diff --git "a/parse/train/HOFxeCutxZR/HOFxeCutxZR_middle.json" "b/parse/train/HOFxeCutxZR/HOFxeCutxZR_middle.json" new file mode 100644--- /dev/null +++ "b/parse/train/HOFxeCutxZR/HOFxeCutxZR_middle.json" @@ -0,0 +1,37336 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 502, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 505, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 505, + 97 + ], + "score": 1.0, + "content": "ENJOY YOUR EDITING: CONTROLLABLE GANS FOR", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 473, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 473, + 118 + ], + "score": 1.0, + "content": "IMAGE EDITING VIA LATENT SPACE NAVIGATION", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 364, + 168 + ], + "lines": [ + { + "bbox": [ + 111, + 134, + 365, + 149 + ], + "spans": [ + { + "bbox": [ + 111, + 134, + 365, + 149 + ], + "score": 1.0, + "content": "Peiye Zhuang, Oluwasanmi Koyejo, Alexander G. Schwing", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 145, + 289, + 159 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 289, + 159 + ], + "score": 1.0, + "content": "University of Illinois at Urbana-Champaign", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 113, + 157, + 334, + 171 + ], + "spans": [ + { + "bbox": [ + 113, + 157, + 334, + 171 + ], + "score": 1.0, + "content": "{peiye, sanmi, aschwing}@illinois.edu", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 278, + 198, + 333, + 209 + ], + "lines": [ + { + "bbox": [ + 276, + 197, + 335, + 210 + ], + "spans": [ + { + "bbox": [ + 276, + 197, + 335, + 210 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 143, + 224, + 468, + 398 + ], + "lines": [ + { + "bbox": [ + 142, + 223, + 469, + 236 + ], + "spans": [ + { + "bbox": [ + 142, + 223, + 469, + 236 + ], + "score": 1.0, + "content": "Controllable semantic image editing enables a user to change entire image at-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 234, + 469, + 246 + ], + "spans": [ + { + "bbox": [ + 141, + 234, + 469, + 246 + ], + "score": 1.0, + "content": "tributes with a few clicks, e.g., gradually making a summer scene look like it was", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 245, + 470, + 258 + ], + "spans": [ + { + "bbox": [ + 141, + 245, + 470, + 258 + ], + "score": 1.0, + "content": "taken in winter. Classic approaches for this task use a Generative Adversarial Net", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 142, + 257, + 470, + 268 + ], + "spans": [ + { + "bbox": [ + 142, + 257, + 470, + 268 + ], + "score": 1.0, + "content": "(GAN) to learn a latent space and suitable latent-space transformations. However,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 268, + 469, + 279 + ], + "spans": [ + { + "bbox": [ + 141, + 268, + 469, + 279 + ], + "score": 1.0, + "content": "current approaches often suffer from attribute edits that are entangled, global im-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 278, + 470, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 278, + 470, + 291 + ], + "score": 1.0, + "content": "age identity changes, and diminished photo-realism. To address these concerns,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 289, + 469, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 289, + 469, + 302 + ], + "score": 1.0, + "content": "we learn multiple attribute transformations simultaneously, integrate attribute re-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 300, + 469, + 313 + ], + "spans": [ + { + "bbox": [ + 141, + 300, + 469, + 313 + ], + "score": 1.0, + "content": "gression into the training of transformation functions, and apply a content loss", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 311, + 469, + 323 + ], + "spans": [ + { + "bbox": [ + 142, + 311, + 469, + 323 + ], + "score": 1.0, + "content": "and an adversarial loss that encourages the maintenance of image identity and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 322, + 470, + 335 + ], + "spans": [ + { + "bbox": [ + 141, + 322, + 470, + 335 + ], + "score": 1.0, + "content": "photo-realism. We propose quantitative evaluation strategies for measuring con-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 333, + 470, + 346 + ], + "spans": [ + { + "bbox": [ + 141, + 333, + 470, + 346 + ], + "score": 1.0, + "content": "trollable editing performance, unlike prior work, which primarily focuses on qual-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 344, + 469, + 356 + ], + "spans": [ + { + "bbox": [ + 142, + 344, + 469, + 356 + ], + "score": 1.0, + "content": "itative evaluation. Our model permits better control for both single- and multiple-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 354, + 469, + 367 + ], + "spans": [ + { + "bbox": [ + 141, + 354, + 469, + 367 + ], + "score": 1.0, + "content": "attribute editing while preserving image identity and realism during transforma-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 366, + 469, + 378 + ], + "spans": [ + { + "bbox": [ + 141, + 366, + 469, + 378 + ], + "score": 1.0, + "content": "tion. We provide empirical results for both natural and synthetic images, high-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 376, + 469, + 390 + ], + "spans": [ + { + "bbox": [ + 141, + 376, + 469, + 390 + ], + "score": 1.0, + "content": "lighting that our model achieves state-of-the-art performance for targeted image", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 388, + 200, + 399 + ], + "spans": [ + { + "bbox": [ + 142, + 388, + 200, + 399 + ], + "score": 1.0, + "content": "manipulation.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 13.5 + }, + { + "type": "title", + "bbox": [ + 108, + 415, + 206, + 427 + ], + "lines": [ + { + "bbox": [ + 105, + 413, + 208, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 208, + 430 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 448 + ], + "score": 1.0, + "content": "Semantic image editing is the task of transforming a source image to a target image while modifying", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 445, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 506, + 459 + ], + "score": 1.0, + "content": "desired semantic attributes, e.g., to make an image taken during summer look like it was captured", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "in winter. The ability to semantically edit images is useful for various real-world tasks, including", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 468, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 480 + ], + "score": 1.0, + "content": "artistic visualization, design, photo enhancement, and targeted data augmentation. To this end,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "semantic image editing has two primary goals: (i) providing continuous manipulation of multiple", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "attributes simultaneously and (ii) maintaining the original image’s identity as much as possible while", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 502, + 204, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 204, + 512 + ], + "score": 1.0, + "content": "ensuring photo-realism.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "Existing GAN-based approaches for semantic image editing can be categorized roughly into two", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "groups: (i) image-space editing methods directly transform one image to another across do-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 538, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 505, + 552 + ], + "score": 1.0, + "content": "mains (Choi et al., 2018; 2020; Isola et al., 2017; Lee et al., 2020; Wu et al., 2019; Zhu et al.,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "2017a;b), usually using variants of generative adversarial nets (GANs) (Goodfellow et al., 2014).", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "score": 1.0, + "content": "These approaches often have high computational cost, and they primarily focus on binary attribute", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 572, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 586 + ], + "score": 1.0, + "content": "(on/off) changes, rather than providing continuous attribute editing abilities. (ii) latent-space edit-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "ing methods focus on discovering latent variable manipulations that permit continuous semantic", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "image edits. The chosen latent space is most often the latent space of GANs. Both unsupervised", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "and (self-)supervised latent space editing methods have been proposed. Unsupervised latent-space", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "editing methods (Hark ¨ onen et al., 2020; Voynov & Babenko, 2020) are often less effective at pro- ¨", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "viding semantically meaningful directions and all too often change image identity during an edit.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 636, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 505, + 652 + ], + "score": 1.0, + "content": "Current (self-)supervised methods (Jahanian et al., 2019; Plumerault et al., 2020) are limited to ge-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "ometric edits such as rotation and scale. To our knowledge, only one supervised approach has been", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "score": 1.0, + "content": "proposed (Shen et al., 2019) – developed to discover semantic latent-space directions for binary", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "attributes. As we show, this method suffers from entangled attributes and often does not preserve", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 682, + 252, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 252, + 694 + ], + "score": 1.0, + "content": "image identity during manipulation.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Contributions. We propose a latent-space editing framework for semantic image manipulation that", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "fulfills the aforementioned primary goals. Specifically, we use a GAN and employ a joint sampling", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "strategy trained to edit multiple attributes simultaneously. To disentangle attribute transformations", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 502, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 505, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 505, + 97 + ], + "score": 1.0, + "content": "ENJOY YOUR EDITING: CONTROLLABLE GANS FOR", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 473, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 473, + 118 + ], + "score": 1.0, + "content": "IMAGE EDITING VIA LATENT SPACE NAVIGATION", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 364, + 168 + ], + "lines": [ + { + "bbox": [ + 111, + 134, + 365, + 149 + ], + "spans": [ + { + "bbox": [ + 111, + 134, + 365, + 149 + ], + "score": 1.0, + "content": "Peiye Zhuang, Oluwasanmi Koyejo, Alexander G. Schwing", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 145, + 289, + 159 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 289, + 159 + ], + "score": 1.0, + "content": "University of Illinois at Urbana-Champaign", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 113, + 157, + 334, + 171 + ], + "spans": [ + { + "bbox": [ + 113, + 157, + 334, + 171 + ], + "score": 1.0, + "content": "{peiye, sanmi, aschwing}@illinois.edu", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 111, + 134, + 365, + 171 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 198, + 333, + 209 + ], + "lines": [ + { + "bbox": [ + 276, + 197, + 335, + 210 + ], + "spans": [ + { + "bbox": [ + 276, + 197, + 335, + 210 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 143, + 224, + 468, + 398 + ], + "lines": [ + { + "bbox": [ + 142, + 223, + 469, + 236 + ], + "spans": [ + { + "bbox": [ + 142, + 223, + 469, + 236 + ], + "score": 1.0, + "content": "Controllable semantic image editing enables a user to change entire image at-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 234, + 469, + 246 + ], + "spans": [ + { + "bbox": [ + 141, + 234, + 469, + 246 + ], + "score": 1.0, + "content": "tributes with a few clicks, e.g., gradually making a summer scene look like it was", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 245, + 470, + 258 + ], + "spans": [ + { + "bbox": [ + 141, + 245, + 470, + 258 + ], + "score": 1.0, + "content": "taken in winter. Classic approaches for this task use a Generative Adversarial Net", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 142, + 257, + 470, + 268 + ], + "spans": [ + { + "bbox": [ + 142, + 257, + 470, + 268 + ], + "score": 1.0, + "content": "(GAN) to learn a latent space and suitable latent-space transformations. However,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 268, + 469, + 279 + ], + "spans": [ + { + "bbox": [ + 141, + 268, + 469, + 279 + ], + "score": 1.0, + "content": "current approaches often suffer from attribute edits that are entangled, global im-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 278, + 470, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 278, + 470, + 291 + ], + "score": 1.0, + "content": "age identity changes, and diminished photo-realism. To address these concerns,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 289, + 469, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 289, + 469, + 302 + ], + "score": 1.0, + "content": "we learn multiple attribute transformations simultaneously, integrate attribute re-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 300, + 469, + 313 + ], + "spans": [ + { + "bbox": [ + 141, + 300, + 469, + 313 + ], + "score": 1.0, + "content": "gression into the training of transformation functions, and apply a content loss", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 311, + 469, + 323 + ], + "spans": [ + { + "bbox": [ + 142, + 311, + 469, + 323 + ], + "score": 1.0, + "content": "and an adversarial loss that encourages the maintenance of image identity and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 322, + 470, + 335 + ], + "spans": [ + { + "bbox": [ + 141, + 322, + 470, + 335 + ], + "score": 1.0, + "content": "photo-realism. We propose quantitative evaluation strategies for measuring con-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 333, + 470, + 346 + ], + "spans": [ + { + "bbox": [ + 141, + 333, + 470, + 346 + ], + "score": 1.0, + "content": "trollable editing performance, unlike prior work, which primarily focuses on qual-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 344, + 469, + 356 + ], + "spans": [ + { + "bbox": [ + 142, + 344, + 469, + 356 + ], + "score": 1.0, + "content": "itative evaluation. Our model permits better control for both single- and multiple-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 354, + 469, + 367 + ], + "spans": [ + { + "bbox": [ + 141, + 354, + 469, + 367 + ], + "score": 1.0, + "content": "attribute editing while preserving image identity and realism during transforma-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 366, + 469, + 378 + ], + "spans": [ + { + "bbox": [ + 141, + 366, + 469, + 378 + ], + "score": 1.0, + "content": "tion. We provide empirical results for both natural and synthetic images, high-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 376, + 469, + 390 + ], + "spans": [ + { + "bbox": [ + 141, + 376, + 469, + 390 + ], + "score": 1.0, + "content": "lighting that our model achieves state-of-the-art performance for targeted image", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 388, + 200, + 399 + ], + "spans": [ + { + "bbox": [ + 142, + 388, + 200, + 399 + ], + "score": 1.0, + "content": "manipulation.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 13.5, + "bbox_fs": [ + 141, + 223, + 470, + 399 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 415, + 206, + 427 + ], + "lines": [ + { + "bbox": [ + 105, + 413, + 208, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 208, + 430 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 448 + ], + "score": 1.0, + "content": "Semantic image editing is the task of transforming a source image to a target image while modifying", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 445, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 506, + 459 + ], + "score": 1.0, + "content": "desired semantic attributes, e.g., to make an image taken during summer look like it was captured", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "in winter. The ability to semantically edit images is useful for various real-world tasks, including", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 468, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 480 + ], + "score": 1.0, + "content": "artistic visualization, design, photo enhancement, and targeted data augmentation. To this end,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "semantic image editing has two primary goals: (i) providing continuous manipulation of multiple", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "attributes simultaneously and (ii) maintaining the original image’s identity as much as possible while", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 502, + 204, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 204, + 512 + ], + "score": 1.0, + "content": "ensuring photo-realism.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 434, + 506, + 512 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "Existing GAN-based approaches for semantic image editing can be categorized roughly into two", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "groups: (i) image-space editing methods directly transform one image to another across do-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 538, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 505, + 552 + ], + "score": 1.0, + "content": "mains (Choi et al., 2018; 2020; Isola et al., 2017; Lee et al., 2020; Wu et al., 2019; Zhu et al.,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "2017a;b), usually using variants of generative adversarial nets (GANs) (Goodfellow et al., 2014).", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "score": 1.0, + "content": "These approaches often have high computational cost, and they primarily focus on binary attribute", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 572, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 586 + ], + "score": 1.0, + "content": "(on/off) changes, rather than providing continuous attribute editing abilities. (ii) latent-space edit-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "ing methods focus on discovering latent variable manipulations that permit continuous semantic", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "image edits. The chosen latent space is most often the latent space of GANs. Both unsupervised", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "and (self-)supervised latent space editing methods have been proposed. Unsupervised latent-space", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "editing methods (Hark ¨ onen et al., 2020; Voynov & Babenko, 2020) are often less effective at pro- ¨", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "viding semantically meaningful directions and all too often change image identity during an edit.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 636, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 505, + 652 + ], + "score": 1.0, + "content": "Current (self-)supervised methods (Jahanian et al., 2019; Plumerault et al., 2020) are limited to ge-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "ometric edits such as rotation and scale. To our knowledge, only one supervised approach has been", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "score": 1.0, + "content": "proposed (Shen et al., 2019) – developed to discover semantic latent-space directions for binary", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "attributes. As we show, this method suffers from entangled attributes and often does not preserve", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 682, + 252, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 252, + 694 + ], + "score": 1.0, + "content": "image identity during manipulation.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 516, + 506, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Contributions. We propose a latent-space editing framework for semantic image manipulation that", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "fulfills the aforementioned primary goals. Specifically, we use a GAN and employ a joint sampling", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "strategy trained to edit multiple attributes simultaneously. To disentangle attribute transformations", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "in the latent space of GANs, we integrate a regressor to predict the attributes that an image exhibits.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "score": 1.0, + "content": "The regressor also permits precise control of the manipulation degree and is easily extended to", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "multiple attributes simultaneously. In addition, we incorporate a perceptual loss (Li et al., 2019) and", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 426, + 480, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 480, + 439 + ], + "score": 1.0, + "content": "an adversarial loss that helps preserve image identity and photo-realism during manipulation.", + "type": "text", + "cross_page": true + } + ], + "index": 12 + } + ], + "index": 47, + "bbox_fs": [ + 105, + 699, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 113, + 71, + 497, + 307 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 71, + 497, + 307 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 71, + 497, + 307 + ], + "spans": [ + { + "bbox": [ + 113, + 71, + 497, + 307 + ], + "score": 0.975, + "type": "image", + "image_path": "fb07bc41864ba9b101d2f40fa0c12c3111aac85a77e90532d3aaba7be5d481e0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 113, + 71, + 497, + 149.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 113, + 149.66666666666669, + 497, + 228.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 113, + 228.33333333333337, + 497, + 307.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 317, + 505, + 383 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "Figure 1: Real image manipulation on scene (top two rows, photo from Flickr) and face (bottom", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "two rows, unseen image from CelebA-HQ) using pretrained StyleGAN2 (Karras et al., 2019b): We", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 339, + 504, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 504, + 351 + ], + "score": 1.0, + "content": "reconstruct the real images (col.1) by finding a latent vector with the best inversion result (col.2) on", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "StyleGAN2 (Abdal et al., 2019). After that, we transform the latent vectors for single- and multiple-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 505, + 373 + ], + "score": 1.0, + "content": "attribute manipulations (col.3-6). Note that unlike ours, the baseline method (Shen et al., 2019)", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 371, + 397, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 397, + 384 + ], + "score": 1.0, + "content": "either changes image identity or confounds semantic properties, or both.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 107, + 393, + 505, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "in the latent space of GANs, we integrate a regressor to predict the attributes that an image exhibits.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "score": 1.0, + "content": "The regressor also permits precise control of the manipulation degree and is easily extended to", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "multiple attributes simultaneously. In addition, we incorporate a perceptual loss (Li et al., 2019) and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 426, + 480, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 480, + 439 + ], + "score": 1.0, + "content": "an adversarial loss that helps preserve image identity and photo-realism during manipulation.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "score": 1.0, + "content": "We compare our method to several popular frameworks, from existing image-to-image translation", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 453, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 467 + ], + "score": 1.0, + "content": "methods (Choi et al., 2020; Wu et al., 2019; Zhu et al., 2017a) to latent space transformation-based", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "approaches (Shen et al., 2019; Voynov & Babenko, 2020). We mention that prior work primarily", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 477, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 489 + ], + "score": 1.0, + "content": "uses qualitative evaluation like the one in Fig. 1. In contrast, we propose a quantitative evaluation to", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "measure controllability. Both qualitative and quantitative results provide evidence that our approach", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 497, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 505, + 512 + ], + "score": 1.0, + "content": "outperforms prior work in terms of quality of the semantic image manipulation while maintaining", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 508, + 168, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 168, + 523 + ], + "score": 1.0, + "content": "image identity.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 108, + 535, + 209, + 547 + ], + "lines": [ + { + "bbox": [ + 104, + 533, + 210, + 550 + ], + "spans": [ + { + "bbox": [ + 104, + 533, + 210, + 550 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 561, + 505, + 638 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 505, + 574 + ], + "score": 1.0, + "content": "Generative Adversarial Networks (GANs) (Goodfellow et al., 2014) have significantly improved", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "realistic image generation in recent years (Brock et al., 2018; Jolicoeur-Martineau, 2019; Karras", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "et al., 2017; 2019a;b; Park et al., 2019; Zhang et al., 2018). For this, a GAN formulates a 2-player", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 595, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 607 + ], + "score": 1.0, + "content": "non-cooperative game between two deep nets: (i) a generator that produces an image given a random", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 606, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 506, + 617 + ], + "score": 1.0, + "content": "noise vector in the latent space, sampled from a known prior distribution, usually a normal or a", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "uniform distribution; (ii) a discriminator whose input is both synthetic and real data, which is to be", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 627, + 164, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 164, + 638 + ], + "score": 1.0, + "content": "differentiated.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "score": 1.0, + "content": "Semantic image editing seeks to automate image manipulation of semantics. Encouraged by the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "success of deep nets, recent works have applied deep learning methods for semantic image editing", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "tasks such as style transfer (Li et al., 2017; Luan et al., 2017), image-to-image translation (Choi et al.,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "2018; 2020; Isola et al., 2017; Lee et al., 2020; Wang et al., 2018; Wu et al., 2019; Zhu et al., 2017a)", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "and discovering semantically meaningful directions in a GAN latent space (Hark ¨ onen et al., 2020; ¨", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "Jahanian et al., 2019; Plumerault et al., 2020; Shen et al., 2019; Voynov & Babenko, 2020). Note", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "that our task is an extended version of semantic image editing that requires more comprehensive", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "control to satisfy user-desired operations. Therefore, most aforementioned approaches do not meet", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 113, + 71, + 497, + 307 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 71, + 497, + 307 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 71, + 497, + 307 + ], + "spans": [ + { + "bbox": [ + 113, + 71, + 497, + 307 + ], + "score": 0.975, + "type": "image", + "image_path": "fb07bc41864ba9b101d2f40fa0c12c3111aac85a77e90532d3aaba7be5d481e0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 113, + 71, + 497, + 149.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 113, + 149.66666666666669, + 497, + 228.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 113, + 228.33333333333337, + 497, + 307.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 317, + 505, + 383 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "Figure 1: Real image manipulation on scene (top two rows, photo from Flickr) and face (bottom", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "two rows, unseen image from CelebA-HQ) using pretrained StyleGAN2 (Karras et al., 2019b): We", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 339, + 504, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 504, + 351 + ], + "score": 1.0, + "content": "reconstruct the real images (col.1) by finding a latent vector with the best inversion result (col.2) on", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "StyleGAN2 (Abdal et al., 2019). After that, we transform the latent vectors for single- and multiple-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 505, + 373 + ], + "score": 1.0, + "content": "attribute manipulations (col.3-6). Note that unlike ours, the baseline method (Shen et al., 2019)", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 371, + 397, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 397, + 384 + ], + "score": 1.0, + "content": "either changes image identity or confounds semantic properties, or both.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 107, + 393, + 505, + 438 + ], + "lines": [], + "index": 10.5, + "bbox_fs": [ + 105, + 393, + 506, + 439 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "score": 1.0, + "content": "We compare our method to several popular frameworks, from existing image-to-image translation", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 453, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 467 + ], + "score": 1.0, + "content": "methods (Choi et al., 2020; Wu et al., 2019; Zhu et al., 2017a) to latent space transformation-based", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "approaches (Shen et al., 2019; Voynov & Babenko, 2020). We mention that prior work primarily", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 477, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 489 + ], + "score": 1.0, + "content": "uses qualitative evaluation like the one in Fig. 1. In contrast, we propose a quantitative evaluation to", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "measure controllability. Both qualitative and quantitative results provide evidence that our approach", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 497, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 505, + 512 + ], + "score": 1.0, + "content": "outperforms prior work in terms of quality of the semantic image manipulation while maintaining", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 508, + 168, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 168, + 523 + ], + "score": 1.0, + "content": "image identity.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 443, + 506, + 523 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 535, + 209, + 547 + ], + "lines": [ + { + "bbox": [ + 104, + 533, + 210, + 550 + ], + "spans": [ + { + "bbox": [ + 104, + 533, + 210, + 550 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 561, + 505, + 638 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 505, + 574 + ], + "score": 1.0, + "content": "Generative Adversarial Networks (GANs) (Goodfellow et al., 2014) have significantly improved", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "realistic image generation in recent years (Brock et al., 2018; Jolicoeur-Martineau, 2019; Karras", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "et al., 2017; 2019a;b; Park et al., 2019; Zhang et al., 2018). For this, a GAN formulates a 2-player", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 595, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 607 + ], + "score": 1.0, + "content": "non-cooperative game between two deep nets: (i) a generator that produces an image given a random", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 606, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 506, + 617 + ], + "score": 1.0, + "content": "noise vector in the latent space, sampled from a known prior distribution, usually a normal or a", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "uniform distribution; (ii) a discriminator whose input is both synthetic and real data, which is to be", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 627, + 164, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 164, + 638 + ], + "score": 1.0, + "content": "differentiated.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 560, + 506, + 638 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "score": 1.0, + "content": "Semantic image editing seeks to automate image manipulation of semantics. Encouraged by the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "success of deep nets, recent works have applied deep learning methods for semantic image editing", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "tasks such as style transfer (Li et al., 2017; Luan et al., 2017), image-to-image translation (Choi et al.,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "2018; 2020; Isola et al., 2017; Lee et al., 2020; Wang et al., 2018; Wu et al., 2019; Zhu et al., 2017a)", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "and discovering semantically meaningful directions in a GAN latent space (Hark ¨ onen et al., 2020; ¨", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "Jahanian et al., 2019; Plumerault et al., 2020; Shen et al., 2019; Voynov & Babenko, 2020). Note", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "that our task is an extended version of semantic image editing that requires more comprehensive", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "control to satisfy user-desired operations. Therefore, most aforementioned approaches do not meet", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 365, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 505, + 377 + ], + "score": 1.0, + "content": "the requirements. Nonetheless, we categorize the approaches most relevant to our task into two", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 376, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 389 + ], + "score": 1.0, + "content": "groups: (i) image editing via manipulation in image space, and (ii) image editing via latent space", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 387, + 154, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 154, + 400 + ], + "score": 1.0, + "content": "navigation.", + "type": "text", + "cross_page": true + } + ], + "index": 11 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 643, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 70, + 496, + 268 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 70, + 496, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 70, + 496, + 268 + ], + "spans": [ + { + "bbox": [ + 108, + 70, + 496, + 268 + ], + "score": 0.973, + "type": "image", + "image_path": "396e29840d9828a2d1068da81859a87debad70ba35b15b3077f0c3d628a6e7d9.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 70, + 496, + 136.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 136.0, + 496, + 202.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 202.0, + 496, + 268.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 277, + 505, + 344 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "score": 1.0, + "content": "Figure 2: Our overall framework (top row) and 3 training strategies different from prior work", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 204, + 302 + ], + "score": 1.0, + "content": "(bottom row). In (a),", + "type": "text" + }, + { + "bbox": [ + 204, + 289, + 213, + 299 + ], + "score": 0.74, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 288, + 240, + 302 + ], + "score": 1.0, + "content": "takes", + "type": "text" + }, + { + "bbox": [ + 240, + 291, + 249, + 299 + ], + "score": 0.73, + "content": "_ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "and an edited latent vector separately to synthesize images.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 182, + 312 + ], + "score": 0.92, + "content": "T = \\{ d _ { 1 } , . . . , d _ { N } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 299, + 362, + 312 + ], + "score": 1.0, + "content": "are the trainable latent-space directions and", + "type": "text" + }, + { + "bbox": [ + 362, + 302, + 369, + 310 + ], + "score": 0.66, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "represents transformation extent.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 258, + 323 + ], + "score": 1.0, + "content": "Original and edited image attributes,", + "type": "text" + }, + { + "bbox": [ + 259, + 313, + 268, + 321 + ], + "score": 0.75, + "content": "_ { \\pmb { \\alpha } }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 311, + 287, + 323 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 287, + 311, + 296, + 321 + ], + "score": 0.81, + "content": "\\hat { \\alpha }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 311, + 471, + 323 + ], + "score": 1.0, + "content": ", are predicted by the pre-trained regressor", + "type": "text" + }, + { + "bbox": [ + 471, + 311, + 480, + 321 + ], + "score": 0.53, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 311, + 505, + 323 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 321, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 104, + 321, + 181, + 336 + ], + "score": 1.0, + "content": "discriminator loss", + "type": "text" + }, + { + "bbox": [ + 182, + 322, + 201, + 333 + ], + "score": 0.9, + "content": "{ \\mathcal { L } } _ { \\mathrm { d i s c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 321, + 279, + 336 + ], + "score": 1.0, + "content": "(the discriminator", + "type": "text" + }, + { + "bbox": [ + 279, + 322, + 289, + 332 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 321, + 506, + 336 + ], + "score": 1.0, + "content": "not shown due to limited space), the regression loss", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 333, + 457, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 124, + 345 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 333, + 221, + 345 + ], + "score": 1.0, + "content": ", and the perceptual loss", + "type": "text" + }, + { + "bbox": [ + 221, + 333, + 250, + 344 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { { c o n t e n t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 333, + 326, + 345 + ], + "score": 1.0, + "content": "are used to update", + "type": "text" + }, + { + "bbox": [ + 326, + 333, + 335, + 343 + ], + "score": 0.68, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 333, + 457, + 345 + ], + "score": 1.0, + "content": ". We explain (b-d) in Sec. 3.2.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 107, + 365, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 106, + 365, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 505, + 377 + ], + "score": 1.0, + "content": "the requirements. Nonetheless, we categorize the approaches most relevant to our task into two", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 376, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 389 + ], + "score": 1.0, + "content": "groups: (i) image editing via manipulation in image space, and (ii) image editing via latent space", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 387, + 154, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 154, + 400 + ], + "score": 1.0, + "content": "navigation.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 504, + 470 + ], + "lines": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "Image-space editing using GANs directly manipulates an image for targeted editing. Early", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "work (Isola et al., 2017) used GANs to implement semantic translations between two image do-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 425, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 506, + 438 + ], + "score": 1.0, + "content": "mains with paired data, e.g., day to night. Follow-up work focused on multi-modal (Bhattarai &", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "score": 1.0, + "content": "Kim, 2020; Choi et al., 2018; 2020; Lee et al., 2020; Wang et al., 2018; Wu et al., 2019) and un-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 447, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 505, + 461 + ], + "score": 1.0, + "content": "paired image domain translation (Zhu et al., 2017a). In this case, they primarily consider binary", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 459, + 502, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 502, + 471 + ], + "score": 1.0, + "content": "(on/off) attribute changes, regardless of whether the process is dynamic, for example, day to night.