliver science results in a timely manner, LIGO collaborated with the Open Science Grid (OSG) to distribute the required computation across a series of dedicated, opportunistic, and allocated resources. To deliver the petabytes necessary for such a large-scale computation, our team deployed a distributed data access infrastructure based on the XRootD server suite and the CernVM File System (CVMFS). This data access strategy grew from simply accessing remote storage to a POSIX-based interface underpinned by distributed, secure caches across the OSG.",
"",
"CyberShake is a computational platform developed by the Southern California Earthquake Center (SCEC) that explicitly incorporates earthquake rupture time histories and deterministic wave propagation effects into seismic hazard calculations through the use of 3D waveform simulations. Using CyberShake, SCEC has created the first physics-based probabilistic seismic hazard analysis (PSHA) models of the Los Angeles region from suites of simulations comprising 108 seismograms. The current models are, however, limited to low seismic frequencies (≤ 0.5 Hz). To increase the maximum simulated frequency to above 1 Hz and produce a California state-wide model, we have transformed SCEC Anelastic Wave Propagation code (AWP-ODC) to include strain Green's tensor (SGT) calculations to accelerate CyberShake calculations. This tensor-valued wave field code has both CPU and GPU components in place for flexibility on different architectures. We have demonstrated the performance and scalability of this solver optimized for the heterogeneous Blue Waters system at NCSA. The high performance of the wave propagation computation, coupled with CPU GPU co-scheduling capabilities of our workflow-managed systems, make a statewide hazard model a goal reachable with existing supercomputers.",
"",
"This paper describes the Pegasus framework that can be used to map complex scientific workflows onto distributed resources. Pegasus enables users to represent the workflows at an abstract level without needing to worry about the particulars of the target execution systems. The paper describes general issues in mapping applications and the functionality of Pegasus. We present the results of improving application performance through workflow restructuring which clusters multiple tasks in a workflow into single entities. A real-life astronomy application is used as the basis for the study."
]
} |
1709.08858 | 2600654830 | In this paper, we propose a statistical test to determine whether a given word is used as a polysemic word or not. The statistic of the word in this test roughly corresponds to the fluctuation in the senses of the neighboring words a nd the word itself. Even though the sense of a word corresponds to a single vector, we discuss how polysemy of the words affects the position of vectors. Finally, we also explain the method to detect this effect. | The distributed word representation can be computed as weight vectors of neurons, which learn language modeling @cite_6 . We can obtain a distributed representation of a word using the Word2Vec software @cite_4 which enable us to perform vector addition subtraction on a word's meaning. The theoretical background is analyzed by @cite_1 , where the operation is to factorize a word-context matrix, where the elements in the matrix are some function of the given word and its context pairs. This analysis gives us insight into how the vector is affected by multiple senses or multiple context sets. If a word has two senses, the obtained representation for the word will be a linearly interpolated point between the two points of their senses. | {
"cite_N": [
"@cite_1",
"@cite_4",
"@cite_6"
],
"mid": [
"2125031621",
"2153579005",
"2132339004"
],
"abstract": [
"We analyze skip-gram with negative-sampling (SGNS), a word embedding method introduced by , and show that it is implicitly factorizing a word-context matrix, whose cells are the pointwise mutual information (PMI) of the respective word and context pairs, shifted by a global constant. We find that another embedding method, NCE, is implicitly factorizing a similar matrix, where each cell is the (shifted) log conditional probability of a word given its context. We show that using a sparse Shifted Positive PMI word-context matrix to represent words improves results on two word similarity tasks and one of two analogy tasks. When dense low-dimensional vectors are preferred, exact factorization with SVD can achieve solutions that are at least as good as SGNS's solutions for word similarity tasks. On analogy questions SGNS remains superior to SVD. We conjecture that this stems from the weighted nature of SGNS's factorization.",
"The recently introduced continuous Skip-gram model is an efficient method for learning high-quality distributed vector representations that capture a large number of precise syntactic and semantic word relationships. In this paper we present several extensions that improve both the quality of the vectors and the training speed. By subsampling of the frequent words we obtain significant speedup and also learn more regular word representations. We also describe a simple alternative to the hierarchical softmax called negative sampling. An inherent limitation of word representations is their indifference to word order and their inability to represent idiomatic phrases. For example, the meanings of \"Canada\" and \"Air\" cannot be easily combined to obtain \"Air Canada\". Motivated by this example, we present a simple method for finding phrases in text, and show that learning good vector representations for millions of phrases is possible.",
"A goal of statistical language modeling is to learn the joint probability function of sequences of words in a language. This is intrinsically difficult because of the curse of dimensionality: a word sequence on which the model will be tested is likely to be different from all the word sequences seen during training. Traditional but very successful approaches based on n-grams obtain generalization by concatenating very short overlapping sequences seen in the training set. We propose to fight the curse of dimensionality by learning a distributed representation for words which allows each training sentence to inform the model about an exponential number of semantically neighboring sentences. The model learns simultaneously (1) a distributed representation for each word along with (2) the probability function for word sequences, expressed in terms of these representations. Generalization is obtained because a sequence of words that has never been seen before gets high probability if it is made of words that are similar (in the sense of having a nearby representation) to words forming an already seen sentence. Training such large models (with millions of parameters) within a reasonable time is itself a significant challenge. We report on experiments using neural networks for the probability function, showing on two text corpora that the proposed approach significantly improves on state-of-the-art n-gram models, and that the proposed approach allows to take advantage of longer contexts."
]
} |
1709.08858 | 2600654830 | In this paper, we propose a statistical test to determine whether a given word is used as a polysemic word or not. The statistic of the word in this test roughly corresponds to the fluctuation in the senses of the neighboring words a nd the word itself. Even though the sense of a word corresponds to a single vector, we discuss how polysemy of the words affects the position of vectors. Finally, we also explain the method to detect this effect. | The importance of multiple senses is well recognized in word sense detection in distributed representation. The usual approach is to compute the corresponding vectors for each sense of a word @cite_5 @cite_3 . In this approach, first, the context is clustered. Then, the vector for each cluster is computed. However, the major problem faced by this approach is that all target words need to be assumed as polysemic words first, and their contexts are always required to be clustered. Another approach is to use external language resources for word sense, and to classify the context @cite_2 . The problem with this approach is that it requires language resources of meanings to obtain the meaning of a polysemic word. If we know whether a given word is polysemic or monosemic thorough a relatively simple method, we can concentrate our attention on polysemic words. | {
"cite_N": [
"@cite_5",
"@cite_3",
"@cite_2"
],
"mid": [
"2164973920",
"2164019165",
"2251537235"
],
"abstract": [
"Current vector-space models of lexical semantics create a single \"prototype\" vector to represent the meaning of a word. However, due to lexical ambiguity, encoding word meaning with a single vector is problematic. This paper presents a method that uses clustering to produce multiple \"sense-specific\" vectors for each word. This approach provides a context-dependent vector representation of word meaning that naturally accommodates homonymy and polysemy. Experimental comparisons to human judgements of semantic similarity for both isolated words as well as words in sentential contexts demonstrate the superiority of this approach over both prototype and exemplar based vector-space models.",
"Unsupervised word representations are very useful in NLP tasks both as inputs to learning algorithms and as extra word features in NLP systems. However, most of these models are built with only local context and one representation per word. This is problematic because words are often polysemous and global context can also provide useful information for learning word meanings. We present a new neural network architecture which 1) learns word embeddings that better capture the semantics of words by incorporating both local and global document context, and 2) accounts for homonymy and polysemy by learning multiple embeddings per word. We introduce a new dataset with human judgments on pairs of words in sentential context, and evaluate our model on it, showing that our model outperforms competitive baselines and other neural language models.",
"Most word representation methods assume that each word owns a single semantic vector. This is usually problematic because lexical ambiguity is ubiquitous, which is also the problem to be resolved by word sense disambiguation. In this paper, we present a unified model for joint word sense representation and disambiguation, which will assign distinct representations for each word sense. 1 The basic idea is that both word sense representation (WSR) and word sense disambiguation (WSD) will benefit from each other: (1) highquality WSR will capture rich information about words and senses, which should be helpful for WSD, and (2) high-quality WSD will provide reliable disambiguated corpora for learning better sense representations. Experimental results show that, our model improves the performance of contextual word similarity compared to existing WSR methods, outperforms stateof-the-art supervised methods on domainspecific WSD, and achieves competitive performance on coarse-grained all-words WSD."
]
} |
1709.08945 | 2758825361 | The uncertainty and variability of underwater environment propose the request to control underwater robots in real time and dynamically, especially in the scenarios where human and robots need to work collaboratively in the field. However, the underwater environment imposes harsh restrictions on the application of typical control and communication methods. Considering that gestures are a natural and efficient interactive way for human, we, utilizing convolution neural network, implement a real-time gesture-based recognition system, who can recognize 50 kinds of gestures from images captured by one normal monocular camera, and apply this recognition system in human and underwater robot interaction. We design A Flexible and Extendable Interaction Scheme (AFEIS) through which underwater robots can be programmed in situ underwater by human operators using customized gesture-based sign language. This paper elaborates the design of gesture recognition system and AFEIS, and presents our field trial results when applying this system and scheme on underwater robots. | In @cite_2 , the authors use ARTags to interact with robots underwater. In order to use this interaction method, operators must print ARTags and bind together into book form before going into the water. ARTags must be waterproofed in advance, and it is also a problem for operators to pick out the expected one from dozens of ARTags, which is unable to provide any intuitive meaning to readers. | {
"cite_N": [
"@cite_2"
],
"mid": [
"2097832243"
],
"abstract": [
"We describe an interaction paradigm for controlling a robot using hand gestures. In particular, we are interested in the control of an underwater robot by an on-site human operator. Under this context, vision-based control is very attractive, and we propose a robot control and programming mechanism based on visual symbols. A human operator presents engineered visual targets to the robotic system, which recognizes and interprets them. This paper describes the approach and proposes a specific gesture language called \"RoboChat\". RoboChat allows an operator to control a robot and even express complex programming concepts, using a sequence of visually presented symbols, encoded into fiducial markers. We evaluate the efficiency and robustness of this symbolic communication scheme by comparing it to traditional gesture-based interaction involving a remote human operator"
]
} |
1709.08945 | 2758825361 | The uncertainty and variability of underwater environment propose the request to control underwater robots in real time and dynamically, especially in the scenarios where human and robots need to work collaboratively in the field. However, the underwater environment imposes harsh restrictions on the application of typical control and communication methods. Considering that gestures are a natural and efficient interactive way for human, we, utilizing convolution neural network, implement a real-time gesture-based recognition system, who can recognize 50 kinds of gestures from images captured by one normal monocular camera, and apply this recognition system in human and underwater robot interaction. We design A Flexible and Extendable Interaction Scheme (AFEIS) through which underwater robots can be programmed in situ underwater by human operators using customized gesture-based sign language. This paper elaborates the design of gesture recognition system and AFEIS, and presents our field trial results when applying this system and scheme on underwater robots. | In @cite_6 , the authors use the motion of the hand and arm instead of the hand gestures to perform interaction. A problem of this interaction scheme is that during interaction, operators have to hold the arm up and draw shapes by moving the hand and arm aloft. Such an interaction method is not so natural and it is unable for operators to draw shapes by the hand and arm precisely. The result provided by the authors only reaches about 97 In @cite_10 , the authors employ gloves with colored markers to implement gesture recognition underwater rather than directly basing the recognition on the whole hand or the glove covering the whole hand. The authors in the paper reveal neither the details of their gesture recognition algorithm nor the performance of their algorithm. However, a conceivable shortage is that the kinds of gestures that their algorithm can cover are limited since the recognition may be achievable only when there is enough space of colored markers shown in the camera. | {
"cite_N": [
"@cite_10",
"@cite_6"
],
"mid": [
"1518898410",
"2166974693"
],
"abstract": [
"Underwater environment is characterized by harsh conditions and is difficult to monitor. The CADDY project deals with the development of a companion robot devoted to support and to monitor human operations and activities during the dive. In this scenario the communication and correct reception of messages between the diver and the robot are essential for success of the dive goals. However, the underwater environment poses a set of technical constraints hardly limiting the communication possibilities. For such reasons the solution proposed is to develop a communication language based on the consolidated and standardized diver gestures, commonly employed during professional and recreational dives, thus leading to the definition of a CADDY language, called CADDIAN, and a communication protocol. This article focuses on the creation of the language providing alphabet, syntax and semantics: future work will explain the part of recognition of gestures that is still in progress.",
"A gesture-based interaction framework is presented for controlling mobile robots. This natural interaction paradigm has few physical requirements, and thus can be deployed in many restrictive and challenging environments. We present an implementation of this scheme in the control of an underwater robot by an on-site human operator. The operator performs discrete gestures using engineered visual targets, which are interpreted by the robot as parametrized actionable commands. By combining the symbolic alphabets resulting from several visual cues, a large vocabulary of statements can be produced. An iterative closest point algorithm is used to detect these observed motions, by comparing them with an established database of gestures. Finally, we present quantitative data collected from human participants indicating accuracy and performance of our proposed scheme."
