-
Notifications
You must be signed in to change notification settings - Fork 0
/
diffmae.html
468 lines (420 loc) · 16.3 KB
/
diffmae.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
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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description"
content="Diffusion Models as Masked Autoencoder">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>DiffMAE</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<link rel="stylesheet" href="./diffmae/css/bulma.min.css">
<link rel="stylesheet" href="./diffmae/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./diffmae/css/bulma-slider.min.css">
<link rel="stylesheet" href="./diffmae/css/fontawesome.all.min.css">
<link rel="stylesheet"
href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./diffmae/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./diffmae/js/fontawesome.all.min.js"></script>
<script src="./diffmae/js/bulma-carousel.min.js"></script>
<script src="./diffmae/js/bulma-slider.min.js"></script>
<script src="./diffmae/js/index.js"></script>
<style>
.src-image{
margin-bottom: -5%;
}
.dst-image {
margin-top:-95%;
opacity: 0;
transition: 1s ease;
}
.overlay-image:hover .dst-image {
opacity: 1;
}
table {
border-collapse: separate;
border-spacing: 3px;
margin: 0 auto;
}
caption {
padding: 10px;
caption-side: bottom;
font-weight: 1000;
}
</style>
</head>
<body>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1 publication-title">
Diffusion Models as Masked Autoencoders
</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://weichen582.github.io/">Chen Wei</a><sup>1,2</sup>,</span>
<span class="author-block">
<a href="https://karttikeya.github.io/">Karttikeya Mangalam</a><sup>1</sup>,</span>
<span class="author-block">
<a href="http://www.cs.cmu.edu/~poyaoh/">Po-Yao Huang</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://lyttonhao.github.io/">Yanghao Li</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://haoqifan.github.io/">Haoqi Fan</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://howardhsu.github.io//">Hu Xu</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://csrhddlam.github.io/">Huiyu Wang</a><sup>1</sup>
</span>
<span class="author-block">
<a href="https://cihangxie.github.io/">Cihang Xie</a><sup>3</sup>,
</span>
<span class="author-block">
<a href="https://www.cs.jhu.edu/~ayuille/">Alan Yuille</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://feichtenhofer.github.io/">Christoph Feichtenhofer</a><sup>1</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>FAIR, Meta AI</span>,
<span class="author-block"><sup>2</sup>Johns Hopkins University</span>,
<span class="author-block"><sup>3</sup>UC Santa Cruz</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<span class="link-block">
<a href="https://arxiv.org/pdf/2304.03283.pdf"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span>
<span class="link-block">
<a href="https://arxiv.org/abs/2304.03283"
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>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<div class="columns is-centered has-text-centered">
<div class="column" style="padding: 1px">
<img src="./diffmae/images/teaser-1.gif">
<table width="100%">
<tr>
<th width="33.3%">ground-truth</th>
<th width="33.3%"><span class="diffmae">DiffMAE</span></th>
<th width="33.3">MAE</th>
</tr>
</table>
</div>
<div class="column" style="padding: 1px">
<img src="./diffmae/images/teaser-2.gif">
<table width="100%">
<tr>
<th width="33.3%">ground-truth</th>
<th width="33.3%"><span class="diffmae">DiffMAE</span></th>
<th width="33.3">MAE</th>
</tr>
</table>
</div>
</div>
<h2 class="subtitle has-text-centered">
<span class="diffmae">DiffMAE</span> gradually adds visual details by diffusion for masked autoencoding.
</h2>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3"><span class="diffmae">DiffMAE</span></h2>
<div class="content has-text-justified">
<p>
It has been a longstanding belief that generation can facilitate a true understanding
of visual data.
</p>
<p>
In line with this, we revisit generatively pre-training visual representations
in light of denoising diffusion models, and build connection between
diffusion models and masked autoencoders.
</p>
<p>
In particular, we condition diffusion models on masked input and formulate diffusion models
as masked autoencoders (<span class="diffmae">DiffMAE</span>).