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 475, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "Latent-space editing via GANs has received an increasing amount of recent interest. Most prior", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "score": 1.0, + "content": "work focused on identifying semantically meaningful directions in the latent space of GANs, so", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "that shifting latent vectors towards these directions achieves the desired image manipulation. Re-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "cent papers found semantics in the latent space of GANs, such as color transformations and camera", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "movements (Jahanian et al., 2019), or face attribute changes (Shen et al., 2019), such as smile. Other", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 530, + 504, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 504, + 543 + ], + "score": 1.0, + "content": "work considered unsupervised methods (Hark ¨ onen et al., 2020; Voynov & Babenko, 2020) to dis-¨", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "score": 1.0, + "content": "cover interpretable latent space directions. However, additional challenges inherent to unsupervised", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 553, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 505, + 565 + ], + "score": 1.0, + "content": "manipulation of the latent space arise, and have not been addressed in prior work, e.g., direction", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 563, + 385, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 385, + 577 + ], + "score": 1.0, + "content": "quality with regard to degree control and image identity preservation.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 108, + 594, + 172, + 608 + ], + "lines": [ + { + "bbox": [ + 104, + 592, + 174, + 610 + ], + "spans": [ + { + "bbox": [ + 104, + 592, + 174, + 610 + ], + "score": 1.0, + "content": "3 METHOD", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 108, + 621, + 227, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 228, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 228, + 633 + ], + "score": 1.0, + "content": "3.1 PROBLEM STATEMENT", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 644, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 106, + 643, + 505, + 657 + ], + "score": 1.0, + "content": "We consider controllable semantic image editing via latent space navigation in GANs. We begin", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 655, + 504, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 327, + 667 + ], + "score": 1.0, + "content": "with a fixed GAN model that consists of a generator", + "type": "text" + }, + { + "bbox": [ + 327, + 656, + 337, + 665 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 655, + 422, + 667 + ], + "score": 1.0, + "content": "and a discriminator", + "type": "text" + }, + { + "bbox": [ + 422, + 656, + 432, + 665 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 655, + 495, + 667 + ], + "score": 1.0, + "content": ". The input of", + "type": "text" + }, + { + "bbox": [ + 495, + 655, + 504, + 665 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 180, + 678 + ], + "score": 1.0, + "content": "is a latent vector", + "type": "text" + }, + { + "bbox": [ + 180, + 666, + 219, + 676 + ], + "score": 0.92, + "content": "z \\in \\mathbb { R } ^ { m }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 666, + 305, + 678 + ], + "score": 1.0, + "content": "from a latent space", + "type": "text" + }, + { + "bbox": [ + 306, + 667, + 315, + 676 + ], + "score": 0.79, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 666, + 349, + 678 + ], + "score": 1.0, + "content": ". Here,", + "type": "text" + }, + { + "bbox": [ + 350, + 668, + 360, + 676 + ], + "score": 0.74, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "is the dimensionality of the latent", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 160, + 690 + ], + "score": 1.0, + "content": "space. Given", + "type": "text" + }, + { + "bbox": [ + 160, + 677, + 171, + 687 + ], + "score": 0.8, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "attributes, we aim to discover semantically meaningful latent-space GAN directions,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 188, + 700 + ], + "score": 0.9, + "content": "\\bar { \\pmb { T } } = \\{ { \\pmb d } _ { 1 } , \\ldots , { \\pmb d } _ { N } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 687, + 390, + 702 + ], + "score": 1.0, + "content": ", to manipulate the attributes of synthetic images", + "type": "text" + }, + { + "bbox": [ + 390, + 688, + 413, + 700 + ], + "score": 0.91, + "content": "G ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 687, + 506, + 702 + ], + "score": 1.0, + "content": "with an assigned step", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 125, + 711 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 126, + 700, + 201, + 711 + ], + "score": 0.9, + "content": "\\boldsymbol \\varepsilon = \\{ \\varepsilon _ { 1 } , \\ldots , \\varepsilon _ { N } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 698, + 234, + 711 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 234, + 699, + 275, + 710 + ], + "score": 0.92, + "content": "\\pmb { d } _ { i } \\in \\mathbb { R } ^ { m }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 698, + 303, + 711 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 303, + 699, + 369, + 711 + ], + "score": 0.92, + "content": "i \\in \\{ 1 , \\ldots , N \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 698, + 505, + 711 + ], + "score": 1.0, + "content": ". In the end, each attribute of an", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 710, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 345, + 723 + ], + "score": 1.0, + "content": "edited image can be changed with the corresponding degree", + "type": "text" + }, + { + "bbox": [ + 345, + 713, + 352, + 720 + ], + "score": 0.77, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 710, + 375, + 723 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 375, + 710, + 453, + 722 + ], + "score": 0.93, + "content": "{ \\pmb { \\alpha } } = \\{ \\alpha _ { 1 } , . . . , \\alpha _ { N } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 710, + 506, + 723 + ], + "score": 1.0, + "content": ", the original", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 721, + 291, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 157, + 733 + ], + "score": 1.0, + "content": "attributes of", + "type": "text" + }, + { + "bbox": [ + 157, + 721, + 179, + 733 + ], + "score": 0.92, + "content": "G ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 721, + 279, + 733 + ], + "score": 1.0, + "content": "predicted by a regressor", + "type": "text" + }, + { + "bbox": [ + 279, + 721, + 287, + 731 + ], + "score": 0.83, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 721, + 291, + 733 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 70, + 496, + 268 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 70, + 496, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 70, + 496, + 268 + ], + "spans": [ + { + "bbox": [ + 108, + 70, + 496, + 268 + ], + "score": 0.973, + "type": "image", + "image_path": "396e29840d9828a2d1068da81859a87debad70ba35b15b3077f0c3d628a6e7d9.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 70, + 496, + 136.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 136.0, + 496, + 202.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 202.0, + 496, + 268.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 277, + 505, + 344 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "score": 1.0, + "content": "Figure 2: Our overall framework (top row) and 3 training strategies different from prior work", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 204, + 302 + ], + "score": 1.0, + "content": "(bottom row). In (a),", + "type": "text" + }, + { + "bbox": [ + 204, + 289, + 213, + 299 + ], + "score": 0.74, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 288, + 240, + 302 + ], + "score": 1.0, + "content": "takes", + "type": "text" + }, + { + "bbox": [ + 240, + 291, + 249, + 299 + ], + "score": 0.73, + "content": "_ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "and an edited latent vector separately to synthesize images.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 182, + 312 + ], + "score": 0.92, + "content": "T = \\{ d _ { 1 } , . . . , d _ { N } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 299, + 362, + 312 + ], + "score": 1.0, + "content": "are the trainable latent-space directions and", + "type": "text" + }, + { + "bbox": [ + 362, + 302, + 369, + 310 + ], + "score": 0.66, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "represents transformation extent.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 258, + 323 + ], + "score": 1.0, + "content": "Original and edited image attributes,", + "type": "text" + }, + { + "bbox": [ + 259, + 313, + 268, + 321 + ], + "score": 0.75, + "content": "_ { \\pmb { \\alpha } }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 311, + 287, + 323 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 287, + 311, + 296, + 321 + ], + "score": 0.81, + "content": "\\hat { \\alpha }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 311, + 471, + 323 + ], + "score": 1.0, + "content": ", are predicted by the pre-trained regressor", + "type": "text" + }, + { + "bbox": [ + 471, + 311, + 480, + 321 + ], + "score": 0.53, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 311, + 505, + 323 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 321, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 104, + 321, + 181, + 336 + ], + "score": 1.0, + "content": "discriminator loss", + "type": "text" + }, + { + "bbox": [ + 182, + 322, + 201, + 333 + ], + "score": 0.9, + "content": "{ \\mathcal { L } } _ { \\mathrm { d i s c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 321, + 279, + 336 + ], + "score": 1.0, + "content": "(the discriminator", + "type": "text" + }, + { + "bbox": [ + 279, + 322, + 289, + 332 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 321, + 506, + 336 + ], + "score": 1.0, + "content": "not shown due to limited space), the regression loss", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 333, + 457, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 124, + 345 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 333, + 221, + 345 + ], + "score": 1.0, + "content": ", and the perceptual loss", + "type": "text" + }, + { + "bbox": [ + 221, + 333, + 250, + 344 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { { c o n t e n t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 333, + 326, + 345 + ], + "score": 1.0, + "content": "are used to update", + "type": "text" + }, + { + "bbox": [ + 326, + 333, + 335, + 343 + ], + "score": 0.68, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 333, + 457, + 345 + ], + "score": 1.0, + "content": ". We explain (b-d) in Sec. 3.2.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 107, + 365, + 505, + 398 + ], + "lines": [], + "index": 10, + "bbox_fs": [ + 105, + 365, + 506, + 400 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 504, + 470 + ], + "lines": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "Image-space editing using GANs directly manipulates an image for targeted editing. Early", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "work (Isola et al., 2017) used GANs to implement semantic translations between two image do-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 425, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 506, + 438 + ], + "score": 1.0, + "content": "mains with paired data, e.g., day to night. Follow-up work focused on multi-modal (Bhattarai &", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "score": 1.0, + "content": "Kim, 2020; Choi et al., 2018; 2020; Lee et al., 2020; Wang et al., 2018; Wu et al., 2019) and un-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 447, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 505, + 461 + ], + "score": 1.0, + "content": "paired image domain translation (Zhu et al., 2017a). In this case, they primarily consider binary", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 459, + 502, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 502, + 471 + ], + "score": 1.0, + "content": "(on/off) attribute changes, regardless of whether the process is dynamic, for example, day to night.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 403, + 506, + 471 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 475, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "Latent-space editing via GANs has received an increasing amount of recent interest. Most prior", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "score": 1.0, + "content": "work focused on identifying semantically meaningful directions in the latent space of GANs, so", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "that shifting latent vectors towards these directions achieves the desired image manipulation. Re-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "cent papers found semantics in the latent space of GANs, such as color transformations and camera", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "movements (Jahanian et al., 2019), or face attribute changes (Shen et al., 2019), such as smile. Other", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 530, + 504, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 504, + 543 + ], + "score": 1.0, + "content": "work considered unsupervised methods (Hark ¨ onen et al., 2020; Voynov & Babenko, 2020) to dis-¨", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "score": 1.0, + "content": "cover interpretable latent space directions. However, additional challenges inherent to unsupervised", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 553, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 505, + 565 + ], + "score": 1.0, + "content": "manipulation of the latent space arise, and have not been addressed in prior work, e.g., direction", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 563, + 385, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 385, + 577 + ], + "score": 1.0, + "content": "quality with regard to degree control and image identity preservation.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 476, + 506, + 577 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 594, + 172, + 608 + ], + "lines": [ + { + "bbox": [ + 104, + 592, + 174, + 610 + ], + "spans": [ + { + "bbox": [ + 104, + 592, + 174, + 610 + ], + "score": 1.0, + "content": "3 METHOD", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 108, + 621, + 227, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 228, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 228, + 633 + ], + "score": 1.0, + "content": "3.1 PROBLEM STATEMENT", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 644, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 106, + 643, + 505, + 657 + ], + "score": 1.0, + "content": "We consider controllable semantic image editing via latent space navigation in GANs. We begin", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 655, + 504, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 327, + 667 + ], + "score": 1.0, + "content": "with a fixed GAN model that consists of a generator", + "type": "text" + }, + { + "bbox": [ + 327, + 656, + 337, + 665 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 655, + 422, + 667 + ], + "score": 1.0, + "content": "and a discriminator", + "type": "text" + }, + { + "bbox": [ + 422, + 656, + 432, + 665 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 655, + 495, + 667 + ], + "score": 1.0, + "content": ". The input of", + "type": "text" + }, + { + "bbox": [ + 495, + 655, + 504, + 665 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 180, + 678 + ], + "score": 1.0, + "content": "is a latent vector", + "type": "text" + }, + { + "bbox": [ + 180, + 666, + 219, + 676 + ], + "score": 0.92, + "content": "z \\in \\mathbb { R } ^ { m }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 666, + 305, + 678 + ], + "score": 1.0, + "content": "from a latent space", + "type": "text" + }, + { + "bbox": [ + 306, + 667, + 315, + 676 + ], + "score": 0.79, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 666, + 349, + 678 + ], + "score": 1.0, + "content": ". Here,", + "type": "text" + }, + { + "bbox": [ + 350, + 668, + 360, + 676 + ], + "score": 0.74, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "is the dimensionality of the latent", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 160, + 690 + ], + "score": 1.0, + "content": "space. Given", + "type": "text" + }, + { + "bbox": [ + 160, + 677, + 171, + 687 + ], + "score": 0.8, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "attributes, we aim to discover semantically meaningful latent-space GAN directions,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 188, + 700 + ], + "score": 0.9, + "content": "\\bar { \\pmb { T } } = \\{ { \\pmb d } _ { 1 } , \\ldots , { \\pmb d } _ { N } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 687, + 390, + 702 + ], + "score": 1.0, + "content": ", to manipulate the attributes of synthetic images", + "type": "text" + }, + { + "bbox": [ + 390, + 688, + 413, + 700 + ], + "score": 0.91, + "content": "G ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 687, + 506, + 702 + ], + "score": 1.0, + "content": "with an assigned step", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 125, + 711 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 126, + 700, + 201, + 711 + ], + "score": 0.9, + "content": "\\boldsymbol \\varepsilon = \\{ \\varepsilon _ { 1 } , \\ldots , \\varepsilon _ { N } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 698, + 234, + 711 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 234, + 699, + 275, + 710 + ], + "score": 0.92, + "content": "\\pmb { d } _ { i } \\in \\mathbb { R } ^ { m }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 698, + 303, + 711 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 303, + 699, + 369, + 711 + ], + "score": 0.92, + "content": "i \\in \\{ 1 , \\ldots , N \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 698, + 505, + 711 + ], + "score": 1.0, + "content": ". In the end, each attribute of an", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 710, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 345, + 723 + ], + "score": 1.0, + "content": "edited image can be changed with the corresponding degree", + "type": "text" + }, + { + "bbox": [ + 345, + 713, + 352, + 720 + ], + "score": 0.77, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 710, + 375, + 723 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 375, + 710, + 453, + 722 + ], + "score": 0.93, + "content": "{ \\pmb { \\alpha } } = \\{ \\alpha _ { 1 } , . . . , \\alpha _ { N } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 710, + 506, + 723 + ], + "score": 1.0, + "content": ", the original", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 721, + 291, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 157, + 733 + ], + "score": 1.0, + "content": "attributes of", + "type": "text" + }, + { + "bbox": [ + 157, + 721, + 179, + 733 + ], + "score": 0.92, + "content": "G ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 721, + 279, + 733 + ], + "score": 1.0, + "content": "predicted by a regressor", + "type": "text" + }, + { + "bbox": [ + 279, + 721, + 287, + 731 + ], + "score": 0.83, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 721, + 291, + 733 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 643, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 226, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 227, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 227, + 95 + ], + "score": 1.0, + "content": "3.2 PROPOSED APPROACH", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 504, + 126 + ], + "lines": [ + { + "bbox": [ + 106, + 102, + 505, + 115 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 115 + ], + "score": 1.0, + "content": "We provide an overview of our method in Fig. 2 (a), and further illustrate how our approach differs", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 114, + 233, + 126 + ], + "spans": [ + { + "bbox": [ + 107, + 114, + 233, + 126 + ], + "score": 1.0, + "content": "from prior work in Fig. 2 (b-d).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 130, + 505, + 244 + ], + "lines": [ + { + "bbox": [ + 106, + 131, + 506, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 131, + 469, + 144 + ], + "score": 1.0, + "content": "Overview. As shown in Fig. 2 (a), we employ a GAN model consisting of a generator", + "type": "text" + }, + { + "bbox": [ + 469, + 132, + 478, + 141 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 131, + 506, + 144 + ], + "score": 1.0, + "content": "and a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 141, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 163, + 155 + ], + "score": 1.0, + "content": "discriminator", + "type": "text" + }, + { + "bbox": [ + 163, + 142, + 173, + 152 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 141, + 405, + 155 + ], + "score": 1.0, + "content": "(not shown due to limited space), as well as a regressor", + "type": "text" + }, + { + "bbox": [ + 406, + 142, + 414, + 152 + ], + "score": 0.79, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 141, + 505, + 155 + ], + "score": 1.0, + "content": ", all of which are pre-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 153, + 504, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 286, + 164 + ], + "score": 1.0, + "content": "trained. Our goal is to find latent directions", + "type": "text" + }, + { + "bbox": [ + 286, + 153, + 296, + 163 + ], + "score": 0.78, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 153, + 504, + 164 + ], + "score": 1.0, + "content": "that provide attribute specific image edits. At each", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 163, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 162, + 176 + ], + "score": 1.0, + "content": "training step,", + "type": "text" + }, + { + "bbox": [ + 162, + 164, + 171, + 174 + ], + "score": 0.79, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 163, + 196, + 176 + ], + "score": 1.0, + "content": "takes", + "type": "text" + }, + { + "bbox": [ + 196, + 165, + 204, + 174 + ], + "score": 0.75, + "content": "_ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 163, + 371, + 176 + ], + "score": 1.0, + "content": "and an edited latent vector referred to as", + "type": "text" + }, + { + "bbox": [ + 372, + 164, + 381, + 174 + ], + "score": 0.86, + "content": "z ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 163, + 505, + 176 + ], + "score": 1.0, + "content": ". We follow prior work which", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 174, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 505, + 187 + ], + "score": 1.0, + "content": "suggests that direction vectors in latent space of GANs permit image editing (Jahanian et al., 2019).", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 186, + 504, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 275, + 199 + ], + "score": 1.0, + "content": "Formally, given a transformation degree", + "type": "text" + }, + { + "bbox": [ + 275, + 188, + 282, + 196 + ], + "score": 0.64, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 186, + 429, + 199 + ], + "score": 1.0, + "content": ", we obtain the latent-space edit as", + "type": "text" + }, + { + "bbox": [ + 430, + 186, + 504, + 197 + ], + "score": 0.89, + "content": "z ^ { \\prime } = z + { \\cal T } \\varepsilon =", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 194, + 508, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 197, + 168, + 211 + ], + "score": 0.91, + "content": "\\begin{array} { r } { z + \\sum _ { i = 1 } ^ { N } \\varepsilon _ { i } d _ { i } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 194, + 173, + 218 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 173, + 199, + 182, + 209 + ], + "score": 0.7, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 194, + 350, + 218 + ], + "score": 1.0, + "content": "provides the recovered and edited images", + "type": "text" + }, + { + "bbox": [ + 351, + 198, + 373, + 210 + ], + "score": 0.92, + "content": "G ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 194, + 391, + 218 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 392, + 198, + 416, + 210 + ], + "score": 0.92, + "content": "G ( z ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 194, + 508, + 218 + ], + "score": 1.0, + "content": ", which are separately", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 209, + 443, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 217, + 222 + ], + "score": 1.0, + "content": "processed by the regressor", + "type": "text" + }, + { + "bbox": [ + 226, + 209, + 443, + 222 + ], + "score": 1.0, + "content": "to predict attribute values. The original attributes of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 219, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 107, + 221, + 144, + 232 + ], + "score": 0.92, + "content": "R ( G ( z ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 219, + 365, + 234 + ], + "score": 1.0, + "content": ". The (pseudo-) ground truth and predicted attribute of", + "type": "text" + }, + { + "bbox": [ + 365, + 221, + 390, + 233 + ], + "score": 0.93, + "content": "G ( z ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 219, + 407, + 234 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 407, + 221, + 419, + 231 + ], + "score": 0.87, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 219, + 437, + 234 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 437, + 221, + 448, + 231 + ], + "score": 0.86, + "content": "\\hat { \\mathbf { \\alpha } } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 219, + 506, + 234 + ], + "score": 1.0, + "content": ", respectively,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 231, + 270, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 133, + 244 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 232, + 183, + 242 + ], + "score": 0.9, + "content": "{ \\pmb { \\alpha } } ^ { \\prime } = { \\pmb { \\alpha } } + { \\pmb { \\varepsilon } }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 231, + 201, + 244 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 202, + 231, + 266, + 244 + ], + "score": 0.93, + "content": "\\hat { \\pmb { \\alpha } } ^ { \\prime } = R ( G ( \\pmb { z } ^ { \\prime } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 231, + 270, + 244 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 107, + 248, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 106, + 249, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 201, + 261 + ], + "score": 1.0, + "content": "Intuitively, the goal of", + "type": "text" + }, + { + "bbox": [ + 201, + 249, + 211, + 258 + ], + "score": 0.78, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 249, + 276, + 261 + ], + "score": 1.0, + "content": "is to transform", + "type": "text" + }, + { + "bbox": [ + 277, + 250, + 284, + 258 + ], + "score": 0.75, + "content": "_ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 249, + 505, + 261 + ], + "score": 1.0, + "content": "by adding semantically meaningful information such", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 258, + 272 + ], + "score": 1.0, + "content": "that the corresponding output image", + "type": "text" + }, + { + "bbox": [ + 258, + 259, + 284, + 271 + ], + "score": 0.92, + "content": "G ( z ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 259, + 393, + 272 + ], + "score": 1.0, + "content": "exhibits attribute changes", + "type": "text" + }, + { + "bbox": [ + 393, + 261, + 400, + 269 + ], + "score": 0.74, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 259, + 425, + 272 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 425, + 261, + 434, + 269 + ], + "score": 0.74, + "content": "_ { \\pmb { \\alpha } }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 259, + 505, + 272 + ], + "score": 1.0, + "content": ". In practice, we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 276, + 284 + ], + "score": 1.0, + "content": "normalize the range of attribute values to", + "type": "text" + }, + { + "bbox": [ + 277, + 271, + 299, + 282 + ], + "score": 0.82, + "content": "[ \\mathbf { 0 } , \\mathbf { 1 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 270, + 342, + 284 + ], + "score": 1.0, + "content": ", i.e., both", + "type": "text" + }, + { + "bbox": [ + 342, + 272, + 351, + 281 + ], + "score": 0.78, + "content": "_ { \\pmb { \\alpha } }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 270, + 370, + 284 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 370, + 270, + 417, + 282 + ], + "score": 0.92, + "content": "{ \\pmb { \\alpha } } ^ { \\prime } \\in [ { \\bf 0 } , { \\bf 1 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 270, + 506, + 284 + ], + "score": 1.0, + "content": ". During training, we", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 280, + 504, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 307, + 295 + ], + "score": 1.0, + "content": "maintain the unit range by controlling the given", + "type": "text" + }, + { + "bbox": [ + 307, + 284, + 313, + 291 + ], + "score": 0.67, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 280, + 363, + 295 + ], + "score": 1.0, + "content": ". Formally,", + "type": "text" + }, + { + "bbox": [ + 363, + 283, + 370, + 291 + ], + "score": 0.72, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 280, + 490, + 295 + ], + "score": 1.0, + "content": "is drawn from a distribution", + "type": "text" + }, + { + "bbox": [ + 491, + 282, + 504, + 293 + ], + "score": 0.88, + "content": "\\mathcal { D } _ { \\varepsilon }", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 291, + 405, + 305 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 151, + 305 + ], + "score": 1.0, + "content": "uniform in", + "type": "text" + }, + { + "bbox": [ + 152, + 292, + 187, + 304 + ], + "score": 0.93, + "content": "[ - 1 , 1 ] ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 291, + 336, + 305 + ], + "score": 1.0, + "content": "while considering the constraint that", + "type": "text" + }, + { + "bbox": [ + 337, + 293, + 401, + 304 + ], + "score": 0.92, + "content": "\\mathbf { 0 } \\leq \\alpha + \\varepsilon \\leq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 291, + 405, + 305 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 309, + 381, + 321 + ], + "lines": [ + { + "bbox": [ + 105, + 307, + 382, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 221, + 324 + ], + "score": 1.0, + "content": "Objective function. To find", + "type": "text" + }, + { + "bbox": [ + 222, + 310, + 231, + 319 + ], + "score": 0.82, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 307, + 382, + 324 + ], + "score": 1.0, + "content": "we minimize the weighted objective:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 223, + 326, + 387, + 345 + ], + "lines": [ + { + "bbox": [ + 223, + 326, + 387, + 345 + ], + "spans": [ + { + "bbox": [ + 223, + 326, + 387, + 345 + ], + "score": 0.93, + "content": "\\operatorname* { m i n } _ { \\pmb { T } } \\pmb { \\mathcal { L } } \\triangleq \\lambda _ { 1 } \\mathcal { L } _ { \\mathrm { r e g } } + \\lambda _ { 2 } \\mathcal { L } _ { \\mathrm { d i s c } } + \\lambda _ { 3 } \\mathcal { L } _ { \\mathrm { c o n t e n t } } .", + "type": "interline_equation", + "image_path": "1cf13ccc6410fc13df4ba994f8afdb46b959e91e9dbf0aae868c1aea712c5536.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 223, + 326, + 387, + 345 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 350, + 504, + 373 + ], + "lines": [ + { + "bbox": [ + 105, + 348, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 295, + 364 + ], + "score": 1.0, + "content": "Note, the objective is only used for optimizing", + "type": "text" + }, + { + "bbox": [ + 296, + 351, + 305, + 361 + ], + "score": 0.81, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 348, + 505, + 364 + ], + "score": 1.0, + "content": ", while the other modules remain fixed. Hyperpa-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 361, + 289, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 143, + 373 + ], + "score": 1.0, + "content": "rameters", + "type": "text" + }, + { + "bbox": [ + 144, + 362, + 169, + 372 + ], + "score": 0.91, + "content": "\\lambda _ { 1 } , \\lambda _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 361, + 190, + 373 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 191, + 362, + 202, + 372 + ], + "score": 0.88, + "content": "\\lambda _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 361, + 289, + 373 + ], + "score": 1.0, + "content": "adjust the loss terms.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 105, + 378, + 504, + 401 + ], + "lines": [ + { + "bbox": [ + 105, + 376, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 184, + 392 + ], + "score": 1.0, + "content": "The regression loss", + "type": "text" + }, + { + "bbox": [ + 184, + 379, + 201, + 390 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 376, + 270, + 392 + ], + "score": 1.0, + "content": "assesses whether", + "type": "text" + }, + { + "bbox": [ + 271, + 379, + 280, + 388 + ], + "score": 0.8, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 376, + 447, + 392 + ], + "score": 1.0, + "content": "performs the transformations indicated by", + "type": "text" + }, + { + "bbox": [ + 447, + 380, + 453, + 388 + ], + "score": 0.73, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 376, + 505, + 392 + ], + "score": 1.0, + "content": ". We express", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 387, + 291, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 291, + 404 + ], + "score": 1.0, + "content": "the regression loss via a binary cross entropy:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "interline_equation", + "bbox": [ + 184, + 405, + 427, + 420 + ], + "lines": [ + { + "bbox": [ + 184, + 405, + 427, + 420 + ], + "spans": [ + { + "bbox": [ + 184, + 405, + 427, + 420 + ], + "score": 0.88, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { \\mathrm { r e g } } = \\ } & { { } \\mathbb { E } _ { z \\sim \\mathcal { Z } , \\varepsilon \\sim \\mathcal { D } _ { \\varepsilon } } [ - \\hat { \\alpha } ^ { \\prime } \\log \\alpha ^ { \\prime } - ( { \\bf 1 } - \\hat { \\alpha } ^ { \\prime } ) \\log \\left( { \\bf 1 } - \\alpha ^ { \\prime } \\right) ] _ { \\mathrm { t } } } \\end{array}", + "type": "interline_equation", + "image_path": "84d343b799f92402b7e68b74529e12d2219071174d3e8d3b12acee3ec7fd0dc8.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 184, + 405, + 427, + 420 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 424, + 506, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 145, + 437 + ], + "score": 1.0, + "content": "Note that", + "type": "text" + }, + { + "bbox": [ + 145, + 425, + 156, + 434 + ], + "score": 0.87, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 423, + 304, + 437 + ], + "score": 1.0, + "content": "is from the distribution generated by", + "type": "text" + }, + { + "bbox": [ + 304, + 426, + 311, + 434 + ], + "score": 0.78, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 423, + 329, + 437 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 329, + 426, + 336, + 434 + ], + "score": 0.76, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 423, + 505, + 437 + ], + "score": 1.0, + "content": ". Please refer to the appendix C.1 for more", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 434, + 214, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 214, + 447 + ], + "score": 1.0, + "content": "details on this distribution.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 451, + 504, + 475 + ], + "lines": [ + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 195, + 465 + ], + "score": 1.0, + "content": "The second loss term", + "type": "text" + }, + { + "bbox": [ + 195, + 452, + 215, + 464 + ], + "score": 0.9, + "content": "{ \\mathcal { L } } _ { \\mathrm { d i s c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 451, + 364, + 465 + ], + "score": 1.0, + "content": "is computed using the discriminator", + "type": "text" + }, + { + "bbox": [ + 365, + 453, + 374, + 462 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 451, + 506, + 465 + ], + "score": 1.0, + "content": "and measures the quality of the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 464, + 198, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 198, + 477 + ], + "score": 1.0, + "content": "generated images, i.e.,", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 479, + 392, + 495 + ], + "lines": [ + { + "bbox": [ + 219, + 479, + 392, + 495 + ], + "spans": [ + { + "bbox": [ + 219, + 479, + 392, + 495 + ], + "score": 0.92, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { \\mathrm { d i s c } } = } & { { } \\ \\mathbb { E } _ { z ^ { \\prime } \\sim \\mathcal { Z } ^ { \\prime } | z } \\left[ \\log ( 1 - D ( G ( z ^ { \\prime } ) ) ) \\right] . } \\end{array}", + "type": "interline_equation", + "image_path": "bbab1a17f99d986c9825ffac191ef03d0aad040386c5c6c81c9a8a84bd98e31d.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 219, + 479, + 392, + 495 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 499, + 486, + 511 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 486, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 232, + 513 + ], + "score": 1.0, + "content": "Here we refer to the domain of", + "type": "text" + }, + { + "bbox": [ + 232, + 500, + 242, + 510 + ], + "score": 0.86, + "content": "z ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 498, + 258, + 513 + ], + "score": 1.0, + "content": "via", + "type": "text" + }, + { + "bbox": [ + 258, + 500, + 270, + 510 + ], + "score": 0.87, + "content": "\\mathcal { Z } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 498, + 371, + 513 + ], + "score": 1.0, + "content": ", which is conditioned on", + "type": "text" + }, + { + "bbox": [ + 372, + 502, + 379, + 510 + ], + "score": 0.74, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 498, + 462, + 513 + ], + "score": 1.0, + "content": ". We write this using", + "type": "text" + }, + { + "bbox": [ + 463, + 500, + 483, + 512 + ], + "score": 0.89, + "content": "\\mathcal { Z } ^ { \\prime } | z", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 498, + 486, + 513 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 516, + 502, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 515, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 229, + 531 + ], + "score": 1.0, + "content": "Lastly, we use a content loss", + "type": "text" + }, + { + "bbox": [ + 230, + 517, + 258, + 528 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { \\mathrm { { c o n t e n t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 515, + 505, + 531 + ], + "score": 1.0, + "content": ", often also referred to as perceptual loss. It is designed to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 525, + 504, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 504, + 541 + ], + "score": 1.0, + "content": "estimate the distance between two images and it is employed to maintain the image identity during", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 539, + 352, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 352, + 550 + ], + "score": 1.0, + "content": "the transformation. Specifically, we use the content loss term", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 555, + 432, + 582 + ], + "lines": [ + { + "bbox": [ + 178, + 555, + 432, + 582 + ], + "spans": [ + { + "bbox": [ + 178, + 555, + 432, + 582 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { \\mathrm { c o n t e n t } } = } & { \\ \\mathbb { E } _ { z \\sim \\mathcal { Z } , z ^ { \\prime } \\sim \\mathcal { Z } ^ { \\prime } | z } \\displaystyle \\sum _ { i \\in \\mathcal { D } _ { \\mathrm { c o n t e n t } } } \\| F _ { i } ( G ( z ^ { \\prime } ) ) - F _ { i } ( G ( z ) ) \\| _ { 2 } ^ { 2 } , } \\end{array}", + "type": "interline_equation", + "image_path": "a63bbf6b2f70798498c313e782c4c925d46483117a840c2b5a5ce26a841dfa2e.jpg" + } + ] + } + ], + "index": 35, + "virtual_lines": [ + { + "bbox": [ + 178, + 555, + 432, + 564.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 178, + 564.0, + 432, + 573.