-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
521 lines (471 loc) · 29.7 KB
/
index.html
File metadata and controls
521 lines (471 loc) · 29.7 KB
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
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
<!DOCTYPE HTML>
<!--
Miniport by HTML5 UP
html5up.net | @ajlkn
Free for personal and commercial use under the CCA 3.0 license (html5up.net/license)
-->
<html>
<head>
<title>SiTH</title>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<link rel="stylesheet" href="assets/css/bulma.min.css">
<link rel="stylesheet" href="assets/css/bulma-carousel.min.css">
<link rel="stylesheet" href="assets/css/bulma-slider.min.css">
<link rel="stylesheet" href="assets/css/bootstrap.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="assets/css/font-awesome.min.css">
<link rel="stylesheet" href="assets/css/index.css">
<link rel="stylesheet" href="assets/css/style.css" type="text/css" media="screen,projection" />
<link rel="stylesheet" href="assets/css/main.css" />
<link rel="shortcut icon" href="images/opt_icon.png" />
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script src="assets/js/bulma-carousel.min.js"></script>
<script src="assets/js/bulma-slider.min.js"></script>
<script src="assets/js/index.js"></script>
<script src="assets/js/video_comparison.js"></script>
</head>
<body>
<!-- Home -->
<div class="wrapper first style1">
<article class="container" id="home">
<img class="responsive-img" src="images/icon/sith.png" width="300px">
<h2> Single-view Textured Human Reconstruction <br> with Image-Conditioned Diffusion</h2>
<br>
<div class="row center">
<div class="author col l4 m6 s12"><a href="https://azuxmioy.github.io/" target="_blank">Hsuan-I Ho</a></div>
<div class="author col l4 m6 s12"><a href="https://ait.ethz.ch/people/song" target="_blank">Jie Song</a></div>
<div class="author col l4 m6 s12"><a href="https://ait.ethz.ch/people/hilliges" target="_blank">Otmar Hilliges</a></div>
</div>
<br>
<div class="col">
<div style="width: 20%; margin: 0em auto 0em auto">
<a href="https://ait.ethz.ch/"><img class="responsive-img" src="images/ait_logo.png"></a>
</div>
<div class="institute">
<a href="https://inf.ethz.ch/" target="_blank">Department of Computer Science</a>,
<a href="https://ethz.ch/en.html" target="_blank">ETH Zürich</a>
</div>
</div>
<br>
<div class="row">
<div class="col-sm-6 col-sm-offset-3 text-center">
<ul class="nav nav-pills nav-justified">
<li>
<a href="https://arxiv.org/abs/2311.15855">
<img src="images/icon/darth.png" width="60px">
<h4><strong>Paper<br>(CVPR 2024)</strong></h4>
</a>
</li>
<li>
<a href="https://youtu.be/gixakzI9UcM">
<img src="images/icon/Yt_icon_2017_print.png" width="60px">
<h4><strong>Video</strong></h4>
</a>
</li>
<li>
<a href="https://github.com/SiTH-Diffusion/SiTH" target="_blank">
<image src="images/icon/github.png" width="60px">
<h4><strong>Code<br></strong></h4>
</a>
</li>
<li>
<a href="https://ait.ethz.ch/sith-demo" target="_blank">
<image src="images/icon/hf.png" width="60px">
<h4><strong>Demo<br></strong></h4>
</a>
</li>
</ul>
</div>
</div>
</article>
<div class="container" style="width:40%; margin: 2em auto 2em auto">
<h2>News:</h2>
<h4>[June 14, 2024] Release training scripts.</h4>
<h4>[May 15, 2024] Update an application of 3D avatar animation.</h4>
<h4>[April 24, 2024] Gradio demo for 3D human creation is now available.</h4>
<h4>[April 15, 2024] Release demo code, models, and the evaluation benchmark.</h4>
</div>
</div>
<!-- Abstract -->
<div class="wrapper style1">
<article id="abstract">
<div class="container">
<h2>TL;DR</h2>
<ul class="default" style="text-align:left">
<li>SiTH is a <b>non-optimization</b> single-view 3D reconstruction pipeline integrating a novel <b>image-conditioned ControlNet model</b> and implicit neural fields.</li>
<li>SiTH reconstructs a <b>fully-textured</b> and high-quality 3D human mesh from a single image in <b>2 minutes</b>.</li>
<li>SiTH can be trained with as few as <b>500 3D human scans</b> and generalizes well to diverse images such as AI-generated images. (<b>Check out our demo!</b>)</li>
<li>To foster future research, we contribute a <b>new evaluation benchmark</b> for single-view 3D human reconstruction.</li>
</ul>
<div class="row center" style="width: 100%; margin: 2em auto 2em auto">
<div class="col s10 offset-s1">
<img class="responsive-img" src="images/teaser.jpg" style="width: 100%"/>
</div>
</div>
<h2>Abstract</h2>
<p style="text-align:justify">
A long-standing goal of 3D human reconstruction is to create lifelike and fully detailed 3D humans from single-view images.
