-
Notifications
You must be signed in to change notification settings - Fork 4
Expand file tree
/
Copy pathindex.html
More file actions
485 lines (437 loc) · 24.1 KB
/
index.html
File metadata and controls
485 lines (437 loc) · 24.1 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
<!DOCTYPE html>
<html lang="en">
<head>
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-M09001M6S2"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-M09001M6S2');
</script>
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
<meta name="description" content="Compositional Visual Generation with Composable Diffusion Models">
<meta name="keywords" content="EBM,Diffusion,GenerativeModels">
<title>Compositional Visual Generation with Composable Diffusion Models</title>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<link rel="stylesheet" href="website_files/bootstrap.css">
<script src="website_files/jquery.js"></script>
<script src="website_files/bootstrap.js"></script>
<style>
/* Remove the navbar's default margin-bottom and rounded borders */
.navbar {
margin-bottom: 0;
border-radius: 0;
}
/* Add a gray background color and some padding to the footer */
footer {
background-color: #f2f2f2;
padding: 25px;
}
</style>
<script>
function setImageGLIDE(select){
var image = document.getElementsByName("image-swap-1")[0];
image.src = select.options[select.selectedIndex].value;
}
function setImageStable(select){
var image = document.getElementsByName("image-swap-2")[0];
image.src = select.options[select.selectedIndex].value;
}
</script>
<link rel="stylesheet" href="website_files/font.css">
<link rel="stylesheet" href="website_files/main.css">
</head>
<body>
<div class="jumbotron">
<div class="container text-center">
<h1 style="color:white;margin-bottom:0;">Composable-Diffusion</h1>
<h3 style="color:white;margin-top:0;">Compositional Visual Generation with Composable Diffusion Models</h3>
<h3 style="color:white;margin-top:0;">ECCV 2022</h3>
<br>
<p style="color:white"><a href="https://nanliu.io/">Nan Liu</a><sup>1*</sup>, <a href="https://shuangli59.github.io/">Shuang Li</a><sup>2*</sup>,
<a href="https://yilundu.github.io">Yilun Du</a><sup>2*</sup>, <a href="https://groups.csail.mit.edu/vision/torralbalab/">Antonio Torralba</a><sup>2</sup>,
<a href="https://mitibmwatsonailab.mit.edu/people/joshua-tenenbaum/">Joshua B. Tenenbaum</a><sup>2</sup>
<br><sup>1</sup>UIUC, <sup>2</sup>MIT
<br>
<a>(* indicate equal contribution)</a>
<br>
</div>
</div>
<div class="container bg-3">
<div class="row text-center">
<!-- <div class="col-sm-2 col-sm-offset-1">
<a href="https://www.youtube.com/watch?v=_zN-wVwPH1s"><img height="78" width="120" src="./NeRF-Supervision_files/video-thumb.png" data-nothumb="" style="border: 1px solid;"><br>Overview<br>Video<br></a>
</div> -->
<div class="col-sm-2 col-sm-offset-1">
<a href="https://arxiv.org/abs/2206.01714"><img height="100" width="78" src="website_files/imgs/paper_page.png" data-nothumb="" style="border: 1px solid"><br>ECCV 2022<br>Paper<br></a>
</div>
<div class="col-sm-2">
<a href="https://github.com/energy-based-model/Compositional-Visual-Generation-with-Composable-Diffusion-Models-PyTorch"><img height="100" width="78" src="website_files/imgs/source_code.png" data-nothumb="" style="border: 1px solid"><br>Source Code<br>Github</a>
</div>
<div class="col-sm-2">
<a href="https://colab.research.google.com/github/energy-based-model/Compositional-Visual-Generation-with-Composable-Diffusion-Models-PyTorch/blob/main/notebooks/demo.ipynb"><img height="78" width="120" src="website_files/imgs/colab.png" data-nothumb="" style="border: 1px solid"><br>Colab Demo:<br>Composable Diffusion Models</a>
</div>
<div class="col-sm-2">
<a href="https://huggingface.co/spaces/Shuang59/Composable-Diffusion"><img height="78" width="78" src="website_files/imgs/huggingface.png" data-nothumb="" style="border: 1px solid"><br>Hugging Face<br>Demo</a>
</div>
<div class="col-sm-2">
<a href="#results"><img height="100" width="130" src="website_files/imgs/website_teaser.png" data-nothumb="" style="border: 1px solid"><br>Results</a>
</div>
</div>
</div><br>
<div class="container bg-3">
<hr>
<div class="row">
<h2 class="text-center">Stable Diffusion Demos - <span class="and">AND</span> + <span class="not">NOT</span> (negative prompts) </h2>
<br />
<p>
<b>Compositional Generation using Stable Diffusion.</b> Our proposed Conjunction (<span class="and">AND</span>) and Negation (<span class="not">NOT</span>)
can be applied to conditional diffusion models for compositional generation. Our operators are added into <a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui">Stable Diffusion WebUI</a>!