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 178, + 573.0, + 432, + 582.0 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 587, + 504, + 622 + ], + "lines": [ + { + "bbox": [ + 105, + 585, + 504, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 134, + 603 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 588, + 155, + 600 + ], + "score": 0.91, + "content": "F _ { i } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 585, + 473, + 603 + ], + "score": 1.0, + "content": "denotes a feature function which extracts intermediate features from images.", + "type": "text" + }, + { + "bbox": [ + 474, + 588, + 504, + 599 + ], + "score": 0.82, + "content": "\\mathcal { D } _ { \\mathrm { c o n t e n t } }", + "type": "inline_equation" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 599, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 505, + 611 + ], + "score": 1.0, + "content": "indicates the layers of a pre-trained model which are used as features. We approximate the afore-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 610, + 470, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 470, + 622 + ], + "score": 1.0, + "content": "mentioned expectations by empirical sampling. We defer algorithm details to Appendix A.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "Joint-distribution sampling and training. The regressor operates in a multi-label setting, i.e.,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 638, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 506, + 650 + ], + "score": 1.0, + "content": "each data possesses multiple attributes. To ensure the disentanglement of attribute edits we sample", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "synthetic images from the entire data distribution and find all transformations at once (illustrated in", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "Fig. 2 (b,c) right). In contrast, Shen et al. (2019) prepare training samples on the two opposing data", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 670, + 504, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 493, + 683 + ], + "score": 1.0, + "content": "subsets with regards to an attribute, e.g., no clouds vs. many clouds, and find the directions with", + "type": "text" + }, + { + "bbox": [ + 493, + 671, + 504, + 681 + ], + "score": 0.77, + "content": "N", + "type": "inline_equation" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 682, + 313, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 313, + 693 + ], + "score": 1.0, + "content": "one-vs-one classifiers (sketched in Fig. 2 (b,c) left).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 210, + 711 + ], + "score": 1.0, + "content": "Transformation module", + "type": "text" + }, + { + "bbox": [ + 210, + 699, + 220, + 709 + ], + "score": 0.51, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 699, + 383, + 711 + ], + "score": 1.0, + "content": ". We study two types of transformations", + "type": "text" + }, + { + "bbox": [ + 383, + 700, + 392, + 709 + ], + "score": 0.75, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 699, + 505, + 711 + ], + "score": 1.0, + "content": ": (i) global and (ii) local. A", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 196, + 723 + ], + "score": 1.0, + "content": "global transformation", + "type": "text" + }, + { + "bbox": [ + 197, + 710, + 206, + 720 + ], + "score": 0.69, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 709, + 467, + 723 + ], + "score": 1.0, + "content": "refers to a semantic latent-space transformation identical for all", + "type": "text" + }, + { + "bbox": [ + 468, + 711, + 475, + 720 + ], + "score": 0.73, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "during", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "inference. This is illustrated via parallel red dashed arrows in Fig. 2 (d) left. These global directions", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 8 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 226, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 227, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 227, + 95 + ], + "score": 1.0, + "content": "3.2 PROPOSED APPROACH", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 504, + 126 + ], + "lines": [ + { + "bbox": [ + 106, + 102, + 505, + 115 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 115 + ], + "score": 1.0, + "content": "We provide an overview of our method in Fig. 2 (a), and further illustrate how our approach differs", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 114, + 233, + 126 + ], + "spans": [ + { + "bbox": [ + 107, + 114, + 233, + 126 + ], + "score": 1.0, + "content": "from prior work in Fig. 2 (b-d).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 106, + 102, + 505, + 126 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 130, + 505, + 244 + ], + "lines": [ + { + "bbox": [ + 106, + 131, + 506, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 131, + 469, + 144 + ], + "score": 1.0, + "content": "Overview. As shown in Fig. 2 (a), we employ a GAN model consisting of a generator", + "type": "text" + }, + { + "bbox": [ + 469, + 132, + 478, + 141 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 131, + 506, + 144 + ], + "score": 1.0, + "content": "and a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 141, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 163, + 155 + ], + "score": 1.0, + "content": "discriminator", + "type": "text" + }, + { + "bbox": [ + 163, + 142, + 173, + 152 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 141, + 405, + 155 + ], + "score": 1.0, + "content": "(not shown due to limited space), as well as a regressor", + "type": "text" + }, + { + "bbox": [ + 406, + 142, + 414, + 152 + ], + "score": 0.79, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 141, + 505, + 155 + ], + "score": 1.0, + "content": ", all of which are pre-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 153, + 504, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 286, + 164 + ], + "score": 1.0, + "content": "trained. Our goal is to find latent directions", + "type": "text" + }, + { + "bbox": [ + 286, + 153, + 296, + 163 + ], + "score": 0.78, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 153, + 504, + 164 + ], + "score": 1.0, + "content": "that provide attribute specific image edits. At each", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 163, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 162, + 176 + ], + "score": 1.0, + "content": "training step,", + "type": "text" + }, + { + "bbox": [ + 162, + 164, + 171, + 174 + ], + "score": 0.79, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 163, + 196, + 176 + ], + "score": 1.0, + "content": "takes", + "type": "text" + }, + { + "bbox": [ + 196, + 165, + 204, + 174 + ], + "score": 0.75, + "content": "_ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 163, + 371, + 176 + ], + "score": 1.0, + "content": "and an edited latent vector referred to as", + "type": "text" + }, + { + "bbox": [ + 372, + 164, + 381, + 174 + ], + "score": 0.86, + "content": "z ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 163, + 505, + 176 + ], + "score": 1.0, + "content": ". We follow prior work which", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 174, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 505, + 187 + ], + "score": 1.0, + "content": "suggests that direction vectors in latent space of GANs permit image editing (Jahanian et al., 2019).", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 186, + 504, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 275, + 199 + ], + "score": 1.0, + "content": "Formally, given a transformation degree", + "type": "text" + }, + { + "bbox": [ + 275, + 188, + 282, + 196 + ], + "score": 0.64, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 186, + 429, + 199 + ], + "score": 1.0, + "content": ", we obtain the latent-space edit as", + "type": "text" + }, + { + "bbox": [ + 430, + 186, + 504, + 197 + ], + "score": 0.89, + "content": "z ^ { \\prime } = z + { \\cal T } \\varepsilon =", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 194, + 508, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 197, + 168, + 211 + ], + "score": 0.91, + "content": "\\begin{array} { r } { z + \\sum _ { i = 1 } ^ { N } \\varepsilon _ { i } d _ { i } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 194, + 173, + 218 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 173, + 199, + 182, + 209 + ], + "score": 0.7, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 194, + 350, + 218 + ], + "score": 1.0, + "content": "provides the recovered and edited images", + "type": "text" + }, + { + "bbox": [ + 351, + 198, + 373, + 210 + ], + "score": 0.92, + "content": "G ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 194, + 391, + 218 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 392, + 198, + 416, + 210 + ], + "score": 0.92, + "content": "G ( z ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 194, + 508, + 218 + ], + "score": 1.0, + "content": ", which are separately", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 209, + 443, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 217, + 222 + ], + "score": 1.0, + "content": "processed by the regressor", + "type": "text" + }, + { + "bbox": [ + 226, + 209, + 443, + 222 + ], + "score": 1.0, + "content": "to predict attribute values. The original attributes of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 219, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 107, + 221, + 144, + 232 + ], + "score": 0.92, + "content": "R ( G ( z ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 219, + 365, + 234 + ], + "score": 1.0, + "content": ". The (pseudo-) ground truth and predicted attribute of", + "type": "text" + }, + { + "bbox": [ + 365, + 221, + 390, + 233 + ], + "score": 0.93, + "content": "G ( z ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 219, + 407, + 234 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 407, + 221, + 419, + 231 + ], + "score": 0.87, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 219, + 437, + 234 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 437, + 221, + 448, + 231 + ], + "score": 0.86, + "content": "\\hat { \\mathbf { \\alpha } } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 219, + 506, + 234 + ], + "score": 1.0, + "content": ", respectively,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 231, + 270, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 133, + 244 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 232, + 183, + 242 + ], + "score": 0.9, + "content": "{ \\pmb { \\alpha } } ^ { \\prime } = { \\pmb { \\alpha } } + { \\pmb { \\varepsilon } }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 231, + 201, + 244 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 202, + 231, + 266, + 244 + ], + "score": 0.93, + "content": "\\hat { \\pmb { \\alpha } } ^ { \\prime } = R ( G ( \\pmb { z } ^ { \\prime } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 231, + 270, + 244 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 131, + 508, + 244 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 248, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 106, + 249, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 201, + 261 + ], + "score": 1.0, + "content": "Intuitively, the goal of", + "type": "text" + }, + { + "bbox": [ + 201, + 249, + 211, + 258 + ], + "score": 0.78, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 249, + 276, + 261 + ], + "score": 1.0, + "content": "is to transform", + "type": "text" + }, + { + "bbox": [ + 277, + 250, + 284, + 258 + ], + "score": 0.75, + "content": "_ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 249, + 505, + 261 + ], + "score": 1.0, + "content": "by adding semantically meaningful information such", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 258, + 272 + ], + "score": 1.0, + "content": "that the corresponding output image", + "type": "text" + }, + { + "bbox": [ + 258, + 259, + 284, + 271 + ], + "score": 0.92, + "content": "G ( z ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 259, + 393, + 272 + ], + "score": 1.0, + "content": "exhibits attribute changes", + "type": "text" + }, + { + "bbox": [ + 393, + 261, + 400, + 269 + ], + "score": 0.74, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 259, + 425, + 272 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 425, + 261, + 434, + 269 + ], + "score": 0.74, + "content": "_ { \\pmb { \\alpha } }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 259, + 505, + 272 + ], + "score": 1.0, + "content": ". In practice, we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 276, + 284 + ], + "score": 1.0, + "content": "normalize the range of attribute values to", + "type": "text" + }, + { + "bbox": [ + 277, + 271, + 299, + 282 + ], + "score": 0.82, + "content": "[ \\mathbf { 0 } , \\mathbf { 1 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 270, + 342, + 284 + ], + "score": 1.0, + "content": ", i.e., both", + "type": "text" + }, + { + "bbox": [ + 342, + 272, + 351, + 281 + ], + "score": 0.78, + "content": "_ { \\pmb { \\alpha } }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 270, + 370, + 284 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 370, + 270, + 417, + 282 + ], + "score": 0.92, + "content": "{ \\pmb { \\alpha } } ^ { \\prime } \\in [ { \\bf 0 } , { \\bf 1 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 270, + 506, + 284 + ], + "score": 1.0, + "content": ". During training, we", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 280, + 504, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 307, + 295 + ], + "score": 1.0, + "content": "maintain the unit range by controlling the given", + "type": "text" + }, + { + "bbox": [ + 307, + 284, + 313, + 291 + ], + "score": 0.67, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 280, + 363, + 295 + ], + "score": 1.0, + "content": ". Formally,", + "type": "text" + }, + { + "bbox": [ + 363, + 283, + 370, + 291 + ], + "score": 0.72, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 280, + 490, + 295 + ], + "score": 1.0, + "content": "is drawn from a distribution", + "type": "text" + }, + { + "bbox": [ + 491, + 282, + 504, + 293 + ], + "score": 0.88, + "content": "\\mathcal { D } _ { \\varepsilon }", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 291, + 405, + 305 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 151, + 305 + ], + "score": 1.0, + "content": "uniform in", + "type": "text" + }, + { + "bbox": [ + 152, + 292, + 187, + 304 + ], + "score": 0.93, + "content": "[ - 1 , 1 ] ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 291, + 336, + 305 + ], + "score": 1.0, + "content": "while considering the constraint that", + "type": "text" + }, + { + "bbox": [ + 337, + 293, + 401, + 304 + ], + "score": 0.92, + "content": "\\mathbf { 0 } \\leq \\alpha + \\varepsilon \\leq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 291, + 405, + 305 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 249, + 506, + 305 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 309, + 381, + 321 + ], + "lines": [ + { + "bbox": [ + 105, + 307, + 382, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 221, + 324 + ], + "score": 1.0, + "content": "Objective function. To find", + "type": "text" + }, + { + "bbox": [ + 222, + 310, + 231, + 319 + ], + "score": 0.82, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 307, + 382, + 324 + ], + "score": 1.0, + "content": "we minimize the weighted objective:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 307, + 382, + 324 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 223, + 326, + 387, + 345 + ], + "lines": [ + { + "bbox": [ + 223, + 326, + 387, + 345 + ], + "spans": [ + { + "bbox": [ + 223, + 326, + 387, + 345 + ], + "score": 0.93, + "content": "\\operatorname* { m i n } _ { \\pmb { T } } \\pmb { \\mathcal { L } } \\triangleq \\lambda _ { 1 } \\mathcal { L } _ { \\mathrm { r e g } } + \\lambda _ { 2 } \\mathcal { L } _ { \\mathrm { d i s c } } + \\lambda _ { 3 } \\mathcal { L } _ { \\mathrm { c o n t e n t } } .", + "type": "interline_equation", + "image_path": "1cf13ccc6410fc13df4ba994f8afdb46b959e91e9dbf0aae868c1aea712c5536.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 223, + 326, + 387, + 345 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 350, + 504, + 373 + ], + "lines": [ + { + "bbox": [ + 105, + 348, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 295, + 364 + ], + "score": 1.0, + "content": "Note, the objective is only used for optimizing", + "type": "text" + }, + { + "bbox": [ + 296, + 351, + 305, + 361 + ], + "score": 0.81, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 348, + 505, + 364 + ], + "score": 1.0, + "content": ", while the other modules remain fixed. Hyperpa-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 361, + 289, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 143, + 373 + ], + "score": 1.0, + "content": "rameters", + "type": "text" + }, + { + "bbox": [ + 144, + 362, + 169, + 372 + ], + "score": 0.91, + "content": "\\lambda _ { 1 } , \\lambda _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 361, + 190, + 373 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 191, + 362, + 202, + 372 + ], + "score": 0.88, + "content": "\\lambda _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 361, + 289, + 373 + ], + "score": 1.0, + "content": "adjust the loss terms.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 348, + 505, + 373 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 378, + 504, + 401 + ], + "lines": [ + { + "bbox": [ + 105, + 376, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 184, + 392 + ], + "score": 1.0, + "content": "The regression loss", + "type": "text" + }, + { + "bbox": [ + 184, + 379, + 201, + 390 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 376, + 270, + 392 + ], + "score": 1.0, + "content": "assesses whether", + "type": "text" + }, + { + "bbox": [ + 271, + 379, + 280, + 388 + ], + "score": 0.8, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 376, + 447, + 392 + ], + "score": 1.0, + "content": "performs the transformations indicated by", + "type": "text" + }, + { + "bbox": [ + 447, + 380, + 453, + 388 + ], + "score": 0.73, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 376, + 505, + 392 + ], + "score": 1.0, + "content": ". We express", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 387, + 291, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 291, + 404 + ], + "score": 1.0, + "content": "the regression loss via a binary cross entropy:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 376, + 505, + 404 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 184, + 405, + 427, + 420 + ], + "lines": [ + { + "bbox": [ + 184, + 405, + 427, + 420 + ], + "spans": [ + { + "bbox": [ + 184, + 405, + 427, + 420 + ], + "score": 0.88, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { \\mathrm { r e g } } = \\ } & { { } \\mathbb { E } _ { z \\sim \\mathcal { Z } , \\varepsilon \\sim \\mathcal { D } _ { \\varepsilon } } [ - \\hat { \\alpha } ^ { \\prime } \\log \\alpha ^ { \\prime } - ( { \\bf 1 } - \\hat { \\alpha } ^ { \\prime } ) \\log \\left( { \\bf 1 } - \\alpha ^ { \\prime } \\right) ] _ { \\mathrm { t } } } \\end{array}", + "type": "interline_equation", + "image_path": "84d343b799f92402b7e68b74529e12d2219071174d3e8d3b12acee3ec7fd0dc8.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 184, + 405, + 427, + 420 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 424, + 506, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 145, + 437 + ], + "score": 1.0, + "content": "Note that", + "type": "text" + }, + { + "bbox": [ + 145, + 425, + 156, + 434 + ], + "score": 0.87, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 423, + 304, + 437 + ], + "score": 1.0, + "content": "is from the distribution generated by", + "type": "text" + }, + { + "bbox": [ + 304, + 426, + 311, + 434 + ], + "score": 0.78, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 423, + 329, + 437 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 329, + 426, + 336, + 434 + ], + "score": 0.76, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 423, + 505, + 437 + ], + "score": 1.0, + "content": ". Please refer to the appendix C.1 for more", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 434, + 214, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 214, + 447 + ], + "score": 1.0, + "content": "details on this distribution.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 423, + 505, + 447 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 451, + 504, + 475 + ], + "lines": [ + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 195, + 465 + ], + "score": 1.0, + "content": "The second loss term", + "type": "text" + }, + { + "bbox": [ + 195, + 452, + 215, + 464 + ], + "score": 0.9, + "content": "{ \\mathcal { L } } _ { \\mathrm { d i s c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 451, + 364, + 465 + ], + "score": 1.0, + "content": "is computed using the discriminator", + "type": "text" + }, + { + "bbox": [ + 365, + 453, + 374, + 462 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 451, + 506, + 465 + ], + "score": 1.0, + "content": "and measures the quality of the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 464, + 198, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 198, + 477 + ], + "score": 1.0, + "content": "generated images, i.e.,", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 451, + 506, + 477 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 479, + 392, + 495 + ], + "lines": [ + { + "bbox": [ + 219, + 479, + 392, + 495 + ], + "spans": [ + { + "bbox": [ + 219, + 479, + 392, + 495 + ], + "score": 0.92, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { \\mathrm { d i s c } } = } & { { } \\ \\mathbb { E } _ { z ^ { \\prime } \\sim \\mathcal { Z } ^ { \\prime } | z } \\left[ \\log ( 1 - D ( G ( z ^ { \\prime } ) ) ) \\right] . } \\end{array}", + "type": "interline_equation", + "image_path": "bbab1a17f99d986c9825ffac191ef03d0aad040386c5c6c81c9a8a84bd98e31d.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 219, + 479, + 392, + 495 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 499, + 486, + 511 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 486, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 232, + 513 + ], + "score": 1.0, + "content": "Here we refer to the domain of", + "type": "text" + }, + { + "bbox": [ + 232, + 500, + 242, + 510 + ], + "score": 0.86, + "content": "z ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 498, + 258, + 513 + ], + "score": 1.0, + "content": "via", + "type": "text" + }, + { + "bbox": [ + 258, + 500, + 270, + 510 + ], + "score": 0.87, + "content": "\\mathcal { Z } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 498, + 371, + 513 + ], + "score": 1.0, + "content": ", which is conditioned on", + "type": "text" + }, + { + "bbox": [ + 372, + 502, + 379, + 510 + ], + "score": 0.74, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 498, + 462, + 513 + ], + "score": 1.0, + "content": ". We write this using", + "type": "text" + }, + { + "bbox": [ + 463, + 500, + 483, + 512 + ], + "score": 0.89, + "content": "\\mathcal { Z } ^ { \\prime } | z", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 498, + 486, + 513 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 498, + 486, + 513 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 516, + 502, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 515, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 229, + 531 + ], + "score": 1.0, + "content": "Lastly, we use a content loss", + "type": "text" + }, + { + "bbox": [ + 230, + 517, + 258, + 528 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { \\mathrm { { c o n t e n t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 515, + 505, + 531 + ], + "score": 1.0, + "content": ", often also referred to as perceptual loss. It is designed to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 525, + 504, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 504, + 541 + ], + "score": 1.0, + "content": "estimate the distance between two images and it is employed to maintain the image identity during", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 539, + 352, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 352, + 550 + ], + "score": 1.0, + "content": "the transformation. Specifically, we use the content loss term", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 515, + 505, + 550 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 555, + 432, + 582 + ], + "lines": [ + { + "bbox": [ + 178, + 555, + 432, + 582 + ], + "spans": [ + { + "bbox": [ + 178, + 555, + 432, + 582 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { \\mathrm { c o n t e n t } } = } & { \\ \\mathbb { E } _ { z \\sim \\mathcal { Z } , z ^ { \\prime } \\sim \\mathcal { Z } ^ { \\prime } | z } \\displaystyle \\sum _ { i \\in \\mathcal { D } _ { \\mathrm { c o n t e n t } } } \\| F _ { i } ( G ( z ^ { \\prime } ) ) - F _ { i } ( G ( z ) ) \\| _ { 2 } ^ { 2 } , } \\end{array}", + "type": "interline_equation", + "image_path": "a63bbf6b2f70798498c313e782c4c925d46483117a840c2b5a5ce26a841dfa2e.jpg" + } + ] + } + ], + "index": 35, + "virtual_lines": [ + { + "bbox": [ + 178, + 555, + 432, + 564.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 178, + 564.0, + 432, + 573.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 178, + 573.0, + 432, + 582.0 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 587, + 504, + 622 + ], + "lines": [ + { + "bbox": [ + 105, + 585, + 504, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 134, + 603 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 588, + 155, + 600 + ], + "score": 0.91, + "content": "F _ { i } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 585, + 473, + 603 + ], + "score": 1.0, + "content": "denotes a feature function which extracts intermediate features from images.", + "type": "text" + }, + { + "bbox": [ + 474, + 588, + 504, + 599 + ], + "score": 0.82, + "content": "\\mathcal { D } _ { \\mathrm { c o n t e n t } }", + "type": "inline_equation" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 599, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 505, + 611 + ], + "score": 1.0, + "content": "indicates the layers of a pre-trained model which are used as features. We approximate the afore-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 610, + 470, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 470, + 622 + ], + "score": 1.0, + "content": "mentioned expectations by empirical sampling. We defer algorithm details to Appendix A.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 585, + 505, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "Joint-distribution sampling and training. The regressor operates in a multi-label setting, i.e.,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 638, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 506, + 650 + ], + "score": 1.0, + "content": "each data possesses multiple attributes. To ensure the disentanglement of attribute edits we sample", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "synthetic images from the entire data distribution and find all transformations at once (illustrated in", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "Fig. 2 (b,c) right). In contrast, Shen et al. (2019) prepare training samples on the two opposing data", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 670, + 504, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 493, + 683 + ], + "score": 1.0, + "content": "subsets with regards to an attribute, e.g., no clouds vs. many clouds, and find the directions with", + "type": "text" + }, + { + "bbox": [ + 493, + 671, + 504, + 681 + ], + "score": 0.77, + "content": "N", + "type": "inline_equation" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 682, + 313, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 313, + 693 + ], + "score": 1.0, + "content": "one-vs-one classifiers (sketched in Fig. 2 (b,c) left).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 626, + 506, + 693 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 210, + 711 + ], + "score": 1.0, + "content": "Transformation module", + "type": "text" + }, + { + "bbox": [ + 210, + 699, + 220, + 709 + ], + "score": 0.51, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 699, + 383, + 711 + ], + "score": 1.0, + "content": ". We study two types of transformations", + "type": "text" + }, + { + "bbox": [ + 383, + 700, + 392, + 709 + ], + "score": 0.75, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 699, + 505, + 711 + ], + "score": 1.0, + "content": ": (i) global and (ii) local. A", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 196, + 723 + ], + "score": 1.0, + "content": "global transformation", + "type": "text" + }, + { + "bbox": [ + 197, + 710, + 206, + 720 + ], + "score": 0.69, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 709, + 467, + 723 + ], + "score": 1.0, + "content": "refers to a semantic latent-space transformation identical for all", + "type": "text" + }, + { + "bbox": [ + 468, + 711, + 475, + 720 + ], + "score": 0.73, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "during", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "inference. This is illustrated via parallel red dashed arrows in Fig. 2 (d) left. These global directions", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "are commonly used (Hark ¨ onen et al., 2020; Jahanian et al., 2019; Shen et al., 2019; Viazovetskyi ¨", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 587, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 505, + 599 + ], + "score": 1.0, + "content": "et al., 2020). However, a globally identical direction might not serve all data. In contrast, a local", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 597, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 168, + 610 + ], + "score": 1.0, + "content": "transformation", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 168, + 598, + 177, + 608 + ], + "score": 0.78, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 178, + 597, + 243, + 610 + ], + "score": 1.0, + "content": "is a function of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 243, + 599, + 250, + 608 + ], + "score": 0.77, + "content": "_ z", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 251, + 597, + 441, + 610 + ], + "score": 1.0, + "content": "which provides various directions for different", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 441, + 599, + 449, + 608 + ], + "score": 0.72, + "content": "_ z", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 450, + 597, + 505, + 610 + ], + "score": 1.0, + "content": "(in Fig. 2 (d)", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 608, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 322, + 621 + ], + "score": 1.0, + "content": "right, shown via non-parallel red arrows). Formally,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 323, + 608, + 372, + 621 + ], + "score": 0.93, + "content": "d _ { i } = f _ { \\theta } ^ { i } ( z )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 372, + 608, + 405, + 621 + ], + "score": 1.0, + "content": ", where", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 405, + 608, + 416, + 621 + ], + "score": 0.9, + "content": "f _ { \\theta } ^ { i }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 416, + 608, + 506, + 621 + ], + "score": 1.0, + "content": "is implemented via a", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 338, + 631 + ], + "score": 1.0, + "content": "deep net. In Sec. 4 we show that the local transformation", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 338, + 621, + 348, + 630 + ], + "score": 0.78, + "content": "\\mathbf { T }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 348, + 620, + 505, + 631 + ], + "score": 1.0, + "content": "succeeds on failure cases of the global", + "type": "text", + "cross_page": true + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 631, + 170, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 170, + 642 + ], + "score": 1.0, + "content": "transformation.", + "type": "text", + "cross_page": true + } + ], + "index": 20 + } + ], + "index": 47, + "bbox_fs": [ + 105, + 699, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 86, + 499, + 275 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 86, + 499, + 275 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 86, + 499, + 275 + ], + "spans": [ + { + "bbox": [ + 111, + 86, + 499, + 275 + ], + "score": 0.976, + "type": "image", + "image_path": "d6d3e6bc1c1ba9ae9062257137ebc1913590a9a30bfac3de55c26f3b4059cfda.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 86, + 499, + 149.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 149.0, + 499, + 212.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 212.0, + 499, + 275.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 286, + 505, + 331 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 285, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 104, + 285, + 506, + 300 + ], + "score": 1.0, + "content": "Figure 3: Comparison of image-space editing approaches with respect to “removing clouds”", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 297, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 506, + 311 + ], + "score": 1.0, + "content": "(top), “enhancing night” (middle) and “adding snow” (bottom). The original images (col.1) are", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 308, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 506, + 321 + ], + "score": 1.0, + "content": "followed by a given editing task. From (a-e): (a) CycleGAN (Zhu et al., 2017a), (b) StarGAN", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 319, + 500, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 334, + 333 + ], + "score": 1.0, + "content": "v2 (Choi et al., 2020), (c) RelGAN (Wu et al., 2019), (d)", + "type": "text" + }, + { + "bbox": [ + 334, + 320, + 371, + 331 + ], + "score": 0.39, + "content": "\\mathrm { D R I T + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 319, + 500, + 333 + ], + "score": 1.0, + "content": "(Lee et al., 2020), and (e) Ours.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "image", + "bbox": [ + 111, + 346, + 499, + 486 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 346, + 499, + 486 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 111, + 346, + 499, + 486 + ], + "spans": [ + { + "bbox": [ + 111, + 346, + 499, + 486 + ], + "score": 0.975, + "type": "image", + "image_path": "8276997cfc6eb8e8ffa6c1fdebf8f9f1218dac268c9fbfd842f4affa8afe9c4e.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 111, + 346, + 499, + 392.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 111, + 392.6666666666667, + 499, + 439.33333333333337 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 111, + 439.33333333333337, + 499, + 486.00000000000006 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 497, + 505, + 552 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 496, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 510 + ], + "score": 1.0, + "content": "Figure 4: Comparison of latent-space editing approaches on StyleGAN2. (a) Shen et al. (2019);", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 508, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 505, + 521 + ], + "score": 1.0, + "content": "(b) Voynov & Babenko (2020); (c) Ours. Original synthetic image (col.1); edited results of a seman-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 519, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 104, + 519, + 505, + 533 + ], + "score": 1.0, + "content": "tic manipulation noted on top (col.2-4; col.5-7). Shen & Zhou (2021) mention that “age,” “gender,”", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 529, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 104, + 529, + 506, + 545 + ], + "score": 1.0, + "content": "and “glasses” directions are hard to disentangle potentially due to data bias, while our results suggest", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 542, + 246, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 246, + 554 + ], + "score": 1.0, + "content": "a better direction disentanglement.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12 + } + ], + "index": 10.0 + }, + { + "type": "text", + "bbox": [ + 107, + 575, + 505, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "are commonly used (Hark ¨ onen et al., 2020; Jahanian et al., 2019; Shen et al., 2019; Viazovetskyi ¨", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 587, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 505, + 599 + ], + "score": 1.0, + "content": "et al., 2020). However, a globally identical direction might not serve all data. In contrast, a local", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 597, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 168, + 610 + ], + "score": 1.0, + "content": "transformation", + "type": "text" + }, + { + "bbox": [ + 168, + 598, + 177, + 608 + ], + "score": 0.78, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 597, + 243, + 610 + ], + "score": 1.0, + "content": "is a function of", + "type": "text" + }, + { + "bbox": [ + 243, + 599, + 250, + 608 + ], + "score": 0.77, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 597, + 441, + 610 + ], + "score": 1.0, + "content": "which provides various directions for different", + "type": "text" + }, + { + "bbox": [ + 441, + 599, + 449, + 608 + ], + "score": 0.72, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 597, + 505, + 610 + ], + "score": 1.0, + "content": "(in Fig. 2 (d)", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 608, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 322, + 621 + ], + "score": 1.0, + "content": "right, shown via non-parallel red arrows). Formally,", + "type": "text" + }, + { + "bbox": [ + 323, + 608, + 372, + 621 + ], + "score": 0.93, + "content": "d _ { i } = f _ { \\theta } ^ { i } ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 608, + 405, + 621 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 405, + 608, + 416, + 621 + ], + "score": 0.9, + "content": "f _ { \\theta } ^ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 608, + 506, + 621 + ], + "score": 1.0, + "content": "is implemented via a", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 338, + 631 + ], + "score": 1.0, + "content": "deep net. In Sec. 4 we show that the local transformation", + "type": "text" + }, + { + "bbox": [ + 338, + 621, + 348, + 630 + ], + "score": 0.78, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 620, + 505, + 631 + ], + "score": 1.0, + "content": "succeeds on failure cases of the global", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 631, + 170, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 170, + 642 + ], + "score": 1.0, + "content": "transformation.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 108, + 660, + 200, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 659, + 201, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 201, + 675 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 105, + 681, + 462, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 681, + 463, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 463, + 694 + ], + "score": 1.0, + "content": "We introduce datasets, implementation details, evaluation metrics, and show results next.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "Datasets. We evaluate the proposed approach on two types of datasets: (i) face datasets –", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "FFHQ (Karras et al., 2019a), CelebA (Liu et al., 2018) and CelebA-HQ (Karras et al., 2017), com-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "monly used in prior work (Hark ¨ onen et al., 2020; Karras et al., 2017; 2019a;b; Shen et al., 2019; ¨", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 86, + 499, + 275 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 86, + 499, + 275 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 86, + 499, + 275 + ], + "spans": [ + { + "bbox": [ + 111, + 86, + 499, + 275 + ], + "score": 0.976, + "type": "image", + "image_path": "d6d3e6bc1c1ba9ae9062257137ebc1913590a9a30bfac3de55c26f3b4059cfda.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 86, + 499, + 149.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 149.0, + 499, + 212.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 212.0, + 499, + 275.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 286, + 505, + 331 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 285, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 104, + 285, + 506, + 300 + ], + "score": 1.0, + "content": "Figure 3: Comparison of image-space editing approaches with respect to “removing clouds”", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 297, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 506, + 311 + ], + "score": 1.0, + "content": "(top), “enhancing night” (middle) and “adding snow” (bottom). The original images (col.1) are", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 308, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 506, + 321 + ], + "score": 1.0, + "content": "followed by a given editing task. From (a-e): (a) CycleGAN (Zhu et al., 2017a), (b) StarGAN", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 319, + 500, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 334, + 333 + ], + "score": 1.0, + "content": "v2 (Choi et al., 2020), (c) RelGAN (Wu et al., 2019), (d)", + "type": "text" + }, + { + "bbox": [ + 334, + 320, + 371, + 331 + ], + "score": 0.39, + "content": "\\mathrm { D R I T + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 319, + 500, + 333 + ], + "score": 1.0, + "content": "(Lee et al., 2020), and (e) Ours.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "image", + "bbox": [ + 111, + 346, + 499, + 486 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 346, + 499, + 486 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 111, + 346, + 499, + 486 + ], + "spans": [ + { + "bbox": [ + 111, + 346, + 499, + 486 + ], + "score": 0.975, + "type": "image", + "image_path": "8276997cfc6eb8e8ffa6c1fdebf8f9f1218dac268c9fbfd842f4affa8afe9c4e.