The main challenge lies in inferring unknown body shapes, appearances, and clothing details in areas not visible in the images.
To address this, we propose SiTH, a novel pipeline that uniquely integrates an image-conditioned diffusion model into a 3D mesh reconstruction workflow.
At the core of our method lies the decomposition of the challenging single-view reconstruction problem into generative hallucination and reconstruction subproblems.
For the former, we employ a powerful generative diffusion model to hallucinate unseen back-view appearance based on the input images.
For the latter, we leverage skinned body meshes as guidance to recover full-body texture meshes from the input and back-view images.
SiTH requires as few as 500 3D human scans for training while maintaining its generality and robustness to diverse images.
Extensive evaluations on two 3D human benchmarks, including our newly created one, highlighted our method's superior accuracy and perceptual quality in 3D textured human reconstruction.
</p>
</div>
</article>
</div>
<!-- Paper -->
<div class="wrapper style1">
<article id="paper">
<div class="container">
<h2>Video</h2>
<div class="row center" style="margin: 2em auto 2em auto">
<div class="col s10 offset-s1">
<div class="youtube-wrapper">
<iframe allowfullscreen="allowfullscreen" src="https://www.youtube.com/embed/gixakzI9UcM"></iframe>
</div>
</div>
</div>
<h2>Method Overview</h2>
<div class="row center" style="width: 100%; margin: 2em auto 2em auto">
<div class="col s10 offset-s1">
<img class="responsive-img" src="images/framework.jpg" style="width: 100%"/>
</div>
</div>
<h3>Hallucination and Reconstruction</h3>
<p style="text-align:justify">
At the core of SiTH is the decomposition of the challenging single-view problem into two subproblems:
generative back-view hallucination and mesh reconstruction. For hallucination, we harness the generative
capabilities of diffusion models to infer unobserved back-view appearances from the input images.
For reconstruction, we utilize a skinned body mesh, providing essential 3D guidance for accurate human mesh reconstruction.
This decomposition strategy allows our pipeline to be trained efficiently with just 500 3D scans,
while still robustly handling unseen images.
</p>
<div class="row center" style="width: 100%; margin: 2em auto 2em auto">
<div class="col s6 offset-s3">
<img class="responsive-img" src="images/diffusion.jpg" style="width: 100%"/>
</div>
</div>
<h3>Image-conditioned Diffusion</h3>
<p style="text-align:justify">
Compared to text-conditioning in traditional diffusion models, our image-conditioning strategy is more consistent and accurate,
allowing for building a data-driven 3D reconstruction pipeline. First, the pretrained CLIP and VAE encoders ensure the output images
maintain visual consistency with the front-view images. Additionally, we render UV maps from the SMPL-X body mesh and extract
silhouette masks from the back-view images. These additional conditional controls ensure the human poses in the output images match
those in the front view. To preserve the diffusion model's generative power, we specifically optimize the ControlNet and
cross-attention layers with only a small amount of 3D human data.
</p>
</div>
</article>
</div>
<!-- Results -->
<div class="wrapper style1">
<article id="results">
<div class="container">
<h2>Results</h2>
<h3>Single-view 3D Human Reconstruction</h3>
<div class="row center" style="width: 100%; margin: 1em auto 1em auto" >
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh1" loop autoplay muted src="videos/000.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh1Merge"></canvas>
<br>
<a onclick="document.getElementById('input000').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/000.jpg" id="input000"/>
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh2" loop autoplay muted src="videos/001.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh2Merge"></canvas>
<br>
<a onclick="document.getElementById('input001').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/001.jpg" id="input001"/>
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh3" loop autoplay muted src="videos/002.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh3Merge"></canvas>
<br>
<a onclick="document.getElementById('input002').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/002.jpg" id="input002"/>
</div>
</div>
<div class="row center" style="width: 100%; margin: 1em auto 1em auto" >
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh4" loop autoplay muted src="videos/003.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh4Merge"></canvas>
<br>