Corresponding pages are as follows: <a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui/commit/c26732fbee2a57e621ac22bf70decf7496daa4cd">Conjunction (<span class="and">AND</span>) </a> and
<a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Negative-prompt">Negation (<span class="not">NOT</span>)</a>.
</p>
<hr>
<div class="col-sm-3 text-center">
<img src="website_files/imgs/videos/example2_A.gif" class="stable_video" alt="">
<p>"mystical trees" <span class="and">AND</span> "A magical pond" <span class="and">AND</span> "Dark"</p>
</div>
<div class="col-sm-3 text-center">
<img src="website_files/imgs/videos/example2_N.gif" class="stable_video" alt="">
<p>"mystical trees" <span class="and">AND</span> "A magical pond" <span class="and">AND</span> <span class="not">NOT</span> "Dark"</p>
</div>
<div class="col-sm-3 text-center">
<img src="website_files/imgs/videos/example1_A.gif" class="stable_video" alt="">
<p>"A stone castle surrounded by lakes and trees, fantasy, wallpaper, concept art, extremely detailed" <span class="and">AND</span> "Black and white"</p>
</div>
<div class="col-sm-3 text-center">
<img src="website_files/imgs/videos/example1_N.gif" class="stable_video" alt="">
<p>"A stone castle surrounded by lakes and trees, fantasy, wallpaper, concept art <span class="and">AND</span> extremely detailed" <span class="and">AND</span> <span class="not">NOT</span> "Black and white"</p>
</div>
</div>
</div>
<br />
<div class="container bg-3">
<hr>
<h2 class="text-center">Point-E Demos - <span class="and">AND</span> + <span class="not">NOT</span> (negative prompts) </h2>
<br />
<p>
<b>Compositional 3D Generation using <a href="https://arxiv.org/abs/2212.08751">Point-E</a></b> using our proposed Conjunction (<span class="and">AND</span>) and and Negation (<span class="not">NOT</span>) operators.
</p>
<div class="row">
<div class="col-sm-4 text-center">
<img src="website_files/imgs/videos/a%20green%20avocado_a%20chair.gif" class="stable_video" alt="">
<p>"A green avocado" <span class="and">AND</span> "A chair"
</div>
<div class="col-sm-4 text-center">
<img src="website_files/imgs/videos/a%20chair_and_not%20chair_legs.gif" class="stable_video" alt="">
<p>"A chair" <span class="and">AND</span> <span class="not">NOT</span> "Chair legs"
</div>
<div class="col-sm-4 text-center">
<img src="website_files/imgs/videos/a%20toilet_a%20chair.gif" class="stable_video" alt="">
<p>"A toilet" <span class="and">AND</span> "A chair"
</div>
</div>
<div class="row">
<div class="col-sm-4 text-center">
<img src="website_files/imgs/videos/a%20couch_a%20boat.gif" class="stable_video" alt="">
<p>"A couch" <span class="and">AND</span> "A boat"
</div>
<div class="col-sm-4 text-center">
<img src="website_files/imgs/videos/a%20monitor_a%20brown%20couch.gif" class="stable_video" alt="">
<p>"A monitor" <span class="and">AND</span> "A brown couch"
</div>
<div class="col-sm-4 text-center">
<img src="website_files/imgs/videos/a%20chair_a%20cake.gif" class="stable_video" alt="">
<p>"A chair" <span class="and">AND</span> "A cake"
</div>
</div>
</div>
<br />
<div class="container bg-3">
<div class="row">
<h2 class="text-center">Video Demos</h2>
<hr>
<div class="col-sm-6 text-center">
<p><b>Compose natural language descriptions:</b></p>
<video autoplay="" muted="" playsinline="" class="img-responsive">
<source src="website_files/imgs/teaser_glide_1.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="col-sm-6 text-center">
<p><b>Compose natural language descriptions:</b></p>
<video autoplay="" muted="" playsinline="" class="img-responsive">
<source src="website_files/imgs/teaser_glide.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<br>
<div class="row">
<div class="col-sm-6 text-center">
<p><b>Compose objects:</b></p>
<video autoplay="" muted="" playsinline="" class="img-responsive">
<source src="website_files/imgs/clevr_pos_teaser.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="col-sm-6 text-center">