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 111, + 346, + 499, + 392.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 111, + 392.6666666666667, + 499, + 439.33333333333337 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 111, + 439.33333333333337, + 499, + 486.00000000000006 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 497, + 505, + 552 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 496, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 510 + ], + "score": 1.0, + "content": "Figure 4: Comparison of latent-space editing approaches on StyleGAN2. (a) Shen et al. (2019);", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 508, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 505, + 521 + ], + "score": 1.0, + "content": "(b) Voynov & Babenko (2020); (c) Ours. Original synthetic image (col.1); edited results of a seman-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 519, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 104, + 519, + 505, + 533 + ], + "score": 1.0, + "content": "tic manipulation noted on top (col.2-4; col.5-7). Shen & Zhou (2021) mention that “age,” “gender,”", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 529, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 104, + 529, + 506, + 545 + ], + "score": 1.0, + "content": "and “glasses” directions are hard to disentangle potentially due to data bias, while our results suggest", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 542, + 246, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 246, + 554 + ], + "score": 1.0, + "content": "a better direction disentanglement.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12 + } + ], + "index": 10.0 + }, + { + "type": "text", + "bbox": [ + 107, + 575, + 505, + 642 + ], + "lines": [], + "index": 17.5, + "bbox_fs": [ + 105, + 576, + 506, + 642 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 660, + 200, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 659, + 201, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 201, + 675 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 105, + 681, + 462, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 681, + 463, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 463, + 694 + ], + "score": 1.0, + "content": "We introduce datasets, implementation details, evaluation metrics, and show results next.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 681, + 463, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "Datasets. We evaluate the proposed approach on two types of datasets: (i) face datasets –", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "FFHQ (Karras et al., 2019a), CelebA (Liu et al., 2018) and CelebA-HQ (Karras et al., 2017), com-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "monly used in prior work (Hark ¨ onen et al., 2020; Karras et al., 2017; 2019a;b; Shen et al., 2019; ¨", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 698, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 115 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Viazovetskyi et al., 2020). (ii) natural scene datasets – Transient Attribute Database (Laffont et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "2014) and MIT Places2 data (Zhou et al., 2017), which contain attributes suitable for continuous", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 366, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 366, + 117 + ], + "score": 1.0, + "content": "semantic image editing. We briefly introduce the scene datasets:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 109, + 124, + 505, + 215 + ], + "lines": [ + { + "bbox": [ + 112, + 123, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 112, + 123, + 505, + 137 + ], + "score": 1.0, + "content": "• Transient Attribute Database (Laffont et al., 2014): It contains 8,571 scene images with 40 at-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 120, + 134, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 120, + 134, + 506, + 149 + ], + "score": 1.0, + "content": "tributes in 5 categories including lighting (e.g., “bright”), weather (e.g., “cloudy”) , seasons (e.g.,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 120, + 145, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 120, + 145, + 505, + 159 + ], + "score": 1.0, + "content": "“winter”), subjective impressions (e.g., “beautiful”), and additional attributes (e.g., “dirty”).", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 120, + 156, + 505, + 170 + ], + "spans": [ + { + "bbox": [ + 120, + 156, + 505, + 170 + ], + "score": 1.0, + "content": "Each attribute is annotated with a real-valued score between 0 and 1, where 0 indicates absence", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 120, + 168, + 184, + 179 + ], + "spans": [ + { + "bbox": [ + 120, + 168, + 184, + 179 + ], + "score": 1.0, + "content": "of the attribute.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 113, + 181, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 113, + 181, + 505, + 195 + ], + "score": 1.0, + "content": "• MIT Places2 data (Zhou et al., 2017): Using the provided category annotations (i.e., in-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 120, + 193, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 120, + 193, + 506, + 205 + ], + "score": 1.0, + "content": "door/outdoor and natural/artificial), we select the natural outdoor scenes, obtaining a total of", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 121, + 203, + 190, + 217 + ], + "spans": [ + { + "bbox": [ + 121, + 203, + 190, + 217 + ], + "score": 1.0, + "content": "144,543 images.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 224, + 505, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 223, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 223, + 259, + 236 + ], + "score": 1.0, + "content": "Implementation details. We choose", + "type": "text" + }, + { + "bbox": [ + 259, + 224, + 392, + 235 + ], + "score": 0.9, + "content": "\\lambda _ { 1 } = 1 0 , \\lambda _ { 2 } = 0 . 0 5 , \\lambda _ { 3 } = 0 . 0 5", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 223, + 505, + 236 + ], + "score": 1.0, + "content": "in Eq. (1) and compute the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 234, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 247 + ], + "score": 1.0, + "content": "perceptual loss using the conv1 2, conv2 2, conv3 2, conv4 2 activations in a VGG-19 network (Si-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 246, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 506, + 258 + ], + "score": 1.0, + "content": "monyan & Zisserman, 2014) pre-trained on the ImageNet dataset (Russakovsky et al., 2015). We", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 256, + 506, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 128, + 270 + ], + "score": 1.0, + "content": "train", + "type": "text" + }, + { + "bbox": [ + 128, + 257, + 137, + 267 + ], + "score": 0.75, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 256, + 153, + 270 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 154, + 257, + 171, + 267 + ], + "score": 0.54, + "content": "5 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 256, + 506, + 270 + ], + "score": 1.0, + "content": "iterations with a batch size of 4. An Adam optimizer is used with a learning rate", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 266, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 104, + 266, + 227, + 281 + ], + "score": 1.0, + "content": "of 1e-4. For the latent space", + "type": "text" + }, + { + "bbox": [ + 228, + 268, + 237, + 278 + ], + "score": 0.74, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 266, + 386, + 281 + ], + "score": 1.0, + "content": "in the proposed method we use the", + "type": "text" + }, + { + "bbox": [ + 386, + 268, + 399, + 278 + ], + "score": 0.7, + "content": "\\mathcal { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 266, + 506, + 281 + ], + "score": 1.0, + "content": "space of StyleGAN2 and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 121, + 291 + ], + "score": 1.0, + "content": "the", + "type": "text" + }, + { + "bbox": [ + 122, + 279, + 131, + 289 + ], + "score": 0.75, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 279, + 316, + 291 + ], + "score": 1.0, + "content": "space of PGGAN. For the attribute regressor", + "type": "text" + }, + { + "bbox": [ + 316, + 279, + 325, + 289 + ], + "score": 0.75, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 279, + 506, + 291 + ], + "score": 1.0, + "content": ", we adopt a ResNet-50 (He et al., 2016) for", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 289, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 506, + 302 + ], + "score": 1.0, + "content": "attribute prediction on the Transient Attribute (Laffont et al., 2014) and the CelebA (Liu et al., 2018)", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "score": 1.0, + "content": "data. The last fully connected layer in the ResNet-50 is replaced by a linear layer with an output", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "score": 1.0, + "content": "dimension of 40. We train the regressors for 500 epochs and use the weights with the best validation", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 104, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "mean squared error (MSE) on CelebA and the best test MSE on the Transient Scene Database. The", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 333, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 369, + 346 + ], + "score": 1.0, + "content": "GAN follows the StyleGAN2 architecture and is pretrained with", + "type": "text" + }, + { + "bbox": [ + 370, + 334, + 392, + 344 + ], + "score": 0.42, + "content": "2 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 333, + 506, + 346 + ], + "score": 1.0, + "content": "iterations on a union of the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 303, + 357 + ], + "score": 1.0, + "content": "two natural scene datasets using a resolution of", + "type": "text" + }, + { + "bbox": [ + 304, + 345, + 348, + 355 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 344, + 505, + 357 + ], + "score": 1.0, + "content": ". The training batch size is 32 and an", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "score": 1.0, + "content": "Adam optimizer is used with a learning rate of 2e-3. For the face datasets, we use FFHQ pre-trained", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 366, + 362, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 318, + 378 + ], + "score": 1.0, + "content": "weights1 of StyleGAN2, and CelebA-HQ weights of", + "type": "text" + }, + { + "bbox": [ + 319, + 366, + 358, + 377 + ], + "score": 0.46, + "content": "\\mathrm { P G G A N } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 366, + 362, + 378 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 384, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 383, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 159, + 394 + ], + "score": 1.0, + "content": "Baselines.", + "type": "text" + }, + { + "bbox": [ + 161, + 383, + 505, + 397 + ], + "score": 1.0, + "content": "We compare our approach to several popular image-to-image translation ap-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 393, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 104, + 393, + 505, + 408 + ], + "score": 1.0, + "content": "proaches (Choi et al., 2020; Wu et al., 2019; Zhu et al., 2017a) and latent space direction discovery", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "score": 1.0, + "content": "methods (Shen et al., 2019; Voynov & Babenko, 2020). Since Choi et al. (2020); Shen et al. (2019);", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 415, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 506, + 429 + ], + "score": 1.0, + "content": "Zhu et al. (2017a) cannot deal with continuous attributes, we split the data into 2 domains using a", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 426, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 506, + 440 + ], + "score": 1.0, + "content": "threshold value of 0.5 for each of the 40 attributes. Afterward, the models are trained on 2 contrast", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "score": 1.0, + "content": "domains, e.g., with or without a certain attribute. We use the official code of all baseline methods", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "score": 1.0, + "content": "and rigorously follow their training steps. Note that work by Voynov & Babenko (2020) is unsuper-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 460, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 506, + 473 + ], + "score": 1.0, + "content": "vised, i.e., the latent-space directions are human interpreted. To avoid bias during the selection of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "directions, we use the attribute regressor to automatically identify the most significant directions that", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 482, + 504, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 504, + 494 + ], + "score": 1.0, + "content": "can edit predetermined attributes. The details of preparing the baselines are given in Appendix C.2.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 499, + 505, + 576 + ], + "lines": [ + { + "bbox": [ + 105, + 498, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 505, + 512 + ], + "score": 1.0, + "content": "Evaluation metrics. There is no good numerical metric to evaluate image editing (Shen et al.,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "score": 1.0, + "content": "2019; Voynov & Babenko, 2020). In an attempt to address this concern, we automate quantitative", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 521, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 533 + ], + "score": 1.0, + "content": "evaluation based on a property that editing of attributes should maintain image identity. To achieve", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "this, we employ a popular image identity recognition model3 pre-trained on the VGGface2 data (Cao", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "score": 1.0, + "content": "et al., 2018). Cosine similarity is used to represent the similarity between paired original and edited", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 553, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 104, + 553, + 506, + 567 + ], + "score": 1.0, + "content": "images (Cao et al., 2018). In addition, we evaluate changes of the other independent attributes using", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 565, + 208, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 208, + 578 + ], + "score": 1.0, + "content": "the pre-trained regressor.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38 + }, + { + "type": "title", + "bbox": [ + 107, + 593, + 303, + 604 + ], + "lines": [ + { + "bbox": [ + 106, + 592, + 305, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 305, + 605 + ], + "score": 1.0, + "content": "4.1 RESULTS ON NATURAL SCENE DATASETS", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 614, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "Comparison to image-to-image translation: As shown in Fig. 3 (a-d), we observe image-to-image", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 625, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 505, + 638 + ], + "score": 1.0, + "content": "translation to perform poorly when editing image details (e.g., removing clouds). Moreover, some-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "times artifacts (orange spots in (a) and (b)) are introduced. In contrast, our model shows overall", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 648, + 329, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 329, + 659 + ], + "score": 1.0, + "content": "compelling transfer performance on all these attributes.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5 + }, + { + "type": "text", + "bbox": [ + 108, + 664, + 504, + 686 + ], + "lines": [ + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "score": 1.0, + "content": "Comparison to latent-space translation: Inspecting Fig. 1, we observe our approach to improve", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 675, + 506, + 688 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 506, + 688 + ], + "score": 1.0, + "content": "upon work by Shen et al. (2019) in attribute edit quality and number of artifacts. Similarly, in", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 699, + 293, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 698, + 293, + 712 + ], + "spans": [ + { + "bbox": [ + 119, + 698, + 293, + 712 + ], + "score": 1.0, + "content": "1https://github.com/rosinality/stylegan2-pytorch", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 709, + 208, + 722 + ], + "spans": [ + { + "bbox": [ + 118, + 709, + 208, + 722 + ], + "score": 1.0, + "content": "2https://pytorch.org/hub", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 720, + 257, + 734 + ], + "spans": [ + { + "bbox": [ + 118, + 720, + 257, + 734 + ], + "score": 1.0, + "content": "3https://github.com/ox-vgg/vgg face2", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 115 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Viazovetskyi et al., 2020). (ii) natural scene datasets – Transient Attribute Database (Laffont et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "2014) and MIT Places2 data (Zhou et al., 2017), which contain attributes suitable for continuous", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 366, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 366, + 117 + ], + "score": 1.0, + "content": "semantic image editing. We briefly introduce the scene datasets:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 82, + 505, + 117 + ] + }, + { + "type": "list", + "bbox": [ + 109, + 124, + 505, + 215 + ], + "lines": [ + { + "bbox": [ + 112, + 123, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 112, + 123, + 505, + 137 + ], + "score": 1.0, + "content": "• Transient Attribute Database (Laffont et al., 2014): It contains 8,571 scene images with 40 at-", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 134, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 120, + 134, + 506, + 149 + ], + "score": 1.0, + "content": "tributes in 5 categories including lighting (e.g., “bright”), weather (e.g., “cloudy”) , seasons (e.g.,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 120, + 145, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 120, + 145, + 505, + 159 + ], + "score": 1.0, + "content": "“winter”), subjective impressions (e.g., “beautiful”), and additional attributes (e.g., “dirty”).", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 120, + 156, + 505, + 170 + ], + "spans": [ + { + "bbox": [ + 120, + 156, + 505, + 170 + ], + "score": 1.0, + "content": "Each attribute is annotated with a real-valued score between 0 and 1, where 0 indicates absence", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 120, + 168, + 184, + 179 + ], + "spans": [ + { + "bbox": [ + 120, + 168, + 184, + 179 + ], + "score": 1.0, + "content": "of the attribute.", + "type": "text" + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 113, + 181, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 113, + 181, + 505, + 195 + ], + "score": 1.0, + "content": "• MIT Places2 data (Zhou et al., 2017): Using the provided category annotations (i.e., in-", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 193, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 120, + 193, + 506, + 205 + ], + "score": 1.0, + "content": "door/outdoor and natural/artificial), we select the natural outdoor scenes, obtaining a total of", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 121, + 203, + 190, + 217 + ], + "spans": [ + { + "bbox": [ + 121, + 203, + 190, + 217 + ], + "score": 1.0, + "content": "144,543 images.", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + } + ], + "index": 6.5, + "bbox_fs": [ + 112, + 123, + 506, + 217 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 224, + 505, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 223, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 223, + 259, + 236 + ], + "score": 1.0, + "content": "Implementation details. We choose", + "type": "text" + }, + { + "bbox": [ + 259, + 224, + 392, + 235 + ], + "score": 0.9, + "content": "\\lambda _ { 1 } = 1 0 , \\lambda _ { 2 } = 0 . 0 5 , \\lambda _ { 3 } = 0 . 0 5", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 223, + 505, + 236 + ], + "score": 1.0, + "content": "in Eq. (1) and compute the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 234, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 247 + ], + "score": 1.0, + "content": "perceptual loss using the conv1 2, conv2 2, conv3 2, conv4 2 activations in a VGG-19 network (Si-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 246, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 506, + 258 + ], + "score": 1.0, + "content": "monyan & Zisserman, 2014) pre-trained on the ImageNet dataset (Russakovsky et al., 2015). We", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 256, + 506, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 128, + 270 + ], + "score": 1.0, + "content": "train", + "type": "text" + }, + { + "bbox": [ + 128, + 257, + 137, + 267 + ], + "score": 0.75, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 256, + 153, + 270 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 154, + 257, + 171, + 267 + ], + "score": 0.54, + "content": "5 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 256, + 506, + 270 + ], + "score": 1.0, + "content": "iterations with a batch size of 4. An Adam optimizer is used with a learning rate", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 266, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 104, + 266, + 227, + 281 + ], + "score": 1.0, + "content": "of 1e-4. For the latent space", + "type": "text" + }, + { + "bbox": [ + 228, + 268, + 237, + 278 + ], + "score": 0.74, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 266, + 386, + 281 + ], + "score": 1.0, + "content": "in the proposed method we use the", + "type": "text" + }, + { + "bbox": [ + 386, + 268, + 399, + 278 + ], + "score": 0.7, + "content": "\\mathcal { W }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 266, + 506, + 281 + ], + "score": 1.0, + "content": "space of StyleGAN2 and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 121, + 291 + ], + "score": 1.0, + "content": "the", + "type": "text" + }, + { + "bbox": [ + 122, + 279, + 131, + 289 + ], + "score": 0.75, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 279, + 316, + 291 + ], + "score": 1.0, + "content": "space of PGGAN. For the attribute regressor", + "type": "text" + }, + { + "bbox": [ + 316, + 279, + 325, + 289 + ], + "score": 0.75, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 279, + 506, + 291 + ], + "score": 1.0, + "content": ", we adopt a ResNet-50 (He et al., 2016) for", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 289, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 506, + 302 + ], + "score": 1.0, + "content": "attribute prediction on the Transient Attribute (Laffont et al., 2014) and the CelebA (Liu et al., 2018)", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "score": 1.0, + "content": "data. The last fully connected layer in the ResNet-50 is replaced by a linear layer with an output", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "score": 1.0, + "content": "dimension of 40. We train the regressors for 500 epochs and use the weights with the best validation", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 104, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "mean squared error (MSE) on CelebA and the best test MSE on the Transient Scene Database. The", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 333, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 369, + 346 + ], + "score": 1.0, + "content": "GAN follows the StyleGAN2 architecture and is pretrained with", + "type": "text" + }, + { + "bbox": [ + 370, + 334, + 392, + 344 + ], + "score": 0.42, + "content": "2 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 333, + 506, + 346 + ], + "score": 1.0, + "content": "iterations on a union of the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 303, + 357 + ], + "score": 1.0, + "content": "two natural scene datasets using a resolution of", + "type": "text" + }, + { + "bbox": [ + 304, + 345, + 348, + 355 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 344, + 505, + 357 + ], + "score": 1.0, + "content": ". The training batch size is 32 and an", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "score": 1.0, + "content": "Adam optimizer is used with a learning rate of 2e-3. For the face datasets, we use FFHQ pre-trained", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 366, + 362, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 318, + 378 + ], + "score": 1.0, + "content": "weights1 of StyleGAN2, and CelebA-HQ weights of", + "type": "text" + }, + { + "bbox": [ + 319, + 366, + 358, + 377 + ], + "score": 0.46, + "content": "\\mathrm { P G G A N } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 366, + 362, + 378 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 17.5, + "bbox_fs": [ + 104, + 223, + 506, + 378 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 384, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 383, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 159, + 394 + ], + "score": 1.0, + "content": "Baselines.", + "type": "text" + }, + { + "bbox": [ + 161, + 383, + 505, + 397 + ], + "score": 1.0, + "content": "We compare our approach to several popular image-to-image translation ap-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 393, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 104, + 393, + 505, + 408 + ], + "score": 1.0, + "content": "proaches (Choi et al., 2020; Wu et al., 2019; Zhu et al., 2017a) and latent space direction discovery", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 418 + ], + "score": 1.0, + "content": "methods (Shen et al., 2019; Voynov & Babenko, 2020). Since Choi et al. (2020); Shen et al. (2019);", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 415, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 506, + 429 + ], + "score": 1.0, + "content": "Zhu et al. (2017a) cannot deal with continuous attributes, we split the data into 2 domains using a", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 426, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 506, + 440 + ], + "score": 1.0, + "content": "threshold value of 0.5 for each of the 40 attributes. Afterward, the models are trained on 2 contrast", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "score": 1.0, + "content": "domains, e.g., with or without a certain attribute. We use the official code of all baseline methods", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "score": 1.0, + "content": "and rigorously follow their training steps. Note that work by Voynov & Babenko (2020) is unsuper-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 460, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 506, + 473 + ], + "score": 1.0, + "content": "vised, i.e., the latent-space directions are human interpreted. To avoid bias during the selection of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "directions, we use the attribute regressor to automatically identify the most significant directions that", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 482, + 504, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 504, + 494 + ], + "score": 1.0, + "content": "can edit predetermined attributes. The details of preparing the baselines are given in Appendix C.2.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 383, + 506, + 494 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 499, + 505, + 576 + ], + "lines": [ + { + "bbox": [ + 105, + 498, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 505, + 512 + ], + "score": 1.0, + "content": "Evaluation metrics. There is no good numerical metric to evaluate image editing (Shen et al.,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "score": 1.0, + "content": "2019; Voynov & Babenko, 2020). In an attempt to address this concern, we automate quantitative", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 521, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 533 + ], + "score": 1.0, + "content": "evaluation based on a property that editing of attributes should maintain image identity. To achieve", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "this, we employ a popular image identity recognition model3 pre-trained on the VGGface2 data (Cao", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "score": 1.0, + "content": "et al., 2018). Cosine similarity is used to represent the similarity between paired original and edited", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 553, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 104, + 553, + 506, + 567 + ], + "score": 1.0, + "content": "images (Cao et al., 2018). In addition, we evaluate changes of the other independent attributes using", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 565, + 208, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 208, + 578 + ], + "score": 1.0, + "content": "the pre-trained regressor.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38, + "bbox_fs": [ + 104, + 498, + 506, + 578 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 593, + 303, + 604 + ], + "lines": [ + { + "bbox": [ + 106, + 592, + 305, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 305, + 605 + ], + "score": 1.0, + "content": "4.1 RESULTS ON NATURAL SCENE DATASETS", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 614, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "Comparison to image-to-image translation: As shown in Fig. 3 (a-d), we observe image-to-image", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 625, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 505, + 638 + ], + "score": 1.0, + "content": "translation to perform poorly when editing image details (e.g., removing clouds). Moreover, some-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "times artifacts (orange spots in (a) and (b)) are introduced. In contrast, our model shows overall", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 648, + 329, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 329, + 659 + ], + "score": 1.0, + "content": "compelling transfer performance on all these attributes.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 613, + 505, + 659 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 664, + 504, + 686 + ], + "lines": [ + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "score": 1.0, + "content": "Comparison to latent-space translation: Inspecting Fig. 1, we observe our approach to improve", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 675, + 506, + 688 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 506, + 688 + ], + "score": 1.0, + "content": "upon work by Shen et al. (2019) in attribute edit quality and number of artifacts. Similarly, in", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5, + "bbox_fs": [ + 106, + 664, + 506, + 688 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 89, + 235, + 147 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 89, + 235, + 147 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 119, + 89, + 235, + 147 + ], + "spans": [ + { + "bbox": [ + 119, + 89, + 235, + 147 + ], + "score": 0.957, + "type": "image", + "image_path": "2a075a8f6392e39967cfd1bca5063ddb38b1894f9895b367a74c0bc9a135410e.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 119, + 89, + 235, + 118.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 118.0, + 235, + 147.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 114, + 152, + 244, + 171 + ], + "lines": [ + { + "bbox": [ + 113, + 151, + 244, + 162 + ], + "spans": [ + { + "bbox": [ + 113, + 151, + 244, + 162 + ], + "score": 1.0, + "content": "Question1: Compared to the picture on the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 114, + 161, + 243, + 171 + ], + "spans": [ + { + "bbox": [ + 114, + 161, + 243, + 171 + ], + "score": 1.0, + "content": "left, how cloudy is the picture on the right?", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.0 + }, + { + "type": "image", + "bbox": [ + 254, + 84, + 499, + 203 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 254, + 84, + 499, + 203 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 254, + 84, + 499, + 203 + ], + "spans": [ + { + "bbox": [ + 254, + 84, + 499, + 203 + ], + "score": 0.952, + "type": "image", + "image_path": "73703449f7e679fa02f343ed6f43883fb95d0b5197a5344311e448950de5bd4b.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 254, + 84, + 499, + 123.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 254, + 123.66666666666666, + 499, + 163.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 254, + 163.33333333333331, + 499, + 202.99999999999997 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 114, + 176, + 238, + 203 + ], + "lines": [ + { + "bbox": [ + 114, + 176, + 238, + 186 + ], + "spans": [ + { + "bbox": [ + 114, + 176, + 238, + 186 + ], + "score": 1.0, + "content": "Question2: Ignoring the attribute clouds,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 113, + 185, + 232, + 195 + ], + "spans": [ + { + "bbox": [ + 113, + 185, + 232, + 195 + ], + "score": 1.0, + "content": "determine if the two pictures record the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 113, + 195, + 150, + 203 + ], + "spans": [ + { + "bbox": [ + 113, + 195, + 150, + 203 + ], + "score": 1.0, + "content": "same scene.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "image", + "bbox": [ + 113, + 268, + 502, + 401 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 108, + 222, + 504, + 255 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 221, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 235 + ], + "score": 1.0, + "content": "Figure 5: User study for image editing and image identity preservation. Blue bars show the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 311, + 246 + ], + "score": 1.0, + "content": "predicted presence of a given attribute for images", + "type": "text" + }, + { + "bbox": [ + 312, + 234, + 326, + 244 + ], + "score": 0.72, + "content": "( I +", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 233, + 505, + 246 + ], + "score": 1.0, + "content": ", higher is better). The orange bars measure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 244, + 326, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 326, + 256 + ], + "score": 1.0, + "content": "the difference between paired images (lower is better).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "image_body", + "bbox": [ + 113, + 268, + 502, + 401 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 268, + 502, + 401 + ], + "spans": [ + { + "bbox": [ + 113, + 268, + 502, + 401 + ], + "score": 0.969, + "type": "image", + "image_path": "779850f4a63da68102681994d49d10d728dc794f9dc3158d85d16ae82a39bfd4.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 113, + 268, + 502, + 312.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 113, + 312.3333333333333, + 502, + 356.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 113, + 356.66666666666663, + 502, + 400.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 409, + 504, + 443 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "score": 1.0, + "content": "Figure 6: Real image synthesis with multiple attributes. Shen et al. (2019) (top row); ours", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 420, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 432 + ], + "score": 1.0, + "content": "(bottom row): original real image (col.1); inverted result (col.2) from Abdal et al. (2019); edited", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 431, + 372, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 372, + 443 + ], + "score": 1.0, + "content": "results (col.3-7) with target attribute changes shown at the bottom.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 455, + 505, + 488 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "score": 1.0, + "content": "Fig. 4, we notice that our method performs well in attribute editing, and has a better ability to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "preserve image identity when adding “night” and “snow.” We show additional analysis and results", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 477, + 204, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 204, + 488 + ], + "score": 1.0, + "content": "in Appendix D and E.1.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 505, + 604 + ], + "lines": [ + { + "bbox": [ + 106, + 493, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 505, + 506 + ], + "score": 1.0, + "content": "Quantitative evaluation: We follow a user study proposed by Kowalski et al. (2020) to evaluate", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 504, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 504, + 352, + 518 + ], + "score": 1.0, + "content": "controllability of our scene edits. Concretely, given an image", + "type": "text" + }, + { + "bbox": [ + 352, + 505, + 393, + 516 + ], + "score": 0.93, + "content": "I = G ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 504, + 505, + 518 + ], + "score": 1.0, + "content": ", we generate paired images", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 515, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 145, + 527 + ], + "score": 0.89, + "content": "( I + , I - )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 515, + 299, + 528 + ], + "score": 1.0, + "content": "that have opposing values for attribute", + "type": "text" + }, + { + "bbox": [ + 299, + 518, + 306, + 525 + ], + "score": 0.54, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 515, + 505, + 528 + ], + "score": 1.0, + "content": ". 40 users took the test to answer: (a) the presence", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 527, + 504, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 218, + 538 + ], + "score": 1.0, + "content": "and strength of the attribute", + "type": "text" + }, + { + "bbox": [ + 218, + 528, + 226, + 536 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 527, + 236, + 538 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 237, + 527, + 251, + 537 + ], + "score": 0.87, + "content": "^ { I + }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 527, + 294, + 538 + ], + "score": 1.0, + "content": ", and (b) if", + "type": "text" + }, + { + "bbox": [ + 294, + 527, + 308, + 537 + ], + "score": 0.88, + "content": "I +", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 527, + 326, + 538 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 326, + 527, + 340, + 537 + ], + "score": 0.87, + "content": "I -", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 527, + 504, + 538 + ], + "score": 1.0, + "content": "are identical when ignoring the attribute.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 536, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 104, + 536, + 506, + 551 + ], + "score": 1.0, + "content": "Both types of questions have options on a 5-level scale from 0 (“not at all”) to 1 (“totally”). Ideally,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 547, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 505, + 561 + ], + "score": 1.0, + "content": "the response to question (a) should be 1 while we expect replies for (b) to be 0. We test performance", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 559, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 505, + 572 + ], + "score": 1.0, + "content": "of image editing with “clouds”, “sunrise&sunset”, “night” and “snow” attributes. A question ex-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 570, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 505, + 583 + ], + "score": 1.0, + "content": "ample and statistical results over 50 image pairs are shown in Fig. 5. Statistical evidence supports", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 505, + 595 + ], + "score": 1.0, + "content": "our claim that the proposed approach performs well with regards to image edits while maintaining", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 592, + 167, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 167, + 605 + ], + "score": 1.0, + "content": "image identity.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5 + }, + { + "type": "title", + "bbox": [ + 108, + 617, + 254, + 628 + ], + "lines": [ + { + "bbox": [ + 105, + 617, + 256, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 256, + 630 + ], + "score": 1.0, + "content": "4.2 RESULTS ON FACE DATASETS", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "Manipulation results on StyleGAN2: The comparison on real images shown in Fig. 1 suggests that", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "score": 1.0, + "content": "our method works well for attribute edits. Further, the synthetic image edit results in Fig. 4 indicate", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "that our edits are disentangled, while the baselines unexpectedly add “glasses” when aging the face.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 669, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 505, + 684 + ], + "score": 1.0, + "content": "To increase the task difficulty, we edit real images with multiple attribute changes simultaneously.