]
} |
1709.08945 | 2758825361 | The uncertainty and variability of underwater environment propose the request to control underwater robots in real time and dynamically, especially in the scenarios where human and robots need to work collaboratively in the field. However, the underwater environment imposes harsh restrictions on the application of typical control and communication methods. Considering that gestures are a natural and efficient interactive way for human, we, utilizing convolution neural network, implement a real-time gesture-based recognition system, who can recognize 50 kinds of gestures from images captured by one normal monocular camera, and apply this recognition system in human and underwater robot interaction. We design A Flexible and Extendable Interaction Scheme (AFEIS) through which underwater robots can be programmed in situ underwater by human operators using customized gesture-based sign language. This paper elaborates the design of gesture recognition system and AFEIS, and presents our field trial results when applying this system and scheme on underwater robots. | Besides, in @cite_10 , the authors define a set of syntax in the form similar to natural language as the way to translate gestures to commands accepted by robots. An effort to interact with robots using natural sentences expressed by gestures are made in the paper. However, in practice, we find that the syntax similar to programming language is more intuitive and acceptable by operators. Therefore, in AFEIS, we define a set of syntax similar to programming language but with simpler form. Besides commands the robot to make action through hand gestures, the syntax of AFEIS also supports function definition and variable setting, which are convenient for doing repetitive tasks. | {
"cite_N": [
"@cite_10"
],
"mid": [
"1518898410"
],
"abstract": [
"Underwater environment is characterized by harsh conditions and is difficult to monitor. The CADDY project deals with the development of a companion robot devoted to support and to monitor human operations and activities during the dive. In this scenario the communication and correct reception of messages between the diver and the robot are essential for success of the dive goals. However, the underwater environment poses a set of technical constraints hardly limiting the communication possibilities. For such reasons the solution proposed is to develop a communication language based on the consolidated and standardized diver gestures, commonly employed during professional and recreational dives, thus leading to the definition of a CADDY language, called CADDIAN, and a communication protocol. This article focuses on the creation of the language providing alphabet, syntax and semantics: future work will explain the part of recognition of gestures that is still in progress."
]
} |
1709.08434 | 2950092267 | Cloud storage services have become accessible and used by everyone. Nevertheless, stored data are dependable on the behavior of the cloud servers, and losses and damages often occur. One solution is to regularly audit the cloud servers in order to check the integrity of the stored data. The Dynamic Provable Data Possession scheme with Public Verifiability and Data Privacy presented in ACISP'15 is a straightforward design of such solution. However, this scheme is threatened by several attacks. In this paper, we carefully recall the definition of this scheme as well as explain how its security is dramatically menaced. Moreover, we proposed two new constructions for Dynamic Provable Data Possession scheme with Public Verifiability and Data Privacy based on the scheme presented in ACISP'15, one using Index Hash Tables and one based on Merkle Hash Trees. We show that the two schemes are secure and privacy-preserving in the random oracle model. | A similar concept to PDP, called Proof of Retrievability (POR), was first defined by Juels and Kaliski @cite_26 . It allows a client to verify the integrity of his her data stored at an untrusted server, to correct the possible errors and to retrieve the entire file. Challenge requests are limited and require randomly-valued check blocks that are called sentinels and added to the file. Thereafter, Shacham and Waters @cite_13 proposed two POR schemes built on BLS signatures and pseudo-random functions. They achieved unlimited number of challenge requests and public verifiability, and reduced communication overhead. Subsequent works on distributed systems followed in @cite_23 @cite_12 @cite_0 , as well as POR protocols with data dynamicity in @cite_31 @cite_24 . | {
"cite_N": [
"@cite_26",
"@cite_0",
"@cite_24",
"@cite_23",
"@cite_31",
"@cite_13",
"@cite_12"
],
"mid": [
"2079493184",
"",
"2144140331",
"2058231031",
"",
"89553247",
"2067410255"
],
"abstract": [
"In this paper, we define and explore proofs of retrievability (PORs). A POR scheme enables an archive or back-up service (prover) to produce a concise proof that a user (verifier) can retrieve a target file F, that is, that the archive retains and reliably transmits file data sufficient for the user to recover F in its entirety. A POR may be viewed as a kind of cryptographic proof of knowledge (POK), but one specially designed to handle a large file (or bitstring) F. We explore POR protocols here in which the communication costs, number of memory accesses for the prover, and storage requirements of the user (verifier) are small parameters essentially independent of the length of F. In addition to proposing new, practical POR constructions, we explore implementation considerations and optimizations that bear on previously explored, related schemes. In a POR, unlike a POK, neither the prover nor the verifier need actually have knowledge of F. PORs give rise to a new and unusual security definition whose formulation is another contribution of our work. We view PORs as an important tool for semi-trusted online archives. Existing cryptographic techniques help users ensure the privacy and integrity of files they retrieve. It is also natural, however, for users to want to verify that archives do not delete or modify files prior to retrieval. The goal of a POR is to accomplish these checks without users having to download the files themselves. A POR can also provide quality-of-service guarantees, i.e., show that a file is retrievable within a certain time bound.",
"",
"Proofs of Retrievability (PoR), proposed by Juels and Kaliski in 2007, enable a client to store n file blocks with a cloud server so that later the server can prove possession of all the data in a very efficient manner (i.e., with constant computation and bandwidth). Although many efficient PoR schemes for static data have been constructed, only two dynamic PoR schemes exist. The scheme by Stefanov et. al. (ACSAC 2012) uses a large of amount of client storage and has a large audit cost. The scheme by Cash (EUROCRYPT 2013) is mostly of theoretical interest, as it employs Oblivious RAM (ORAM) as a black box, leading to increased practical overhead (e.g., it requires about 300 times more bandwidth than our construction). We propose a dynamic PoR scheme with constant client storage whose bandwidth cost is comparable to a Merkle hash tree, thus being very practical. Our construction outperforms the constructions of Stefanov et. al. and Cash et. al., both in theory and in practice. Specifically, for n outsourced blocks of beta bits each, writing a block requires beta+O(lambdalog n) bandwidth and O(bet alog n) server computation (lambda is the security parameter). Audits are also very efficient, requiring beta+O(lambda^2log n) bandwidth. We also show how to make our scheme publicly verifiable, providing the first dynamic PoR scheme with such a property. We finally provide a very efficient implementation of our scheme.",
"A proof of retrievability (POR) is a compact proof by a file system (prover) to a client (verifier) that a target file F is intact, in the sense that the client can fully recover it. As PORs incur lower communication complexity than transmission of F itself, they are an attractive building block for high-assurance remote storage systems. In this paper, we propose a theoretical framework for the design of PORs. Our framework improves the previously proposed POR constructions of Juels-Kaliski and Shacham-Waters, and also sheds light on the conceptual limitations of previous theoretical models for PORs. It supports a fully Byzantine adversarial model, carrying only the restriction---fundamental to all PORs---that the adversary's error rate be bounded when the client seeks to extract F. We propose a new variant on the Juels-Kaliski protocol and describe a prototype implementation. We demonstrate practical encoding even for files F whose size exceeds that of client main memory.",
"",
"In a proof-of-retrievability system, a data storage center convinces a verifier that he is actually storing all of a client's data. The central challenge is to build systems that are both efficient and provably secure--that is, it should be possible to extract the client's data from any prover that passes a verification check. In this paper, we give the first proof-of-retrievability schemes with full proofs of security against arbitrary adversaries in the strongest model, that of Juels and Kaliski. Our first scheme, built from BLS signatures and secure in the random oracle model, has the shortest query and response of any proof-of-retrievability with public verifiability. Our second scheme, which builds elegantly on pseudorandom functions (PRFs) and is secure in the standard model, has the shortest response of any proof-of-retrievability scheme with private verifiability (but a longer query). Both schemes rely on homomorphic properties to aggregate a proof into one small authenticator value.",
"We introduce HAIL (High-Availability and Integrity Layer), a distributed cryptographic system that allows a set of servers to prove to a client that a stored file is intact and retrievable. HAIL strengthens, formally unifies, and streamlines distinct approaches from the cryptographic and distributed-systems communities. Proofs in HAIL are efficiently computable by servers and highly compact---typically tens or hundreds of bytes, irrespective of file size. HAIL cryptographically verifies and reactively reallocates file shares. It is robust against an active, mobile adversary, i.e., one that may progressively corrupt the full set of servers. We propose a strong, formal adversarial model for HAIL, and rigorous analysis and parameter choices. We show how HAIL improves on the security and efficiency of existing tools, like Proofs of Retrievability (PORs) deployed on individual servers. We also report on a prototype implementation."
]
} |
1709.08325 | 2760299260 | Feature extraction and matching are two crucial components in person Re-Identification (ReID). The large pose deformations and the complex view variations exhibited by the captured person images significantly increase the difficulty of learning and matching of the features from person images. To overcome these difficulties, in this work we propose a Pose-driven Deep Convolutional (PDC) model to learn improved feature extraction and matching models from end to end. Our deep architecture explicitly leverages the human part cues to alleviate the pose variations and learn robust feature representations from both the global image and different local parts. To match the features from global human body and local body parts, a pose driven feature weighting sub-network is further designed to learn adaptive feature fusions. Extensive experimental analyses and results on three popular datasets demonstrate significant performance improvements of our model over all published state-of-the-art methods. | Deep learning is commonly used to either learn a person's representation or the distance metric. When handling a pair of person images, existing deep learning methods usually learn feature representations of each person by using a deep matching function from convolutional features @cite_80 @cite_51 @cite_92 @cite_95 or from the Fully Connected (FC) features @cite_15 @cite_14 @cite_97 . Apart from deep metric learning methods, some algorithms first learn image representations directly with the Triplet Loss or the Siamese Contrastive Loss, then utilize Euclidean distance for comparison @cite_47 @cite_10 @cite_11 @cite_1 . Wang al @cite_47 use a joint learning framework to unify single-image representation and cross-image representation using a doublet or triplet CNN. Shi al @cite_14 propose a moderate positive mining method to use deep distance metric learning for person ReID. Another novel method @cite_14 learns deep attributes feature for ReID with semi-supervised learning. Xiao al @cite_92 train one network with several person ReID datasets using a Domain Guided Dropout algorithm. | {
"cite_N": [
"@cite_47",
"@cite_14",
"@cite_92",
"@cite_97",
"@cite_10",
"@cite_1",
"@cite_95",
"@cite_80",
"@cite_15",
"@cite_51",
"@cite_11"
],
"mid": [
"",
"2519373641",
"2342611082",
"",
"",
"",
"2953350812",
"1928419358",
"",
"1982925187",
"1971955426"
],
"abstract": [
"",
"Person re-identification is challenging due to the large variations of pose, illumination, occlusion and camera view. Owing to these variations, the pedestrian data is distributed as highly-curved manifolds in the feature space, despite the current convolutional neural networks (CNN)’s capability of feature extraction. However, the distribution is unknown, so it is difficult to use the geodesic distance when comparing two samples. In practice, the current deep embedding methods use the Euclidean distance for the training and test. On the other hand, the manifold learning methods suggest to use the Euclidean distance in the local range, combining with the graphical relationship between samples, for approximating the geodesic distance. From this point of view, selecting suitable positive (i.e. intra-class) training samples within a local range is critical for training the CNN embedding, especially when the data has large intra-class variations. In this paper, we propose a novel moderate positive sample mining method to train robust CNN for person re-identification, dealing with the problem of large variation. In addition, we improve the learning by a metric weight constraint, so that the learned metric has a better generalization ability. Experiments show that these two strategies are effective in learning robust deep metrics for person re-identification, and accordingly our deep model significantly outperforms the state-of-the-art methods on several benchmarks of person re-identification. Therefore, the study presented in this paper may be useful in inspiring new designs of deep models for person re-identification.",
"Learning generic and robust feature representations with data from multiple domains for the same problem is of great value, especially for the problems that have multiple datasets but none of them are large enough to provide abundant data variations. In this work, we present a pipeline for learning deep feature representations from multiple domains with Convolutional Neural Networks (CNNs). When training a CNN with data from all the domains, some neurons learn representations shared across several domains, while some others are effective only for a specific one. Based on this important observation, we propose a Domain Guided Dropout algorithm to improve the feature learning procedure. Experiments show the effectiveness of our pipeline and the proposed algorithm. Our methods on the person re-identification problem outperform stateof-the-art methods on multiple datasets by large margins.",
"",
"",
"",
"Person re-identification (Re-ID) poses a unique challenge to deep learning: how to learn a deep model with millions of parameters on a small training set of few or no labels. In this paper, a number of deep transfer learning models are proposed to address the data sparsity problem. First, a deep network architecture is designed which differs from existing deep Re-ID models in that (a) it is more suitable for transferring representations learned from large image classification datasets, and (b) classification loss and verification loss are combined, each of which adopts a different dropout strategy. Second, a two-stepped fine-tuning strategy is developed to transfer knowledge from auxiliary datasets. Third, given an unlabelled Re-ID dataset, a novel unsupervised deep transfer learning model is developed based on co-training. The proposed models outperform the state-of-the-art deep Re-ID models by large margins: we achieve Rank-1 accuracy of 85.4 , 83.7 and 56.3 on CUHK03, Market1501, and VIPeR respectively, whilst on VIPeR, our unsupervised model (45.1 ) beats most supervised models.",
"In this work, we propose a method for simultaneously learning features and a corresponding similarity metric for person re-identification. We present a deep convolutional architecture with layers specially designed to address the problem of re-identification. Given a pair of images as input, our network outputs a similarity value indicating whether the two input images depict the same person. Novel elements of our architecture include a layer that computes cross-input neighborhood differences, which capture local relationships between the two input images based on mid-level features from each input image. A high-level summary of the outputs of this layer is computed by a layer of patch summary features, which are then spatially integrated in subsequent layers. Our method significantly outperforms the state of the art on both a large data set (CUHK03) and a medium-sized data set (CUHK01), and is resistant to over-fitting. We also demonstrate that by initially training on an unrelated large data set before fine-tuning on a small target data set, our network can achieve results comparable to the state of the art even on a small data set (VIPeR).",
"",
"Person re-identification is to match pedestrian images from disjoint camera views detected by pedestrian detectors. Challenges are presented in the form of complex variations of lightings, poses, viewpoints, blurring effects, image resolutions, camera settings, occlusions and background clutter across camera views. In addition, misalignment introduced by the pedestrian detector will affect most existing person re-identification methods that use manually cropped pedestrian images and assume perfect detection. In this paper, we propose a novel filter pairing neural network (FPNN) to jointly handle misalignment, photometric and geometric transforms, occlusions and background clutter. All the key components are jointly optimized to maximize the strength of each component when cooperating with others. In contrast to existing works that use handcrafted features, our method automatically learns features optimal for the re-identification task from data. The learned filter pairs encode photometric transforms. Its deep architecture makes it possible to model a mixture of complex photometric and geometric transforms. We build the largest benchmark re-id dataset with 13, 164 images of 1, 360 pedestrians. Unlike existing datasets, which only provide manually cropped pedestrian images, our dataset provides automatically detected bounding boxes for evaluation close to practical applications. Our neural network significantly outperforms state-of-the-art methods on this dataset.",
"Identifying the same individual across different scenes is an important yet difficult task in intelligent video surveillance. Its main difficulty lies in how to preserve similarity of the same person against large appearance and structure variation while discriminating different individuals. In this paper, we present a scalable distance driven feature learning framework based on the deep neural network for person re-identification, and demonstrate its effectiveness to handle the existing challenges. Specifically, given the training images with the class labels (person IDs), we first produce a large number of triplet units, each of which contains three images, i.e. one person with a matched reference and a mismatched reference. Treating the units as the input, we build the convolutional neural network to generate the layered representations, and follow with the L 2 distance metric. By means of parameter optimization, our framework tends to maximize the relative distance between the matched pair and the mismatched pair for each triplet unit. Moreover, a nontrivial issue arising with the framework is that the triplet organization cubically enlarges the number of training triplets, as one image can be involved into several triplet units. To overcome this problem, we develop an effective triplet generation scheme and an optimized gradient descent algorithm, making the computational load mainly depend on the number of original images instead of the number of triplets. On several challenging databases, our approach achieves very promising results and outperforms other state-of-the-art approaches. HighlightsWe present a novel feature learning framework for person re-identification.Our framework is based on the maximum relative distance comparison.The learning algorithm is scalable to process large amount of data.We demonstrate superior performances over other state-of-the-arts."