<p><b>Compose object relational descriptions:</b></p>
<video autoplay="" muted="" playsinline="" class="img-responsive">
<source src="website_files/imgs/clevr_rel_teaser.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div><br>
</div>
<br><br>
<div class="container">
<h2 id="interactive_demo_stable_diffusion" class="text-center">Interactive Demos (Stable Diffusion)</h2>
<hr>
<div class="row">
<div class="col-sm-12 text-center">
<div class="section">
<div class="section" style="text-align: center">
<span><strong>Text Prompt: </strong></span>
<label for="composed_stable_diffusion">
<select id='composed_stable_diffusion' class="dropselect" style="text-align: center" name="composed_stale_diffusion_generation" onchange="setImageStable(this);">
<option value="website_files/imgs/stable_4.png" selected="">A cloudy blue sky AND A mountain in the horizon AND Cherry Blossoms around the mountain</option>
<option value="website_files/imgs/stable_1.png">A horse AND a yellow flower field</option>
<option value="website_files/imgs/stable_2.png">A train on a bridge AND a river under the bridge</option>
<option value="website_files/imgs/stable_3.png">A river leading into mountains AND red trees on the side</option>
<option value="website_files/imgs/stable_5.png">A house on a hill AND clouds in the background</option>
<option value="website_files/imgs/stable_6.png">A red house next to a pond AND yellow trees next to the pond</option>
<option value="website_files/imgs/stable_7.png">A church on a hill AND a river next to the church</option>
</select>
</label>
</div>
<div class="row align-items-center">
<div class="col justify-content-center text-center">
<img src="website_files/imgs/stable_4.png"
class="img-fluid" alt="" style="width:100%; height:100%" name="image-swap-2">
</div>
</div>
</div>
</div>
</div>
</div><br>
<div class="container">
<h2 id="interactive_demo_glide" class="text-center">Interactive Demos (GLIDE)</h2>
<hr>
<div class="row">
<div class="col-sm-12 text-center">
<div class="section">
<div class="section" style="text-align: center">
<span><strong>Text Prompt: </strong></span>
<label for="composed_glide">
<select id='composed_glide' class="dropselect" style="text-align: center" name="composed_glide_generation" onchange="setImageGLIDE(this);">
<option value="website_files/imgs/1.png" selected="">A horse AND a yellow flower field</option>
<option value="website_files/imgs/2.png">A train on a bridge AND a river under the bridge</option>
<option value="website_files/imgs/3.png">A river leading into mountains AND red trees on the side</option>
<option value="website_files/imgs/4.png">A cloudy blue sky AND A mountain in the horizon AND Cherry Blossoms in front of the mountain</option>
<option value="website_files/imgs/5.png">A house on a hill AND clouds in the background</option>
<option value="website_files/imgs/6.png">A red house next to a pond AND green trees next to the pond</option>
<option value="website_files/imgs/7.png">A church on a hill AND a road in front of a hill</option>
<option value="website_files/imgs/8.png">A tree on a hill AND a wheat field in front of a hill</option>
</select>
</label>
</div>
<div class="row align-items-center">
<div class="col justify-content-center text-center">
<img src="website_files/imgs/1.png"
class="img-fluid" alt="" style="width:100%; height:100%" name="image-swap-1">
</div>
</div>
</div>
</div>
</div>
</div><br>
<div class="container bg-3">
<div class="row">
<h2 class="text-center">Method</h2>
<hr>
<p><b>Compositional generation.</b> Our method can compose multiple diffusion models during inference and generate images containing all the concepts described in the inputs without further training. We first send an image from iteration \(t\) and each individual concept \(c_i\) to the diffusion model to generate a set of scores \(\{\epsilon_\theta(\mathbf{x}_t, t|c_1), \ldots, \epsilon_\theta(\mathbf{x}_t, t|c_n)\}\). We then compose different concepts using the proposed compositional operators, such as conjunction, to denoise the generated images. The final image is obtained after \(T\) iterations.</p>