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 682, + 435, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 435, + 694 + ], + "score": 1.0, + "content": "Results are summarized in Fig. 6, which highlights the controllability of our edits.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "Quantitative evaluation: We measure the changing degrees of the other independent attributes and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 365, + 723 + ], + "score": 1.0, + "content": "the image identity when editing attributes with various degrees", + "type": "text" + }, + { + "bbox": [ + 366, + 711, + 371, + 720 + ], + "score": 0.76, + "content": "\\hat { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 709, + 400, + 723 + ], + "score": 1.0, + "content": ". Here", + "type": "text" + }, + { + "bbox": [ + 400, + 710, + 407, + 720 + ], + "score": 0.77, + "content": "\\hat { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "is a predicted changing", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 504, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 226, + 734 + ], + "score": 1.0, + "content": "degree on a target attribute by", + "type": "text" + }, + { + "bbox": [ + 226, + 721, + 235, + 730 + ], + "score": 0.78, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 720, + 504, + 734 + ], + "score": 1.0, + "content": ". For a thorough comparison, we evaluate the performance on 3 seg-", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 89, + 235, + 147 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 89, + 235, + 147 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 119, + 89, + 235, + 147 + ], + "spans": [ + { + "bbox": [ + 119, + 89, + 235, + 147 + ], + "score": 0.957, + "type": "image", + "image_path": "2a075a8f6392e39967cfd1bca5063ddb38b1894f9895b367a74c0bc9a135410e.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 119, + 89, + 235, + 118.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 118.0, + 235, + 147.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 114, + 152, + 244, + 171 + ], + "lines": [ + { + "bbox": [ + 113, + 151, + 244, + 162 + ], + "spans": [ + { + "bbox": [ + 113, + 151, + 244, + 162 + ], + "score": 1.0, + "content": "Question1: Compared to the picture on the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 114, + 161, + 243, + 171 + ], + "spans": [ + { + "bbox": [ + 114, + 161, + 243, + 171 + ], + "score": 1.0, + "content": "left, how cloudy is the picture on the right?", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.0, + "bbox_fs": [ + 113, + 151, + 244, + 171 + ] + }, + { + "type": "image", + "bbox": [ + 254, + 84, + 499, + 203 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 254, + 84, + 499, + 203 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 254, + 84, + 499, + 203 + ], + "spans": [ + { + "bbox": [ + 254, + 84, + 499, + 203 + ], + "score": 0.952, + "type": "image", + "image_path": "73703449f7e679fa02f343ed6f43883fb95d0b5197a5344311e448950de5bd4b.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 254, + 84, + 499, + 123.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 254, + 123.66666666666666, + 499, + 163.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 254, + 163.33333333333331, + 499, + 202.99999999999997 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 114, + 176, + 238, + 203 + ], + "lines": [ + { + "bbox": [ + 114, + 176, + 238, + 186 + ], + "spans": [ + { + "bbox": [ + 114, + 176, + 238, + 186 + ], + "score": 1.0, + "content": "Question2: Ignoring the attribute clouds,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 113, + 185, + 232, + 195 + ], + "spans": [ + { + "bbox": [ + 113, + 185, + 232, + 195 + ], + "score": 1.0, + "content": "determine if the two pictures record the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 113, + 195, + 150, + 203 + ], + "spans": [ + { + "bbox": [ + 113, + 195, + 150, + 203 + ], + "score": 1.0, + "content": "same scene.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 113, + 176, + 238, + 203 + ] + }, + { + "type": "image", + "bbox": [ + 113, + 268, + 502, + 401 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 108, + 222, + 504, + 255 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 221, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 235 + ], + "score": 1.0, + "content": "Figure 5: User study for image editing and image identity preservation. Blue bars show the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 311, + 246 + ], + "score": 1.0, + "content": "predicted presence of a given attribute for images", + "type": "text" + }, + { + "bbox": [ + 312, + 234, + 326, + 244 + ], + "score": 0.72, + "content": "( I +", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 233, + 505, + 246 + ], + "score": 1.0, + "content": ", higher is better). The orange bars measure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 244, + 326, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 326, + 256 + ], + "score": 1.0, + "content": "the difference between paired images (lower is better).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "image_body", + "bbox": [ + 113, + 268, + 502, + 401 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 268, + 502, + 401 + ], + "spans": [ + { + "bbox": [ + 113, + 268, + 502, + 401 + ], + "score": 0.969, + "type": "image", + "image_path": "779850f4a63da68102681994d49d10d728dc794f9dc3158d85d16ae82a39bfd4.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 113, + 268, + 502, + 312.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 113, + 312.3333333333333, + 502, + 356.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 113, + 356.66666666666663, + 502, + 400.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 409, + 504, + 443 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "score": 1.0, + "content": "Figure 6: Real image synthesis with multiple attributes. Shen et al. (2019) (top row); ours", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 420, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 432 + ], + "score": 1.0, + "content": "(bottom row): original real image (col.1); inverted result (col.2) from Abdal et al. (2019); edited", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 431, + 372, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 372, + 443 + ], + "score": 1.0, + "content": "results (col.3-7) with target attribute changes shown at the bottom.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 455, + 505, + 488 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "score": 1.0, + "content": "Fig. 4, we notice that our method performs well in attribute editing, and has a better ability to", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "preserve image identity when adding “night” and “snow.” We show additional analysis and results", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 477, + 204, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 204, + 488 + ], + "score": 1.0, + "content": "in Appendix D and E.1.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 454, + 505, + 488 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 505, + 604 + ], + "lines": [ + { + "bbox": [ + 106, + 493, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 505, + 506 + ], + "score": 1.0, + "content": "Quantitative evaluation: We follow a user study proposed by Kowalski et al. (2020) to evaluate", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 504, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 504, + 352, + 518 + ], + "score": 1.0, + "content": "controllability of our scene edits. Concretely, given an image", + "type": "text" + }, + { + "bbox": [ + 352, + 505, + 393, + 516 + ], + "score": 0.93, + "content": "I = G ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 504, + 505, + 518 + ], + "score": 1.0, + "content": ", we generate paired images", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 515, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 145, + 527 + ], + "score": 0.89, + "content": "( I + , I - )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 515, + 299, + 528 + ], + "score": 1.0, + "content": "that have opposing values for attribute", + "type": "text" + }, + { + "bbox": [ + 299, + 518, + 306, + 525 + ], + "score": 0.54, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 515, + 505, + 528 + ], + "score": 1.0, + "content": ". 40 users took the test to answer: (a) the presence", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 527, + 504, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 218, + 538 + ], + "score": 1.0, + "content": "and strength of the attribute", + "type": "text" + }, + { + "bbox": [ + 218, + 528, + 226, + 536 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 527, + 236, + 538 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 237, + 527, + 251, + 537 + ], + "score": 0.87, + "content": "^ { I + }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 527, + 294, + 538 + ], + "score": 1.0, + "content": ", and (b) if", + "type": "text" + }, + { + "bbox": [ + 294, + 527, + 308, + 537 + ], + "score": 0.88, + "content": "I +", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 527, + 326, + 538 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 326, + 527, + 340, + 537 + ], + "score": 0.87, + "content": "I -", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 527, + 504, + 538 + ], + "score": 1.0, + "content": "are identical when ignoring the attribute.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 536, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 104, + 536, + 506, + 551 + ], + "score": 1.0, + "content": "Both types of questions have options on a 5-level scale from 0 (“not at all”) to 1 (“totally”). Ideally,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 547, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 505, + 561 + ], + "score": 1.0, + "content": "the response to question (a) should be 1 while we expect replies for (b) to be 0. We test performance", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 559, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 505, + 572 + ], + "score": 1.0, + "content": "of image editing with “clouds”, “sunrise&sunset”, “night” and “snow” attributes. A question ex-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 570, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 505, + 583 + ], + "score": 1.0, + "content": "ample and statistical results over 50 image pairs are shown in Fig. 5. Statistical evidence supports", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 505, + 595 + ], + "score": 1.0, + "content": "our claim that the proposed approach performs well with regards to image edits while maintaining", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 592, + 167, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 167, + 605 + ], + "score": 1.0, + "content": "image identity.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5, + "bbox_fs": [ + 104, + 493, + 506, + 605 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 617, + 254, + 628 + ], + "lines": [ + { + "bbox": [ + 105, + 617, + 256, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 256, + 630 + ], + "score": 1.0, + "content": "4.2 RESULTS ON FACE DATASETS", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "Manipulation results on StyleGAN2: The comparison on real images shown in Fig. 1 suggests that", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "score": 1.0, + "content": "our method works well for attribute edits. Further, the synthetic image edit results in Fig. 4 indicate", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "that our edits are disentangled, while the baselines unexpectedly add “glasses” when aging the face.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 669, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 505, + 684 + ], + "score": 1.0, + "content": "To increase the task difficulty, we edit real images with multiple attribute changes simultaneously.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 682, + 435, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 435, + 694 + ], + "score": 1.0, + "content": "Results are summarized in Fig. 6, which highlights the controllability of our edits.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 637, + 506, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "Quantitative evaluation: We measure the changing degrees of the other independent attributes and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 365, + 723 + ], + "score": 1.0, + "content": "the image identity when editing attributes with various degrees", + "type": "text" + }, + { + "bbox": [ + 366, + 711, + 371, + 720 + ], + "score": 0.76, + "content": "\\hat { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 709, + 400, + 723 + ], + "score": 1.0, + "content": ". Here", + "type": "text" + }, + { + "bbox": [ + 400, + 710, + 407, + 720 + ], + "score": 0.77, + "content": "\\hat { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "is a predicted changing", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 504, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 226, + 734 + ], + "score": 1.0, + "content": "degree on a target attribute by", + "type": "text" + }, + { + "bbox": [ + 226, + 721, + 235, + 730 + ], + "score": 0.78, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 720, + 504, + 734 + ], + "score": 1.0, + "content": ". For a thorough comparison, we evaluate the performance on 3 seg-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 428, + 504, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 272, + 440 + ], + "score": 1.0, + "content": "ments according to the absolute value of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 272, + 428, + 278, + 438 + ], + "score": 0.63, + "content": "\\hat { \\varepsilon }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 279, + 428, + 300, + 440 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 300, + 428, + 311, + 440 + ], + "score": 0.87, + "content": "| \\hat { \\varepsilon } |", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 312, + 428, + 375, + 440 + ], + "score": 1.0, + "content": "in the range of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 375, + 428, + 405, + 440 + ], + "score": 0.33, + "content": "( 0 , 0 . 3 ]", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 405, + 428, + 504, + 440 + ], + "score": 1.0, + "content": ", (0.3, 0.6] and (0.6, 0.9].", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "score": 1.0, + "content": "Evaluation on “smile,” “hair color,” and “smile+hair color” attributes are shown in Tab. 1 and Tab. 2.", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 449, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 505, + 462 + ], + "score": 1.0, + "content": "The “hair color” attribute includes “blond” and “black” colors where we average the results on both", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 461, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 280, + 473 + ], + "score": 1.0, + "content": "cases. For multiple attribute editing, “smile", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 281, + 462, + 289, + 471 + ], + "score": 0.75, + "content": "^ +", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 290, + 461, + 506, + 473 + ], + "score": 1.0, + "content": "hair color,” we evaluate the case that both the two tar-", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 243, + 484 + ], + "score": 1.0, + "content": "geted attributes change within the", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 244, + 472, + 255, + 484 + ], + "score": 0.89, + "content": "| \\hat { \\varepsilon } |", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 255, + 471, + 461, + 484 + ], + "score": 1.0, + "content": "range. We use around 1k original images, generate", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 461, + 472, + 477, + 482 + ], + "score": 0.31, + "content": "1 0 \\mathrm { k }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 478, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "edited", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "images with regard to each target attribute, and repeat the experiment 3 times. The averaged results", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 494, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 505, + 506 + ], + "score": 1.0, + "content": "and the standard deviations presented in Tab. 1 and Tab. 2 suggest that our model outperforms the", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 504, + 402, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 402, + 518 + ], + "score": 1.0, + "content": "baselines with regard to disentanglement and image identity preservation.", + "type": "text", + "cross_page": true + } + ], + "index": 19 + } + ], + "index": 39, + "bbox_fs": [ + 104, + 698, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 124, + 79, + 487, + 192 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 124, + 79, + 487, + 192 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 79, + 487, + 192 + ], + "spans": [ + { + "bbox": [ + 124, + 79, + 487, + 192 + ], + "score": 0.98, + "html": "
SmileHair colorSmile + Hair color
(0,.3](.3,.6](.6,.9](0,.3](.3, 6] (.6,.9](0,.3](.3,.6](.6,.9]
Shen et al..202 ±5e-2.204 ±6e-2.224 ±4e-2.256 ±1e-2.272 ±2e-2.277 ±2e-2.299 ±5e-3.318 ±1e-2.329 ±3e-2
Voynov et al..115 ± 6e-3.211 ±4e-2.277.162.166.177.155.220 ±4e-2.284 ±2e-2
Ours.085 ±4e-2.084 ±3e-3±7e-3 .098 ±4e-3±1e-2 .075 ±7e-4±3e-2 .083 ±5e-3±9e-3 .084 ±5e-3±4e-3 .088 ±7e-3.111 ±4e-3.134 ±4e-3
", + "type": "table", + "image_path": "6befb6d6ec2e53040e29b5d9fd95e4787eb57a5a0b06a99590158e80057fe3af.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 124, + 79, + 487, + 116.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 124, + 116.66666666666666, + 487, + 154.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 124, + 154.33333333333331, + 487, + 191.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 200, + 506, + 244 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 200, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 505, + 212 + ], + "score": 1.0, + "content": "Table 1: Quantitative evaluation of numerical changes on the other semantically independent", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 210, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 506, + 225 + ], + "score": 1.0, + "content": "attributes (lower is better). (first row) edited attributes; (second row) the absolute changing range", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "score": 1.0, + "content": "of the edited attributes; (bottom three rows) averages (up) and standard deviations (down) in each", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 235, + 127, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 127, + 245 + ], + "score": 1.0, + "content": "row.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "table", + "bbox": [ + 129, + 258, + 482, + 370 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 129, + 258, + 482, + 370 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 129, + 258, + 482, + 370 + ], + "spans": [ + { + "bbox": [ + 129, + 258, + 482, + 370 + ], + "score": 0.979, + "html": "
SmileHair colorSmile+Haircolor
(0,.3] (.3,.6] (.6,.9] (0,.3] (.3,.6] (.6,.9] (0,.3] (.3,.6] (.6,.9)
Shen et al..918.916.907.887.877.874.877.819.801 ± 4e-3 ±1e-3 ± 5e-3 ±3e-2 ± 4e-2 ±4e-2 ± 6e-3 ± 3e-2 ±4e-2
Voynov et al..979.896.869.955.909.904.940.829.811
Ours.995.994.992.993.993.993.992.986.984± 2e-3 ±4e-3 ± 9e-3 ± 6e-2 ±4e-2 ±3e-2 ±9e-3 ± 4e-2 ±3e-2 ± 7e-4 ± 1e-3 ± 9e-4 ± 5e-4 ±1e-4 ± 2e-4 ± 9e-4 ± 2e-3 ± 4e-4
", + "type": "table", + "image_path": "f56e615daf9cbffd26483bde8850fc7d63006749c13e85f773ea4e03fdb4da8f.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 129, + 258, + 482, + 295.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 129, + 295.3333333333333, + 482, + 332.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 129, + 332.66666666666663, + 482, + 369.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 108, + 378, + 504, + 401 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 376, + 506, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 391 + ], + "score": 1.0, + "content": "Table 2: Quantitative evaluation on image identity preservation (higher is better). Notation is", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 389, + 183, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 183, + 401 + ], + "score": 1.0, + "content": "identical to Tab. 1.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + } + ], + "index": 9.25 + }, + { + "type": "text", + "bbox": [ + 106, + 427, + 505, + 516 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 504, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 272, + 440 + ], + "score": 1.0, + "content": "ments according to the absolute value of", + "type": "text" + }, + { + "bbox": [ + 272, + 428, + 278, + 438 + ], + "score": 0.63, + "content": "\\hat { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 428, + 300, + 440 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + }, + { + "bbox": [ + 300, + 428, + 311, + 440 + ], + "score": 0.87, + "content": "| \\hat { \\varepsilon } |", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 428, + 375, + 440 + ], + "score": 1.0, + "content": "in the range of", + "type": "text" + }, + { + "bbox": [ + 375, + 428, + 405, + 440 + ], + "score": 0.33, + "content": "( 0 , 0 . 3 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 428, + 504, + 440 + ], + "score": 1.0, + "content": ", (0.3, 0.6] and (0.6, 0.9].", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "score": 1.0, + "content": "Evaluation on “smile,” “hair color,” and “smile+hair color” attributes are shown in Tab. 1 and Tab. 2.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 449, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 505, + 462 + ], + "score": 1.0, + "content": "The “hair color” attribute includes “blond” and “black” colors where we average the results on both", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 461, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 280, + 473 + ], + "score": 1.0, + "content": "cases. For multiple attribute editing, “smile", + "type": "text" + }, + { + "bbox": [ + 281, + 462, + 289, + 471 + ], + "score": 0.75, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 461, + 506, + 473 + ], + "score": 1.0, + "content": "hair color,” we evaluate the case that both the two tar-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 243, + 484 + ], + "score": 1.0, + "content": "geted attributes change within the", + "type": "text" + }, + { + "bbox": [ + 244, + 472, + 255, + 484 + ], + "score": 0.89, + "content": "| \\hat { \\varepsilon } |", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 471, + 461, + 484 + ], + "score": 1.0, + "content": "range. We use around 1k original images, generate", + "type": "text" + }, + { + "bbox": [ + 461, + 472, + 477, + 482 + ], + "score": 0.31, + "content": "1 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "edited", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "images with regard to each target attribute, and repeat the experiment 3 times. The averaged results", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 494, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 505, + 506 + ], + "score": 1.0, + "content": "and the standard deviations presented in Tab. 1 and Tab. 2 suggest that our model outperforms the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 504, + 402, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 402, + 518 + ], + "score": 1.0, + "content": "baselines with regard to disentanglement and image identity preservation.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 521, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "Manipulation results on PGGAN: For PGGAN, we use local transformations parameterized by", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 532, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 506, + 545 + ], + "score": 1.0, + "content": "three-layer perceptrons with leaky ReLU activations. We show the detailed MLP architecture in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "Appendix B.1. Editing results in Fig. 7 suggest that the proposed approach is able to continuously", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 555, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 566 + ], + "score": 1.0, + "content": "edit faces with various semantic attributes. We also study the differences between local and global", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 564, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 275, + 578 + ], + "score": 1.0, + "content": "directions in Fig. 11. Particularly, a global", + "type": "text" + }, + { + "bbox": [ + 275, + 566, + 284, + 576 + ], + "score": 0.8, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 564, + 466, + 578 + ], + "score": 1.0, + "content": "(in Fig. 11 (a)) is a matrix where each column", + "type": "text" + }, + { + "bbox": [ + 467, + 566, + 471, + 575 + ], + "score": 0.7, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 564, + 506, + 578 + ], + "score": 1.0, + "content": "is given", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 576, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 119, + 589 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 119, + 577, + 129, + 587 + ], + "score": 0.87, + "content": "\\mathbf { \\ b { d } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 576, + 180, + 589 + ], + "score": 1.0, + "content": ". For a local", + "type": "text" + }, + { + "bbox": [ + 180, + 577, + 190, + 587 + ], + "score": 0.79, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 576, + 253, + 589 + ], + "score": 1.0, + "content": "(in Fig. 11 (b))", + "type": "text" + }, + { + "bbox": [ + 254, + 577, + 264, + 587 + ], + "score": 0.86, + "content": "\\mathbf { \\ b { d } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 576, + 506, + 589 + ], + "score": 1.0, + "content": "is given by a fully-connected layer with identical input and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "score": 1.0, + "content": "output dimensions. Fig. 11 (c) and (d) use the MLP components with the aforementioned structure.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 597, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 104, + 597, + 138, + 612 + ], + "score": 1.0, + "content": "We use", + "type": "text" + }, + { + "bbox": [ + 139, + 599, + 156, + 610 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 597, + 189, + 612 + ], + "score": 1.0, + "content": "to train", + "type": "text" + }, + { + "bbox": [ + 190, + 599, + 199, + 609 + ], + "score": 0.74, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 597, + 307, + 612 + ], + "score": 1.0, + "content": "in (c), while the total loss", + "type": "text" + }, + { + "bbox": [ + 307, + 599, + 316, + 608 + ], + "score": 0.75, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 597, + 506, + 612 + ], + "score": 1.0, + "content": "is used for (d). The results in Fig. 11 indicate", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 608, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 152, + 622 + ], + "score": 1.0, + "content": "that a local", + "type": "text" + }, + { + "bbox": [ + 153, + 610, + 162, + 619 + ], + "score": 0.8, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 608, + 506, + 622 + ], + "score": 1.0, + "content": ", with either linear or MLP structure works well for face attribute edits. Moreover, our", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 621, + 277, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 124, + 632 + ], + "score": 1.0, + "content": "loss", + "type": "text" + }, + { + "bbox": [ + 124, + 621, + 133, + 630 + ], + "score": 0.79, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 621, + 264, + 632 + ], + "score": 1.0, + "content": "stabilizes the training process of", + "type": "text" + }, + { + "bbox": [ + 265, + 621, + 274, + 630 + ], + "score": 0.74, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 621, + 277, + 632 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5 + }, + { + "type": "image", + "bbox": [ + 106, + 655, + 512, + 710 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 655, + 512, + 710 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 655, + 512, + 710 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 512, + 710 + ], + "score": 0.683, + "type": "image", + "image_path": "4fa6e5104333e00a1e806c7aae8b7cd5a1fcdbf707a582fbe68a63e86a89de99.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 106, + 655, + 512, + 673.3333333333334 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 106, + 673.3333333333334, + 512, + 691.6666666666667 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 106, + 691.6666666666667, + 512, + 710.0000000000001 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 713, + 504, + 736 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 713, + 505, + 726 + ], + "spans": [ + { + "bbox": [ + 106, + 713, + 505, + 726 + ], + "score": 1.0, + "content": "Figure 7: Continuous editing results on PGGAN (Karras et al., 2017) with the “black hair” (left)", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 723, + 237, + 737 + ], + "spans": [ + { + "bbox": [ + 106, + 723, + 237, + 737 + ], + "score": 1.0, + "content": "and the “smile” (right) attribute.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + } + ], + "index": 32.25 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 124, + 79, + 487, + 192 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 124, + 79, + 487, + 192 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 79, + 487, + 192 + ], + "spans": [ + { + "bbox": [ + 124, + 79, + 487, + 192 + ], + "score": 0.98, + "html": "
SmileHair colorSmile + Hair color
(0,.3](.3,.6](.6,.9](0,.3](.3, 6] (.6,.9](0,.3](.3,.6](.6,.9]
Shen et al..202 ±5e-2.204 ±6e-2.224 ±4e-2.256 ±1e-2.272 ±2e-2.277 ±2e-2.299 ±5e-3.318 ±1e-2.329 ±3e-2
Voynov et al..115 ± 6e-3.211 ±4e-2.277.162.166.177.155.220 ±4e-2.284 ±2e-2
Ours.085 ±4e-2.084 ±3e-3±7e-3 .098 ±4e-3±1e-2 .075 ±7e-4±3e-2 .083 ±5e-3±9e-3 .084 ±5e-3±4e-3 .088 ±7e-3.111 ±4e-3.134 ±4e-3
", + "type": "table", + "image_path": "6befb6d6ec2e53040e29b5d9fd95e4787eb57a5a0b06a99590158e80057fe3af.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 124, + 79, + 487, + 116.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 124, + 116.66666666666666, + 487, + 154.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 124, + 154.33333333333331, + 487, + 191.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 200, + 506, + 244 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 200, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 505, + 212 + ], + "score": 1.0, + "content": "Table 1: Quantitative evaluation of numerical changes on the other semantically independent", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 210, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 506, + 225 + ], + "score": 1.0, + "content": "attributes (lower is better). (first row) edited attributes; (second row) the absolute changing range", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "score": 1.0, + "content": "of the edited attributes; (bottom three rows) averages (up) and standard deviations (down) in each", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 235, + 127, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 127, + 245 + ], + "score": 1.0, + "content": "row.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "table", + "bbox": [ + 129, + 258, + 482, + 370 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 129, + 258, + 482, + 370 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 129, + 258, + 482, + 370 + ], + "spans": [ + { + "bbox": [ + 129, + 258, + 482, + 370 + ], + "score": 0.979, + "html": "
SmileHair colorSmile+Haircolor
(0,.3] (.3,.6] (.6,.9] (0,.3] (.3,.6] (.6,.9] (0,.3] (.3,.6] (.6,.9)
Shen et al..918.916.907.887.877.874.877.819.801 ± 4e-3 ±1e-3 ± 5e-3 ±3e-2 ± 4e-2 ±4e-2 ± 6e-3 ± 3e-2 ±4e-2
Voynov et al..979.896.869.955.909.904.940.829.811
Ours.995.994.992.993.993.993.992.986.984± 2e-3 ±4e-3 ± 9e-3 ± 6e-2 ±4e-2 ±3e-2 ±9e-3 ± 4e-2 ±3e-2 ± 7e-4 ± 1e-3 ± 9e-4 ± 5e-4 ±1e-4 ± 2e-4 ± 9e-4 ± 2e-3 ± 4e-4
", + "type": "table", + "image_path": "f56e615daf9cbffd26483bde8850fc7d63006749c13e85f773ea4e03fdb4da8f.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 129, + 258, + 482, + 295.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 129, + 295.3333333333333, + 482, + 332.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 129, + 332.66666666666663, + 482, + 369.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 108, + 378, + 504, + 401 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 376, + 506, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 391 + ], + "score": 1.0, + "content": "Table 2: Quantitative evaluation on image identity preservation (higher is better). Notation is", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 389, + 183, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 183, + 401 + ], + "score": 1.0, + "content": "identical to Tab. 1.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + } + ], + "index": 9.25 + }, + { + "type": "text", + "bbox": [ + 106, + 427, + 505, + 516 + ], + "lines": [], + "index": 15.5, + "bbox_fs": [ + 105, + 428, + 506, + 518 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 521, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "Manipulation results on PGGAN: For PGGAN, we use local transformations parameterized by", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 532, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 506, + 545 + ], + "score": 1.0, + "content": "three-layer perceptrons with leaky ReLU activations. We show the detailed MLP architecture in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "Appendix B.1. Editing results in Fig. 7 suggest that the proposed approach is able to continuously", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 555, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 566 + ], + "score": 1.0, + "content": "edit faces with various semantic attributes. We also study the differences between local and global", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 564, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 275, + 578 + ], + "score": 1.0, + "content": "directions in Fig. 11. Particularly, a global", + "type": "text" + }, + { + "bbox": [ + 275, + 566, + 284, + 576 + ], + "score": 0.8, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 564, + 466, + 578 + ], + "score": 1.0, + "content": "(in Fig. 11 (a)) is a matrix where each column", + "type": "text" + }, + { + "bbox": [ + 467, + 566, + 471, + 575 + ], + "score": 0.7, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 564, + 506, + 578 + ], + "score": 1.0, + "content": "is given", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 576, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 119, + 589 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 119, + 577, + 129, + 587 + ], + "score": 0.87, + "content": "\\mathbf { \\ b { d } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 576, + 180, + 589 + ], + "score": 1.0, + "content": ". For a local", + "type": "text" + }, + { + "bbox": [ + 180, + 577, + 190, + 587 + ], + "score": 0.79, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 576, + 253, + 589 + ], + "score": 1.0, + "content": "(in Fig. 11 (b))", + "type": "text" + }, + { + "bbox": [ + 254, + 577, + 264, + 587 + ], + "score": 0.86, + "content": "\\mathbf { \\ b { d } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 576, + 506, + 589 + ], + "score": 1.0, + "content": "is given by a fully-connected layer with identical input and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "score": 1.0, + "content": "output dimensions. Fig. 11 (c) and (d) use the MLP components with the aforementioned structure.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 597, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 104, + 597, + 138, + 612 + ], + "score": 1.0, + "content": "We use", + "type": "text" + }, + { + "bbox": [ + 139, + 599, + 156, + 610 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 597, + 189, + 612 + ], + "score": 1.0, + "content": "to train", + "type": "text" + }, + { + "bbox": [ + 190, + 599, + 199, + 609 + ], + "score": 0.74, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 597, + 307, + 612 + ], + "score": 1.0, + "content": "in (c), while the total loss", + "type": "text" + }, + { + "bbox": [ + 307, + 599, + 316, + 608 + ], + "score": 0.75, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 597, + 506, + 612 + ], + "score": 1.0, + "content": "is used for (d). The results in Fig. 11 indicate", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 608, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 152, + 622 + ], + "score": 1.0, + "content": "that a local", + "type": "text" + }, + { + "bbox": [ + 153, + 610, + 162, + 619 + ], + "score": 0.8, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 608, + 506, + 622 + ], + "score": 1.0, + "content": ", with either linear or MLP structure works well for face attribute edits. Moreover, our", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 621, + 277, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 124, + 632 + ], + "score": 1.0, + "content": "loss", + "type": "text" + }, + { + "bbox": [ + 124, + 621, + 133, + 630 + ], + "score": 0.79, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 621, + 264, + 632 + ], + "score": 1.0, + "content": "stabilizes the training process of", + "type": "text" + }, + { + "bbox": [ + 265, + 621, + 274, + 630 + ], + "score": 0.74, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 621, + 277, + 632 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5, + "bbox_fs": [ + 104, + 521, + 506, + 632 + ] + }, + { + "type": "image", + "bbox": [ + 106, + 655, + 512, + 710 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 655, + 512, + 710 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 655, + 512, + 710 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 512, + 710 + ], + "score": 0.683, + "type": "image", + "image_path": "4fa6e5104333e00a1e806c7aae8b7cd5a1fcdbf707a582fbe68a63e86a89de99.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 106, + 655, + 512, + 673.3333333333334 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 106, + 673.3333333333334, + 512, + 691.6666666666667 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 106, + 691.6666666666667, + 512, + 710.0000000000001 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 713, + 504, + 736 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 713, + 505, + 726 + ], + "spans": [ + { + "bbox": [ + 106, + 713, + 505, + 726 + ], + "score": 1.0, + "content": "Figure 7: Continuous editing results on PGGAN (Karras et al., 2017) with the “black hair” (left)", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 723, + 237, + 737 + ], + "spans": [ + { + "bbox": [ + 106, + 723, + 237, + 737 + ], + "score": 1.0, + "content": "and the “smile” (right) attribute.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + } + ], + "index": 32.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 149, + 75, + 460, + 213 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 149, + 75, + 460, + 213 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 149, + 75, + 460, + 213 + ], + "spans": [ + { + "bbox": [ + 149, + 75, + 460, + 213 + ], + "score": 0.972, + "type": "image", + "image_path": "65a0ce13abac868def241a0d1ed0540afe7ab5e45559a10ee00dfd0c82016709.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 149, + 75, + 460, + 121.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 149, + 121.0, + 460, + 167.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 149, + 167.0, + 460, + 213.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 218, + 506, + 262 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "score": 1.0, + "content": "Figure 8: Comparison of global and local transformations on PGGAN. Original synthetic images", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 387, + 242 + ], + "score": 1.0, + "content": "(col.1); the edited results on enhancing “smile” (col.2-5): (a) a global", + "type": "text" + }, + { + "bbox": [ + 387, + 230, + 396, + 239 + ], + "score": 0.68, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 228, + 443, + 242 + ], + "score": 1.0, + "content": ", (b) a local", + "type": "text" + }, + { + "bbox": [ + 443, + 230, + 452, + 239 + ], + "score": 0.77, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 228, + 506, + 242 + ], + "score": 1.0, + "content": ", with linear-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 252, + 253 + ], + "score": 1.0, + "content": "layer direction functions, (c) a local", + "type": "text" + }, + { + "bbox": [ + 252, + 240, + 262, + 250 + ], + "score": 0.7, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 240, + 429, + 253 + ], + "score": 1.0, + "content": "with MLP direction functions trained via", + "type": "text" + }, + { + "bbox": [ + 429, + 240, + 447, + 252 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 240, + 506, + 253 + ], + "score": 1.0, + "content": "loss only, and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 251, + 399, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 149, + 264 + ], + "score": 1.0, + "content": "(d) a local", + "type": "text" + }, + { + "bbox": [ + 149, + 252, + 159, + 261 + ], + "score": 0.75, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 251, + 386, + 264 + ], + "score": 1.0, + "content": "with MLP direction functions training via the entire loss", + "type": "text" + }, + { + "bbox": [ + 387, + 252, + 394, + 261 + ], + "score": 0.72, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 251, + 399, + 264 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "title", + "bbox": [ + 107, + 282, + 195, + 294 + ], + "lines": [ + { + "bbox": [ + 104, + 280, + 198, + 298 + ], + "spans": [ + { + "bbox": [ + 104, + 280, + 198, + 298 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 299, + 505, + 343 + ], + "lines": [ + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "We propose an effective approach to semantically edit images by transferring latent vectors towards", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 311, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 505, + 322 + ], + "score": 1.0, + "content": "meaningful latent space directions. The proposed method enables continuous image manipulations", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 321, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 505, + 334 + ], + "score": 1.0, + "content": "with respect to various attributes. Extensive experiments highlight that the model achieves state-of-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 332, + 320, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 320, + 344 + ], + "score": 1.0, + "content": "the-art performance for targeted image manipulation.