]
} |
1709.08325 | 2760299260 | Feature extraction and matching are two crucial components in person Re-Identification (ReID). The large pose deformations and the complex view variations exhibited by the captured person images significantly increase the difficulty of learning and matching of the features from person images. To overcome these difficulties, in this work we propose a Pose-driven Deep Convolutional (PDC) model to learn improved feature extraction and matching models from end to end. Our deep architecture explicitly leverages the human part cues to alleviate the pose variations and learn robust feature representations from both the global image and different local parts. To match the features from global human body and local body parts, a pose driven feature weighting sub-network is further designed to learn adaptive feature fusions. Extensive experimental analyses and results on three popular datasets demonstrate significant performance improvements of our model over all published state-of-the-art methods. | : For the dataset, we compare our PDC with some recent methods, including distance metric learning methods: MLAPG @cite_66 , LOMO + XQDA @cite_85 , BoW+HS @cite_28 , WARCA @cite_42 , LDNS @cite_7 , feature extraction method: GOG @cite_33 and deep learning based methods: IDLA @cite_80 , PersonNet @cite_3 , DGDropout @cite_92 , SI+CI @cite_47 , Gate S-CNN @cite_1 , LSTM S-CNN @cite_62 , EDM @cite_14 , PIE @cite_93 and Spindle @cite_78 . We conduct experiments on both the detected dataset and the labeled dataset. Experimental results are presented in Table and Table . | {
"cite_N": [
"@cite_62",
"@cite_14",
"@cite_33",
"@cite_7",
"@cite_78",
"@cite_28",
"@cite_92",
"@cite_85",
"@cite_42",
"@cite_1",
"@cite_3",
"@cite_93",
"@cite_80",
"@cite_47",
"@cite_66"
],
"mid": [
"",
"2519373641",
"",
"2300840837",
"",
"",
"2342611082",
"1949591461",
"2289130514",
"",
"2259687230",
"2583528645",
"1928419358",
"",
""
],
"abstract": [
"",
"Person re-identification is challenging due to the large variations of pose, illumination, occlusion and camera view. Owing to these variations, the pedestrian data is distributed as highly-curved manifolds in the feature space, despite the current convolutional neural networks (CNN)’s capability of feature extraction. However, the distribution is unknown, so it is difficult to use the geodesic distance when comparing two samples. In practice, the current deep embedding methods use the Euclidean distance for the training and test. On the other hand, the manifold learning methods suggest to use the Euclidean distance in the local range, combining with the graphical relationship between samples, for approximating the geodesic distance. From this point of view, selecting suitable positive (i.e. intra-class) training samples within a local range is critical for training the CNN embedding, especially when the data has large intra-class variations. In this paper, we propose a novel moderate positive sample mining method to train robust CNN for person re-identification, dealing with the problem of large variation. In addition, we improve the learning by a metric weight constraint, so that the learned metric has a better generalization ability. Experiments show that these two strategies are effective in learning robust deep metrics for person re-identification, and accordingly our deep model significantly outperforms the state-of-the-art methods on several benchmarks of person re-identification. Therefore, the study presented in this paper may be useful in inspiring new designs of deep models for person re-identification.",
"",
"Most existing person re-identification (re-id) methods focus on learning the optimal distance metrics across camera views. Typically a person's appearance is represented using features of thousands of dimensions, whilst only hundreds of training samples are available due to the difficulties in collecting matched training images. With the number of training samples much smaller than the feature dimension, the existing methods thus face the classic small sample size (SSS) problem and have to resort to dimensionality reduction techniques and or matrix regularisation, which lead to loss of discriminative power. In this work, we propose to overcome the SSS problem in re-id distance metric learning by matching people in a discriminative null space of the training data. In this null space, images of the same person are collapsed into a single point thus minimising the within-class scatter to the extreme and maximising the relative between-class separation simultaneously. Importantly, it has a fixed dimension, a closed-form solution and is very efficient to compute. Extensive experiments carried out on five person re-identification benchmarks including VIPeR, PRID2011, CUHK01, CUHK03 and Market1501 show that such a simple approach beats the state-of-the-art alternatives, often by a big margin.",
"",
"",
"Learning generic and robust feature representations with data from multiple domains for the same problem is of great value, especially for the problems that have multiple datasets but none of them are large enough to provide abundant data variations. In this work, we present a pipeline for learning deep feature representations from multiple domains with Convolutional Neural Networks (CNNs). When training a CNN with data from all the domains, some neurons learn representations shared across several domains, while some others are effective only for a specific one. Based on this important observation, we propose a Domain Guided Dropout algorithm to improve the feature learning procedure. Experiments show the effectiveness of our pipeline and the proposed algorithm. Our methods on the person re-identification problem outperform stateof-the-art methods on multiple datasets by large margins.",
"Person re-identification is an important technique towards automatic search of a person's presence in a surveillance video. Two fundamental problems are critical for person re-identification, feature representation and metric learning. An effective feature representation should be robust to illumination and viewpoint changes, and a discriminant metric should be learned to match various person images. In this paper, we propose an effective feature representation called Local Maximal Occurrence (LOMO), and a subspace and metric learning method called Cross-view Quadratic Discriminant Analysis (XQDA). The LOMO feature analyzes the horizontal occurrence of local features, and maximizes the occurrence to make a stable representation against viewpoint changes. Besides, to handle illumination variations, we apply the Retinex transform and a scale invariant texture operator. To learn a discriminant metric, we propose to learn a discriminant low dimensional subspace by cross-view quadratic discriminant analysis, and simultaneously, a QDA metric is learned on the derived subspace. We also present a practical computation method for XQDA, as well as its regularization. Experiments on four challenging person re-identification databases, VIPeR, QMUL GRID, CUHK Campus, and CUHK03, show that the proposed method improves the state-of-the-art rank-1 identification rates by 2.2 , 4.88 , 28.91 , and 31.55 on the four databases, respectively.",
"We are interested in the large-scale learning of Mahalanobis distances, with a particular focus on person re-identification. We propose a metric learning formulation called Weighted Approximate Rank Component Analysis (WARCA). WARCA optimizes the precision at top ranks by combining the WARP loss with a regularizer that favors orthonormal linear mappings and avoids rank-deficient embeddings. Using this new regularizer allows us to adapt the large-scale WSABIE procedure and to leverage the Adam stochastic optimization algorithm, which results in an algorithm that scales gracefully to very large data-sets. Also, we derive a kernelized version which allows to take advantage of state-of-the-art features for re-identification when data-set size permits kernel computation. Benchmarks on recent and standard re-identification data-sets show that our method beats existing state-of-the-art techniques both in terms of accuracy and speed. We also provide experimental analysis to shade lights on the properties of the regularizer we use, and how it improves performance.",
"",
"In this paper, we propose a deep end-to-end neu- ral network to simultaneously learn high-level features and a corresponding similarity metric for person re-identification. The network takes a pair of raw RGB images as input, and outputs a similarity value indicating whether the two input images depict the same person. A layer of computing neighborhood range differences across two input images is employed to capture local relationship between patches. This operation is to seek a robust feature from input images. By increasing the depth to 10 weight layers and using very small (3 @math 3) convolution filters, our architecture achieves a remarkable improvement on the prior-art configurations. Meanwhile, an adaptive Root- Mean-Square (RMSProp) gradient decent algorithm is integrated into our architecture, which is beneficial to deep nets. Our method consistently outperforms state-of-the-art on two large datasets (CUHK03 and Market-1501), and a medium-sized data set (CUHK01).",
"Pedestrian misalignment, which mainly arises from detector errors and pose variations, is a critical problem for a robust person re-identification (re-ID) system. With bad alignment, the background noise will significantly compromise the feature learning and matching process. To address this problem, this paper introduces the pose invariant embedding (PIE) as a pedestrian descriptor. First, in order to align pedestrians to a standard pose, the PoseBox structure is introduced, which is generated through pose estimation followed by affine transformations. Second, to reduce the impact of pose estimation errors and information loss during PoseBox construction, we design a PoseBox fusion (PBF) CNN architecture that takes the original image, the PoseBox, and the pose estimation confidence as input. The proposed PIE descriptor is thus defined as the fully connected layer of the PBF network for the retrieval task. Experiments are conducted on the Market-1501, CUHK03, and VIPeR datasets. We show that PoseBox alone yields decent re-ID accuracy and that when integrated in the PBF network, the learned PIE descriptor produces competitive performance compared with the state-of-the-art approaches.",
"In this work, we propose a method for simultaneously learning features and a corresponding similarity metric for person re-identification. We present a deep convolutional architecture with layers specially designed to address the problem of re-identification. Given a pair of images as input, our network outputs a similarity value indicating whether the two input images depict the same person. Novel elements of our architecture include a layer that computes cross-input neighborhood differences, which capture local relationships between the two input images based on mid-level features from each input image. A high-level summary of the outputs of this layer is computed by a layer of patch summary features, which are then spatially integrated in subsequent layers. Our method significantly outperforms the state of the art on both a large data set (CUHK03) and a medium-sized data set (CUHK01), and is resistant to over-fitting. We also demonstrate that by initially training on an unrelated large data set before fine-tuning on a small target data set, our network can achieve results comparable to the state of the art even on a small data set (VIPeR).",
"",
""
]
} |
1709.08325 | 2760299260 | Feature extraction and matching are two crucial components in person Re-Identification (ReID). The large pose deformations and the complex view variations exhibited by the captured person images significantly increase the difficulty of learning and matching of the features from person images. To overcome these difficulties, in this work we propose a Pose-driven Deep Convolutional (PDC) model to learn improved feature extraction and matching models from end to end. Our deep architecture explicitly leverages the human part cues to alleviate the pose variations and learn robust feature representations from both the global image and different local parts. To match the features from global human body and local body parts, a pose driven feature weighting sub-network is further designed to learn adaptive feature fusions. Extensive experimental analyses and results on three popular datasets demonstrate significant performance improvements of our model over all published state-of-the-art methods. | : On , the compared works that learn distance metrics for person ReID include LOMO + XQDA @cite_85 , BoW+Kissme @cite_28 , WARCA @cite_42 , LDNS @cite_7 , TMA @cite_86 and HVIL @cite_37 . Compared works based on deep learning are PersonNet @cite_3 , Gate S-CNN @cite_1 , LSTM S-CNN @cite_62 , PIE @cite_93 and Spindle @cite_78 . DGDropout @cite_92 does not report performance on Market1501. So we implemented DGDroput and show experimental results in Table . | {
"cite_N": [
"@cite_37",
"@cite_62",
"@cite_7",
"@cite_78",
"@cite_28",
"@cite_92",
"@cite_85",
"@cite_42",
"@cite_1",
"@cite_3",
"@cite_86",
"@cite_93"
],
"mid": [
"",
"",
"2300840837",
"",
"",
"2342611082",
"1949591461",
"2289130514",
"",
"2259687230",
"",
"2583528645"
],
"abstract": [
"",
"",
"Most existing person re-identification (re-id) methods focus on learning the optimal distance metrics across camera views. Typically a person's appearance is represented using features of thousands of dimensions, whilst only hundreds of training samples are available due to the difficulties in collecting matched training images. With the number of training samples much smaller than the feature dimension, the existing methods thus face the classic small sample size (SSS) problem and have to resort to dimensionality reduction techniques and or matrix regularisation, which lead to loss of discriminative power. In this work, we propose to overcome the SSS problem in re-id distance metric learning by matching people in a discriminative null space of the training data. In this null space, images of the same person are collapsed into a single point thus minimising the within-class scatter to the extreme and maximising the relative between-class separation simultaneously. Importantly, it has a fixed dimension, a closed-form solution and is very efficient to compute. Extensive experiments carried out on five person re-identification benchmarks including VIPeR, PRID2011, CUHK01, CUHK03 and Market1501 show that such a simple approach beats the state-of-the-art alternatives, often by a big margin.",
"",
"",
"Learning generic and robust feature representations with data from multiple domains for the same problem is of great value, especially for the problems that have multiple datasets but none of them are large enough to provide abundant data variations. In this work, we present a pipeline for learning deep feature representations from multiple domains with Convolutional Neural Networks (CNNs). When training a CNN with data from all the domains, some neurons learn representations shared across several domains, while some others are effective only for a specific one. Based on this important observation, we propose a Domain Guided Dropout algorithm to improve the feature learning procedure. Experiments show the effectiveness of our pipeline and the proposed algorithm. Our methods on the person re-identification problem outperform stateof-the-art methods on multiple datasets by large margins.",
"Person re-identification is an important technique towards automatic search of a person's presence in a surveillance video. Two fundamental problems are critical for person re-identification, feature representation and metric learning. An effective feature representation should be robust to illumination and viewpoint changes, and a discriminant metric should be learned to match various person images. In this paper, we propose an effective feature representation called Local Maximal Occurrence (LOMO), and a subspace and metric learning method called Cross-view Quadratic Discriminant Analysis (XQDA). The LOMO feature analyzes the horizontal occurrence of local features, and maximizes the occurrence to make a stable representation against viewpoint changes. Besides, to handle illumination variations, we apply the Retinex transform and a scale invariant texture operator. To learn a discriminant metric, we propose to learn a discriminant low dimensional subspace by cross-view quadratic discriminant analysis, and simultaneously, a QDA metric is learned on the derived subspace. We also present a practical computation method for XQDA, as well as its regularization. Experiments on four challenging person re-identification databases, VIPeR, QMUL GRID, CUHK Campus, and CUHK03, show that the proposed method improves the state-of-the-art rank-1 identification rates by 2.2 , 4.88 , 28.91 , and 31.55 on the four databases, respectively.",
