-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
427 lines (378 loc) · 18.8 KB
/
index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description" content="CosmicMan">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>CosmicMan</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
<script src="https://kit.fontawesome.com/ec39017701.js" crossorigin="anonymous"></script>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./static/js/fontawesome.all.min.js"></script>
<script src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-slider.min.js"></script>
<script src="./static/js/index.js"></script>
<style>
.image-group {
display: flex;
justify-content: space-between;
}
.image-group img {
flex: 1;
}
.image-group {
display: flex;
justify-content: center;
}
</style>
</head>
<body>
<div id="loading">
<div class="loader"></div>
<!-- <img id="loading-image" src="path/to/ajax-loader.gif" alt="Loading..." /> -->
</div>
<!-- <script> magnify("myimage", 3); </script> -->
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<br><br><br>
<h1 class="title is-1 publication-title">CosmicMan: A Text-to-Image Foundation Model for Humans</h1>
<h4 class="title is-4 publication-title" style="color:#5a6268;"> CVPR 2024 <span style="color: red;">(Highlight)</span></h4>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="mailto:lishikai@pjlab.org.cn" target="_blank">Shikai Li</a>*, </span>
<!-- <sup>1</sup> -->
<span class="author-block">
<a href="mailto:arlenefu@outlook.com" target="_blank">Jianglin Fu</a>*, </span>
<span class="author-block">
<a href="mailto:1154864382@mail.dlut.edu.cn" target="_blank">Kaiyuan Liu</a>*, </span>
<span class="author-block">
<a href="mailto:wangwentao@pjlab.org.cn" target="_blank">Wentao Wang</a>*, </span>
<span class="author-block">
<a href="https://kwanyeelin.github.io/" target="_blank">Kwan-Yee Lin</a><sup>†</sup>, </span>
<span class="author-block">
<a href="https://wywu.github.io/" target="_blank">Wayne Wu</a><sup>†</sup> </span>
<!-- <sup>1†</sup> -->
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">Shanghai AI Laboratory </span>
<!-- <sup>1</sup> -->
</div>
<div class="column has-text-centered">
<div class="publication-links">
<span class="link-block">
<a href="https://arxiv.org/abs/2404.01294" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>Arxiv</span>
</a>
</span>
<span class="link-block">
<a href="https://huggingface.co/datasets/cosmicman/CosmicManHQ-1.0" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fa-solid fa-database"></i>
</span>
<span>Data </span>
</a>
</span>
<!-- Video Link. -->
<span class="link-block">
<a href="https://www.youtube.com/watch?v=CsZKA27tQDA" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-youtube"></i>
</span>
<span>Video</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/cosmicman-cvpr2024/CosmicMan" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>GitHub</span>
</a>
</span>
<!-- photo gallery. -->
<!-- <span class="link-block">
<a href="" target="_blank" onclick="event.preventDefault();"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fa-solid fa-image"></i>
</span>
<span>Photo Gallery (Coming Soon)</span>
</a>
</span> -->
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<br><br>
<!-- Teaser Picture -->
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<img class="cover" src="./static/imgs/teaser.png" style="width:1000px; margin-bottom:20px;">
</div>
</div>
</section>
<!-- Abstract. -->
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
<p>
We present <strong>CosmicMan</strong>, a text-to-image foundation model specialized for generating high-fidelity human images.
Unlike current general-purpose foundation models that are stuck in the dilemma of inferior quality and text-image misalignment for humans,
CosmicMan enables generating photo-realistic human images with meticulous appearance, reasonable structure,
and precise text-image alignment with detailed dense descriptions.
At the heart of CosmicMan's success are the new reflections and perspectives on data and model:
(1) We found that data quality and a scalable data production flow are essential for the final results from trained models.
Hence, we propose a new data production paradigm, <strong>Annotate Anyone</strong>, which serves as a perpetual data flywheel to produce high-quality data with accurate yet
cost-effective annotations over time. Based on this, we constructed a large-scale dataset <strong>CosmicMan-HQ 1.0</strong>,
with 6 Million high-quality real-world human images in a mean resolution of 1488x1255,
and attached with precise text annotations deriving from 115 Million attributes in diverse granularities.