<div class="col-sm-12 text-center">
<!-- <img src="website_files/imgs/models-medium.gif" alt="Computer man" style="width:1250px;height:250px;border:0.5px solid gray"> -->
<!-- <img src="website_files/imgs/models-medium.gif" alt="Computer man" style="width:1200px;height:250px;"> -->
<!-- <img src="website_files/imgs/models-medium.gif" alt="Computer man" style="width:100%;"> -->
<video autoplay="" muted="" playsinline="" class="img-responsive" controls="">
<source src="website_files/imgs/models-medium.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</div><br>
<div class="container bg-3">
<div class="row">
<h2 class="text-center" id="results">More Results</h2>
<hr>
<p>
<b>Composing Language Descriptions.</b> We develop Composed GLIDE (Ours), a version of GLIDE
that utilizes our compositional operators to combine textual descriptions, without further training. We compare it
to the original GLIDE, which directly encodes the descriptions as a single long sentence. Our approach more
accurately captures text details, such as the "overwater bungalows" in the third example.
</p>
<div class="col-sm-12 text-center">
<img src="website_files/imgs/glide.png" class="img-fluid" alt="glide" style="width:88%">
</div>
</div>
<div class="row">
<hr>
<p>
<b>Composing Objects.</b> Our method can compose multiple objects while baselines either miss or generate more objects.
</p>
<div class="col-sm-12 text-center">
<img src="website_files/imgs/clevr_object.png" class="img-fluid" alt="clevr_objects" style="width:100%; height:100%">
</div>
</div>
<div class="row">
<hr>
<p>
<b>Composing Visual Relations.</b> Image generation results on the relational CLEVR dataset.
Our model is trained to generate images conditioned on a single object relation, but during inference, our model can compose multiple object relations.
The baseline methods either miss objects or generate more object relations.
</p>
<div class="col-sm-12 text-center">
<img src="website_files/imgs/clevr_rel.png" class="img-fluid" alt="clevr_relations" style="width:80%; height:100%">
</div>
</div>
<div class="row">
<hr>
<p>
<b>Composing Facial Attributes.</b> Image generation results on the FFHQ dataset.
Our model is trained to generate images conditioned on a single human facial attribute, but during inference,
our model can recursively compose multiple facial attributes using the proposed compositional operators.
The baselines either fail to compose attributes (StyleGAN2 and LACE) or generate low-quality images (EBM).
</p>
<div class="col-sm-12 text-center">
<img src="website_files/imgs/ffhq.png" class="img-fluid" alt="ffhq" style="width:80%">
</div>
</div>
</div><br>
<div class="container bg-3">
<div class="row">
<h2 class="text-center">Result Analysis</h2>
<hr>
<p>
<b>Success Examples.</b> In each example, the first two images are generated conditioned on each individual sentence description and the last image is generated by composing the sentences.
</p>
<div class="col-sm-6 text-center">
<img src="website_files/imgs/success_cases.png" class="img-fluid" alt="failures" style="width:100%">
</div>
<div class="col-sm-6 text-center">
<img src="website_files/imgs/success_cases2.png" class="img-fluid" alt="failures" style="width:100%">
</div>
</div>
<div class="row">
<hr>
<p>
<b>Failure Examples</b>. There are three main types of failures:<br>
(1) The pre-trained diffusion model does not understand certain concepts, such as "person".<br>
(2) The pre-trained duffision model confuses objects' attributes.<br>
(3) The composition fails. This usually happens when the objects are in the center of the images.