Our approach can:
<ul>
<li>Serve as a strong initialization for downstream <em>recognition</em> tasks;</li>
<li>Conduct <em>generative</em> image inpainting;</li>
<li>Be effortlessly extended to <em>video</em>.</li>
</ul>
</p>
</div>
</div>
</div>
<!--/ Abstract. -->
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column is-full-width">
<!-- <h2 class="title is-3">Inpainting</h2> -->
<!-- Random Masking. -->
<h3 class="title is-4">Random Mask Inpainting</h3>
<div class="content has-text-justified has-text-centered">
<p>
The images are from ImageNet-1K validation set.
</p>
<p>
Use the slider to see generations from different inference steps.
</p>
</div>
<!-- Interpolating. -->
<div class="columns is-vcentered is-centered interpolation-panel">
<div class="column is-3 has-text-centered">
<img src="./diffmae/interpolation/input.png"
class="interpolation-image"
alt="Interpolate start reference image."/>
<p>masked input</p>
</div>
<div class="column is-3 is-centered">
<div id="interpolation-image-wrapper">
Loading...
</div>
<input class="slider is-fullwidth is-large is-info"
id="interpolation-slider"
step="1" min="0" max="2" value="0" type="range">
</div>
<div class="column is-3 has-text-centered">
<img src="./diffmae/interpolation/gt.png"
class="interpolation-image"
alt="Interpolation end reference image."/>
<p class="is-bold">ground-truth</p>
</div>
</div>
<br/>
<!--/ Interpolating. -->
<div class="content has-text-justified has-text-centered">
<p>
Hover to view the masked inputs.
</p>
</div>
<div class="columns">
<div class="column overlay-image has-text-centered">
<img src="./diffmae/images/random-0-2.png" class="src-image">
<img src="./diffmae/images/random-0-1.png" class="dst-image">
</div>
<div class="column overlay-image has-text-centered">
<img src="./diffmae/images/random-1-2.png" class="src-image">
<img src="./diffmae/images/random-1-1.png" class="dst-image">
</div>
</div>
<div class="columns">
<div class="column overlay-image has-text-centered">
<img src="./diffmae/images/random-2-2.png" class="src-image">
<img src="./diffmae/images/random-2-1.png" class="dst-image">
<table width="100%">
<tr>
<td width="50%">ground-truth</th>
<td width="50%"><span class="diffmae">DiffMAE</span></th>
</tr>
</table>
</div>
<div class="column overlay-image has-text-centered">
<img src="./diffmae/images/random-3-2.png" class="src-image">
<img src="./diffmae/images/random-3-1.png" class="dst-image">
<table width="100%">
<tr>
<td width="50%">ground-truth</th>
<td width="50%"><span class="diffmae">DiffMAE</span></th>
</tr>
</table>
</div>
</div>
<!--/ Random Masking. -->
<!-- Center Masking. -->
<h3 class="title is-4">Center Mask Inpainting</h3>
<div class="content has-text-justified has-text-centered">
<p>
The images are from ImageNet-1K validation set.
Hover to view the masked inputs.
</p>
</div>
<div class="columns">
<div class="column overlay-image has-text-centered">
<img src="./diffmae/images/center-0-2.png" class="src-image" width="94%" height="94%" >
<img src="./diffmae/images/center-0-1.png" class="dst-image" width="94%" height="94%">
</div>
<div class="column overlay-image has-text-centered">
<img src="./diffmae/images/center-1-2.png" class="src-image" width="94%" height="94%">
<img src="./diffmae/images/center-1-1.png" class="dst-image" width="94%" height="94%">
</div>
</div>
<div class="columns">
<div class="column overlay-image has-text-centered">
<img src="./diffmae/images/center-2-2.png" class="src-image" width="94%" height="94%">
<img src="./diffmae/images/center-2-1.png" class="dst-image" width="94%" height="94%">
<table width="100%">
<tr>
<td width="50%">ground-truth</th>
<td width="50%"><span class="diffmae">DiffMAE</span></th>
</tr>
</table>
</div>
<div class="column overlay-image has-text-centered">
<img src="./diffmae/images/center-3-2.png" class="src-image" width="94%" height="94%">
<img src="./diffmae/images/center-3-1.png" class="dst-image" width="94%" height="94%">
<table width="100%">
<tr>
<td width="50%">ground-truth</th>
<td width="50%"><span class="diffmae">DiffMAE</span></th>
</tr>
</table>
</div>
</div>
<!--/ Center Masking. -->
<!-- Video Inpainting. -->
<h3 class="title is-4">Video Inpainting</h3>
<div class="content has-text-justified has-text-centered">
<p>
The videos are from Kinetics-400 validation set.