(2) We argue that a text-to-image foundation model specialized for humans must be pragmatic - easy to integrate into down-streaming tasks while effective
in producing high-quality human images. Hence, we propose to model the relationship between dense text descriptions and image pixels in
a decomposed manner, and present <strong>D</strong>ecomposed-<strong>A</strong>ttention-<strong>R</strong>efocus<strong>ing</strong> (<strong>Daring</strong>) training framework.
It seamlessly decomposes the cross-attention features in existing text-to-image diffusion model, and enforces attention refocusing without adding extra modules.
Through Daring, we show that explicitly discretizing continuous text space into several basic groups that align with human body structure is the key to tackling the misalignment problem in a breeze.
</p>
</div>
</div>
</div>
</section>
<!--/ Abstract. -->
<!-- Paper video. -->
<section>
<div class="columns is-centered has-text-centered">
<div class="column is-three-fifths">
<h2 class="title is-3">Overview Video</h2>
<div class="publication-video">
<iframe src="https://www.youtube.com/embed/CsZKA27tQDA?autoplay=0" frameborder="10"
allow="autoplay; encrypted-media" width="50%" allowfullscreen></iframe>
</div>
<!-- https://www.youtube.com/embed/nIrb9hwsdcI?autoplay=0 -->
</div>
</div>
</section>
<!-- / Paper video. -->
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Data pipeline - Annotate Anyone</h2>
<div class="content has-text-justified">
<img src="./static/imgs/data_pipeline.png" style="width:80%; display: block; margin: 0 auto;">
<p>
There are two representative data production paradigms: (a) data production by humans and (b) data production by AI.
(c) Our proposed new data production paradigm by Human-AI cooperation, named Annotate Anyone.
It serves as a <strong>data flywheel</strong> to produce dynamic up-to-date high-quality data at a low cost.
</p>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Data - CosmicMan-HQ 1.0</h2>
<div class="content has-text-justified">
<img src="./static/imgs/dataset_stat_table.png" alt="Image 1">
<p>
The dataset CosmicMan-HQ 1.0, with 6 Million high-quality real-world human images, will be released soon.
The statistical comparison between publicly available human-related datasets and CosmicMan-HQ 1.0.
</p>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3"><strong>Daring</strong> - <strong>D</strong>ecomposed-<strong>A</strong>ttention-<strong>R</strong>efocus<strong>ing</strong> </h2>
<div class="content has-text-justified">
<img src="./static/imgs/daring.png" alt="Image 1" style="width:80%; display: block; margin: 0 auto;">
<p>
Here we presend the core element of our network. The proposed training framework includes two parts:
(1) data discretion for decomposing text-human data into fixed groups that obey human structure;
(2) a new loss, called HOLA, to enforce the cross-attention features actively response in the proper spatial region with respect
to the scale of body structure and outfit arrangement.
</p>
</div>
</div>
</div>
</section>
<!-- Results. -->
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Foundation Models - Compare with SOTAs</h2>
<h2 class="title is-5">Quantitative Results</h2>
<div class="content has-text-justified">
<img src="./static/imgs/quantitative_comp.png" alt="Image 1">
<p>
</p>
</div>
<h2 class="title is-5">Qualitative Results</h2>
<div class="content has-text-justified">
<img src="./static/imgs/qualitative_comp.png" alt="Image 1">
<img src="./static/imgs/qualitative_comp2.png" alt="Image 1">
<p>
</p>
</div>
</div>
</div>
</section>
<!-- Supp Visualization. -->
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">More Generated Results</h2>
<div class="content has-text-justified">
<div class="image-group">
<img src="./static/imgs/supp_vis/cat1.png" alt="Image 1" style="width:60%;">
<img src="./static/imgs/supp_vis/cat2.png" alt="Image 2" style="width:60%;">
</div>
<div class="image-group">
<img src="./static/imgs/supp_vis/cat3.png" alt="Image 3" style="width:60%;">
<img src="./static/imgs/supp_vis/cat4.png" alt="Image 4" style="width:60%;">
</div>