"We are interested in the large-scale learning of Mahalanobis distances, with a particular focus on person re-identification. We propose a metric learning formulation called Weighted Approximate Rank Component Analysis (WARCA). WARCA optimizes the precision at top ranks by combining the WARP loss with a regularizer that favors orthonormal linear mappings and avoids rank-deficient embeddings. Using this new regularizer allows us to adapt the large-scale WSABIE procedure and to leverage the Adam stochastic optimization algorithm, which results in an algorithm that scales gracefully to very large data-sets. Also, we derive a kernelized version which allows to take advantage of state-of-the-art features for re-identification when data-set size permits kernel computation. Benchmarks on recent and standard re-identification data-sets show that our method beats existing state-of-the-art techniques both in terms of accuracy and speed. We also provide experimental analysis to shade lights on the properties of the regularizer we use, and how it improves performance.",
"",
"In this paper, we propose a deep end-to-end neu- ral network to simultaneously learn high-level features and a corresponding similarity metric for person re-identification. The network takes a pair of raw RGB images as input, and outputs a similarity value indicating whether the two input images depict the same person. A layer of computing neighborhood range differences across two input images is employed to capture local relationship between patches. This operation is to seek a robust feature from input images. By increasing the depth to 10 weight layers and using very small (3 @math 3) convolution filters, our architecture achieves a remarkable improvement on the prior-art configurations. Meanwhile, an adaptive Root- Mean-Square (RMSProp) gradient decent algorithm is integrated into our architecture, which is beneficial to deep nets. Our method consistently outperforms state-of-the-art on two large datasets (CUHK03 and Market-1501), and a medium-sized data set (CUHK01).",
"",
"Pedestrian misalignment, which mainly arises from detector errors and pose variations, is a critical problem for a robust person re-identification (re-ID) system. With bad alignment, the background noise will significantly compromise the feature learning and matching process. To address this problem, this paper introduces the pose invariant embedding (PIE) as a pedestrian descriptor. First, in order to align pedestrians to a standard pose, the PoseBox structure is introduced, which is generated through pose estimation followed by affine transformations. Second, to reduce the impact of pose estimation errors and information loss during PoseBox construction, we design a PoseBox fusion (PBF) CNN architecture that takes the original image, the PoseBox, and the pose estimation confidence as input. The proposed PIE descriptor is thus defined as the fully connected layer of the PBF network for the retrieval task. Experiments are conducted on the Market-1501, CUHK03, and VIPeR datasets. We show that PoseBox alone yields decent re-ID accuracy and that when integrated in the PBF network, the learned PIE descriptor produces competitive performance compared with the state-of-the-art approaches."
]
} |
1709.08325 | 2760299260 | Feature extraction and matching are two crucial components in person Re-Identification (ReID). The large pose deformations and the complex view variations exhibited by the captured person images significantly increase the difficulty of learning and matching of the features from person images. To overcome these difficulties, in this work we propose a Pose-driven Deep Convolutional (PDC) model to learn improved feature extraction and matching models from end to end. Our deep architecture explicitly leverages the human part cues to alleviate the pose variations and learn robust feature representations from both the global image and different local parts. To match the features from global human body and local body parts, a pose driven feature weighting sub-network is further designed to learn adaptive feature fusions. Extensive experimental analyses and results on three popular datasets demonstrate significant performance improvements of our model over all published state-of-the-art methods. | : We also evaluate our method by comparing it with several existing methods on . The compared methods include distance metric learning ones: MLAPG @cite_66 , LOMO + XQDA @cite_85 , BoW @cite_28 , WARCA @cite_42 and LDNS @cite_7 , and deep learning based ones: IDLA @cite_80 , DGDropout @cite_92 , SI+CI @cite_47 , Gate S-CNN @cite_1 , LSTM S-CNN @cite_62 , MTL-LORAE @cite_70 and Spindle @cite_78 . | {
"cite_N": [
"@cite_62",
"@cite_7",
"@cite_78",
"@cite_28",
"@cite_92",
"@cite_70",
"@cite_85",
"@cite_42",
"@cite_1",
"@cite_80",
"@cite_47",
"@cite_66"
],
"mid": [
"",
"2300840837",
"",
"",
"2342611082",
"2203864774",
"1949591461",
"2289130514",
"",
"1928419358",
"",
""
],
"abstract": [
"",
"Most existing person re-identification (re-id) methods focus on learning the optimal distance metrics across camera views. Typically a person's appearance is represented using features of thousands of dimensions, whilst only hundreds of training samples are available due to the difficulties in collecting matched training images. With the number of training samples much smaller than the feature dimension, the existing methods thus face the classic small sample size (SSS) problem and have to resort to dimensionality reduction techniques and or matrix regularisation, which lead to loss of discriminative power. In this work, we propose to overcome the SSS problem in re-id distance metric learning by matching people in a discriminative null space of the training data. In this null space, images of the same person are collapsed into a single point thus minimising the within-class scatter to the extreme and maximising the relative between-class separation simultaneously. Importantly, it has a fixed dimension, a closed-form solution and is very efficient to compute. Extensive experiments carried out on five person re-identification benchmarks including VIPeR, PRID2011, CUHK01, CUHK03 and Market1501 show that such a simple approach beats the state-of-the-art alternatives, often by a big margin.",
"",
"",
"Learning generic and robust feature representations with data from multiple domains for the same problem is of great value, especially for the problems that have multiple datasets but none of them are large enough to provide abundant data variations. In this work, we present a pipeline for learning deep feature representations from multiple domains with Convolutional Neural Networks (CNNs). When training a CNN with data from all the domains, some neurons learn representations shared across several domains, while some others are effective only for a specific one. Based on this important observation, we propose a Domain Guided Dropout algorithm to improve the feature learning procedure. Experiments show the effectiveness of our pipeline and the proposed algorithm. Our methods on the person re-identification problem outperform stateof-the-art methods on multiple datasets by large margins.",
"We propose a novel Multi-Task Learning with Low Rank Attribute Embedding (MTL-LORAE) framework for person re-identification. Re-identifications from multiple cameras are regarded as related tasks to exploit shared information to improve re-identification accuracy. Both low level features and semantic data-driven attributes are utilized. Since attributes are generally correlated, we introduce a low rank attribute embedding into the MTL formulation to embed original binary attributes to a continuous attribute space, where incorrect and incomplete attributes are rectified and recovered to better describe people. The learning objective function consists of a quadratic loss regarding class labels and an attribute embedding error, which is solved by an alternating optimization procedure. Experiments on three person re-identification datasets have demonstrated that MTL-LORAE outperforms existing approaches by a large margin and produces state-of-the-art results.",
"Person re-identification is an important technique towards automatic search of a person's presence in a surveillance video. Two fundamental problems are critical for person re-identification, feature representation and metric learning. An effective feature representation should be robust to illumination and viewpoint changes, and a discriminant metric should be learned to match various person images. In this paper, we propose an effective feature representation called Local Maximal Occurrence (LOMO), and a subspace and metric learning method called Cross-view Quadratic Discriminant Analysis (XQDA). The LOMO feature analyzes the horizontal occurrence of local features, and maximizes the occurrence to make a stable representation against viewpoint changes. Besides, to handle illumination variations, we apply the Retinex transform and a scale invariant texture operator. To learn a discriminant metric, we propose to learn a discriminant low dimensional subspace by cross-view quadratic discriminant analysis, and simultaneously, a QDA metric is learned on the derived subspace. We also present a practical computation method for XQDA, as well as its regularization. Experiments on four challenging person re-identification databases, VIPeR, QMUL GRID, CUHK Campus, and CUHK03, show that the proposed method improves the state-of-the-art rank-1 identification rates by 2.2 , 4.88 , 28.91 , and 31.55 on the four databases, respectively.",
"We are interested in the large-scale learning of Mahalanobis distances, with a particular focus on person re-identification. We propose a metric learning formulation called Weighted Approximate Rank Component Analysis (WARCA). WARCA optimizes the precision at top ranks by combining the WARP loss with a regularizer that favors orthonormal linear mappings and avoids rank-deficient embeddings. Using this new regularizer allows us to adapt the large-scale WSABIE procedure and to leverage the Adam stochastic optimization algorithm, which results in an algorithm that scales gracefully to very large data-sets. Also, we derive a kernelized version which allows to take advantage of state-of-the-art features for re-identification when data-set size permits kernel computation. Benchmarks on recent and standard re-identification data-sets show that our method beats existing state-of-the-art techniques both in terms of accuracy and speed. We also provide experimental analysis to shade lights on the properties of the regularizer we use, and how it improves performance.",
"",
"In this work, we propose a method for simultaneously learning features and a corresponding similarity metric for person re-identification. We present a deep convolutional architecture with layers specially designed to address the problem of re-identification. Given a pair of images as input, our network outputs a similarity value indicating whether the two input images depict the same person. Novel elements of our architecture include a layer that computes cross-input neighborhood differences, which capture local relationships between the two input images based on mid-level features from each input image. A high-level summary of the outputs of this layer is computed by a layer of patch summary features, which are then spatially integrated in subsequent layers. Our method significantly outperforms the state of the art on both a large data set (CUHK03) and a medium-sized data set (CUHK01), and is resistant to over-fitting. We also demonstrate that by initially training on an unrelated large data set before fine-tuning on a small target data set, our network can achieve results comparable to the state of the art even on a small data set (VIPeR).",
"",
""
]
} |
1709.08600 | 2760228341 | Many real-world applications require automated data annotation, such as identifying tissue origins based on gene expressions and classifying images into semantic categories. Annotation classes are often numerous and subject to changes over time, and annotating examples has become the major bottleneck for supervised learning methods. In science and other high-value domains, large repositories of data samples are often available, together with two sources of organic supervision: a lexicon for the annotation classes, and text descriptions that accompany some data samples. Distant supervision has emerged as a promising paradigm for exploiting such indirect supervision by automatically annotating examples where the text description contains a class mention in the lexicon. However, due to linguistic variations and ambiguities, such training data is inherently noisy, which limits the accuracy of this approach. In this paper, we introduce an auxiliary natural language processing system for the text modality, and incorporate co-training to reduce noise and augment signal in distant supervision. Without using any manually labeled data, our EZLearn system learned to accurately annotate data samples in functional genomics and scientific figure comprehension, substantially outperforming state-of-the-art supervised methods trained on tens of thousands of annotated examples. | In semi-supervised learning, direct supervision is augmented by annotating unlabeled examples using either a learned model @cite_8 @cite_35 or similarity between examples @cite_12 . It is an effective paradigm to refine learned models, but still requires initialization with sufficient labeled examples for all classes. Zero-shot learning or few-shot learning relax the requirement of labeled examples for some classes, but still need to have sufficient labeled examples for classes @cite_36 @cite_18 . In this regard, they bear resemblance with domain adaptation @cite_15 @cite_20 and transfer learning @cite_3 @cite_34 . Zero-shot learning also faces additional challenges such as novelty detection to distinguish between known classes and new ones. | {
"cite_N": [
"@cite_35",
"@cite_18",
"@cite_8",
"@cite_36",
"@cite_3",
"@cite_34",
"@cite_15",
"@cite_20",
"@cite_12"
],
"mid": [
"2048679005",
"2950276680",
"2037603696",
"2150295085",
"",
"2122922389",
"2163302275",
"2120354757",
"1630959083"
],
"abstract": [
"We consider the problem of using a large unlabeled sample to boost performance of a learning algorit,hrn when only a small set of labeled examples is available. In particular, we consider a problem setting motivated by the task of learning to classify web pages, in which the description of each example can be partitioned into two distinct views. For example, the description of a web page can be partitioned into the words occurring on that page, and the words occurring in hyperlinks t,hat point to that page. We assume that either view of the example would be sufficient for learning if we had enough labeled data, but our goal is to use both views together to allow inexpensive unlabeled data to augment, a much smaller set of labeled examples. Specifically, the presence of two distinct views of each example suggests strategies in which two learning algorithms are trained separately on each view, and then each algorithm’s predictions on new unlabeled examples are used to enlarge the training set of the other. Our goal in this paper is to provide a PAC-style analysis for this setting, and, more broadly, a PAC-style framework for the general problem of learning from both labeled and unlabeled data. We also provide empirical results on real web-page data indicating that this use of unlabeled examples can lead to significant improvement of hypotheses in practice. *This research was supported in part by the DARPA HPKB program under contract F30602-97-1-0215 and by NSF National Young investigator grant CCR-9357793. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. TO copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and or a fee. COLT 98 Madison WI USA Copyright ACM 1998 l-58113-057--0 98 7... 5.00 92 Tom Mitchell School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213-3891 mitchell+@cs.cmu.edu",
"This work introduces a model that can recognize objects in images even if no training data is available for the objects. The only necessary knowledge about the unseen categories comes from unsupervised large text corpora. In our zero-shot framework distributional information in language can be seen as spanning a semantic basis for understanding what objects look like. Most previous zero-shot learning models can only differentiate between unseen classes. In contrast, our model can both obtain state of the art performance on classes that have thousands of training images and obtain reasonable performance on unseen classes. This is achieved by first using outlier detection in the semantic space and then two separate recognition models. Furthermore, our model does not require any manually defined semantic features for either words or images.",
"",