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 349, + 505, + 393 + ], + "lines": [ + { + "bbox": [ + 105, + 348, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 505, + 361 + ], + "score": 1.0, + "content": "Acknowledgements: This work is supported in part by NSF under Grant No. 1718221, 2008387,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 359, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 505, + 372 + ], + "score": 1.0, + "content": "1934986 and MRI #1725729, NIFA award 2020-67021-32799, UIUC, Samsung, Amazon, 3M, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "Cisco Systems Inc. (Gift Award CG 1377144). We thank Cisco for access to the Arcetri cluster. We", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 381, + 234, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 234, + 393 + ], + "score": 1.0, + "content": "thank Amazon for EC2 credits.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "title", + "bbox": [ + 107, + 410, + 175, + 423 + ], + "lines": [ + { + "bbox": [ + 106, + 411, + 176, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 176, + 424 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 105, + 429, + 506, + 733 + ], + "lines": [ + { + "bbox": [ + 104, + 428, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 104, + 428, + 505, + 444 + ], + "score": 1.0, + "content": "Rameen Abdal, Yipeng Qin, and Peter Wonka. Image2stylegan: How to embed images into the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 441, + 271, + 453 + ], + "spans": [ + { + "bbox": [ + 116, + 441, + 271, + 453 + ], + "score": 1.0, + "content": "stylegan latent space? In ICCV, 2019.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 459, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 104, + 459, + 505, + 475 + ], + "score": 1.0, + "content": "Binod Bhattarai and Tae-Kyun Kim. Inducing optimal attribute representations for conditional gans.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 472, + 172, + 484 + ], + "spans": [ + { + "bbox": [ + 116, + 472, + 172, + 484 + ], + "score": 1.0, + "content": "ECCV, 2020.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "score": 1.0, + "content": "Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 503, + 238, + 516 + ], + "spans": [ + { + "bbox": [ + 116, + 503, + 238, + 516 + ], + "score": 1.0, + "content": "image synthesis. arXiv, 2018.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 522, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 506, + 537 + ], + "score": 1.0, + "content": "Q. Cao, L. Shen, W. Xie, et al. Vggface2: A dataset for recognising faces across pose and age. In", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 533, + 184, + 547 + ], + "spans": [ + { + "bbox": [ + 115, + 533, + 184, + 547 + ], + "score": 1.0, + "content": "IEEE FG, 2018.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "score": 1.0, + "content": "Yunjey Choi, Minje Choi, Munyoung Kim, et al. Stargan: Unified generative adversarial networks", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 114, + 564, + 367, + 579 + ], + "spans": [ + { + "bbox": [ + 114, + 564, + 367, + 579 + ], + "score": 1.0, + "content": "for multi-domain image-to-image translation. In CVPR, 2018.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 506, + 599 + ], + "score": 1.0, + "content": "Yunjey Choi, Youngjung Uh, Jaejun Yoo, et al. Stargan v2: Diverse image synthesis for multiple", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 596, + 211, + 609 + ], + "spans": [ + { + "bbox": [ + 115, + 596, + 211, + 609 + ], + "score": 1.0, + "content": "domains. CVPR, 2020.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 616, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 630 + ], + "score": 1.0, + "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, et al. Generative adversarial nets. In NeurIPS,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 627, + 142, + 640 + ], + "spans": [ + { + "bbox": [ + 115, + 627, + 142, + 640 + ], + "score": 1.0, + "content": "2014.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 646, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 104, + 646, + 505, + 662 + ], + "score": 1.0, + "content": "Erik Hark ¨ onen, Aaron Hertzmann, Jaakko Lehtinen, et al. Ganspace: Discovering interpretable gan ¨", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 659, + 207, + 672 + ], + "spans": [ + { + "bbox": [ + 116, + 659, + 207, + 672 + ], + "score": 1.0, + "content": "controls. arXiv, 2020.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 677, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 104, + 677, + 505, + 694 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 690, + 211, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 211, + 702 + ], + "score": 1.0, + "content": "nition. In CVPR, 2016.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 709, + 505, + 725 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 505, + 725 + ], + "score": 1.0, + "content": "Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, et al. Image-to-image translation with conditional adver-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 721, + 249, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 721, + 249, + 733 + ], + "score": 1.0, + "content": "sarial networks. In CVPR, 2017.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 26.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 149, + 75, + 460, + 213 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 149, + 75, + 460, + 213 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 149, + 75, + 460, + 213 + ], + "spans": [ + { + "bbox": [ + 149, + 75, + 460, + 213 + ], + "score": 0.972, + "type": "image", + "image_path": "65a0ce13abac868def241a0d1ed0540afe7ab5e45559a10ee00dfd0c82016709.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 149, + 75, + 460, + 121.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 149, + 121.0, + 460, + 167.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 149, + 167.0, + 460, + 213.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 218, + 506, + 262 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "score": 1.0, + "content": "Figure 8: Comparison of global and local transformations on PGGAN. Original synthetic images", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 387, + 242 + ], + "score": 1.0, + "content": "(col.1); the edited results on enhancing “smile” (col.2-5): (a) a global", + "type": "text" + }, + { + "bbox": [ + 387, + 230, + 396, + 239 + ], + "score": 0.68, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 228, + 443, + 242 + ], + "score": 1.0, + "content": ", (b) a local", + "type": "text" + }, + { + "bbox": [ + 443, + 230, + 452, + 239 + ], + "score": 0.77, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 228, + 506, + 242 + ], + "score": 1.0, + "content": ", with linear-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 252, + 253 + ], + "score": 1.0, + "content": "layer direction functions, (c) a local", + "type": "text" + }, + { + "bbox": [ + 252, + 240, + 262, + 250 + ], + "score": 0.7, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 240, + 429, + 253 + ], + "score": 1.0, + "content": "with MLP direction functions trained via", + "type": "text" + }, + { + "bbox": [ + 429, + 240, + 447, + 252 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 240, + 506, + 253 + ], + "score": 1.0, + "content": "loss only, and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 251, + 399, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 149, + 264 + ], + "score": 1.0, + "content": "(d) a local", + "type": "text" + }, + { + "bbox": [ + 149, + 252, + 159, + 261 + ], + "score": 0.75, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 251, + 386, + 264 + ], + "score": 1.0, + "content": "with MLP direction functions training via the entire loss", + "type": "text" + }, + { + "bbox": [ + 387, + 252, + 394, + 261 + ], + "score": 0.72, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 251, + 399, + 264 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "title", + "bbox": [ + 107, + 282, + 195, + 294 + ], + "lines": [ + { + "bbox": [ + 104, + 280, + 198, + 298 + ], + "spans": [ + { + "bbox": [ + 104, + 280, + 198, + 298 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 299, + 505, + 343 + ], + "lines": [ + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "We propose an effective approach to semantically edit images by transferring latent vectors towards", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 311, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 505, + 322 + ], + "score": 1.0, + "content": "meaningful latent space directions. The proposed method enables continuous image manipulations", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 321, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 505, + 334 + ], + "score": 1.0, + "content": "with respect to various attributes. Extensive experiments highlight that the model achieves state-of-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 332, + 320, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 320, + 344 + ], + "score": 1.0, + "content": "the-art performance for targeted image manipulation.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 106, + 299, + 505, + 344 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 349, + 505, + 393 + ], + "lines": [ + { + "bbox": [ + 105, + 348, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 505, + 361 + ], + "score": 1.0, + "content": "Acknowledgements: This work is supported in part by NSF under Grant No. 1718221, 2008387,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 359, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 505, + 372 + ], + "score": 1.0, + "content": "1934986 and MRI #1725729, NIFA award 2020-67021-32799, UIUC, Samsung, Amazon, 3M, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "Cisco Systems Inc. (Gift Award CG 1377144). We thank Cisco for access to the Arcetri cluster. We", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 381, + 234, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 234, + 393 + ], + "score": 1.0, + "content": "thank Amazon for EC2 credits.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 348, + 505, + 393 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 410, + 175, + 423 + ], + "lines": [ + { + "bbox": [ + 106, + 411, + 176, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 176, + 424 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "list", + "bbox": [ + 105, + 429, + 506, + 733 + ], + "lines": [ + { + "bbox": [ + 104, + 428, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 104, + 428, + 505, + 444 + ], + "score": 1.0, + "content": "Rameen Abdal, Yipeng Qin, and Peter Wonka. Image2stylegan: How to embed images into the", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 441, + 271, + 453 + ], + "spans": [ + { + "bbox": [ + 116, + 441, + 271, + 453 + ], + "score": 1.0, + "content": "stylegan latent space? In ICCV, 2019.", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 459, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 104, + 459, + 505, + 475 + ], + "score": 1.0, + "content": "Binod Bhattarai and Tae-Kyun Kim. Inducing optimal attribute representations for conditional gans.", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 472, + 172, + 484 + ], + "spans": [ + { + "bbox": [ + 116, + 472, + 172, + 484 + ], + "score": 1.0, + "content": "ECCV, 2020.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "score": 1.0, + "content": "Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 503, + 238, + 516 + ], + "spans": [ + { + "bbox": [ + 116, + 503, + 238, + 516 + ], + "score": 1.0, + "content": "image synthesis. arXiv, 2018.", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 522, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 506, + 537 + ], + "score": 1.0, + "content": "Q. Cao, L. Shen, W. Xie, et al. Vggface2: A dataset for recognising faces across pose and age. In", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 533, + 184, + 547 + ], + "spans": [ + { + "bbox": [ + 115, + 533, + 184, + 547 + ], + "score": 1.0, + "content": "IEEE FG, 2018.", + "type": "text" + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "score": 1.0, + "content": "Yunjey Choi, Minje Choi, Munyoung Kim, et al. Stargan: Unified generative adversarial networks", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 114, + 564, + 367, + 579 + ], + "spans": [ + { + "bbox": [ + 114, + 564, + 367, + 579 + ], + "score": 1.0, + "content": "for multi-domain image-to-image translation. In CVPR, 2018.", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 506, + 599 + ], + "score": 1.0, + "content": "Yunjey Choi, Youngjung Uh, Jaejun Yoo, et al. Stargan v2: Diverse image synthesis for multiple", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 596, + 211, + 609 + ], + "spans": [ + { + "bbox": [ + 115, + 596, + 211, + 609 + ], + "score": 1.0, + "content": "domains. CVPR, 2020.", + "type": "text" + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 616, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 630 + ], + "score": 1.0, + "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, et al. Generative adversarial nets. In NeurIPS,", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 627, + 142, + 640 + ], + "spans": [ + { + "bbox": [ + 115, + 627, + 142, + 640 + ], + "score": 1.0, + "content": "2014.", + "type": "text" + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 646, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 104, + 646, + 505, + 662 + ], + "score": 1.0, + "content": "Erik Hark ¨ onen, Aaron Hertzmann, Jaakko Lehtinen, et al. Ganspace: Discovering interpretable gan ¨", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 659, + 207, + 672 + ], + "spans": [ + { + "bbox": [ + 116, + 659, + 207, + 672 + ], + "score": 1.0, + "content": "controls. arXiv, 2020.", + "type": "text" + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 677, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 104, + 677, + 505, + 694 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 690, + 211, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 211, + 702 + ], + "score": 1.0, + "content": "nition. In CVPR, 2016.", + "type": "text" + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 709, + 505, + 725 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 505, + 725 + ], + "score": 1.0, + "content": "Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, et al. Image-to-image translation with conditional adver-", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 721, + 249, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 721, + 249, + 733 + ], + "score": 1.0, + "content": "sarial networks. In CVPR, 2017.", + "type": "text" + } + ], + "index": 36, + "is_list_end_line": true + } + ], + "index": 26.5, + "bbox_fs": [ + 104, + 428, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Ali Jahanian, Lucy Chai, and Phillip Isola. On the”steerability” of generative adversarial networks.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 169, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 169, + 105 + ], + "score": 1.0, + "content": "ICLR, 2019.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 112, + 504, + 136 + ], + "lines": [ + { + "bbox": [ + 106, + 113, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 113, + 505, + 126 + ], + "score": 1.0, + "content": "Alexia Jolicoeur-Martineau. The relativistic discriminator: a key element missing from standard", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 124, + 189, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 124, + 189, + 135 + ], + "score": 1.0, + "content": "gan. ICLR, 2019.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 142, + 504, + 166 + ], + "lines": [ + { + "bbox": [ + 105, + 141, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 505, + 157 + ], + "score": 1.0, + "content": "Tero Karras, Timo Aila, Samuli Laine, et al. Progressive growing of gans for improved quality,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 154, + 264, + 166 + ], + "spans": [ + { + "bbox": [ + 116, + 154, + 264, + 166 + ], + "score": 1.0, + "content": "stability, and variation. arXiv, 2017.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 173, + 504, + 197 + ], + "lines": [ + { + "bbox": [ + 106, + 173, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 505, + 186 + ], + "score": 1.0, + "content": "Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 185, + 275, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 185, + 275, + 196 + ], + "score": 1.0, + "content": "adversarial networks. In CVPR, 2019a.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 204, + 504, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 218 + ], + "score": 1.0, + "content": "Tero Karras, Samuli Laine, Miika Aittala, et al. Analyzing and improving the image quality of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 216, + 213, + 227 + ], + "spans": [ + { + "bbox": [ + 115, + 216, + 213, + 227 + ], + "score": 1.0, + "content": "stylegan. arXiv, 2019b.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 105, + 234, + 504, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 235, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 235, + 505, + 248 + ], + "score": 1.0, + "content": "Marek Kowalski, Stephan J Garbin, Virginia Estellers, Tadas Baltrusaitis, Matthew Johnson, and ˇ", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 246, + 439, + 259 + ], + "spans": [ + { + "bbox": [ + 115, + 246, + 439, + 259 + ], + "score": 1.0, + "content": "Jamie Shotton. Config: Controllable neural face image generation. ICML, 2020.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 264, + 504, + 289 + ], + "lines": [ + { + "bbox": [ + 105, + 263, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 279 + ], + "score": 1.0, + "content": "Pierre-Yves Laffont, Zhile Ren, Xiaofeng Tao, et al. Transient attributes for high-level understanding", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 276, + 289, + 288 + ], + "spans": [ + { + "bbox": [ + 116, + 276, + 289, + 288 + ], + "score": 1.0, + "content": "and editing of outdoor scenes. TOG, 2014.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 504, + 318 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 309 + ], + "score": 1.0, + "content": "H.Y. Lee, H.Y. Tseng, Q. Mao, et al. Drit++: Diverse image-to-image translation via disentangled", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 307, + 233, + 318 + ], + "spans": [ + { + "bbox": [ + 115, + 307, + 233, + 318 + ], + "score": 1.0, + "content": "representations. IJCV, 2020.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 107, + 326, + 503, + 349 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 504, + 340 + ], + "score": 1.0, + "content": "Bowen Li, Xiaojuan Qi, Thomas Lukasiewicz, and Others. Controllable text-to-image generation.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 336, + 191, + 348 + ], + "spans": [ + { + "bbox": [ + 115, + 336, + 191, + 348 + ], + "score": 1.0, + "content": "In NeurIPS, 2019.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 105, + 356, + 488, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 356, + 489, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 489, + 369 + ], + "score": 1.0, + "content": "Y. Li, C. Fang, J. Yang, et al. Universal style transfer via feature transforms. In NeurIPS, 2017.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 105, + 376, + 502, + 389 + ], + "lines": [ + { + "bbox": [ + 106, + 376, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 390 + ], + "score": 1.0, + "content": "Ziwei Liu, Ping Luo, Xiaogang Wang, et al. Large-scale celebfaces attributes (celeba) dataset. 2018.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 395, + 475, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 475, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 475, + 409 + ], + "score": 1.0, + "content": "F. Luan, S. Paris, and E. andothers Shechtman. Deep photo style transfer. In PCVPR, 2017.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 415, + 505, + 439 + ], + "lines": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "Taesung Park, Ming-Yu Liu, Ting-Chun Wang, and Jun-Yan Zhu. Semantic image synthesis with", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 427, + 316, + 439 + ], + "spans": [ + { + "bbox": [ + 116, + 427, + 316, + 439 + ], + "score": 1.0, + "content": "spatially-adaptive normalization. In CVPR, 2019.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 445, + 504, + 469 + ], + "lines": [ + { + "bbox": [ + 106, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "Antoine Plumerault, Herve Le Borgne, and C ´ eline Hudelot. Controlling generative models with ´", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 457, + 299, + 468 + ], + "spans": [ + { + "bbox": [ + 116, + 457, + 299, + 468 + ], + "score": 1.0, + "content": "continuous factors of variations. ICLR, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 108, + 476, + 503, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 474, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 505, + 490 + ], + "score": 1.0, + "content": "Olga Russakovsky, Jia Deng, Hao Su, et al. Imagenet large scale visual recognition challenge. ICCV,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 486, + 143, + 500 + ], + "spans": [ + { + "bbox": [ + 115, + 486, + 143, + 500 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 108, + 506, + 495, + 519 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 496, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 496, + 520 + ], + "score": 1.0, + "content": "Yujun Shen and Bolei Zhou. Closed-form factorization of latent semantics in gans. CVPR, 2021.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 526, + 505, + 550 + ], + "lines": [ + { + "bbox": [ + 106, + 525, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 506, + 540 + ], + "score": 1.0, + "content": "Yujun Shen, Jinjin Gu, Xiaoou Tang, et al. Interpreting the latent space of gans for semantic face", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 537, + 176, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 537, + 176, + 550 + ], + "score": 1.0, + "content": "editing. 2019.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 105, + 556, + 505, + 580 + ], + "lines": [ + { + "bbox": [ + 105, + 555, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 571 + ], + "score": 1.0, + "content": "Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 568, + 221, + 580 + ], + "spans": [ + { + "bbox": [ + 115, + 568, + 221, + 580 + ], + "score": 1.0, + "content": "recognition. arXiv, 2014.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 105, + 587, + 505, + 611 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "score": 1.0, + "content": "Yuri Viazovetskyi, Vladimir Ivashkin, and Evgeny Kashin. Stylegan2 distillation for feed-forward", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 599, + 253, + 610 + ], + "spans": [ + { + "bbox": [ + 116, + 599, + 253, + 610 + ], + "score": 1.0, + "content": "image manipulation. arXiv, 2020.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 106, + 617, + 504, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 616, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 506, + 632 + ], + "score": 1.0, + "content": "Andrey Voynov and Artem Babenko. Unsupervised discovery of interpretable directions in the gan", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 628, + 223, + 642 + ], + "spans": [ + { + "bbox": [ + 115, + 628, + 223, + 642 + ], + "score": 1.0, + "content": "latent space. ICML, 2020.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 105, + 648, + 504, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 647, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 505, + 662 + ], + "score": 1.0, + "content": "Sheng-Yu Wang, Oliver Wang, Richard Zhang, et al. Cnn-generated images are surprisingly easy to", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 660, + 238, + 671 + ], + "spans": [ + { + "bbox": [ + 116, + 660, + 238, + 671 + ], + "score": 1.0, + "content": "spot... for now. CVPR, 2019a.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 678, + 503, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "Sheng-Yu Wang, Oliver Wang, et al. Detecting photoshopped faces by scripting photoshop. In", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 690, + 174, + 702 + ], + "spans": [ + { + "bbox": [ + 115, + 690, + 174, + 702 + ], + "score": 1.0, + "content": "ICCV, 2019b.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, et al. High-resolution image synthesis and semantic", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 720, + 329, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 329, + 732 + ], + "score": 1.0, + "content": "manipulation with conditional gans. In CVPR, 2018.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Ali Jahanian, Lucy Chai, and Phillip Isola. On the”steerability” of generative adversarial networks.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 169, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 169, + 105 + ], + "score": 1.0, + "content": "ICLR, 2019.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 112, + 504, + 136 + ], + "lines": [ + { + "bbox": [ + 106, + 113, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 113, + 505, + 126 + ], + "score": 1.0, + "content": "Alexia Jolicoeur-Martineau. The relativistic discriminator: a key element missing from standard", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 124, + 189, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 124, + 189, + 135 + ], + "score": 1.0, + "content": "gan. ICLR, 2019.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 106, + 113, + 505, + 135 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 142, + 504, + 166 + ], + "lines": [ + { + "bbox": [ + 105, + 141, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 505, + 157 + ], + "score": 1.0, + "content": "Tero Karras, Timo Aila, Samuli Laine, et al. Progressive growing of gans for improved quality,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 154, + 264, + 166 + ], + "spans": [ + { + "bbox": [ + 116, + 154, + 264, + 166 + ], + "score": 1.0, + "content": "stability, and variation. arXiv, 2017.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 141, + 505, + 166 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 173, + 504, + 197 + ], + "lines": [ + { + "bbox": [ + 106, + 173, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 505, + 186 + ], + "score": 1.0, + "content": "Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 185, + 275, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 185, + 275, + 196 + ], + "score": 1.0, + "content": "adversarial networks. In CVPR, 2019a.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 173, + 505, + 196 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 204, + 504, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 218 + ], + "score": 1.0, + "content": "Tero Karras, Samuli Laine, Miika Aittala, et al. Analyzing and improving the image quality of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 216, + 213, + 227 + ], + "spans": [ + { + "bbox": [ + 115, + 216, + 213, + 227 + ], + "score": 1.0, + "content": "stylegan. arXiv, 2019b.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 203, + 505, + 227 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 234, + 504, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 235, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 235, + 505, + 248 + ], + "score": 1.0, + "content": "Marek Kowalski, Stephan J Garbin, Virginia Estellers, Tadas Baltrusaitis, Matthew Johnson, and ˇ", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 246, + 439, + 259 + ], + "spans": [ + { + "bbox": [ + 115, + 246, + 439, + 259 + ], + "score": 1.0, + "content": "Jamie Shotton. Config: Controllable neural face image generation. ICML, 2020.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 106, + 235, + 505, + 259 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 264, + 504, + 289 + ], + "lines": [ + { + "bbox": [ + 105, + 263, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 279 + ], + "score": 1.0, + "content": "Pierre-Yves Laffont, Zhile Ren, Xiaofeng Tao, et al. Transient attributes for high-level understanding", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 276, + 289, + 288 + ], + "spans": [ + { + "bbox": [ + 116, + 276, + 289, + 288 + ], + "score": 1.0, + "content": "and editing of outdoor scenes. TOG, 2014.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 263, + 505, + 288 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 504, + 318 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 309 + ], + "score": 1.0, + "content": "H.Y. Lee, H.Y. Tseng, Q. Mao, et al. Drit++: Diverse image-to-image translation via disentangled", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 307, + 233, + 318 + ], + "spans": [ + { + "bbox": [ + 115, + 307, + 233, + 318 + ], + "score": 1.0, + "content": "representations. IJCV, 2020.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 294, + 505, + 318 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 326, + 503, + 349 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 504, + 340 + ], + "score": 1.0, + "content": "Bowen Li, Xiaojuan Qi, Thomas Lukasiewicz, and Others. Controllable text-to-image generation.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 336, + 191, + 348 + ], + "spans": [ + { + "bbox": [ + 115, + 336, + 191, + 348 + ], + "score": 1.0, + "content": "In NeurIPS, 2019.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 325, + 504, + 348 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 356, + 488, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 356, + 489, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 489, + 369 + ], + "score": 1.0, + "content": "Y. Li, C. Fang, J. Yang, et al. Universal style transfer via feature transforms. In NeurIPS, 2017.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 356, + 489, + 369 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 376, + 502, + 389 + ], + "lines": [ + { + "bbox": [ + 106, + 376, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 390 + ], + "score": 1.0, + "content": "Ziwei Liu, Ping Luo, Xiaogang Wang, et al. Large-scale celebfaces attributes (celeba) dataset. 2018.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 106, + 376, + 505, + 390 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 395, + 475, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 475, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 475, + 409 + ], + "score": 1.0, + "content": "F. Luan, S. Paris, and E. andothers Shechtman. Deep photo style transfer. In PCVPR, 2017.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 395, + 475, + 409 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 415, + 505, + 439 + ], + "lines": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "Taesung Park, Ming-Yu Liu, Ting-Chun Wang, and Jun-Yan Zhu. Semantic image synthesis with", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 427, + 316, + 439 + ], + "spans": [ + { + "bbox": [ + 116, + 427, + 316, + 439 + ], + "score": 1.0, + "content": "spatially-adaptive normalization. In CVPR, 2019.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 106, + 415, + 505, + 439 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 445, + 504, + 469 + ], + "lines": [ + { + "bbox": [ + 106, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "Antoine Plumerault, Herve Le Borgne, and C ´ eline Hudelot. Controlling generative models with ´", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 457, + 299, + 468 + ], + "spans": [ + { + "bbox": [ + 116, + 457, + 299, + 468 + ], + "score": 1.0, + "content": "continuous factors of variations. ICLR, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 446, + 505, + 468 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 476, + 503, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 474, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 505, + 490 + ], + "score": 1.0, + "content": "Olga Russakovsky, Jia Deng, Hao Su, et al. Imagenet large scale visual recognition challenge. ICCV,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 486, + 143, + 500 + ], + "spans": [ + { + "bbox": [ + 115, + 486, + 143, + 500 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 474, + 505, + 500 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 506, + 495, + 519 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 496, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 496, + 520 + ], + "score": 1.0, + "content": "Yujun Shen and Bolei Zhou. Closed-form factorization of latent semantics in gans. CVPR, 2021.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 506, + 496, + 520 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 526, + 505, + 550 + ], + "lines": [ + { + "bbox": [ + 106, + 525, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 506, + 540 + ], + "score": 1.0, + "content": "Yujun Shen, Jinjin Gu, Xiaoou Tang, et al. Interpreting the latent space of gans for semantic face", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 537, + 176, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 537, + 176, + 550 + ], + "score": 1.0, + "content": "editing. 2019.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 106, + 525, + 506, + 550 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 556, + 505, + 580 + ], + "lines": [ + { + "bbox": [ + 105, + 555, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 571 + ], + "score": 1.0, + "content": "Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 568, + 221, + 580 + ], + "spans": [ + { + "bbox": [ + 115, + 568, + 221, + 580 + ], + "score": 1.0, + "content": "recognition. arXiv, 2014.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 555, + 505, + 580 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 587, + 505, + 611 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "score": 1.0, + "content": "Yuri Viazovetskyi, Vladimir Ivashkin, and Evgeny Kashin. Stylegan2 distillation for feed-forward", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 599, + 253, + 610 + ], + "spans": [ + { + "bbox": [ + 116, + 599, + 253, + 610 + ], + "score": 1.0, + "content": "image manipulation. arXiv, 2020.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 106, + 587, + 505, + 610 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 617, + 504, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 616, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 506, + 632 + ], + "score": 1.0, + "content": "Andrey Voynov and Artem Babenko. Unsupervised discovery of interpretable directions in the gan", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 628, + 223, + 642 + ], + "spans": [ + { + "bbox": [ + 115, + 628, + 223, + 642 + ], + "score": 1.0, + "content": "latent space. ICML, 2020.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 616, + 506, + 642 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 648, + 504, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 647, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 505, + 662 + ], + "score": 1.0, + "content": "Sheng-Yu Wang, Oliver Wang, Richard Zhang, et al. Cnn-generated images are surprisingly easy to", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 660, + 238, + 671 + ], + "spans": [ + { + "bbox": [ + 116, + 660, + 238, + 671 + ], + "score": 1.0, + "content": "spot... for now. CVPR, 2019a.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 647, + 505, + 671 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 678, + 503, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "Sheng-Yu Wang, Oliver Wang, et al. Detecting photoshopped faces by scripting photoshop. In", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 690, + 174, + 702 + ], + "spans": [ + { + "bbox": [ + 115, + 690, + 174, + 702 + ], + "score": 1.0, + "content": "ICCV, 2019b.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 677, + 505, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, et al. High-resolution image synthesis and semantic", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 720, + 329, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 329, + 732 + ], + "score": 1.0, + "content": "manipulation with conditional gans. In CVPR, 2018.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 106, + 709, + 504, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Po-Wei Wu, Yu-Jing Lin, Che-Han Chang, et al. Relgan: Multi-domain image-to-image translation", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 117, + 93, + 269, + 105 + ], + "spans": [ + { + "bbox": [ + 117, + 93, + 269, + 105 + ], + "score": 1.0, + "content": "via relative attributes. In ICCV, 2019.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 105, + 111, + 504, + 135 + ], + "lines": [ + { + "bbox": [ + 106, + 112, + 504, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 504, + 123 + ], + "score": 1.0, + "content": "Han Zhang, Ian Goodfellow, Dimitris Metaxas, et al. Self-attention generative adversarial networks.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 173, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 173, + 135 + ], + "score": 1.0, + "content": "PMLR, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 141, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 506, + 156 + ], + "score": 1.0, + "content": "Bolei Zhou, Agata Lapedriza, Aditya Khosla, et al. Places: A 10 million image database for scene", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 154, + 226, + 165 + ], + "spans": [ + { + "bbox": [ + 116, + 154, + 226, + 165 + ], + "score": 1.0, + "content": "recognition. TPAMI, 2017.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 104, + 172, + 504, + 195 + ], + "lines": [ + { + "bbox": [ + 105, + 170, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 505, + 186 + ], + "score": 1.0, + "content": "Jun-Yan Zhu, Taesung Park, Phillip Isola, et al. Unpaired image-to-image translation using cycle-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 316, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 316, + 196 + ], + "score": 1.0, + "content": "consistent adversarial networks. In ICCV, 2017a.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 104, + 201, + 504, + 225 + ], + "lines": [ + { + "bbox": [ + 104, + 200, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 104, + 200, + 505, + 216 + ], + "score": 1.0, + "content": "Jun-Yan Zhu, Richard Zhang, Deepak Pathak, et al. Toward multimodal image-to-image translation.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 213, + 196, + 224 + ], + "spans": [ + { + "bbox": [ + 115, + 213, + 196, + 224 + ], + "score": 1.0, + "content": "In NeurIPS, 2017b.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Po-Wei Wu, Yu-Jing Lin, Che-Han Chang, et al. Relgan: Multi-domain image-to-image translation", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 117, + 93, + 269, + 105 + ], + "spans": [ + { + "bbox": [ + 117, + 93, + 269, + 105 + ], + "score": 1.0, + "content": "via relative attributes. In ICCV, 2019.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 111, + 504, + 135 + ], + "lines": [ + { + "bbox": [ + 106, + 112, + 504, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 504, + 123 + ], + "score": 1.0, + "content": "Han Zhang, Ian Goodfellow, Dimitris Metaxas, et al. Self-attention generative adversarial networks.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 173, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 173, + 135 + ], + "score": 1.0, + "content": "PMLR, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 106, + 112, + 504, + 135 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 141, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 506, + 156 + ], + "score": 1.0, + "content": "Bolei Zhou, Agata Lapedriza, Aditya Khosla, et al. Places: A 10 million image database for scene", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 154, + 226, + 165 + ], + "spans": [ + { + "bbox": [ + 116, + 154, + 226, + 165 + ], + "score": 1.0, + "content": "recognition. TPAMI, 2017.