<div class="image-group">
<img src="./static/imgs/supp_vis/cat5.png" alt="Image 5" style="width:60%;">
<img src="./static/imgs/supp_vis/cat6.png" alt="Image 6" style="width:60%;">
</div>
</div>
</div>
</section>
<!-- Results. -->
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column" style="width: 83.333333%;">
<h2 class="title is-3">Potential Downstream Applications</h2>
<h2 class="title is-5">Gen-2 video produced by CosmicMan and Art Designers</h2>
<div class="content has-text-justified">
<div class="hero-body " padding=-10rem>
<div class="container">
<div id="results-carousel" class="carousel results-carousel">
<div class="item">
<video id="replay-video" controls muted preload autoplay loop width="100%">
<source src="./static/videos/134_0_V.mp4" type="video/mp4">
</video>
<br><br><br>
<div class="text-small" style="margin-top: 10px;">A close up portrait shot of a Caucasian woman, wavy above chest blonde hair, short sleeve white silk floral dress.<br><br> </div>
</div>
<div class="item ">
<video id="replay-video" controls muted preload autoplay loop width="100%">
<source src="./static/videos/166_0_V.mp4" type="video/mp4">
</video>
<br><br><br><br>
<div class="text-small" style="margin-top: 10px;">A headshot, an Asian adult female, a white wall, wavy brown above chest hair. <br><br><br></div>
</div>
<div class="item">
<video id="replay-video" controls muted preload autoplay loop width="100%">
<source src="./static/videos/698_0_V.mp4" type="video/mp4">
</video>
<br><br><br>
<div class="text-small" style="margin-top: 10px;">A headshot, an adult Asian female, fit, a blurry image of a woman's face, floral mesh graphic black normal short sleeve shirt, wavy brown above chest hair.</div>
</div>
<div class="item ">
<video id="replay-video" controls muted preload autoplay loop width="100%">
<source src="./static/videos/754_1_V.mp4" type="video/mp4">
</video>
<br>
<div class="text-small" style="margin-top: 10px;">A headshot, an adult Latino male, fit, a black background, above eyes black hair, normal black cotton solid color short sleeve t-shirt.<br><br>
</div>
</div>
<div class="item ">
<video id="replay-video" controls muted preload autoplay loop width="100%">
<source src="./static/videos/811_0_V.mp4" type="video/mp4">
</video>
<br>
<div class="text-small" style="margin-top: 10px;">A headshot, an adult Caucasian female, fit, a gray wall, black normal solid color cotton long sleeve sweaters, above shoulders straight gray hair.
</div>
</div>
<div class="item ">
<video id="replay-video" controls muted preload autoplay loop width="100%">
<source src="./static/videos/829_0_V.mp4" type="video/mp4">
</video>
<br>
<div class="text-small" style="margin-top: 10px;">A headshot, an adult Caucasian female, fit, a gray wall, black normal solid color cotton long sleeve sweaters, above shoulders straight gray hair.
</div>
</div>
</div>
</div>
</div>
<p>
</p>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Related Work </h2>
<div class="content has-text-justified is-centered">
<ul>
<li>[ECCV 2022] <a href="https://stylegan-human.github.io/" target="_blank" >StyleGAN-Human: A Data-Centric Odyssey of Human Generation</a>, (Jianglin Fu et al.) </li>
</ul>
<!-- <p style=" text-align: center;">Click <a href="" target="_blank" onclick="event.preventDefault();">here</a> to see more images generated by our foundation models.</p> -->
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Acknowledgements </h2>
<div class="content has-text-justified is-centered">
<p>Special thanks to Rongzhang Gu for helping creating demo videos and Yi Zhang for her project management. </p>
</div>
</div>
</div>
</section>
<section class="section" id="Citaton">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code> @article{li2024cosmicman,
title={CosmicMan: A Text-to-Image Foundation Model for Humans},
author={Li, Shikai and Fu, Jianglin and Liu, Kaiyuan and Wang, Wentao and Lin, Kwan-Yee and Wu, Wayne},
year={2024},
journal = {arXiv preprint},
volume = {arXiv:2404.01294}}
</code></pre>
</div>
</section>
<script>
$(window).on('load', function () {
$('#loading').hide();
})
</script>
</body>
</html>