"We consider the problem of zero-shot learning, where the goal is to learn a classifier f : X → Y that must predict novel values of Y that were omitted from the training set. To achieve this, we define the notion of a semantic output code classifier (SOC) which utilizes a knowledge base of semantic properties of Y to extrapolate to novel classes. We provide a formalism for this type of classifier and study its theoretical properties in a PAC framework, showing conditions under which the classifier can accurately predict novel classes. As a case study, we build a SOC classifier for a neural decoding task and show that it can often predict words that people are thinking about from functional magnetic resonance images (fMRI) of their neural activity, even without training examples for those words.",
"",
"We present a new machine learning framework called \"self-taught learning\" for using unlabeled data in supervised classification tasks. We do not assume that the unlabeled data follows the same class labels or generative distribution as the labeled data. Thus, we would like to use a large number of unlabeled images (or audio samples, or text documents) randomly downloaded from the Internet to improve performance on a given image (or audio, or text) classification task. Such unlabeled data is significantly easier to obtain than in typical semi-supervised or transfer learning settings, making self-taught learning widely applicable to many practical learning problems. We describe an approach to self-taught learning that uses sparse coding to construct higher-level features using the unlabeled data. These features form a succinct input representation and significantly improve classification performance. When using an SVM for classification, we further show how a Fisher kernel can be learned for this representation.",
"Automatic sentiment classification has been extensively studied and applied in recent years. However, sentiment is expressed differently in different domains, and annotating corpora for every possible domain of interest is impractical. We investigate domain adaptation for sentiment classifiers, focusing on online reviews for different types of products. First, we extend to sentiment classification the recently-proposed structural correspondence learning (SCL) algorithm, reducing the relative error due to adaptation between domains by an average of 30 over the original SCL algorithm and 46 over a supervised baseline. Second, we identify a measure of domain similarity that correlates well with the potential for adaptation of a classifier from one domain to another. This measure could for instance be used to select a small set of domains to annotate whose trained classifiers would transfer well to many other domains.",
"We describe an approach to domain adaptation that is appropriate exactly in the case when one has enough “target” data to do slightly better than just using only “source” data. Our approach is incredibly simple, easy to implement as a preprocessing step (10 lines of Perl!) and outperforms stateof-the-art approaches on a range of datasets. Moreover, it is trivially extended to a multidomain adaptation problem, where one has data from a variety of different domains.",
""
]
} |
1709.08309 | 2564416464 | There are several estimators of conditional probability from observed frequencies of features. In this paper, we propose using the lower limit of confidence interval on posterior distribution determined by the observed frequencies to ascertain conditional probability. In our experiments, this method outperformed other popular estimators. | Church and Hanks @cite_8 , proposed the concept of mutual information to measure the association between words. This method solves low-frequency problem, because infrequent features cannot sum up to a large mutual information quantity. However, if the true strength of the relationship between two features can be regarded as conditional probability, it may not be an appropriate measure to use because mutual information is symmetric towards two features, whereas conditional probability is asymmetric. | {
"cite_N": [
"@cite_8"
],
"mid": [
"1593045043"
],
"abstract": [
"The term word association is used in a very particular sense in the psycholinguistic literature. (Generally speaking, subjects respond quicker than normal to the word nurse if it follows a highly associated word such as doctor. ) We will extend the term to provide the basis for a statistical description of a variety of interesting linguistic phenomena, ranging from semantic relations of the doctor nurse type (content word content word) to lexico-syntactic co-occurrence constraints between verbs and prepositions (content word function word). This paper will propose an objective measure based on the information theoretic notion of mutual information, for estimating word association norms from computer readable corpora. (The standard method of obtaining word association norms, testing a few thousand subjects on a few hundred words, is both costly and unreliable.) The proposed measure, the association ratio, estimates word association norms directly from computer readable corpora, making it possible to estimate norms for tens of thousands of words."
]
} |
1709.08309 | 2564416464 | There are several estimators of conditional probability from observed frequencies of features. In this paper, we propose using the lower limit of confidence interval on posterior distribution determined by the observed frequencies to ascertain conditional probability. In our experiments, this method outperformed other popular estimators. | Conditional probability is more popular in the database field than NLP. Apriori @cite_6 is the most practically used method for finding the relationship between features, when the true strength between features is known to be conditional probability. It uses MLE as the measure for the relationship. To overcome the problem of low frequency, Apriori ignores the rare features using a threshold value, which is called minimum supports. Although Apriori is efficient by ignoring low-frequency features, we sometimes need to compare a high frequency but low MLE value relationship with a low frequency but high MLE value relationship. Apriori simply ignores relation of the low frequency but high MLE value relationship. | {
"cite_N": [
"@cite_6"
],
"mid": [
"1484413656"
],
"abstract": [
"We consider the problem of discovering association rules between items in a large database of sales transactions. We present two new algorithms for solving thii problem that are fundamentally different from the known algorithms. Empirical evaluation shows that these algorithms outperform the known algorithms by factors ranging from three for small problems to more than an order of magnitude for large problems. We also show how the best features of the two proposed algorithms can be combined into a hybrid algorithm, called AprioriHybrid. Scale-up experiments show that AprioriHybrid scales linearly with the number of transactions. AprioriHybrid also has excellent scale-up properties with respect to the transaction size and the number of items in the database."
]
} |
1709.08430 | 2756604803 | In this paper we focus on developing a control algorithm for multi-terrain tracked robots with flippers using a reinforcement learning (RL) approach. The work is based on the deep deterministic policy gradient (DDPG) algorithm, proven to be very successful in simple simulation environments. The algorithm works in an end-to-end fashion in order to control the continuous position of the flippers. This end-to-end approach makes it easy to apply the controller to a wide array of circumstances, but the huge flexibility comes to the cost of an increased difficulty of solution. The complexity of the task is enlarged even more by the fact that real multi-terrain robots move in partially observable environments. Notwithstanding these complications, being able to smoothly control a multi-terrain robot can produce huge benefits in impaired people daily lives or in search and rescue situations. | The state of the art deeprl algorithms manage to solve many different tasks. Among them, the two most widely used are dqn @cite_3 and ddpg @cite_8 . The former managed to master, often better than human players, many Atari games just using raw pixel images as input and outputting discrete actions. The latter instead achieved something more complex: the use of continuous actions. This could be the key to extend deeprl to real environments: in these cases in fact, just using discrete actions might not be enough to guarantee good performances. Another important result was obtained in @cite_1 , where the authors applied deeprl to a real and complex robot to make it learn how to move in the most efficient way. It needs to be noticed though among these works, only the first two used deeprl in an end-to-end fashion only in game-style simulators with mostly fully observable environments. | {
"cite_N": [
"@cite_1",
"@cite_3",
"@cite_8"
],
"mid": [
"2526152612",
"2145339207",
"2173248099"
],
"abstract": [
"Tensegrity robots, composed of rigid rods connected by elastic cables, have a number of unique properties that make them appealing for use as planetary exploration rovers. However, control of tensegrity robots remains a difficult problem due to their unusual structures and complex dynamics. In this work, we show how locomotion gaits can be learned automatically using a novel extension of mirror descent guided policy search (MDGPS) applied to periodic locomotion movements, and we demonstrate the effectiveness of our approach on tensegrity robot locomotion. We evaluate our method with real-world and simulated experiments on the SUPERball tensegrity robot, showing that the learned policies generalize to changes in system parameters, unreliable sensor measurements, and variation in environmental conditions, including varied terrains and a range of different gravities. Our experiments demonstrate that our method not only learns fast, power-efficient feedback policies for rolling gaits, but that these policies can succeed with only the limited onboard sensing provided by SUPERball's accelerometers. We compare the learned feedback policies to learned open-loop policies and hand-engineered controllers, and demonstrate that the learned policy enables the first continuous, reliable locomotion gait for the real SUPERball robot. Our code and other supplementary materials are available from this http URL",
"An artificial agent is developed that learns to play a diverse range of classic Atari 2600 computer games directly from sensory experience, achieving a performance comparable to that of an expert human player; this work paves the way to building general-purpose learning algorithms that bridge the divide between perception and action.",
"We adapt the ideas underlying the success of Deep Q-Learning to the continuous action domain. We present an actor-critic, model-free algorithm based on the deterministic policy gradient that can operate over continuous action spaces. Using the same learning algorithm, network architecture and hyper-parameters, our algorithm robustly solves more than 20 simulated physics tasks, including classic problems such as cartpole swing-up, dexterous manipulation, legged locomotion and car driving. Our algorithm is able to find policies whose performance is competitive with those found by a planning algorithm with full access to the dynamics of the domain and its derivatives. We further demonstrate that for many of the tasks the algorithm can learn policies end-to-end: directly from raw pixel inputs."
]
} |
1709.08430 | 2756604803 | In this paper we focus on developing a control algorithm for multi-terrain tracked robots with flippers using a reinforcement learning (RL) approach. The work is based on the deep deterministic policy gradient (DDPG) algorithm, proven to be very successful in simple simulation environments. The algorithm works in an end-to-end fashion in order to control the continuous position of the flippers. This end-to-end approach makes it easy to apply the controller to a wide array of circumstances, but the huge flexibility comes to the cost of an increased difficulty of solution. The complexity of the task is enlarged even more by the fact that real multi-terrain robots move in partially observable environments. Notwithstanding these complications, being able to smoothly control a multi-terrain robot can produce huge benefits in impaired people daily lives or in search and rescue situations. | A problem of rl is the time required for the training and for the collection of a sufficient number of samples. This issue has been addressed in @cite_9 , where agents working in parallel have been used for sample collection, thus significantly speeding up the network's training. In @cite_4 the authors applied an asynchronous deeprl algorithm to a mobile robot to navigate it through unseen environments, obtaining better results than previously available algorithms. | {
"cite_N": [
"@cite_9",
"@cite_4"
],
"mid": [
"2260756217",
"2592873849"
],
"abstract": [
"We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input.",
"We present a learning-based mapless motion planner by taking the sparse 10-dimensional range findings and the target position with respect to the mobile robot coordinate frame as input and the continuous steering commands as output. Traditional motion planners for mobile ground robots with a laser range sensor mostly depend on the obstacle map of the navigation environment where both the highly precise laser sensor and the obstacle map building work of the environment are indispensable. We show that, through an asynchronous deep reinforcement learning method, a mapless motion planner can be trained end-to-end without any manually designed features and prior demonstrations. The trained planner can be directly applied in unseen virtual and real environments. The experiments show that the proposed mapless motion planner can navigate the nonholonomic mobile robot to the desired targets without colliding with any obstacles."
]
} |
1709.08430 | 2756604803 | In this paper we focus on developing a control algorithm for multi-terrain tracked robots with flippers using a reinforcement learning (RL) approach. The work is based on the deep deterministic policy gradient (DDPG) algorithm, proven to be very successful in simple simulation environments. The algorithm works in an end-to-end fashion in order to control the continuous position of the flippers. This end-to-end approach makes it easy to apply the controller to a wide array of circumstances, but the huge flexibility comes to the cost of an increased difficulty of solution. The complexity of the task is enlarged even more by the fact that real multi-terrain robots move in partially observable environments. Notwithstanding these complications, being able to smoothly control a multi-terrain robot can produce huge benefits in impaired people daily lives or in search and rescue situations. | When solving complex problems with rl , having the algorithm converge can be tricky. To solve this issue many tricks have been developed, like the use of Replay Memory @cite_2 and auxiliary target networks @cite_3 . Another trick used in @cite_0 is to give the agent other auxiliary tasks, which differ from the main goal, to solve; this pushes the agent to explore more, thus leading to better solutions. | {
"cite_N": [
"@cite_0",
"@cite_3",
"@cite_2"
],
"mid": [
"2952791429",
"2145339207",
"1757796397"
],
"abstract": [
"Learning to navigate in complex environments with dynamic elements is an important milestone in developing AI agents. In this work we formulate the navigation question as a reinforcement learning problem and show that data efficiency and task performance can be dramatically improved by relying on additional auxiliary tasks leveraging multimodal sensory inputs. In particular we consider jointly learning the goal-driven reinforcement learning problem with auxiliary depth prediction and loop closure classification tasks. This approach can learn to navigate from raw sensory input in complicated 3D mazes, approaching human-level performance even under conditions where the goal location changes frequently. We provide detailed analysis of the agent behaviour, its ability to localise, and its network activity dynamics, showing that the agent implicitly learns key navigation abilities.",
"An artificial agent is developed that learns to play a diverse range of classic Atari 2600 computer games directly from sensory experience, achieving a performance comparable to that of an expert human player; this work paves the way to building general-purpose learning algorithms that bridge the divide between perception and action.",
"We present the first deep learning model to successfully learn control policies directly from high-dimensional sensory input using reinforcement learning. The model is a convolutional neural network, trained with a variant of Q-learning, whose input is raw pixels and whose output is a value function estimating future rewards. We apply our method to seven Atari 2600 games from the Arcade Learning Environment, with no adjustment of the architecture or learning algorithm. We find that it outperforms all previous approaches on six of the games and surpasses a human expert on three of them."