<a onclick="document.getElementById('input003').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/003.jpg" id="input003"/>
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh5" loop autoplay muted src="videos/004.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh5Merge"></canvas>
<br>
<a onclick="document.getElementById('input004').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/004.jpg" id="input004"/>
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh6" loop autoplay muted src="videos/005.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh6Merge"></canvas>
<br>
<a onclick="document.getElementById('input005').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/005.jpg" id="input005"/>
</div>
</div>
<div class="row center" style="width: 100%; margin: 1em auto 1em auto" >
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh7" loop autoplay muted src="videos/006.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh7Merge"></canvas>
<br>
<a onclick="document.getElementById('input006').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/006.jpg" id="input006"/>
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh8" loop autoplay muted src="videos/007.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh8Merge"></canvas>
<br>
<a onclick="document.getElementById('input007').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/007.jpg" id="input007"/>
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh9" loop autoplay muted src="videos/008.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh9Merge"></canvas>
<br>
<a onclick="document.getElementById('input008').click();">Show input image</a>
<img class="materialboxed responsive-img" src="images/008.jpg" id="input008"/>
</div>
</div>
<h3>Text-guided 3D Human Creation</h3>
<div class="row center" style="width: 50%; margin: 2em auto 2em auto">
<div class="col s6 offset-s3">
<img class="responsive-img" src="images/sith_demo.gif" style="width: 100%"/>
</div>
</div>
<h4 style="text-align: center">SiTH can be easily combined with powerful 2D text-to-image models.</h4>
<div class="row center" style="width: 100%; margin: 1em auto 1em auto" >
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh10" loop autoplay muted src="videos/011.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh10Merge"></canvas>
<br>
"A male wearing a white suit"
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh11" loop autoplay muted src="videos/012.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh11Merge"></canvas>
<br>
"A man wearing brown t-shirt and pants"
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesh12" loop autoplay muted src="videos/013.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesh12Merge"></canvas>
<br>
"A woman in a white t-shirt and a tennis skirt"
</div>
</div>
<h3>3D Avatars</h3>
<div class="row center " style="width: 75%; margin: 2em auto 2em auto">
<div class="col s12">
<img class="responsive-img" src="images/animation.gif" style="width: 100%"/>
</div>
</div>
<h4 style="text-align: center">SiTH can easily create animation-ready 3D avatars from images.</h4>
</div>
</article>
</div>
<!-- Download -->
<div class="wrapper style1">
<article id="benchmark">
<div class="container">
<h2>Benchmark Evaluation</h2>
The benchmark is based on the <a href="https://custom-humans.github.io/#download" target="_blank">CustomHumans</a> dataet. To access the benchmark, please apply the dataset directly.
<br>
<br>
<h3>Single-view 3D Human Reconstruction</h3>
<div class="table-wrapper" align="center">
<table>
<thead>
<tr>
<th>Methods</th>
<th>P-to-S (cm) ↓</th>
<th>S-to-P (cm) ↓</th>
<th>NC ↑</th>
<th>f-Score ↑</th>
</tr>
</thead>
<tbody>
<tr>
<td><h6>PIFu <a href="https://github.com/shunsukesaito/PIFu">[Saito2019]</a></h6></td>
<td>2.209</td>
<td>2.582</td>
<td>0.805</td>
<td>34.881</td>
</tr>
<tr>
<td><h6>PIFuHD <a href="https://github.com/facebookresearch/pifuhd">[Saito2020]</a></h6></td>
<td>2.107</td>
<td><ins>2.228</ins></td>
<td>0.804</td>
<td><strong>39.076</strong></td>
</tr>
<tr>
<td><h6>PaMIR <a href="https://github.com/ZhengZerong/PaMIR">[Zheng2021]</a></h6></td>
<td>2.181</td>
<td>2.507</td>
<td><ins>0.813</ins></td>
<td>35.847</td>
</tr>
<tr>
<td><h6>FOF <a href="https://github.com/fengq1a0/FOF">[Feng2022]</a></h6></td>
<td><ins>2.079</ins></td>
<td>2.644</td>
<td>0.808</td>
<td>36.013</td>
</tr>
<tr>
<td><h6>2K2K <a href="https://github.com/SangHunHan92/2K2K">[Han2023]</a></h6></td>
<td>2.488</td>
<td>3.292</td>
<td>0.796</td>
<td>30.186</td>
</tr>
<tr>
<td><h6>ICON* <a href="https://github.com/YuliangXiu/ICON">[Xiu2022]</a></h6></td>
<td>2.256</td>
<td>2.795</td>
<td>0.791</td>
<td>30.437</td>
</tr>
<tr>
<td><h6>ECON* <a href="https://github.com/fengq1a0/FOF">[Xiu2023]</a></h6></td>