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 140, + 506, + 165 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 172, + 504, + 195 + ], + "lines": [ + { + "bbox": [ + 105, + 170, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 505, + 186 + ], + "score": 1.0, + "content": "Jun-Yan Zhu, Taesung Park, Phillip Isola, et al. Unpaired image-to-image translation using cycle-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 183, + 316, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 316, + 196 + ], + "score": 1.0, + "content": "consistent adversarial networks. In ICCV, 2017a.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 170, + 505, + 196 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 201, + 504, + 225 + ], + "lines": [ + { + "bbox": [ + 104, + 200, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 104, + 200, + 505, + 216 + ], + "score": 1.0, + "content": "Jun-Yan Zhu, Richard Zhang, Deepak Pathak, et al. Toward multimodal image-to-image translation.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 213, + 196, + 224 + ], + "spans": [ + { + "bbox": [ + 115, + 213, + 196, + 224 + ], + "score": 1.0, + "content": "In NeurIPS, 2017b.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 104, + 200, + 505, + 224 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 172, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 173, + 99 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 173, + 99 + ], + "score": 1.0, + "content": "APPENDIX", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 105, + 200, + 117 + ], + "lines": [ + { + "bbox": [ + 105, + 104, + 202, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 202, + 120 + ], + "score": 1.0, + "content": "A. ALGORITHM", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 108, + 126, + 505, + 148 + ], + "lines": [ + { + "bbox": [ + 106, + 125, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 471, + 139 + ], + "score": 1.0, + "content": "The overall procedure of the proposed method in Algorithm 1, taking local transformation", + "type": "text" + }, + { + "bbox": [ + 471, + 127, + 481, + 136 + ], + "score": 0.78, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 125, + 505, + 139 + ], + "score": 1.0, + "content": "as an", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 136, + 146, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 146, + 151 + ], + "score": 1.0, + "content": "example.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 159, + 240, + 171 + ], + "lines": [ + { + "bbox": [ + 106, + 158, + 240, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 158, + 240, + 173 + ], + "score": 1.0, + "content": "Algorithm 1 Training Procedure", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 108, + 173, + 505, + 294 + ], + "lines": [ + { + "bbox": [ + 106, + 172, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 240, + 185 + ], + "score": 1.0, + "content": "Input: A pre-trained GAN with", + "type": "text" + }, + { + "bbox": [ + 240, + 173, + 248, + 183 + ], + "score": 0.58, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 172, + 252, + 185 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 253, + 173, + 262, + 182 + ], + "score": 0.6, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 172, + 379, + 185 + ], + "score": 1.0, + "content": ", and input noise distribution", + "type": "text" + }, + { + "bbox": [ + 379, + 173, + 408, + 183 + ], + "score": 0.9, + "content": "z \\sim \\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 172, + 505, + 185 + ], + "score": 1.0, + "content": "; a pre-trained regressor", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 124, + 183, + 431, + 196 + ], + "spans": [ + { + "bbox": [ + 124, + 184, + 133, + 194 + ], + "score": 0.74, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 183, + 207, + 196 + ], + "score": 1.0, + "content": "for predictions on", + "type": "text" + }, + { + "bbox": [ + 207, + 184, + 216, + 194 + ], + "score": 0.43, + "content": "_ \\mathrm { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 183, + 313, + 196 + ], + "score": 1.0, + "content": "attributes; an initialized", + "type": "text" + }, + { + "bbox": [ + 313, + 184, + 322, + 194 + ], + "score": 0.8, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 183, + 414, + 196 + ], + "score": 1.0, + "content": "; max iteration number", + "type": "text" + }, + { + "bbox": [ + 415, + 184, + 426, + 194 + ], + "score": 0.83, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 183, + 431, + 196 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 110, + 194, + 250, + 207 + ], + "spans": [ + { + "bbox": [ + 110, + 194, + 174, + 207 + ], + "score": 1.0, + "content": "1: for iteration", + "type": "text" + }, + { + "bbox": [ + 174, + 195, + 235, + 206 + ], + "score": 0.91, + "content": "m = 1 , \\ldots , M", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 194, + 250, + 207 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 111, + 205, + 283, + 218 + ], + "spans": [ + { + "bbox": [ + 111, + 205, + 121, + 218 + ], + "score": 1.0, + "content": "2:", + "type": "text" + }, + { + "bbox": [ + 137, + 205, + 227, + 218 + ], + "score": 1.0, + "content": "Sample random noise", + "type": "text" + }, + { + "bbox": [ + 227, + 206, + 255, + 216 + ], + "score": 0.89, + "content": "z \\sim \\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 205, + 276, + 218 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 276, + 208, + 283, + 216 + ], + "score": 0.69, + "content": "\\varepsilon", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 110, + 215, + 414, + 230 + ], + "spans": [ + { + "bbox": [ + 110, + 217, + 121, + 230 + ], + "score": 1.0, + "content": "3:", + "type": "text" + }, + { + "bbox": [ + 137, + 215, + 356, + 229 + ], + "score": 1.0, + "content": "Compute internal attributes for a synthetic image, i.e.,", + "type": "text" + }, + { + "bbox": [ + 356, + 216, + 414, + 228 + ], + "score": 0.92, + "content": "{ \\pmb \\alpha } = R ( G ( { \\pmb z } ) )", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 110, + 227, + 389, + 241 + ], + "spans": [ + { + "bbox": [ + 110, + 227, + 121, + 239 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 137, + 227, + 262, + 241 + ], + "score": 1.0, + "content": "Compute the actual shift value", + "type": "text" + }, + { + "bbox": [ + 263, + 227, + 389, + 239 + ], + "score": 0.91, + "content": "\\pmb { \\delta } = C L I P ( \\pmb { \\alpha } + \\pmb { \\varepsilon } , ( 0 , 1 ) ) - \\pmb { \\alpha }", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 110, + 237, + 363, + 251 + ], + "spans": [ + { + "bbox": [ + 110, + 239, + 121, + 250 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 137, + 237, + 294, + 251 + ], + "score": 1.0, + "content": "Compute the transformed latent vector", + "type": "text" + }, + { + "bbox": [ + 295, + 239, + 363, + 250 + ], + "score": 0.88, + "content": "z ^ { \\prime } = z + { \\cal T } ( z ) \\delta", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 110, + 249, + 279, + 262 + ], + "spans": [ + { + "bbox": [ + 110, + 250, + 121, + 261 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 137, + 249, + 177, + 261 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 177, + 249, + 224, + 262 + ], + "score": 0.89, + "content": "I ^ { \\prime } = G ( z ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 249, + 227, + 261 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 228, + 249, + 279, + 260 + ], + "score": 0.83, + "content": "{ \\pmb { \\alpha } } ^ { \\prime } = { \\pmb { \\alpha } } + { \\pmb { \\delta } }", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 110, + 259, + 308, + 274 + ], + "spans": [ + { + "bbox": [ + 110, + 261, + 121, + 272 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 137, + 259, + 260, + 274 + ], + "score": 1.0, + "content": "Compute attribute predictions", + "type": "text" + }, + { + "bbox": [ + 260, + 261, + 308, + 272 + ], + "score": 0.91, + "content": "\\hat { \\pmb { \\alpha } } ^ { \\prime } = { \\cal R } ( { \\cal I ^ { \\prime } } )", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 110, + 271, + 218, + 283 + ], + "spans": [ + { + "bbox": [ + 110, + 271, + 121, + 283 + ], + "score": 1.0, + "content": "8:", + "type": "text" + }, + { + "bbox": [ + 138, + 272, + 210, + 283 + ], + "score": 1.0, + "content": "Compute the loss", + "type": "text" + }, + { + "bbox": [ + 210, + 272, + 218, + 281 + ], + "score": 0.74, + "content": "\\mathcal { L }", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 282, + 179, + 295 + ], + "spans": [ + { + "bbox": [ + 110, + 282, + 121, + 294 + ], + "score": 1.0, + "content": "9:", + "type": "text" + }, + { + "bbox": [ + 137, + 282, + 169, + 295 + ], + "score": 1.0, + "content": "Update", + "type": "text" + }, + { + "bbox": [ + 170, + 282, + 179, + 292 + ], + "score": 0.57, + "content": "\\mathbf { T }", + "type": "inline_equation" + } + ], + "index": 15 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 294, + 155, + 316 + ], + "lines": [ + { + "bbox": [ + 106, + 292, + 157, + 305 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 157, + 305 + ], + "score": 1.0, + "content": "10: end for", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 303, + 156, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 156, + 316 + ], + "score": 1.0, + "content": "Return: T", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 108, + 340, + 227, + 353 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 227, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 227, + 356 + ], + "score": 1.0, + "content": "B. METHOD DETAILS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 107, + 361, + 247, + 373 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 248, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 248, + 376 + ], + "score": 1.0, + "content": "B.1. METHOD ON PGGAN", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 380, + 504, + 403 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 394 + ], + "score": 1.0, + "content": "We apply a multi-layer perceptron (MLP) to parameterize a latent space path on PGGAN. Fig. 9", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 392, + 270, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 270, + 405 + ], + "score": 1.0, + "content": "shows the MLP architecture that we use.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "image", + "bbox": [ + 252, + 437, + 350, + 619 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 252, + 437, + 350, + 619 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 252, + 437, + 350, + 619 + ], + "spans": [ + { + "bbox": [ + 252, + 437, + 350, + 619 + ], + "score": 0.934, + "type": "image", + "image_path": "b73e22acdc9562bd91a2933f94ca6e6ff199eae5f9734ff99eb3251d59132010.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 252, + 437, + 350, + 619 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 199, + 639, + 411, + 651 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 199, + 638, + 412, + 654 + ], + "spans": [ + { + "bbox": [ + 199, + 638, + 412, + 654 + ], + "score": 1.0, + "content": "Figure 9: Example MLP architecture on PGGAN.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 668, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 341, + 682 + ], + "score": 1.0, + "content": "In practice, we find that normalizing the direction vector", + "type": "text" + }, + { + "bbox": [ + 342, + 669, + 352, + 680 + ], + "score": 0.88, + "content": "\\mathbf { \\ b { d } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 667, + 506, + 682 + ], + "score": 1.0, + "content": "helps preserve image identity during", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 680, + 442, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 442, + 692 + ], + "score": 1.0, + "content": "manipulation on PGGAN. Formally, the transformed latent vector can be written as", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "interline_equation", + "bbox": [ + 267, + 693, + 344, + 719 + ], + "lines": [ + { + "bbox": [ + 267, + 693, + 344, + 719 + ], + "spans": [ + { + "bbox": [ + 267, + 693, + 344, + 719 + ], + "score": 0.94, + "content": "z ^ { \\prime } = z + \\lambda \\varepsilon { \\frac { d _ { i } } { \\parallel d _ { i } \\parallel } } ,", + "type": "interline_equation", + "image_path": "8846eaed78b1a1bc3282a3120eb7d291f1baebaea53810bfb7dc5a4e9a836f1f.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 267, + 693, + 344, + 719 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 362, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 719, + 362, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 133, + 733 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 721, + 140, + 730 + ], + "score": 0.8, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 719, + 333, + 733 + ], + "score": 1.0, + "content": "is a weight adjusting the direction scale. We set", + "type": "text" + }, + { + "bbox": [ + 334, + 721, + 359, + 731 + ], + "score": 0.88, + "content": "\\lambda = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 719, + 362, + 733 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 172, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 173, + 99 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 173, + 99 + ], + "score": 1.0, + "content": "APPENDIX", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 105, + 200, + 117 + ], + "lines": [ + { + "bbox": [ + 105, + 104, + 202, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 202, + 120 + ], + "score": 1.0, + "content": "A. ALGORITHM", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 104, + 202, + 120 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 126, + 505, + 148 + ], + "lines": [ + { + "bbox": [ + 106, + 125, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 471, + 139 + ], + "score": 1.0, + "content": "The overall procedure of the proposed method in Algorithm 1, taking local transformation", + "type": "text" + }, + { + "bbox": [ + 471, + 127, + 481, + 136 + ], + "score": 0.78, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 125, + 505, + 139 + ], + "score": 1.0, + "content": "as an", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 136, + 146, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 146, + 151 + ], + "score": 1.0, + "content": "example.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 125, + 505, + 151 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 159, + 240, + 171 + ], + "lines": [ + { + "bbox": [ + 106, + 158, + 240, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 158, + 240, + 173 + ], + "score": 1.0, + "content": "Algorithm 1 Training Procedure", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "index", + "bbox": [ + 108, + 173, + 505, + 294 + ], + "lines": [ + { + "bbox": [ + 106, + 172, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 240, + 185 + ], + "score": 1.0, + "content": "Input: A pre-trained GAN with", + "type": "text" + }, + { + "bbox": [ + 240, + 173, + 248, + 183 + ], + "score": 0.58, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 172, + 252, + 185 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 253, + 173, + 262, + 182 + ], + "score": 0.6, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 172, + 379, + 185 + ], + "score": 1.0, + "content": ", and input noise distribution", + "type": "text" + }, + { + "bbox": [ + 379, + 173, + 408, + 183 + ], + "score": 0.9, + "content": "z \\sim \\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 172, + 505, + 185 + ], + "score": 1.0, + "content": "; a pre-trained regressor", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 183, + 431, + 196 + ], + "spans": [ + { + "bbox": [ + 124, + 184, + 133, + 194 + ], + "score": 0.74, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 183, + 207, + 196 + ], + "score": 1.0, + "content": "for predictions on", + "type": "text" + }, + { + "bbox": [ + 207, + 184, + 216, + 194 + ], + "score": 0.43, + "content": "_ \\mathrm { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 183, + 313, + 196 + ], + "score": 1.0, + "content": "attributes; an initialized", + "type": "text" + }, + { + "bbox": [ + 313, + 184, + 322, + 194 + ], + "score": 0.8, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 183, + 414, + 196 + ], + "score": 1.0, + "content": "; max iteration number", + "type": "text" + }, + { + "bbox": [ + 415, + 184, + 426, + 194 + ], + "score": 0.83, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 183, + 431, + 196 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 194, + 250, + 207 + ], + "spans": [ + { + "bbox": [ + 110, + 194, + 174, + 207 + ], + "score": 1.0, + "content": "1: for iteration", + "type": "text" + }, + { + "bbox": [ + 174, + 195, + 235, + 206 + ], + "score": 0.91, + "content": "m = 1 , \\ldots , M", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 194, + 250, + 207 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 205, + 283, + 218 + ], + "spans": [ + { + "bbox": [ + 111, + 205, + 121, + 218 + ], + "score": 1.0, + "content": "2:", + "type": "text" + }, + { + "bbox": [ + 137, + 205, + 227, + 218 + ], + "score": 1.0, + "content": "Sample random noise", + "type": "text" + }, + { + "bbox": [ + 227, + 206, + 255, + 216 + ], + "score": 0.89, + "content": "z \\sim \\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 205, + 276, + 218 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 276, + 208, + 283, + 216 + ], + "score": 0.69, + "content": "\\varepsilon", + "type": "inline_equation" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 215, + 414, + 230 + ], + "spans": [ + { + "bbox": [ + 110, + 217, + 121, + 230 + ], + "score": 1.0, + "content": "3:", + "type": "text" + }, + { + "bbox": [ + 137, + 215, + 356, + 229 + ], + "score": 1.0, + "content": "Compute internal attributes for a synthetic image, i.e.,", + "type": "text" + }, + { + "bbox": [ + 356, + 216, + 414, + 228 + ], + "score": 0.92, + "content": "{ \\pmb \\alpha } = R ( G ( { \\pmb z } ) )", + "type": "inline_equation" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 227, + 389, + 241 + ], + "spans": [ + { + "bbox": [ + 110, + 227, + 121, + 239 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 137, + 227, + 262, + 241 + ], + "score": 1.0, + "content": "Compute the actual shift value", + "type": "text" + }, + { + "bbox": [ + 263, + 227, + 389, + 239 + ], + "score": 0.91, + "content": "\\pmb { \\delta } = C L I P ( \\pmb { \\alpha } + \\pmb { \\varepsilon } , ( 0 , 1 ) ) - \\pmb { \\alpha }", + "type": "inline_equation" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 237, + 363, + 251 + ], + "spans": [ + { + "bbox": [ + 110, + 239, + 121, + 250 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 137, + 237, + 294, + 251 + ], + "score": 1.0, + "content": "Compute the transformed latent vector", + "type": "text" + }, + { + "bbox": [ + 295, + 239, + 363, + 250 + ], + "score": 0.88, + "content": "z ^ { \\prime } = z + { \\cal T } ( z ) \\delta", + "type": "inline_equation" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 249, + 279, + 262 + ], + "spans": [ + { + "bbox": [ + 110, + 250, + 121, + 261 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 137, + 249, + 177, + 261 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 177, + 249, + 224, + 262 + ], + "score": 0.89, + "content": "I ^ { \\prime } = G ( z ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 249, + 227, + 261 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 228, + 249, + 279, + 260 + ], + "score": 0.83, + "content": "{ \\pmb { \\alpha } } ^ { \\prime } = { \\pmb { \\alpha } } + { \\pmb { \\delta } }", + "type": "inline_equation" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 259, + 308, + 274 + ], + "spans": [ + { + "bbox": [ + 110, + 261, + 121, + 272 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 137, + 259, + 260, + 274 + ], + "score": 1.0, + "content": "Compute attribute predictions", + "type": "text" + }, + { + "bbox": [ + 260, + 261, + 308, + 272 + ], + "score": 0.91, + "content": "\\hat { \\pmb { \\alpha } } ^ { \\prime } = { \\cal R } ( { \\cal I ^ { \\prime } } )", + "type": "inline_equation" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 271, + 218, + 283 + ], + "spans": [ + { + "bbox": [ + 110, + 271, + 121, + 283 + ], + "score": 1.0, + "content": "8:", + "type": "text" + }, + { + "bbox": [ + 138, + 272, + 210, + 283 + ], + "score": 1.0, + "content": "Compute the loss", + "type": "text" + }, + { + "bbox": [ + 210, + 272, + 218, + 281 + ], + "score": 0.74, + "content": "\\mathcal { L }", + "type": "inline_equation" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 282, + 179, + 295 + ], + "spans": [ + { + "bbox": [ + 110, + 282, + 121, + 294 + ], + "score": 1.0, + "content": "9:", + "type": "text" + }, + { + "bbox": [ + 137, + 282, + 169, + 295 + ], + "score": 1.0, + "content": "Update", + "type": "text" + }, + { + "bbox": [ + 170, + 282, + 179, + 292 + ], + "score": 0.57, + "content": "\\mathbf { T }", + "type": "inline_equation" + } + ], + "index": 15, + "is_list_start_line": true + } + ], + "index": 10, + "bbox_fs": [ + 106, + 172, + 505, + 295 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 294, + 155, + 316 + ], + "lines": [ + { + "bbox": [ + 106, + 292, + 157, + 305 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 157, + 305 + ], + "score": 1.0, + "content": "10: end for", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 303, + 156, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 156, + 316 + ], + "score": 1.0, + "content": "Return: T", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 292, + 157, + 316 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 340, + 227, + 353 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 227, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 227, + 356 + ], + "score": 1.0, + "content": "B. METHOD DETAILS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 107, + 361, + 247, + 373 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 248, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 248, + 376 + ], + "score": 1.0, + "content": "B.1. METHOD ON PGGAN", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 380, + 504, + 403 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 394 + ], + "score": 1.0, + "content": "We apply a multi-layer perceptron (MLP) to parameterize a latent space path on PGGAN. Fig. 9", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 392, + 270, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 270, + 405 + ], + "score": 1.0, + "content": "shows the MLP architecture that we use.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 106, + 380, + 505, + 405 + ] + }, + { + "type": "image", + "bbox": [ + 252, + 437, + 350, + 619 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 252, + 437, + 350, + 619 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 252, + 437, + 350, + 619 + ], + "spans": [ + { + "bbox": [ + 252, + 437, + 350, + 619 + ], + "score": 0.934, + "type": "image", + "image_path": "b73e22acdc9562bd91a2933f94ca6e6ff199eae5f9734ff99eb3251d59132010.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 252, + 437, + 350, + 619 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 199, + 639, + 411, + 651 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 199, + 638, + 412, + 654 + ], + "spans": [ + { + "bbox": [ + 199, + 638, + 412, + 654 + ], + "score": 1.0, + "content": "Figure 9: Example MLP architecture on PGGAN.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 668, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 341, + 682 + ], + "score": 1.0, + "content": "In practice, we find that normalizing the direction vector", + "type": "text" + }, + { + "bbox": [ + 342, + 669, + 352, + 680 + ], + "score": 0.88, + "content": "\\mathbf { \\ b { d } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 667, + 506, + 682 + ], + "score": 1.0, + "content": "helps preserve image identity during", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 680, + 442, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 442, + 692 + ], + "score": 1.0, + "content": "manipulation on PGGAN. Formally, the transformed latent vector can be written as", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 667, + 506, + 692 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 267, + 693, + 344, + 719 + ], + "lines": [ + { + "bbox": [ + 267, + 693, + 344, + 719 + ], + "spans": [ + { + "bbox": [ + 267, + 693, + 344, + 719 + ], + "score": 0.94, + "content": "z ^ { \\prime } = z + \\lambda \\varepsilon { \\frac { d _ { i } } { \\parallel d _ { i } \\parallel } } ,", + "type": "interline_equation", + "image_path": "8846eaed78b1a1bc3282a3120eb7d291f1baebaea53810bfb7dc5a4e9a836f1f.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 267, + 693, + 344, + 719 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 362, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 719, + 362, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 133, + 733 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 721, + 140, + 730 + ], + "score": 0.8, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 719, + 333, + 733 + ], + "score": 1.0, + "content": "is a weight adjusting the direction scale. We set", + "type": "text" + }, + { + "bbox": [ + 334, + 721, + 359, + 731 + ], + "score": 0.88, + "content": "\\lambda = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 719, + 362, + 733 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 719, + 362, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 265, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 266, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 120, + 96 + ], + "score": 1.0, + "content": "C.", + "type": "text" + }, + { + "bbox": [ + 135, + 82, + 266, + 96 + ], + "score": 1.0, + "content": "EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 102, + 240, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 239, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 218, + 118 + ], + "score": 1.0, + "content": "C.1 THE EXPECTATION IN", + "type": "text" + }, + { + "bbox": [ + 218, + 104, + 239, + 114 + ], + "score": 0.83, + "content": "\\mathcal { L } _ { \\mathrm { { R E G } } }", + "type": "inline_equation" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 108, + 123, + 254, + 135 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 255, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 225, + 138 + ], + "score": 1.0, + "content": "We define the regression loss", + "type": "text" + }, + { + "bbox": [ + 225, + 124, + 242, + 136 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 121, + 255, + 138 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "interline_equation", + "bbox": [ + 183, + 137, + 425, + 151 + ], + "lines": [ + { + "bbox": [ + 183, + 137, + 425, + 151 + ], + "spans": [ + { + "bbox": [ + 183, + 137, + 425, + 151 + ], + "score": 0.85, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { \\mathrm { r e g } } = \\ } & { { } \\mathbb { E } _ { z \\sim \\mathcal { Z } , \\varepsilon \\sim \\mathcal { D } _ { \\varepsilon } } [ - \\hat { \\alpha } ^ { \\prime } \\log \\alpha ^ { \\prime } - ( { \\bf 1 } - \\hat { \\alpha } ^ { \\prime } ) \\log \\left( { \\bf 1 } - \\alpha ^ { \\prime } \\right) ] , } \\end{array}", + "type": "interline_equation", + "image_path": "7d290e5bf22b3369bd3b12d080a587bb5027cd3440c4c919326b9ad0a137e90a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 183, + 137, + 425, + 151 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 109, + 151, + 504, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 133, + 162 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 150, + 145, + 160 + ], + "score": 0.87, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 149, + 297, + 162 + ], + "score": 1.0, + "content": "is from the distribution generated by", + "type": "text" + }, + { + "bbox": [ + 297, + 152, + 304, + 160 + ], + "score": 0.77, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 149, + 322, + 162 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 322, + 152, + 329, + 160 + ], + "score": 0.77, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 149, + 505, + 162 + ], + "score": 1.0, + "content": ". We now discuss how we derive the distri-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 146, + 173 + ], + "score": 1.0, + "content": "bution of", + "type": "text" + }, + { + "bbox": [ + 147, + 161, + 158, + 171 + ], + "score": 0.87, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 159, + 205, + 173 + ], + "score": 1.0, + "content": ". To begin,", + "type": "text" + }, + { + "bbox": [ + 206, + 163, + 213, + 171 + ], + "score": 0.77, + "content": "_ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 159, + 284, + 173 + ], + "score": 1.0, + "content": "is sampled from", + "type": "text" + }, + { + "bbox": [ + 284, + 161, + 293, + 171 + ], + "score": 0.82, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 159, + 312, + 173 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 312, + 163, + 319, + 171 + ], + "score": 0.78, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 159, + 389, + 173 + ], + "score": 1.0, + "content": "is sampled from", + "type": "text" + }, + { + "bbox": [ + 389, + 161, + 402, + 172 + ], + "score": 0.89, + "content": "\\mathcal { D } _ { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 159, + 506, + 173 + ], + "score": 1.0, + "content": ". The other variables are", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 172, + 196, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 160, + 183 + ], + "score": 1.0, + "content": "generated by", + "type": "text" + }, + { + "bbox": [ + 160, + 173, + 167, + 181 + ], + "score": 0.78, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 172, + 185, + 183 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 186, + 173, + 192, + 181 + ], + "score": 0.78, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 172, + 196, + 183 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 264, + 180, + 347, + 226 + ], + "lines": [ + { + "bbox": [ + 264, + 180, + 347, + 226 + ], + "spans": [ + { + "bbox": [ + 264, + 180, + 347, + 226 + ], + "score": 0.83, + "content": "\\begin{array} { r l } & { z ^ { \\prime } = z + \\pmb { T } \\pmb { \\varepsilon } } \\\\ & { \\pmb { \\alpha } ^ { \\prime } = \\boldsymbol { R } ( G ( z ) ) + \\pmb { \\varepsilon } } \\\\ & { \\hat { \\pmb { \\alpha } } ^ { \\prime } = \\boldsymbol { R } ( G ( z ^ { \\prime } ) ) . } \\end{array}", + "type": "interline_equation", + "image_path": "54c2f56fa908bf04ffb48007d9cfaa8a6a15f62faca42b8bb9f92c992c27eb50.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 264, + 180, + 347, + 203.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 264, + 203.0, + 347, + 226.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 226, + 349, + 238 + ], + "lines": [ + { + "bbox": [ + 106, + 225, + 350, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 228, + 240 + ], + "score": 1.0, + "content": "As a result, the distribution of", + "type": "text" + }, + { + "bbox": [ + 228, + 227, + 239, + 236 + ], + "score": 0.88, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 225, + 306, + 240 + ], + "score": 1.0, + "content": "is dependent on", + "type": "text" + }, + { + "bbox": [ + 306, + 227, + 315, + 236 + ], + "score": 0.82, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 225, + 333, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 333, + 227, + 346, + 237 + ], + "score": 0.89, + "content": "\\mathcal { D } _ { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 225, + 350, + 240 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 109, + 251, + 329, + 262 + ], + "lines": [ + { + "bbox": [ + 107, + 251, + 330, + 263 + ], + "spans": [ + { + "bbox": [ + 107, + 251, + 330, + 263 + ], + "score": 1.0, + "content": "C.2 UNSUPERVISED LATENT-SPACE EDITS OF GANS", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 271, + 505, + 349 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "Voynov & Babenko (2020) learn a matrix where each column is a direction. Yet, Voynov & Babenko", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 282, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 506, + 295 + ], + "score": 1.0, + "content": "(2020) require a human to interpret the learnt directions. To avoid bias during the selection of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "score": 1.0, + "content": "directions in our comparison, we use the pre-trained attribute regressor to automatically identify", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "score": 1.0, + "content": "the most significant directions. Concretely, we first generate 100 images and edit them with target", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 315, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 330, + 327 + ], + "score": 1.0, + "content": "attributes of various degrees. Next, we use the regressor", + "type": "text" + }, + { + "bbox": [ + 330, + 316, + 339, + 325 + ], + "score": 0.8, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 315, + 505, + 327 + ], + "score": 1.0, + "content": "on the edited images to predict attributes.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 326, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 505, + 339 + ], + "score": 1.0, + "content": "We choose for an attribute edit the direction on which all the edited images have the overall highest", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 337, + 273, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 273, + 349 + ], + "score": 1.0, + "content": "changing response on the target attribute.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 107, + 361, + 240, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 241, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 122, + 374 + ], + "score": 1.0, + "content": "D.", + "type": "text" + }, + { + "bbox": [ + 135, + 361, + 241, + 374 + ], + "score": 1.0, + "content": "ADDITIONAL ANALYSIS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 108, + 382, + 266, + 393 + ], + "lines": [ + { + "bbox": [ + 106, + 382, + 267, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 382, + 267, + 394 + ], + "score": 1.0, + "content": "D.1 GAN INVERSION PERFORMANCE", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 402, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 106, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "We use an optimization-based GAN inversion method (Abdal et al., 2019) to find optimal latent", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 414, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 414, + 505, + 426 + ], + "score": 1.0, + "content": "vectors that can best reconstruct the real images via the generator. To examine the effect of GAN", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 425, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 506, + 437 + ], + "score": 1.0, + "content": "inversion performance on our approach, we terminated the GAN inversion approach at different", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "training steps, i.e., 500 and 4,000 iterations. We show averaged reconstruction MSE loss in Tab. 3.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 446, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 458 + ], + "score": 1.0, + "content": "In this case, we reconstructed 20 real face images and edited their “Smile” and the “Blond hair”", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 457, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 244, + 470 + ], + "score": 1.0, + "content": "attribute with 10 different degrees", + "type": "text" + }, + { + "bbox": [ + 244, + 459, + 250, + 468 + ], + "score": 0.59, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 457, + 506, + 470 + ], + "score": 1.0, + "content": ", i.e., 200 images in total for each attribute editing. Quantitative", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 468, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 481 + ], + "score": 1.0, + "content": "evaluation on image identity preservation and numerical changes on the other semantically indepen-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 479, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 506, + 492 + ], + "score": 1.0, + "content": "dent attributes for the reconstructed face images are given in Tab. 4. The results in Tab. 4 suggest", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 490, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 505, + 502 + ], + "score": 1.0, + "content": "that the performance of the GAN inversion method affects our method to some degree. Visualized", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 501, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 506, + 514 + ], + "score": 1.0, + "content": "inversion and editing results are shown in Fig. 10. The qualitative results suggest that our method", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 511, + 340, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 340, + 525 + ], + "score": 1.0, + "content": "still works remarkably well on the worse inversion image.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25 + }, + { + "type": "table", + "bbox": [ + 238, + 574, + 370, + 621 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 541, + 505, + 574 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "Table 3: Averaged MSE loss of reconstructing 20 real face images. The GAN inversion", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 352, + 564 + ], + "score": 1.0, + "content": "method (Abdal et al., 2019) was trained and terminated at", + "type": "text" + }, + { + "bbox": [ + 353, + 553, + 365, + 563 + ], + "score": 0.45, + "content": "4 \\mathrm { k \\Omega }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "and 500 iterations with averaged", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 563, + 228, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 228, + 575 + ], + "score": 1.0, + "content": "MSE loss in the right column.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "table_body", + "bbox": [ + 238, + 574, + 370, + 621 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 238, + 574, + 370, + 621 + ], + "spans": [ + { + "bbox": [ + 238, + 574, + 370, + 621 + ], + "score": 0.972, + "html": "
Training iterationsMSE
4k iters1657.40
500 iters3096.75