]
} |
1709.08430 | 2756604803 | In this paper we focus on developing a control algorithm for multi-terrain tracked robots with flippers using a reinforcement learning (RL) approach. The work is based on the deep deterministic policy gradient (DDPG) algorithm, proven to be very successful in simple simulation environments. The algorithm works in an end-to-end fashion in order to control the continuous position of the flippers. This end-to-end approach makes it easy to apply the controller to a wide array of circumstances, but the huge flexibility comes to the cost of an increased difficulty of solution. The complexity of the task is enlarged even more by the fact that real multi-terrain robots move in partially observable environments. Notwithstanding these complications, being able to smoothly control a multi-terrain robot can produce huge benefits in impaired people daily lives or in search and rescue situations. | If rl algorithms are to be applied to real-world robots, the risk that they can cause damage to the environment and the robot itself while exploring different policies, needs to be considered. To overcome this problem @cite_6 and @cite_12 added some constraints to the exploration policies. Although they also used a tracked robot with flippers, this solution is not feasible when using nn because, as of now, it is not possible to control their learning process through constraints. Thus, in our work we relied on training the agent in a simulation environment. This allowed the algorithm to test all the policies needed with no real harm. | {
"cite_N": [
"@cite_12",
"@cite_6"
],
"mid": [
"2509374375",
"2324567380"
],
"abstract": [
"Most learning algorithms are not invariant to the scale of the function that is being approximated. We propose to adaptively normalize the targets used in learning. This is useful in value-based reinforcement learning, where the magnitude of appropriate value approximations can change over time when we update the policy of behavior. Our main motivation is prior work on learning to play Atari games, where the rewards were all clipped to a predetermined range. This clipping facilitates learning across many different games with a single learning algorithm, but a clipped reward function can result in qualitatively different behavior. Using the adaptive normalization we can remove this domain-specific heuristic without diminishing overall performance.",
"Received Feb 6, 2012 Revised Mar 12, 2012 Accepted Mar 21, 2012 This paper proposes a behavior-switching control strategy of an evolutionary robotics based on Artificial Neural Network (ANN) and Genetic Algorithms (GA). This method is able not only to construct the reinforcement learning models for autonomous robots and evolutionary robot modules that control behaviors and reinforcement learning environments, and but also to perform the behavior-switching control and obstacle avoidance of an evolutionary robotics (ER) in time-varying environments with static and moving obstacles by combining ANN and GA. The experimental results on the basic behaviors and behavior-switching control have demonstrated that our method can perform the decision-making strategy and parameters set optimization of FNN and GA by learning and can escape successfully from the trap of a local minima and avoid “motion deadlock” status of humanoid soccer robotics agents, and reduce the oscillation of the planned trajectory between the multiple obstacles by crossover and mutation. Some results of the proposed algorithm have been successfully applied to our simulation humanoid robotics soccer team CIT3D which won the 1st prize of RoboCup Championship and China Open 2010 (July 2010) and the 2nd place of the official Robo Cup World Championship (2011) on 5-11 July, 2011 in Istanbul, Turkey. As compared with the conventional behavior network and the adaptive behavior method, the genetic encoding complexity of our algorithm is simplified, and the network performance and the convergence rate ρ have been greatly improved. Keyword:"
]
} |
1709.08460 | 2599933303 | Discrete wavelet transform of finite-length signals must necessarily handle the signal boundaries. The state-of-the-art approaches treat such boundaries in a complicated and inflexible way, using special prolog or epilog phases. This holds true in particular for images decomposed into a number of scales, exemplary in JPEG 2000 coding system. In this paper, the state-of-the-art approaches are extended to perform the treatment using a compact streaming core, possibly in multi-scale fashion. We present the core focused on CDF 5 3 wavelet and the symmetric border extension method, both employed in the JPEG 2000. As a result of our work, every input sample is visited only once, while the results are produced immediately, i.e. without buffering. | The polyphase matrix @cite_5 @cite_9 @cite_1 is a convenient tool to express the transform filters. The lifting scheme factorizes this matrix into a series of shorter filterings. In detail, the polyphase matrix can be factorized @cite_13 , so that , where @math is a non-zero constant, and polynomials @math represent the individual lifting steps. Focusing on the CDF 5 3 wavelet as an example, the forward transform in can be expressed @cite_9 by the dual polyphase matrix , where @math are real constants, and the @math is called the scaling factor. For details about the lifting scheme, see @cite_9 @cite_12 . Unlike a convolution scheme, the lifting allows @cite_10 formulation of the transforms mapping integers to integers. | {
"cite_N": [
"@cite_10",
"@cite_9",
"@cite_1",
"@cite_5",
"@cite_13",
"@cite_12"
],
"mid": [
"1975358183",
"2015370045",
"658512522",
"",
"565312106",
"2117188745"
],
"abstract": [
"Abstract Invertible wavelet transforms that map integers to integers have important applications in lossless coding. In this paper we present two approaches to build integer to integer wavelet transforms. The first approach is to adapt the precoder of Laroiaet al,which is used in information transmission; we combine it with expansion factors for the high and low pass band in subband filtering. The second approach builds upon the idea of factoring wavelet transforms into so-called lifting steps. This allows the construction of an integer version of every wavelet transform. Finally, we use these approaches in a lossless image coder and compare the results to those given in the literature.",
"This article is essentially tutorial in nature. We show how any discrete wavelet transform or two band subband filtering with finite filters can be decomposed into a finite sequence of simple filtering steps, which we call lifting steps but that are also known as ladder structures. This decomposition corresponds to a factorization of the polyphase matrix of the wavelet or subband filters into elementary matrices. That such a factorization is possible is well-known to algebraists (and expressed by the formula SL(n; R(Z, z-I)) = E(n; R(z, z-l))); it is also used in linear systems theory in the electrical engineering community. We present here a self-contained derivation, building the decomposition from basic principles such as the Euclidean algorithm, with a focus on applying it to wavelet filtering. This factorization provides an alternative for the lattice factorization, with the advantage that it can also be used in the biorthogonal, i.e., non-unitary case. Like the lattice factorization, the decomposition presented here asymptotically reduces the computational complexity of the transform by a factor two. It has other applications, such as the possibility of defining a wavelet-like transform that maps integers to integers.",
"1. On rainbows and spectra 2. From Euclid to Hilbert: 2.1 Introduction 2.2 Vector spaces 2.3 Hilbert spaces 2.4 Approximations, projections, and decompositions 2.5 Bases and frames 2.6 Computational aspects 2.A Elements of analysis and topology 2.B Elements of linear algebra 2.C Elements of probability 2.D Basis concepts Exercises with solutions Exercises 3. Sequences and discrete-time systems: 3.1 Introduction 3.2 Sequences 3.3 Systems 3.4 Discrete-time Fourier Transform 3.5 z-Transform 3.6 Discrete Fourier Transform 3.7 Multirate sequences and systems 3.8 Stochastic processes and systems 3.9 Computational aspects 3.A Elements of analysis 3.B Elements of algebra Exercises with solutions Exercises 4. Functions and continuous-time systems: 4.1 Introduction 4.2 Functions 4.3 Systems 4.4 Fourier Transform 4.5 Fourier series 4.6 Stochastic processes and systems Exercises with solutions Exercises 5. Sampling and interpolation: 5.1 Introduction 5.2 Finite-dimensional vectors 5.3 Sequences 5.4 Functions 5.5 Periodic functions 5.6 Computational aspects Exercises with solutions Exercises 6. Approximation and compression: 6.1 Introduction 6.2 Approximation of functions on finite intervals by polynomials 6.3 Approximation of functions by splines 6.4 Approximation of functions and sequences by series truncation 6.5 Compression 6.6 Computational aspects Exercises with solutions Exercises 7. Localization and uncertainty: 7.1 Introduction 7.2 Localization for functions 7.3 Localization for sequences 7.4 Tiling the time-frequency plane 7.5 Examples of local Fourier and wavelet bases 7.6 Recap and a glimpse forward Exercises with solutions Exercises.",
"",
"1. Introduction 2. Introduction to abstract algebra 3. Fast algorithms for the discrete Fourier transform 4. Fast algorithms based on doubling strategies 5. Fast algorithms for short convolutions 6. Architecture of filters and transforms 7. Fast algorithms for solving Toeplitz systems 8. Fast algorithms for trellis search 9. Numbers and fields 10. Computation in finite fields and rings 11. Fast algorithms and multidimensional convolutions 12. Fast algorithms and multidimensional transforms Appendices: A. A collection of cyclic convolution algorithms B. A collection of Winograd small FFT algorithms.",
"Abstract We present the lifting scheme, a new idea for constructing compactly supported wavelets with compactly supported duals. The lifting scheme uses a simple relationship between all multiresolution analyses with the same scaling function. It isolates the degrees of freedom remaining after fixing the biorthogonality relations. Then one has full control over these degrees of freedom to custom design the wavelet for a particular application. The lifting scheme can also speed up the fast wavelet transform. We illustrate the use of the lifting scheme in the construction of wavelets with interpolating scaling functions."