<td>2.483</td>
<td>2.680</td>
<td>0.797</td>
<td>30.894</td>
</tr>
<tr>
<td><h6>SiTH* (Ours)</h6></td>
<td><strong>1.871</strong></td>
<td><strong>2.045</strong></td>
<td><strong>0.826</strong></td>
<td><ins>37.029</ins></td>
</tr>
</tbody>
</table>
<h4>*indicates methods trained on the same THuman2.0 dataset.</h4>
</div>
<div class="row center" style="width: 100%; margin: 2em auto 2em auto">
<div class="col s12">
<img class="responsive-img" src="images/benchmark_3D.jpg" style="width: 100%"/>
</div>
</div>
<h3>Comparison with Optimization-based (Score Distillation) Approaches</h3>
<div class="row center" style="width: 100%; margin: 1em auto 1em auto" >
<div class="col l4 m6 s12 text-center">
<video class="video" id="mesha" loop autoplay muted src="videos/a.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="meshaMerge"></canvas>
<br>
<h5>Zero-1-to-3, 10 mins<h5>
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="meshb" loop autoplay muted src="videos/b.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="meshbMerge"></canvas>
<br>
<h5>Magic-1-to-3, 6 hours<h5>
</div>
<div class="col l4 m6 s12 text-center">
<video class="video" id="meshc" loop autoplay muted src="videos/c.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="meshcMerge"></canvas>
<br>
<h5>TeCH, 6 hours<h5>
</div>
</div>
<div class="row center" style="width: 100%; margin: 1em auto 1em auto" >
<div class="col l6 m6 s12 text-center">
<video class="video" id="meshd" loop autoplay muted src="videos/d.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="meshdMerge"></canvas>
<br>
<h5>SiTH (Ours), 2 mins<h5>
</div>
<div class="col l6 m6 s12 text-center">
<video class="video" id="meshe" loop autoplay muted src="videos/e.mp4" onplay="resizeAndPlay(this)"></video>
<canvas height=0 class="videoMerge" id="mesheMerge"></canvas>
<br>
<h5>GT Reference<h5>
</div>
</div>
<div class="row center" style="width: 70%; margin: 2em auto 2em auto">
<div class="col s12">
<img class="responsive-img" src="images/vs_tech.jpg" style="width: 100%"/>
</div>
<h4>Compared to optimization-based methods (Score Distillation),
SiTH reconstructs consistent facial textures that align with the input images and their underlying geometry.</h4>
</div>
</article>
</div>
<!-- Reference -->
<div class="wrapper style2">
<article id="references" class="container">
<header>
<h3>References</h3>
</header>
<ul class="default" style="text-align:left">
<li>
Ho et. al, "<a href="https://custom-humans.github.io/" target="_blank">Learning Locally Editable Virtual Humans</a>." CVPR, 2023.
</li>
<li>
Kim et. al, "<a href="https://snuvclab.github.io/chupa/" target="_blank">Chupa : Carving 3D Clothed Humans from Skinned Shape Priors using 2D Diffusion Probabilistic Models</a>." ICCV, 2023.
</li>
<li>
Liu et. al, "<a href="https://zero123.cs.columbia.edu/" target="_blank">Zero-1-to-3: Zero-shot One Image to 3D Object</a>." ICCV, 2023.
</li>
<li>
Qian et. al, "<a href="https://guochengqian.github.io/project/magic123/" target="_blank">Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors</a>." ICLR, 2024.
</li>
<li>
Huang et. al, "<a href="https://huangyangyi.github.io/TeCH/" target="_blank">TeCH: Text-guided Reconstruction of Lifelike Clothed Humans</a>." 3DV, 2024.
</li>
</ul>
<div class="subtitle">
<h3>Bibtex</h3>
<pre><cite>@inproceedings{ho2024sith,
title={SiTH: Single-view Textured Human Reconstruction with Image-Conditioned Diffusion},
author={Ho, Hsuan-I and Song, Jie and Hilliges, Otmar},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2024}
}</cite></pre>
</div>
</article>
</div>
<!-- Acknowledgement -->
<div class="wrapper style3">
<article>
<div class="container">
<h2>Acknowledgement</h2>
<p style="text-align:left">
This work was partially supported by the Swiss SERI Consolidation Grant "AI-PERCEIVE".
We thank Xu Chen for insightful discussions, Manuel Kaufmann for suggestions on writing and the title,
and Christoph Gebhardt, Marcel Buehler, and Juan-Ting Lin for their writing advice.</p>
</div>
</article>
</div>
<!-- Copyright -->
<div class="wrapper last style4">
<article class="container">
<footer>
<ul id="copyright">
<li>© 2023 Hsuan-I Ho.</li><li>Thanks: <a href="http://html5up.net" target="_blank">HTML5 UP</a> and Po-Chen Wu provides this template</li>
</ul>
</footer>
</article>
</div>
<!-- Scripts -->
<script src="assets/js/jquery.min.js"></script>
<script src="assets/js/jquery.scrolly.min.js"></script>
<script src="assets/js/skel.min.js"></script>
<script src="assets/js/skel-viewport.min.js"></script>
<script src="assets/js/util.js"></script>
<script src="assets/js/main.js"></script>
<script src="assets/js/materialize.js"></script>
</body>
</html>