", + "type": "table", + "image_path": "b353d3242d046ba95c2ab2b07c2fac855867210eb3d383b29086be109ac4732e.jpg" + } + ] + } + ], + "index": 34.5, + "virtual_lines": [ + { + "bbox": [ + 238, + 574, + 370, + 597.5 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 238, + 597.5, + 370, + 621.0 + ], + "spans": [], + "index": 35 + } + ] + } + ], + "index": 33.25 + }, + { + "type": "title", + "bbox": [ + 107, + 635, + 203, + 646 + ], + "lines": [ + { + "bbox": [ + 106, + 634, + 204, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 204, + 648 + ], + "score": 1.0, + "content": "D.2 ABLATION STUDY", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "We conduct an ablation study with single- and joint-distribution training strategies in our method.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "Specifically, single-distribution sampling refers to learn one attribute direction at a time. In contrast,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "joint-distribution training means to train multiple attribute directions simultaneously. Fig. 11 shows", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "the visualized results with the two training strategies on the face and the scene dataset. We observe", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "that the model with the single-distribution training strategy generates more unexpected changes as", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "the manipulation degree is getting large, e.g., darker scene colors by the model trained for a single", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 720, + 231, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 231, + 733 + ], + "score": 1.0, + "content": "attribute, shown in Fig. 11 (b).", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 265, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 266, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 120, + 96 + ], + "score": 1.0, + "content": "C.", + "type": "text" + }, + { + "bbox": [ + 135, + 82, + 266, + 96 + ], + "score": 1.0, + "content": "EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 102, + 240, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 239, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 218, + 118 + ], + "score": 1.0, + "content": "C.1 THE EXPECTATION IN", + "type": "text" + }, + { + "bbox": [ + 218, + 104, + 239, + 114 + ], + "score": 0.83, + "content": "\\mathcal { L } _ { \\mathrm { { R E G } } }", + "type": "inline_equation" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 100, + 239, + 118 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 123, + 254, + 135 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 255, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 225, + 138 + ], + "score": 1.0, + "content": "We define the regression loss", + "type": "text" + }, + { + "bbox": [ + 225, + 124, + 242, + 136 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { r e g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 121, + 255, + 138 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 121, + 255, + 138 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 183, + 137, + 425, + 151 + ], + "lines": [ + { + "bbox": [ + 183, + 137, + 425, + 151 + ], + "spans": [ + { + "bbox": [ + 183, + 137, + 425, + 151 + ], + "score": 0.85, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { \\mathrm { r e g } } = \\ } & { { } \\mathbb { E } _ { z \\sim \\mathcal { Z } , \\varepsilon \\sim \\mathcal { D } _ { \\varepsilon } } [ - \\hat { \\alpha } ^ { \\prime } \\log \\alpha ^ { \\prime } - ( { \\bf 1 } - \\hat { \\alpha } ^ { \\prime } ) \\log \\left( { \\bf 1 } - \\alpha ^ { \\prime } \\right) ] , } \\end{array}", + "type": "interline_equation", + "image_path": "7d290e5bf22b3369bd3b12d080a587bb5027cd3440c4c919326b9ad0a137e90a.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 183, + 137, + 425, + 151 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 109, + 151, + 504, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 133, + 162 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 150, + 145, + 160 + ], + "score": 0.87, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 149, + 297, + 162 + ], + "score": 1.0, + "content": "is from the distribution generated by", + "type": "text" + }, + { + "bbox": [ + 297, + 152, + 304, + 160 + ], + "score": 0.77, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 149, + 322, + 162 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 322, + 152, + 329, + 160 + ], + "score": 0.77, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 149, + 505, + 162 + ], + "score": 1.0, + "content": ". We now discuss how we derive the distri-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 146, + 173 + ], + "score": 1.0, + "content": "bution of", + "type": "text" + }, + { + "bbox": [ + 147, + 161, + 158, + 171 + ], + "score": 0.87, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 159, + 205, + 173 + ], + "score": 1.0, + "content": ". To begin,", + "type": "text" + }, + { + "bbox": [ + 206, + 163, + 213, + 171 + ], + "score": 0.77, + "content": "_ { z }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 159, + 284, + 173 + ], + "score": 1.0, + "content": "is sampled from", + "type": "text" + }, + { + "bbox": [ + 284, + 161, + 293, + 171 + ], + "score": 0.82, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 159, + 312, + 173 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 312, + 163, + 319, + 171 + ], + "score": 0.78, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 159, + 389, + 173 + ], + "score": 1.0, + "content": "is sampled from", + "type": "text" + }, + { + "bbox": [ + 389, + 161, + 402, + 172 + ], + "score": 0.89, + "content": "\\mathcal { D } _ { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 159, + 506, + 173 + ], + "score": 1.0, + "content": ". The other variables are", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 172, + 196, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 160, + 183 + ], + "score": 1.0, + "content": "generated by", + "type": "text" + }, + { + "bbox": [ + 160, + 173, + 167, + 181 + ], + "score": 0.78, + "content": "_ z", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 172, + 185, + 183 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 186, + 173, + 192, + 181 + ], + "score": 0.78, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 172, + 196, + 183 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 106, + 149, + 506, + 183 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 264, + 180, + 347, + 226 + ], + "lines": [ + { + "bbox": [ + 264, + 180, + 347, + 226 + ], + "spans": [ + { + "bbox": [ + 264, + 180, + 347, + 226 + ], + "score": 0.83, + "content": "\\begin{array} { r l } & { z ^ { \\prime } = z + \\pmb { T } \\pmb { \\varepsilon } } \\\\ & { \\pmb { \\alpha } ^ { \\prime } = \\boldsymbol { R } ( G ( z ) ) + \\pmb { \\varepsilon } } \\\\ & { \\hat { \\pmb { \\alpha } } ^ { \\prime } = \\boldsymbol { R } ( G ( z ^ { \\prime } ) ) . } \\end{array}", + "type": "interline_equation", + "image_path": "54c2f56fa908bf04ffb48007d9cfaa8a6a15f62faca42b8bb9f92c992c27eb50.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 264, + 180, + 347, + 203.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 264, + 203.0, + 347, + 226.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 226, + 349, + 238 + ], + "lines": [ + { + "bbox": [ + 106, + 225, + 350, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 228, + 240 + ], + "score": 1.0, + "content": "As a result, the distribution of", + "type": "text" + }, + { + "bbox": [ + 228, + 227, + 239, + 236 + ], + "score": 0.88, + "content": "\\alpha ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 225, + 306, + 240 + ], + "score": 1.0, + "content": "is dependent on", + "type": "text" + }, + { + "bbox": [ + 306, + 227, + 315, + 236 + ], + "score": 0.82, + "content": "\\mathcal { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 225, + 333, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 333, + 227, + 346, + 237 + ], + "score": 0.89, + "content": "\\mathcal { D } _ { \\varepsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 225, + 350, + 240 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 106, + 225, + 350, + 240 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 251, + 329, + 262 + ], + "lines": [ + { + "bbox": [ + 107, + 251, + 330, + 263 + ], + "spans": [ + { + "bbox": [ + 107, + 251, + 330, + 263 + ], + "score": 1.0, + "content": "C.2 UNSUPERVISED LATENT-SPACE EDITS OF GANS", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 271, + 505, + 349 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "Voynov & Babenko (2020) learn a matrix where each column is a direction. Yet, Voynov & Babenko", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 282, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 506, + 295 + ], + "score": 1.0, + "content": "(2020) require a human to interpret the learnt directions. To avoid bias during the selection of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "score": 1.0, + "content": "directions in our comparison, we use the pre-trained attribute regressor to automatically identify", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "score": 1.0, + "content": "the most significant directions. Concretely, we first generate 100 images and edit them with target", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 315, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 330, + 327 + ], + "score": 1.0, + "content": "attributes of various degrees. Next, we use the regressor", + "type": "text" + }, + { + "bbox": [ + 330, + 316, + 339, + 325 + ], + "score": 0.8, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 315, + 505, + 327 + ], + "score": 1.0, + "content": "on the edited images to predict attributes.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 326, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 505, + 339 + ], + "score": 1.0, + "content": "We choose for an attribute edit the direction on which all the edited images have the overall highest", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 337, + 273, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 273, + 349 + ], + "score": 1.0, + "content": "changing response on the target attribute.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 271, + 506, + 349 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 361, + 240, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 241, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 122, + 374 + ], + "score": 1.0, + "content": "D.", + "type": "text" + }, + { + "bbox": [ + 135, + 361, + 241, + 374 + ], + "score": 1.0, + "content": "ADDITIONAL ANALYSIS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 108, + 382, + 266, + 393 + ], + "lines": [ + { + "bbox": [ + 106, + 382, + 267, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 382, + 267, + 394 + ], + "score": 1.0, + "content": "D.1 GAN INVERSION PERFORMANCE", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 402, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 106, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "We use an optimization-based GAN inversion method (Abdal et al., 2019) to find optimal latent", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 414, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 414, + 505, + 426 + ], + "score": 1.0, + "content": "vectors that can best reconstruct the real images via the generator. To examine the effect of GAN", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 425, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 506, + 437 + ], + "score": 1.0, + "content": "inversion performance on our approach, we terminated the GAN inversion approach at different", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "training steps, i.e., 500 and 4,000 iterations. We show averaged reconstruction MSE loss in Tab. 3.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 446, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 458 + ], + "score": 1.0, + "content": "In this case, we reconstructed 20 real face images and edited their “Smile” and the “Blond hair”", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 457, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 244, + 470 + ], + "score": 1.0, + "content": "attribute with 10 different degrees", + "type": "text" + }, + { + "bbox": [ + 244, + 459, + 250, + 468 + ], + "score": 0.59, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 457, + 506, + 470 + ], + "score": 1.0, + "content": ", i.e., 200 images in total for each attribute editing. Quantitative", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 468, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 481 + ], + "score": 1.0, + "content": "evaluation on image identity preservation and numerical changes on the other semantically indepen-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 479, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 506, + 492 + ], + "score": 1.0, + "content": "dent attributes for the reconstructed face images are given in Tab. 4. The results in Tab. 4 suggest", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 490, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 505, + 502 + ], + "score": 1.0, + "content": "that the performance of the GAN inversion method affects our method to some degree. Visualized", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 501, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 506, + 514 + ], + "score": 1.0, + "content": "inversion and editing results are shown in Fig. 10. The qualitative results suggest that our method", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 511, + 340, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 340, + 525 + ], + "score": 1.0, + "content": "still works remarkably well on the worse inversion image.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 402, + 506, + 525 + ] + }, + { + "type": "table", + "bbox": [ + 238, + 574, + 370, + 621 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 541, + 505, + 574 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "Table 3: Averaged MSE loss of reconstructing 20 real face images. The GAN inversion", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 352, + 564 + ], + "score": 1.0, + "content": "method (Abdal et al., 2019) was trained and terminated at", + "type": "text" + }, + { + "bbox": [ + 353, + 553, + 365, + 563 + ], + "score": 0.45, + "content": "4 \\mathrm { k \\Omega }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "and 500 iterations with averaged", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 563, + 228, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 228, + 575 + ], + "score": 1.0, + "content": "MSE loss in the right column.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "table_body", + "bbox": [ + 238, + 574, + 370, + 621 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 238, + 574, + 370, + 621 + ], + "spans": [ + { + "bbox": [ + 238, + 574, + 370, + 621 + ], + "score": 0.972, + "html": "
Training iterationsMSE
4k iters1657.40
500 iters3096.75
", + "type": "table", + "image_path": "b353d3242d046ba95c2ab2b07c2fac855867210eb3d383b29086be109ac4732e.jpg" + } + ] + } + ], + "index": 34.5, + "virtual_lines": [ + { + "bbox": [ + 238, + 574, + 370, + 597.5 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 238, + 597.5, + 370, + 621.0 + ], + "spans": [], + "index": 35 + } + ] + } + ], + "index": 33.25 + }, + { + "type": "title", + "bbox": [ + 107, + 635, + 203, + 646 + ], + "lines": [ + { + "bbox": [ + 106, + 634, + 204, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 204, + 648 + ], + "score": 1.0, + "content": "D.2 ABLATION STUDY", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "We conduct an ablation study with single- and joint-distribution training strategies in our method.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "Specifically, single-distribution sampling refers to learn one attribute direction at a time. In contrast,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "joint-distribution training means to train multiple attribute directions simultaneously. Fig. 11 shows", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "the visualized results with the two training strategies on the face and the scene dataset. We observe", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "that the model with the single-distribution training strategy generates more unexpected changes as", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "the manipulation degree is getting large, e.g., darker scene colors by the model trained for a single", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 720, + 231, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 231, + 733 + ], + "score": 1.0, + "content": "attribute, shown in Fig. 11 (b).", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 654, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 113, + 508, + 174 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 504, + 112 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 89, + 506, + 102 + ], + "spans": [ + { + "bbox": [ + 106, + 89, + 345, + 102 + ], + "score": 1.0, + "content": "Table 4: Quantitative evaluation of identity preservation", + "type": "text" + }, + { + "bbox": [ + 345, + 90, + 363, + 100 + ], + "score": 0.29, + "content": "\\mathbf { \\Pi } ^ { ( \\mathbf { I I D } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 89, + 506, + 102 + ], + "score": 1.0, + "content": "and attribute changes (Attr). We", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 100, + 492, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 492, + 113 + ], + "score": 1.0, + "content": "edited two attributes for the real face images, i.e., “Smile” (col.2-7) and “Blond hair” (col.8-13).", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 113, + 508, + 174 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 113, + 508, + 174 + ], + "spans": [ + { + "bbox": [ + 106, + 113, + 508, + 174 + ], + "score": 0.971, + "html": "
SmileBlond hair
ID (↑)Attr (↓)ID(↑)Attr (↓)
国(0,.3](.3,.6](.6,.9] (0, .3](.3,.6] (.6,.9](0,.3] (.3,.6](.6,.9] (0,.3] (.3,.6] (.6,.9]
4k iters 500 iters.997 .996.994 .99.986 .982.077 .087.108 .138.147 .165.988 .978.979 .936.94 .914.1 .122.112 .158.16 .185
", + "type": "table", + "image_path": "a994c5ec868514ac66d0a20ab1aacec8f673989a2bc20a53b74b5fe12fcd4cec.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 106, + 113, + 508, + 133.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 133.33333333333334, + 508, + 153.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 153.66666666666669, + 508, + 174.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "image", + "bbox": [ + 149, + 189, + 462, + 326 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 149, + 189, + 462, + 326 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 149, + 189, + 462, + 326 + ], + "spans": [ + { + "bbox": [ + 149, + 189, + 462, + 326 + ], + "score": 0.97, + "type": "image", + "image_path": "af8db2995983534031404349d44950bd87eef7c591b07f228e520b77de7cd045.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 149, + 189, + 462, + 234.66666666666666 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 149, + 234.66666666666666, + 462, + 280.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 149, + 280.3333333333333, + 462, + 326.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 339, + 504, + 374 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 339, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 506, + 353 + ], + "score": 1.0, + "content": "Figure 10: Visual comparisons of “Smile” editing with different reconstructed images: (col.1)", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 350, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 363, + 363 + ], + "score": 1.0, + "content": "Original image; (col.2) reconstructed images with 500 (top) and", + "type": "text" + }, + { + "bbox": [ + 363, + 351, + 375, + 361 + ], + "score": 0.32, + "content": "4 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 350, + 505, + 363 + ], + "score": 1.0, + "content": "(bottom) inversion optimization", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 362, + 326, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 326, + 375 + ], + "score": 1.0, + "content": "steps; (col.3-6) results of editing the “Smile” attribute.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 108, + 393, + 249, + 405 + ], + "lines": [ + { + "bbox": [ + 103, + 391, + 251, + 410 + ], + "spans": [ + { + "bbox": [ + 103, + 391, + 251, + 410 + ], + "score": 1.0, + "content": "E. ADDITIONAL RESULTS", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 107, + 413, + 327, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 328, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 128, + 427 + ], + "score": 1.0, + "content": "E.1", + "type": "text" + }, + { + "bbox": [ + 136, + 414, + 328, + 428 + ], + "score": 1.0, + "content": "SCENE IMAGE EDITS ON STYLEGAN2", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 505, + 468 + ], + "lines": [ + { + "bbox": [ + 106, + 435, + 504, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 504, + 447 + ], + "score": 1.0, + "content": "We show more results of our method on continuous scene image edits in Fig. 12 - 15. The seman-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 445, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 506, + 459 + ], + "score": 1.0, + "content": "tically edited attributes include “clouds” (Fig. 12), “brightness” (Fig. 13), “snow” (Fig. 14), and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 456, + 190, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 190, + 470 + ], + "score": 1.0, + "content": "“summer” (Fig. 15).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 481, + 210, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 481, + 212, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 119, + 493 + ], + "score": 1.0, + "content": "F.", + "type": "text" + }, + { + "bbox": [ + 129, + 481, + 212, + 494 + ], + "score": 1.0, + "content": "BROADER IMPACT", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 502, + 505, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "score": 1.0, + "content": "From an application perspective, our method is effective and efficient with regard to image manip-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 512, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 527 + ], + "score": 1.0, + "content": "ulation and photo-realism, which we hope will contribute to 2D and 3D controllable image editing", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 525, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 505, + 536 + ], + "score": 1.0, + "content": "tasks. Moreover, usage of deep nets to learn mappings between spaces is still not well understood,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 535, + 506, + 548 + ], + "spans": [ + { + "bbox": [ + 104, + 535, + 506, + 548 + ], + "score": 1.0, + "content": "e.g., mappings between low-dimensional space and image space in classification and generation", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "tasks. We hope our method provides inspiration for representation learning and a first step for a new", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 557, + 285, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 285, + 570 + ], + "score": 1.0, + "content": "view with regard to deep net interpretability.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 574, + 505, + 629 + ], + "lines": [ + { + "bbox": [ + 106, + 573, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 505, + 587 + ], + "score": 1.0, + "content": "Obviously, we are aware of the dangers of automated image manipulation. Similar to deepfake tasks", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 584, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 505, + 598 + ], + "score": 1.0, + "content": "whose aim is to produce fabricated images and videos that appear to be real, improper use of image", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "score": 1.0, + "content": "manipulation approaches might raise negative issues with regard to information security, property,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 606, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 621 + ], + "score": 1.0, + "content": "etc. Beyond that, edited image detection techniques Wang et al. (2019a;b) are proposed recently to", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 616, + 405, + 632 + ], + "spans": [ + { + "bbox": [ + 104, + 616, + 405, + 632 + ], + "score": 1.0, + "content": "avoid the aforementioned issues, which promotes growth in both domains.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 113, + 508, + 174 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 504, + 112 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 89, + 506, + 102 + ], + "spans": [ + { + "bbox": [ + 106, + 89, + 345, + 102 + ], + "score": 1.0, + "content": "Table 4: Quantitative evaluation of identity preservation", + "type": "text" + }, + { + "bbox": [ + 345, + 90, + 363, + 100 + ], + "score": 0.29, + "content": "\\mathbf { \\Pi } ^ { ( \\mathbf { I I D } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 89, + 506, + 102 + ], + "score": 1.0, + "content": "and attribute changes (Attr). We", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 100, + 492, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 492, + 113 + ], + "score": 1.0, + "content": "edited two attributes for the real face images, i.e., “Smile” (col.2-7) and “Blond hair” (col.8-13).", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 113, + 508, + 174 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 113, + 508, + 174 + ], + "spans": [ + { + "bbox": [ + 106, + 113, + 508, + 174 + ], + "score": 0.971, + "html": "
SmileBlond hair
ID (↑)Attr (↓)ID(↑)Attr (↓)
国(0,.3](.3,.6](.6,.9] (0, .3](.3,.6] (.6,.9](0,.3] (.3,.6](.6,.9] (0,.3] (.3,.6] (.6,.9]
4k iters 500 iters.997 .996.994 .99.986 .982.077 .087.108 .138.147 .165.988 .978.979 .936.94 .914.1 .122.112 .158.16 .185
", + "type": "table", + "image_path": "a994c5ec868514ac66d0a20ab1aacec8f673989a2bc20a53b74b5fe12fcd4cec.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 106, + 113, + 508, + 133.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 133.33333333333334, + 508, + 153.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 153.66666666666669, + 508, + 174.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "image", + "bbox": [ + 149, + 189, + 462, + 326 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 149, + 189, + 462, + 326 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 149, + 189, + 462, + 326 + ], + "spans": [ + { + "bbox": [ + 149, + 189, + 462, + 326 + ], + "score": 0.97, + "type": "image", + "image_path": "af8db2995983534031404349d44950bd87eef7c591b07f228e520b77de7cd045.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 149, + 189, + 462, + 234.66666666666666 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 149, + 234.66666666666666, + 462, + 280.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 149, + 280.3333333333333, + 462, + 326.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 339, + 504, + 374 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 339, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 506, + 353 + ], + "score": 1.0, + "content": "Figure 10: Visual comparisons of “Smile” editing with different reconstructed images: (col.1)", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 350, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 363, + 363 + ], + "score": 1.0, + "content": "Original image; (col.2) reconstructed images with 500 (top) and", + "type": "text" + }, + { + "bbox": [ + 363, + 351, + 375, + 361 + ], + "score": 0.32, + "content": "4 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 350, + 505, + 363 + ], + "score": 1.0, + "content": "(bottom) inversion optimization", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 362, + 326, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 326, + 375 + ], + "score": 1.0, + "content": "steps; (col.3-6) results of editing the “Smile” attribute.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 108, + 393, + 249, + 405 + ], + "lines": [ + { + "bbox": [ + 103, + 391, + 251, + 410 + ], + "spans": [ + { + "bbox": [ + 103, + 391, + 251, + 410 + ], + "score": 1.0, + "content": "E. ADDITIONAL RESULTS", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 107, + 413, + 327, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 328, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 128, + 427 + ], + "score": 1.0, + "content": "E.1", + "type": "text" + }, + { + "bbox": [ + 136, + 414, + 328, + 428 + ], + "score": 1.0, + "content": "SCENE IMAGE EDITS ON STYLEGAN2", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 505, + 468 + ], + "lines": [ + { + "bbox": [ + 106, + 435, + 504, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 504, + 447 + ], + "score": 1.0, + "content": "We show more results of our method on continuous scene image edits in Fig. 12 - 15. The seman-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 445, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 506, + 459 + ], + "score": 1.0, + "content": "tically edited attributes include “clouds” (Fig. 12), “brightness” (Fig. 13), “snow” (Fig. 14), and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 456, + 190, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 190, + 470 + ], + "score": 1.0, + "content": "“summer” (Fig. 15).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 435, + 506, + 470 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 481, + 210, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 481, + 212, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 119, + 493 + ], + "score": 1.0, + "content": "F.", + "type": "text" + }, + { + "bbox": [ + 129, + 481, + 212, + 494 + ], + "score": 1.0, + "content": "BROADER IMPACT", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 502, + 505, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "score": 1.0, + "content": "From an application perspective, our method is effective and efficient with regard to image manip-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 512, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 527 + ], + "score": 1.0, + "content": "ulation and photo-realism, which we hope will contribute to 2D and 3D controllable image editing", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 525, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 505, + 536 + ], + "score": 1.0, + "content": "tasks. Moreover, usage of deep nets to learn mappings between spaces is still not well understood,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 535, + 506, + 548 + ], + "spans": [ + { + "bbox": [ + 104, + 535, + 506, + 548 + ], + "score": 1.0, + "content": "e.g., mappings between low-dimensional space and image space in classification and generation", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "tasks. We hope our method provides inspiration for representation learning and a first step for a new", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 557, + 285, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 285, + 570 + ], + "score": 1.0, + "content": "view with regard to deep net interpretability.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19.5, + "bbox_fs": [ + 104, + 501, + 506, + 570 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 574, + 505, + 629 + ], + "lines": [ + { + "bbox": [ + 106, + 573, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 505, + 587 + ], + "score": 1.0, + "content": "Obviously, we are aware of the dangers of automated image manipulation. Similar to deepfake tasks", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 584, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 505, + 598 + ], + "score": 1.0, + "content": "whose aim is to produce fabricated images and videos that appear to be real, improper use of image", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "score": 1.0, + "content": "manipulation approaches might raise negative issues with regard to information security, property,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 606, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 621 + ], + "score": 1.0, + "content": "etc. Beyond that, edited image detection techniques Wang et al. (2019a;b) are proposed recently to", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 616, + 405, + 632 + ], + "spans": [ + { + "bbox": [ + 104, + 616, + 405, + 632 + ], + "score": 1.0, + "content": "avoid the aforementioned issues, which promotes growth in both domains.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 104, + 573, + 505, + 632 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 82, + 501, + 215 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 82, + 501, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 82, + 501, + 215 + ], + "spans": [ + { + "bbox": [ + 112, + 82, + 501, + 215 + ], + "score": 0.965, + "type": "image", + "image_path": "070499f2f63fa653dfec515acb1f5e8b3c6dcbe6b26fef8e4ea673ab6cbf7190.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 82, + 501, + 126.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 126.33333333333334, + 501, + 170.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 170.66666666666669, + 501, + 215.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 223, + 505, + 257 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 222, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 505, + 236 + ], + "score": 1.0, + "content": "Figure 11: Visual comparisons of single- and joint-distribution training. Single-distribution", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 234, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 505, + 246 + ], + "score": 1.0, + "content": "training refers to train one attribute direction at a time, while joint-distribution training means to", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 246, + 304, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 304, + 258 + ], + "score": 1.0, + "content": "train multiple attribute directions simultaneously.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 107, + 268, + 504, + 389 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 268, + 504, + 389 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 268, + 504, + 389 + ], + "spans": [ + { + "bbox": [ + 107, + 268, + 504, + 389 + ], + "score": 0.967, + "type": "image", + "image_path": "191cda68622e417c6d8420dd5a11163fe54fe664b24042948c573d3330369ea7.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 107, + 268, + 504, + 308.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 107, + 308.3333333333333, + 504, + 348.66666666666663 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 107, + 348.66666666666663, + 504, + 388.99999999999994 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 398, + 504, + 420 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "score": 1.0, + "content": "Figure 12: Additional results. Continuous image edits using StyleGAN2 (Karras et al., 2019b) on", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 408, + 196, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 196, + 421 + ], + "score": 1.0, + "content": "the “clouds” attribute.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + } + ], + "index": 8.25 + }, + { + "type": "image", + "bbox": [ + 107, + 425, + 504, + 546 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 425, + 504, + 546 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 107, + 425, + 504, + 546 + ], + "spans": [ + { + "bbox": [ + 107, + 425, + 504, + 546 + ], + "score": 0.971, + "type": "image", + "image_path": "18b8b7a561e6889f43597a128284efa207895a9e0477c5d4698ddd06161657af.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 107, + 425, + 504, + 465.3333333333333 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 107, + 465.3333333333333, + 504, + 505.66666666666663 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 107, + 505.66666666666663, + 504, + 546.0 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 554, + 505, + 577 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "Figure 13: Additional results. Continuous image edits using StyleGAN2 (Karras et al., 2019b) on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 565, + 212, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 212, + 578 + ], + "score": 1.0, + "content": "the “brightness” attribute.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + } + ], + "index": 13.25 + }, + { + "type": "image", + "bbox": [ + 107, + 582, + 504, + 704 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 582, + 504, + 704 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 107, + 582, + 504, + 704 + ], + "spans": [ + { + "bbox": [ + 107, + 582, + 504, + 704 + ], + "score": 0.969, + "type": "image", + "image_path": "ff2234514908b69cd7c90051e7852b66cdc1b435be6dd2d622ff6f37ebbd9971.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 107, + 582, + 504, + 622.6666666666666 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 107, + 622.6666666666666, + 504, + 663.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 107, + 663.3333333333333, + 504, + 703.9999999999999 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 713, + 506, + 735 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 105, + 712, + 505, + 726 + ], + "spans": [ + { + "bbox": [ + 105, + 712, + 505, + 726 + ], + "score": 1.0, + "content": "Figure 14: Additional results. Continuous image edits using StyleGAN2 (Karras et al., 2019b) on", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 722, + 191, + 735 + ], + "spans": [ + { + "bbox": [ + 106, + 722, + 191, + 735 + ], + "score": 1.0, + "content": "the “snow” attribute.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + } + ], + "index": 18.25 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 82, + 501, + 215 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 82, + 501, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 82, + 501, + 215 + ], + "spans": [ + { + "bbox": [ + 112, + 82, + 501, + 215 + ], + "score": 0.965, + "type": "image", + "image_path": "070499f2f63fa653dfec515acb1f5e8b3c6dcbe6b26fef8e4ea673ab6cbf7190.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 82, + 501, + 126.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 126.33333333333334, + 501, + 170.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 170.66666666666669, + 501, + 215.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 223, + 505, + 257 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 222, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 505, + 236 + ], + "score": 1.0, + "content": "Figure 11: Visual comparisons of single- and joint-distribution training. Single-distribution", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 234, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 505, + 246 + ], + "score": 1.0, + "content": "training refers to train one attribute direction at a time, while joint-distribution training means to", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 246, + 304, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 304, + 258 + ], + "score": 1.0, + "content": "train multiple attribute directions simultaneously.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 107, + 268, + 504, + 389 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 268, + 504, + 389 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 268, + 504, + 389 + ], + "spans": [ + { + "bbox": [ + 107, + 268, + 504, + 389 + ], + "score": 0.967, + "type": "image", + "image_path": "191cda68622e417c6d8420dd5a11163fe54fe664b24042948c573d3330369ea7.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 107, + 268, + 504, + 308.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 107, + 308.3333333333333, + 504, + 348.66666666666663 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 107, + 348.66666666666663, + 504, + 388.99999999999994 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 398, + 504, + 420 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "score": 1.0, + "content": "Figure 12: Additional results. Continuous image edits using StyleGAN2 (Karras et al., 2019b) on", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 408, + 196, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 196, + 421 + ], + "score": 1.0, + "content": "the “clouds” attribute.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + } + ], + "index": 8.25 + }, + { + "type": "image", + "bbox": [ + 107, + 425, + 504, + 546 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 425, + 504, + 546 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 107, + 425, + 504, + 546 + ], + "spans": [ + { + "bbox": [ + 107, + 425, + 504, + 546 + ], + "score": 0.971, + "type": "image", + "image_path": "18b8b7a561e6889f43597a128284efa207895a9e0477c5d4698ddd06161657af.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 107, + 425, + 504, + 465.3333333333333 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 107, + 465.3333333333333, + 504, + 505.66666666666663 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 107, + 505.66666666666663, + 504, + 546.0 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 554, + 505, + 577 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "Figure 13: Additional results. Continuous image edits using StyleGAN2 (Karras et al., 2019b) on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 565, + 212, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 212, + 578 + ], + "score": 1.0, + "content": "the “brightness” attribute.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + } + ], + "index": 13.25 + }, + { + "type": "image", + "bbox": [ + 107, + 582, + 504, + 704 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 582, + 504, + 704 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 107, + 582, + 504, + 704 + ], + "spans": [ + { + "bbox": [ + 107, + 582, + 504, + 704 + ], + "score": 0.969, + "type": "image", + "image_path": "ff2234514908b69cd7c90051e7852b66cdc1b435be6dd2d622ff6f37ebbd9971.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 107, + 582, + 504, + 622.6666666666666 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 107, + 622.6666666666666, + 504, + 663.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 107, + 663.3333333333333, + 504, + 703.9999999999999 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 713, + 506, + 735 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 105, + 712, + 505, + 726 + ], + "spans": [ + { + "bbox": [ + 105, + 712, + 505, + 726 + ], + "score": 1.0, + "content": "Figure 14: Additional results. Continuous image edits using StyleGAN2 (Karras et al., 2019b) on", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 722, + 191, + 735 + ], + "spans": [ + { + "bbox": [ + 106, + 722, + 191, + 735 + ], + "score": 1.0, + "content": "the “snow” attribute.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + } + ], + "index": 18.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 330, + 504, + 451 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 330, + 504, + 451 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 330, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 107, + 330, + 504, + 451 + ], + "score": 0.969, + "type": "image", + "image_path": "683ee00e8b0863b6e305268599fb5308be987905f6b832d5f8490b4a0911e2b8.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 330, + 504, + 370.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 370.3333333333333, + 504, + 410.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 410.66666666666663, + 504, + 450.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 460, + 505, + 482 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "score": 1.0, + "content": "Figure 15: Additional results. Continuous image edits using StyleGAN2 (Karras et al., 2019b) on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 470, + 203, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 203, + 483 + ], + "score": 1.0, + "content": "the “summer” attribute.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 330, + 504, + 451 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 330, + 504, + 451 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 330, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 107, + 330, + 504, + 451 + ], + "score": 0.969, + "type": "image", + "image_path": "683ee00e8b0863b6e305268599fb5308be987905f6b832d5f8490b4a0911e2b8.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 330, + 504, + 370.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 370.3333333333333, + 504, + 410.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 410.66666666666663, + 504, + 450.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 460, + 505, + 482 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "score": 1.0, + "content": "Figure 15: Additional results. Continuous image edits using StyleGAN2 (Karras et al., 2019b) on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 470, + 203, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 203, + 483 + ], + "score": 1.0, + "content": "the “summer” attribute.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file