]
} |
1709.08460 | 2599933303 | Discrete wavelet transform of finite-length signals must necessarily handle the signal boundaries. The state-of-the-art approaches treat such boundaries in a complicated and inflexible way, using special prolog or epilog phases. This holds true in particular for images decomposed into a number of scales, exemplary in JPEG 2000 coding system. In this paper, the state-of-the-art approaches are extended to perform the treatment using a compact streaming core, possibly in multi-scale fashion. We present the core focused on CDF 5 3 wavelet and the symmetric border extension method, both employed in the JPEG 2000. As a result of our work, every input sample is visited only once, while the results are produced immediately, i.e. without buffering. | Originally, the problem of efficient implementation of the 1-D lifting scheme was addressed in @cite_24 by Ch. Chrysafis and A. Ortega. Their general approach transforms the input data in a single loop. Nonetheless, this is essentially the same method as the on-line or pipelined method mentioned in other papers (although not necessarily using lifting scheme nor 1-D transform). The key idea is to make the lifting scheme causal, so that it may be evaluated as a running scheme without buffering of the whole signal. However, due to borders of the finite signals, some buffering is necessary at least at the beginning and end of the single-loop processing. For 2-D signals, the same idea was also discovered in another papers under various names, e.g. in @cite_25 @cite_7 @cite_21 @cite_8 @cite_14 @cite_23 . The most sophisticated techniques, that we are aware of, were investigated by R. Kutil in @cite_3 . The author emphasizes that the main issue are the arduous prolog and epilog phases. Many authors @cite_6 @cite_19 @cite_4 @cite_11 also discovered a similar technique with a slightly different goal (a better utilization of cache lines). This technique is referred to as the aggregation, strip-mine, or loop tiling. | {
"cite_N": [
"@cite_14",
"@cite_4",
"@cite_7",
"@cite_8",
"@cite_21",
"@cite_3",
"@cite_6",
"@cite_24",
"@cite_19",
"@cite_23",
"@cite_25",
"@cite_11"
],
"mid": [
"2082055510",
"2096921464",
"",
"2124994383",
"2098217375",
"2106145405",
"2585518016",
"2589762227",
"1624490870",
"1966659712",
"2623413973",
"2134718751"
],
"abstract": [
"In this paper, a new algorithm to efficiently implement the two-dimensional lifting scheme is presented. The 1D lifting-scheme performs in-place processing of the input samples, and hence it provides reduction in memory requirements. However, for image processing (2D), in-place computation is not enough, resulting in a memory-intensive algorithm, since it has to keep the whole image in memory. We propose the use of line-by-line processing algorithm for the lifting scheme, and we address some issues on how to perform synchronization among different buffer levels, so that an implementation can be easily written. Experimental results show that, for a 5-megapixel image, our algorithm requires 200 times less memory and it is more than 3 times faster than the usual one.",
"The discrete wavelet transform (DWT), the technology at the heart of the JPEG 2000 image compression system, operates on user-definable tiles of the image, as opposed to fixed-size blocks of the image as does the discrete cosine transform (DCT) used in JPEG. This difference reduces artificial blocking effects but can severely stress the memory system. We examine the interaction of the DWT and the memory hierarchy, modify the structure of the DWT computation and the layout of the image data to improve cache and translation lookaside buffer (TLB) locality, and demonstrate significant performance improvements of the DWT over a baseline implementation. Our optimized DWT implementation exhibits speedups of up to 4 spl times over the DWT in a JPEG 2000 reference implementation.",
"",
"This paper addresses the problem of low memory wavelet image compression. While wavelet or subband coding of images has been shown to be superior to more traditional transform coding techniques, little attention has been paid until recently to the important issue of whether both the wavelet transforms and the subsequent coding can be implemented in low memory without significant loss in performance. We present a complete system to perform low memory wavelet image coding. Our approach is \"line-based\" in that the images are read line by line and only the minimum required number of lines is kept in memory. There are two main contributions of our work. First, we introduce a line-based approach for the implementation of the wavelet transform, which yields the same results as a \"normal\" implementation, but where, unlike prior work, we address memory issues arising from the need to synchronize encoder and decoder. Second, we propose a novel context-based encoder which requires no global information and stores only a local set of wavelet coefficients. This low memory coder achieves performance comparable to state of the art coders at a fraction of their memory utilization.",
"The JPEG2000 standard uses the 2D Discrete Wavelet Transform (2D DWT), while the JPEG standard uses the 2D Discrete Cosine Transform (DCT). However, the 2D DWT has higher computational requirements than the 2D DCT and consumes a significant part of the total JPEG2000 encoding time. One way to improve the performance of the 2D DWT is using parallel techniques on an SIMD-enhanced architecture. In this paper, we apply data-level parallelism technique to exploit available parallelism of the 2D DWT. We focus on the two algorithms to traverse an image to implement the 2D Discrete Wavelet Transform (DWT), namely Row-Column Wavelet Transform (RCWT) and Line-Based Wavelet Transform (LBWT). Our experimental results show that the SIMD implementation of the LBWT algorithm is more complicated than the SIMD implementation of the RCWT algorithm, while the former algorithm is 1.60 times faster than the latter algorithm for an image of size 4096 × 4096.",
"Widespread use of wavelet transforms as in JPEG2000 demands efficient implementations on general purpose computers as well as dedicated hardware. The increasing availability of SIMD technologies is a great challenge since efficient SIMD parallelizations are not trivial. This work presents a parallelized 2D wavelet transform following a single-loop approach, i.e. a loop fusion of the lifting steps of horizontal filtering, and interleaving horizontal and vertical filtering for optimal temporal locality. In this way, each input value is read only once and each output value is written once without subsequent updates. Such an approach turns out to be a necessary basis for an efficient SIMD parallelization. Results are obtained on a general purpose processor with a 4-fold single-precision SIMD extension. Speedups of about 3.7 due to the use of SIMD, 2.55 due to the single-loop approach and up to 6 due to cache effects for pathologic data sizes are obtained, giving total speedups of up to 56.",
"In this paper, we have a close look at the runtime performance of the intra-component transform employed in the reference implementations of the JPEG2000 image coding standard. Typically, wavelet lifting is used to obtain a wavelet decomposition of the source image in a computationally efficient way. However, so far no attention has been paid to the impact of the CPU's memory cache on the overall performance. We propose two simple techniques that dramatically reduce the number of cache misses and cut column filtering runtime by a factor of 10. Theoretical estimates as well as experimental results on a number of hardware platforms show the effectivity of our approach.",
"",
"This paper addresses the vectorization of the lifting-based wavelet transform on general-purpose microprocessors in the context of JPEG2000. Since SIMD exploitation strongly depends on an efficient memory hierarchy usage, this research is based on previous work about cache-conscious DWT implementations. The experimental platform on which we have chosen to study the benefits of the SIMD extensions is an Intel Pentium-4 (P-4) based PC. However, unlike other authors, the vectorization has been performed avoiding assembler language programming in order to improve both code portability and development cost.",
"With the start of the widespread use of discrete wavelet transform in image processing, the need for its efficient implementation is becoming increasingly more important. This work presents several novel SIMD-vectorized algorithms of 2-D discrete wavelet transform, using a lifting scheme. At the beginning, a stand-alone core of an already known single-loop approach is extracted. This core is further simplified by an appropriate reorganization of operations. Furthermore, the influence of the CPU cache on a 2-D processing order is examined. Finally, SIMD-vectorizations and parallelizations of the proposed approaches are evaluated. The best of the proposed algorithms scale almost linearly with the number of threads. For all of the platforms used in the tests, these algorithms are significantly faster than other known methods, as shown in the experimental sections of the paper.",
"This paper addresses the implementation of a 2-D Discrete Wavelet Transform on general-purpose microprocessors, focusing on both memory hierarchy and SIMD parallelization issues. Both topics are somewhat related, since SIMD extensions are only useful if the memory hierarchy is efficiently exploited. In this work, locality has been significantly improved by means of a novel approach called pipelined computation, which complements previous techniques based on loop tiling and non-linear layouts. As experimental platforms we have employed a Pentium-III (P-III) and a Pentium-4 (P-4) microprocessor. However, our SIMD-oriented tuning has been exclusively performed at source code level. Basically, we have reordered some loops and introduced some modifications that allow automatic vectorization. Taking into account the abstraction level at which the optimizations are carried out, the speedups obtained on the investigated platforms are quite satisfactory, even though further improvement can be obtained by dropping the level of abstraction (compiler intrinsics or assembly code).",
"The overall performance of a computing system increasingly depends on the efficient use of the cache memories. Traditional approaches for cache tuning deploy performance tools to help the user optimize the source program towards a better runtime data locality. Following this conventional way, we developed a set of such toolkits including data profiling, pattern analysis, and performance visualization tools. This paper demonstrates how the toolset can be used step-by-step to understand the cache access behavior of the applications and then achieve optimized program code. The Discrete Wavelet Transform, a common used algorithm for image and video compression, is applied as an example. Our initial experimental results with this sample application show an up to 3.19 speedup in execution time compared to the original implementation."
]
} |
1709.08460 | 2599933303 | Discrete wavelet transform of finite-length signals must necessarily handle the signal boundaries. The state-of-the-art approaches treat such boundaries in a complicated and inflexible way, using special prolog or epilog phases. This holds true in particular for images decomposed into a number of scales, exemplary in JPEG 2000 coding system. In this paper, the state-of-the-art approaches are extended to perform the treatment using a compact streaming core, possibly in multi-scale fashion. We present the core focused on CDF 5 3 wavelet and the symmetric border extension method, both employed in the JPEG 2000. As a result of our work, every input sample is visited only once, while the results are produced immediately, i.e. without buffering. | Since this work is based on our previous works in @cite_0 @cite_15 @cite_20 , it should be explained what the difference between this work and the referenced papers is. In @cite_0 , we formulated two-dimensional cores using a different notation. However, they were not designed for any effective treatment of signal boundaries. In @cite_15 @cite_20 , we have proposed DWT transform engine for JPEG 2000 encoders. Also in this work we have not overcame the problem with effective treatment of signal boundaries. Instead, we have introduced several buffers in order to circumvent the problem. | {
"cite_N": [
"@cite_0",
"@cite_15",
"@cite_20"
],
"mid": [
"2294910995",
"2565089243",
"2467458447"
],
"abstract": [
"A novel approach to 2-D single-loop wavelet lifting with compatibility to JPEG 2000 is presented in this paper. A newly developed 2-D core of CDF 5 3 wavelet filter is presented that, using a new sequence of operations, simplify the design. Moreover, the proposed approach, that uses one pass for 2-D transform, directly produces final output and reduces significantly the need for storing intermediate results into memory. All the proposed structures can be efficiently pipelined in hardware. This paper describes the proposed approach, its implementation in FPGA, cost of such implementation, and brings an experimental evaluation as well as discussion of the features of the approach.",
"JPEG 2000 is an image coding system based on the discrete wavelet transform. Unfortunately, there exist several major issues with the effective implementation of the codec. For high resolution data decomposed by a separable transform, immensely many CPU cache misses occur. Following the procedure as defined in the standard, the coefficients of a single resolution appears all at once. Consequently, the entropy coder (EBCOT) needs to once again return to the data already touched. We present a software architecture designed for JPEG 2000 coders. The proposed method employs a strip-based data processing technique while it performs a single-pass multi-scale wavelet transform. The overall compression chain is driven by incoming data while the fragments of the resulting bitstream can be produced immediately after loading the corresponding data and additionally in parallel. The method is friendly to the CPU cache and can nicely exploit the SIMD extensions.",
"We present a novel and very efficient software architecture designed for JPEG 2000 coders. The proposed method employs a strip-based data processing technique while performing a single-pass multi-scale wavelet transform. The overall compression chain is driven by incoming data while the fragments of the resulting bitstream are produced immediately after loading the corresponding data and additionally in parallel. The method is friendly to the CPU cache and nicely exploits the SIMD capabilities of the modern CPUs. Implanted into reference OpenJPEG implementation, our method has significantly better performance in terms of the execution time."
]
} |
1709.08374 | 2759684359 | In this paper, we present a deep extension of Sparse Subspace Clustering, termed Deep Sparse Subspace Clustering (DSSC). Regularized by the unit sphere distribution assumption for the learned deep features, DSSC can infer a new data affinity matrix by simultaneously satisfying the sparsity principle of SSC and the nonlinearity given by neural networks. One of the appealing advantages brought by DSSC is: when original real-world data do not meet the class-specific linear subspace distribution assumption, DSSC can employ neural networks to make the assumption valid with its hierarchical nonlinear transformations. To the best of our knowledge, this is among the first deep learning based subspace clustering methods. Extensive experiments are conducted on four real-world datasets to show the proposed DSSC is significantly superior to 12 existing methods for subspace clustering. | Aimed at learning high-level features from inputs, deep learning has shown promising results in numerous computer vision tasks in the scenario of supervised learning, such as image classification @cite_82 . In contrast, less attention @cite_79 @cite_8 @cite_27 has been paid to the applications with unsupervised learning scheme. Recently, some works @cite_17 @cite_81 @cite_29 @cite_62 @cite_10 @cite_50 @cite_60 have devoted to combining deep learning and unsupervised clustering and shown impressive results over the traditional clustering approaches. These methods share the same basic idea, , using deep learning to learn a good representation and then achieving clustering with existing clustering methods like k-means. The major differences among them reside on the neural network structure and the objective function. | {
"cite_N": [
"@cite_62",
"@cite_82",
"@cite_8",
"@cite_60",
"@cite_29",
"@cite_79",
"@cite_27",
"@cite_81",
"@cite_50",
"@cite_10",
"@cite_17"
],
"mid": [
"",
"2163605009",
"1971014294",
"2753276256",
"2571899125",
"2163922914",
"2795467458",
"2173649752",
"",
"2337374958",
"2120303002"
],
"abstract": [
"",
"We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5 and 17.0 which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called \"dropout\" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3 , compared to 26.2 achieved by the second-best entry.",
"There has been much interest in unsupervised learning of hierarchical generative models such as deep belief networks (DBNs); however, scaling such models to full-sized, high-dimensional images remains a difficult problem. To address this problem, we present the convolutional deep belief network, a hierarchical generative model that scales to realistic image sizes. This model is translation-invariant and supports efficient bottom-up and top-down probabilistic inference. Key to our approach is probabilistic max-pooling, a novel technique that shrinks the representations of higher layers in a probabilistically sound way. Our experiments show that the algorithm learns useful high-level visual features, such as object parts, from unlabeled images of objects and natural scenes. We demonstrate excellent performance on several visual recognition tasks and show that our model can perform hierarchical (bottom-up and top-down) inference over full-sized images.",
"We present a novel deep neural network architecture for unsupervised subspace clustering. This architecture is built upon deep auto-encoders, which non-linearly map the input data into a latent space. Our key idea is to introduce a novel self-expressive layer between the encoder and the decoder to mimic the \"self-expressiveness\" property that has proven effective in traditional subspace clustering. Being differentiable, our new self-expressive layer provides a simple but effective way to learn pairwise affinities between all data points through a standard back-propagation procedure. Being nonlinear, our neural-network based method is able to cluster data points having complex (often nonlinear) structures. We further propose pre-training and fine-tuning strategies that let us effectively learn the parameters of our subspace clustering networks. Our experiments show that the proposed method significantly outperforms the state-of-the-art unsupervised subspace clustering methods.",
"Subspace clustering aims to cluster unlabeled samples into multiple groups by implicitly seeking a subspace to fit each group. Most of existing methods are based on a shallow linear model, which may fail in handling data with nonlinear structure. In this paper, we propose a novel subspace clustering method -- deeP subspAce clusteRing with sparsiTY prior (PARTY) -- based on a new deep learning architecture. PARTY explicitly learns to progressively transform input data into nonlinear latent space and to be adaptive to the local and global subspace structure simultaneously. In particular, considering local structure, PARTY learns representation for the input data with minimal reconstruction error. Moreover, PARTY incorporates a prior sparsity information into the hidden representation learning to preserve the sparse reconstruction relation over the whole data set. To the best of our knowledge, PARTY is the first deep learning based subspace clustering method. Extensive experiments verify the effectiveness of our method.",
"The success of machine learning algorithms generally depends on data representation, and we hypothesize that this is because different representations can entangle and hide more or less the different explanatory factors of variation behind the data. Although specific domain knowledge can be used to help design representations, learning with generic priors can also be used, and the quest for AI is motivating the design of more powerful representation-learning algorithms implementing such priors. This paper reviews recent work in the area of unsupervised feature learning and deep learning, covering advances in probabilistic models, autoencoders, manifold learning, and deep networks. This motivates longer term unanswered questions about the appropriate objectives for learning good representations, for computing representations (i.e., inference), and the geometrical connections between representation learning, density estimation, and manifold learning.",
"",
"Clustering is central to many data-driven application domains and has been studied extensively in terms of distance functions and grouping algorithms. Relatively little work has focused on learning representations for clustering. In this paper, we propose Deep Embedded Clustering (DEC), a method that simultaneously learns feature representations and cluster assignments using deep neural networks. DEC learns a mapping from the data space to a lower-dimensional feature space in which it iteratively optimizes a clustering objective. Our experimental evaluations on image and text corpora show significant improvement over state-of-the-art methods.",
"",
"In this paper, we propose a recurrent framework for Joint Unsupervised LEarning (JULE) of deep representations and image clusters. In our framework, successive operations in a clustering algorithm are expressed as steps in a recurrent process, stacked on top of representations output by a Convolutional Neural Network (CNN). During training, image clusters and representations are updated jointly: image clustering is conducted in the forward pass, while representation learning in the backward pass. Our key idea behind this framework is that good representations are beneficial to image clustering and clustering results provide supervisory signals to representation learning. By integrating two processes into a single model with a unified weighted triplet loss and optimizing it end-to-end, we can obtain not only more powerful representations, but also more precise image clusters. Extensive experiments show that our method outperforms the state-of-the-art on image clustering across a variety of image datasets. Moreover, the learned representations generalize well when transferred to other tasks.",
"Clustering is a fundamental technique widely used for exploring the inherent data structure in pattern recognition and machine learning. Most of the existing methods focus on modeling the similarity dissimilarity relationship among instances, such as k-means and spectral clustering, and ignore to extract more effective representation for clustering. In this paper, we propose a deep embedding network for representation learning, which is more beneficial for clustering by considering two constraints on learned representations. We first utilize a deep auto encoder to learn the reduced representations from the raw data. To make the learned representations suitable for clustering, we first impose a locality-persevering constraint on the learned representations, which aims to embed original data into its underlying manifold space. Then, different from spectral clustering which extracts representations from the block diagonal similarity matrix, we apply a group sparsity constraint for the learned representations, and aim to learn block diagonal representations in which the nonzero groups correspond to its cluster. After obtaining the learned representations, we use k-means to cluster them. To evaluate the proposed deep embedding network, we compare its performance with k-means and spectral clustering on three commonly-used datasets. The experiments demonstrate that the proposed method achieves promising performance."