</p>
</div>
<div class="columns">
<div class="column has-text-centered">
<img src="./diffmae/images/video-0.gif" width="95%" height="95%" >
<table width="100%">
<tr>
<td width="33.3%">ground-truth</td>
<td width="33.3%">inputs</td>
<td width="33.3"><span class="diffmae">DiffMAE</span></td>
</tr>
</table>
</div>
<div class="column has-text-centered">
<img src="./diffmae/images/video-1.gif" width="95%" height="95%">
<table width="100%">
<tr>
<td width="33.3%">ground-truth</td>
<td width="33.3%">inputs</td>
<td width="33.3"><span class="diffmae">DiffMAE</span></td>
</tr>
</table>
</div>
</div>
<!--/ Video Inpainting. -->
<!-- Fine-Tuning. -->
<h3 class="title is-4">Fine-Tuning Generative Models</h3>
<div class="content has-text-justified has-text-centered">
<p>
While being able to generatively inpaint images,
<strong><span class="diffmae">DiffMAE</span></strong> is a strong
self-supervised pre-training approach. The performance is:
<ul>
<li>Comparable to leading self-supervised algorithms that focus solely on recognition;</li>
<li>Stronger than other generative based algorithms by a large margin.</li>
</ul>
</p>
</div>
<div class="is-centered has-text-centered">
<table style="width: 70%">
<tr>
<th>pre-train</th>
<th>architecture</th>
<th>params. (M)</th>
<th>fine-tuned</th>
</tr>
<tr>
<td colspan="4" style="border-bottom: 1px solid #ddd;"></td>
</tr>
<tr>
<td style="color:lightgray">MAE</td>
<td style="color:lightgray">ViT-L</td>
<td style="color:lightgray">304</td>
<td style="color:lightgray">85.9</td>
</tr>
<tr>
<td colspan="4" style="border-bottom: 1px solid #ddd;"></td>
</tr>
<tr>
<td>iGPT</td>
<td>iGPT-L</td>
<td>1362</td>
<td>72.6</td>
</tr>
<tr>
<td>ADM</td>
<td>U-Net</td>
<td>211</td>
<td>83.3</td>
</tr>
<tr>
<td>DDPM</td>
<td>ViT-L</td>
<td>304</td>
<td>83.4</td>
</tr>
<tr>
<td><strong><span class="diffmae">DiffMAE</span></strong></td>
<td>ViT-L</td>
<td>304</td>
<td><strong>85.8</strong></td>
</tr>
<caption>
Fine-tuning generative models on ImageNet-1K, a system-level comparison.
</caption>
</table>
</div>
</div>
<!--/ Fine-Tuning. -->
<br/>
</div>
</section>
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>@inproceedings{wei2023diffusion,
author = {Wei, Chen and Mangalam, Karttikeya and Huang, Po-Yao and Li, Yanghao and Fan, Haoqi and Xu, Hu and Wang, Huiyu and Xie, Cihang and Yuille, Alan and Feichtenhofer, Christoph},
title = {Diffusion Models as Masked Autoencoder},
booktitle = {ICCV},
year = {2023},
}</code></pre>
</div>
</section>
<footer class="footer">
<div class="container">
<div class="columns is-centered">
<div class="content">
<p>
<a href="https://github.com/nerfies/nerfies.github.io">Template</a>
</p>
</div>
</div>
</div>
</footer>
</body>
</html>