</p>
<br>
<div class="col-sm-6 text-center">
<img src="website_files/imgs/failure1.png" class="img-fluid" alt="failures" style="width:90%">
</div>
<div class="col-sm-6 text-center">
<img src="website_files/imgs/failure2.png" class="img-fluid" alt="failures" style="width:90%">
</div>
<br>
<div class="col-sm-12 text-center">
<img src="website_files/imgs/failure3.png" class="img-fluid" alt="failures" style="width:95%">
</div>
</div>
<div class="row">
<hr>
<p>
<b>Interesting Examples</b>. Our method, which combines multiple textual descriptions, can generate different styles of images compared to GLIDE, which directly encodes the descriptions as a single long sentence. Prompted with 'a dog' and 'the sky', our method generates a dog-shaped cloud, whereas GLIDE generates a dog under the sky from the prompt 'a dog and the sky'.
</p>
<br>
<div class="col-sm-6 text-center">
<img src="website_files/imgs/interesting_cases.png" class="img-fluid" alt="failures" style="width:70%">
</div>
<div class="col-sm-6 text-center">
<img src="website_files/imgs/interesting_cases2.png" class="img-fluid" alt="failures" style="width:70%">
</div>
</div>
</div><br>
<!-- related work-->
<div class="container bg-3" id="Related">
<h2 class="text-center">Related Projects</h2>
<hr>
Check out a full list of our papers on energy-based models at this <a href="https://energy-based-model.github.io/Energy-based-Model-MIT/">link</a>!
<br>
<br>
<div class="row vspace-top">
<div class="col-sm-3">
<video width="100%" playsinline="" autoplay="" preload="" muted="">
<source src="website_files/imgs/clevr_teaser.mp4" type="video/mp4">
</video>
</div>
<div class="col-sm-9">
<div class="paper-title">
<a href="https://composevisualrelations.github.io/">Learning to Compose Visual Relations</a>
</div>
<div>
We propose to represent each relation as an unnormalized density (an energy-based model),
enabling us to compose separate relations in a factorized manner.
We show that such a factorized decomposition allows the model to both generate and edit scenes that have multiple sets of relations more faithfully.
We further show that decomposition enables our model to effectively understand the underlying relational scene structure.
</div>
</div>
</div>
<br>
<div class="row vspace-top">
<div class="col-sm-3">
<img src="website_files/imgs/comet_teaser.gif" class="img-fluid" alt="comet" style="width:100%">
</div>
<div class="col-sm-9">
<div class="paper-title">
<a href="https://energy-based-model.github.io/comet/">Unsupervised Learning of Compositional Energy Concepts</a>
</div>
<div>
We show COMET provides a unified framework enabling us to decompose images into both global factors
of variation as well as local factors of variation. Second, we show that COMET enables us to
scale to more realistic datasets than previous work. Finally, we show that components obtained by
COMET generalize well, and are amenable to compositions across different modes of data, and with
components discovered by other instances of COMET.
</div>
</div>
</div>
<br>
<div class="row vspace-top">
<div class="col-sm-3">
<img src="website_files/imgs/comp_cartoon.png" class="img-fluid" alt="ffhq" style="width:100%">
</div>
<div class="col-sm-9">
<div class="paper-title">
<a href="https://energy-based-model.github.io/compositional-generation-inference/">Compositional Visual Generation with Energy Based Models</a>
</div>
<div>
We show how EBMs enable <b>zero-shot compositional</b> visual generation, enabling us to compose visual concepts
(through operators of conjunction, disjunction, or negation) together in a zero-shot manner.
Our approach enables us to generate faces given a description
((Smiling AND Female) OR (NOT Smiling AND Male)) or to combine several different objects together.
</div>
</div>
</div>
</div>
<br>
<div class="container bg-3">
<div class="row">
<h2 class="text-center">Acknowledgement</h2>
<hr>
<p>Thanks <a href="https://baulab.info/">David Bau</a> for proofreading the paper and providing suggestions,
and <a href="https://scholar.google.com/citations?user=5fU-QMwAAAAJ&hl=en">Mark Chen</a> for running DALLE-2 examples for our paper.
This webpage template was recycled from <a href="https://yenchenlin.me/nerf-supervision/">here</a>. </p>
</div>
</div><br><br>
</div>
</body>
<script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
</html>