]
} |
1709.08340 | 2561574324 | When we watch a video, in which human hands manipulate objects, these hands may obscure some parts of those objects. We are willing to make clear how the objects are manipulated by making the image of hands semi-transparent, and showing the complete images of the hands and the object. By carefully choosing a Half-Diminished Reality method, this paper proposes a method that can process the video in real time and verifies that the proposed method works well. | He and Zhang @cite_3 use the difference between the current image and the previous image but not the fixed background images. For a certain time, if there is no significant difference, the part of the image is regarded as stable. If the image is changing during a certain period, the corresponding part of the image is treated as a foreground region. The key concept of this approach is that the foreground image is unstable. Please note that an unstable region may not be a foreground region, but that a stable region is always a background region. In this approach, the background region, which is uncovered by a foreground object, is also treated the same way as the moving foreground region. In this sense, this approach does not correctly detect a particular foreground region, but successfully detects the region that contains all of foreground regions. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2100286961"
],
"abstract": [
"This paper describes our recently developed system which captures pen strokes on physical whiteboards in real time using an off-the-shelf video camera. Unlike many existing tools, our system does not instrument the pens or the whiteboard. It analyzes the sequence of captured video images in real time, classifies the pixels into whiteboard background, pen strokes and foreground objects (e.g., people in front of the whiteboard), extracts newly written pen strokes, and corrects the color to make the whiteboard completely white. This allows us to transmit whiteboard contents using very low bandwidth to remote meeting participants. Combined with other teleconferencing tools such as voice conference and application sharing, our system becomes a powerful tool to share ideas during online meetings"
]
} |
1709.08340 | 2561574324 | When we watch a video, in which human hands manipulate objects, these hands may obscure some parts of those objects. We are willing to make clear how the objects are manipulated by making the image of hands semi-transparent, and showing the complete images of the hands and the object. By carefully choosing a Half-Diminished Reality method, this paper proposes a method that can process the video in real time and verifies that the proposed method works well. | He and Zhang @cite_3 use a time shift approach. Since only a single camera is used in this approach, the information complementing the background image comes from the previous images. In this approach, the region that is regarded as the foreground part is not updated. As a result, the past images are used to complement the region that is currently concealed by a foreground object. The background image is updated by the current image in only the part where the image is regarded as stable. A nice feature of He and Zhang's approach is the ability to update the background. Please note that the change of background is inevitably delayed in order to judge its stability. Since they aimed to recover the information from a white board, in other words, they aimed at DR only, they did not need to consider the interaction between the foreground image and the background image. Therefore, this delay was not an issue for them. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2100286961"
],
"abstract": [
"This paper describes our recently developed system which captures pen strokes on physical whiteboards in real time using an off-the-shelf video camera. Unlike many existing tools, our system does not instrument the pens or the whiteboard. It analyzes the sequence of captured video images in real time, classifies the pixels into whiteboard background, pen strokes and foreground objects (e.g., people in front of the whiteboard), extracts newly written pen strokes, and corrects the color to make the whiteboard completely white. This allows us to transmit whiteboard contents using very low bandwidth to remote meeting participants. Combined with other teleconferencing tools such as voice conference and application sharing, our system becomes a powerful tool to share ideas during online meetings"
]
} |
1709.08546 | 2759666721 | Visualizations of tabular data are widely used; understanding their effectiveness in different task and data contexts is fundamental to scaling their impact. However, little is known about how basic tabular data visualizations perform across varying data analysis tasks and data attribute types. In this paper, we report results from a crowdsourced experiment to evaluate the effectiveness of five visualization types --- Table, Line Chart, Bar Chart, Scatterplot, and Pie Chart --- across ten common data analysis tasks and three data attribute types using two real world datasets. We found the effectiveness of these visualization types significantly varies across task and data attribute types, suggesting that visualization design would benefit from considering context dependent effectiveness. Based on our findings, we derive recommendations on which visualizations to choose based on different task and data contexts. | While these independent studies provide helpful generic guidelines, they were conducted under different conditions, varying sample sizes, datasets, and for a disperse set of tasks. In fact, several of these studies used manually created visualizations in their experiments without using actual datasets @cite_39 @cite_47 @cite_19 @cite_42 or created visualizations using artificial datasets @cite_50 . Also, these earlier studies have conducted experiments typically using atomic generic tasks such as comparison of data values (e.g., @cite_9 @cite_47 ) or estimation of proportions (e.g., @cite_39 @cite_6 @cite_43 ). However, many visual analysis tasks (e.g., filtering, finding clusters) require integration of results from multiple atomic tasks, limiting the applicability of earlier findings @cite_15 @cite_0 . Inconsistency in experimental settings and limited atomic tasks used in previous work encourages studying the effectiveness of visualization types for larger spectrum of tasks in a more consistence setting. | {
"cite_N": [
"@cite_15",
"@cite_9",
"@cite_42",
"@cite_6",
"@cite_39",
"@cite_0",
"@cite_19",
"@cite_43",
"@cite_50",
"@cite_47"
],
"mid": [
"1512287614",
"2095738647",
"",
"2059636449",
"2060296236",
"",
"",
"2469679165",
"2043523210",
""
],
"abstract": [
"Existing system-level taxonomies of visualization tasks are geared more towards the design of particular representations than the facilitation of user analytic activity. We present a set of ten low-level analysis tasks that largely capture people's activities while employing information visualization tools for understanding data. To help develop these tasks, we collected nearly 200 sample questions from students about how they would analyze five particular data sets from different domains. The questions, while not being totally comprehensive, illustrated the sheer variety of analytic questions typically posed by users when employing information visualization systems. We hope that the presented set of tasks is useful for information visualization system designers as a kind of common substrate to discuss the relative analytic capabilities of the systems. Further, the tasks may provide a form of checklist for system designers.",
"Interpretations of graphs seem to be rooted in principles of cognitive naturalness and information processing rather than arbitrary correspondences. These predict that people should more readily associate bars with discrete comparisons between data points because bars are discrete entities and facilitate point estimates. They should more readily associate lines with trends because lines connect discrete entities and directly represent slope. The predictions were supported in three experiments—two examining comprehension and one production. The correspondence does not seem to depend on explicit knowledge of rules. Instead, it may reflect the influence of the communicative situation as well as the perceptual properties of graphs.",
"",
"Pie and bar charts are commonly used to display percentage or proportional data, but professional data analysts have frowned on the use of the pie chart on the grounds that judgements of area are less accurate than judgements of lenth. Thus the bar chart has been favoured. When the amount of data to be communicated is small, some authorities have advocated the use of properly constructed tables, as another option. The series of experiments reported here suggests that there is little to choose between the pie and the bar chart, with the former enjoying a slight advantage if the required judgement is a complicated one, but that both forms of chart are superior to the table. Thus our results do not support the commonly expressed opinion that pie charts are inferior. An analysis of the nature of the task and a review of the psychophysical literature suggest that the traditional prejudice against the pie chart is misguided.",
"",
"",
"",
"Pie and donut charts have been a hotly debated topic in the visualization community for some time now. Even though pie charts have been around for over 200 years, our understanding of the perceptual factors used to read data in them is still limited. Data is encoded in pie and donut charts in three ways: arc length, center angle, and segment area. For our first study, we designed variations of pie charts to test the importance of individual encodings for reading accuracy. In our second study, we varied the inner radius of a donut chart from a filled pie to a thin outline to test the impact of removing the central angle. Both studies point to angle being the least important visual cue for both charts, and the donut chart being as accurate as the traditional pie chart.",
"Despite years of research yielding systems and guidelines to aid visualization design, practitioners still face the challenge of identifying the best visualization for a given dataset and task. One promising approach to circumvent this problem is to leverage perceptual laws to quantitatively evaluate the effectiveness of a visualization design. Following previously established methodologies, we conduct a large scale (n=1687) crowdsourced experiment to investigate whether the perception of correlation in nine commonly used visualizations can be modeled using Weber's law. The results of this experiment contribute to our understanding of information visualization by establishing that: 1) for all tested visualizations, the precision of correlation judgment could be modeled by Weber's law, 2) correlation judgment precision showed striking variation between negatively and positively correlated data, and 3) Weber models provide a concise means to quantify, compare, and rank the perceptual precision afforded by a visualization. Index Terms—Perception, Visualization, Evaluation.",
""
]
} |
1709.08571 | 2758197809 | The Hessian-vector product has been utilized to find a second-order stationary solution with strong complexity guarantee (e.g., almost linear time complexity in the problem's dimensionality). In this paper, we propose to further reduce the number of Hessian-vector products for faster non-convex optimization. Previous algorithms need to approximate the smallest eigen-value with a sufficient precision (e.g., @math ) in order to achieve a sufficiently accurate second-order stationary solution (i.e., @math . In contrast, the proposed algorithms only need to compute the smallest eigen-vector approximating the corresponding eigen-value up to a small power of current gradient's norm. As a result, it can dramatically reduce the number of Hessian-vector products during the course of optimization before reaching first-order stationary points (e.g., saddle points). The key building block of the proposed algorithms is a novel updating step named the NCG step, which lets a noisy negative curvature descent compete with the gradient descent. We show that the worst-case time complexity of the proposed algorithms with their favorable prescribed accuracy requirements can match the best in literature for achieving a second-order stationary point but with an arguably smaller per-iteration cost. We also show that the proposed algorithms can benefit from inexact Hessian by developing their variants accepting inexact Hessian under a mild condition for achieving the same goal. Moreover, we develop a stochastic algorithm for a finite or infinite sum non-convex optimization problem. To the best of our knowledge, the proposed stochastic algorithm is the first one that converges to a second-order stationary point in high probability with a time complexity independent of the sample size and almost linear in dimensionality. | Earlier work using second-order information for non-convex optimization revolves around trust-region methods , in which each iteration solves a non-convex constrained quadratic problem known as the trust-region problem. The classical iteration complexity of trust-region methods for finding an @math -second-order stationary point is @math . Solving the trust-region problem may take a super-liner time in the dimensionality of the problem. For example, @cite_6 designed a first-order method that finds an @math -approximate solution to the trust-region problem with @math time complexity, where @math is the number of non-zero elements in the involved Hessian matrix that could be @math . Recently, @cite_8 designed an efficient variant of the trust-region method that allows for an inexact Hessian with the same iteration complexity @math . They also showed that the trust-region problem can be solved approximately by searching a direction in a two dimentional space consisting of the negative gradient and the negative curvature. Recent proposals on different variants of trust-region methods match the iteration complexity of 's cubic regularization method, i.e., @math . Nonetheless, their per-iteration costs could be still super-linear of the problem's dimensionality. | {
"cite_N": [
"@cite_6",
"@cite_8"
],
"mid": [
"1483062147",
"2747280560"
],
"abstract": [
"We consider the fundamental problem of minimizing a general quadratic function over an ellipsoidal domain, also known as the trust region (sub)problem. We give the first provable linear-time (in the number of non-zero entries of the input) algorithm for approximately solving this problem. Specifically, our algorithm returns an @math ∈-approximate solution in runtime of order N @math ∈, where N is the number of non-zero entries in the input. This matches the runtime of Nesterov's accelerated gradient descent, suitable for the special case in which the quadratic objective is convex, and the runtime of the Lanczos method which is applicable when the problem is purely quadratic.",
"We consider variants of trust-region and cubic regularization methods for non-convex optimization, in which the Hessian matrix is approximated. Under mild conditions on the inexact Hessian, and using approximate solution of the corresponding sub-problems, we provide iteration complexity to achieve @math -approximate second-order optimality which have shown to be tight. Our Hessian approximation conditions constitute a major relaxation over the existing ones in the literature. Consequently, we are able to show that such mild conditions allow for the construction of the approximate Hessian through various random sampling methods. In this light, we consider the canonical problem of finite-sum minimization, provide appropriate uniform and non-uniform sub-sampling strategies to construct such Hessian approximations, and obtain optimal iteration complexity for the corresponding sub-sampled trust-region and cubic regularization methods."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.