-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
453 lines (410 loc) · 19.2 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
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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
<!-- Replace the content tag with appropriate information -->
<meta name="description" content="Camera Motion Guidance">
<meta property="og:title" content="Boosting Camera Motion Control for Video Diffusion Transformers"/>
<meta property="og:description" content="Project Page."/>
<meta property="og:url" content="https://www.linkedin.com/in/soonyau/"/>
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
<!-- <meta property="og:image" content="static/image/9_texture_transfer.png" />
<meta property="og:image:width" content="1200"/>
<meta property="og:image:height" content="630"/> -->
<!-- Keywords for your paper to be indexed by-->
<meta name="keywords" content="generative AI, computer vision, diffusion models, video diffusion">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Boosting Camera Motion Control for Video Diffusion Transformers</title>
<link rel="icon" type="image/x-icon" >
<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">
<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 src="https://documentcloud.adobe.com/view-sdk/main.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>
</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">Boosting Camera Motion Control for Video Diffusion Transformers</h1>
<div class="is-size-5 publication-authors">
<!-- Paper authors -->
<span class="author-block">
<a href="https://scholar.google.com/citations?user=dRot7GUAAAAJ&hl=en" target="_blank">Soon Yau Cheong <sup>1,2</sup></SUB></a>
<a href="https://www.linkedin.com/in/soonyau"><img src="static/images/LI-In-Bug.png" alt="Linkedin Icon" width="24" height="24"></a>
<a href="https://scholar.google.com/citations?user=56Kj2QoAAAAJ&hl=en" target="_blank"> , Duygu Ceylan <sup>2</sup> </a>
<a href="https://scholar.google.com/citations?user=0xOHqkMAAAAJ&hl=en" target="_blank"> , Armin Mustafa <sup>1</sup> </a>
<a href="https://scholar.google.com/citations?user=NNhnVwoAAAAJ&hl=en" target="_blank"> , Andrew Gilbert <sup>1</sup></a>
<a href="https://scholar.google.com/citations?user=LphRgywAAAAJ" target="_blank"> , Chun-Hao Paul Huang <sup>2</sup></a></span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">University of Surrey<sup>1</sup>. Adobe Research<sup>2</sup>. <br>October, 2024</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- Arxiv PDF link -->
<span class="link-block">
<a href="https://arxiv.org/abs/2410.10802" target="_blank"
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://github.com/soon-yau/CameraMotionGuidance/tree/web/supplementary/" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Supplementary</span>
</a>
</span>
<span class="link-block">
<a href="https://github.com/soon-yau/CameraMotionGuidance" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
</div>
</div>
<!-- <span> Let us know if Demo link is not running.</span> -->
</div>
</div>
</div>
</div>
</section>
<!-- Teaser video-->
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<span style="font-size: 20px;">
<b>TL;DR: </b> Existing camera control methods do not work for diffusion transformers. We developed the first camera control for space-time diffuser transformer. Our method, based on classifier-free guidance, restore controllability and boost motion by over 400%. </span>
<!--<div class="publication-video">
<iframe src="https://www.youtube.com/embed/J2QYb2F-sxg" frameborder="0" allow="autoplay; encrypted-media" ></iframe>
</div>
-->
<!-- <video poster="" id="tree" autoplay muted controls controlsList="nofullscreen" height="100%">
<source src="static/videos/visconet_banner_full.mp4"
type="video/mp4">
</video> -->
<h2 class="subtitle has-text-centered">
</h2>
</div>
</div>
</section>
<!-- End teaser video -->
<!-- Paper abstract -->
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column ">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
<p style="font-size: 20px;">
Recent advancements in diffusion models have significantly enhanced the quality of video generation. However, fine-grained control over camera pose remains a challenge. While U-Net-based models have shown promising results for camera control, transformer-based diffusion models (DiT)—the preferred architecture for large-scale video generation—suffer from severe degradation in camera motion accuracy. In this paper, we investigate the underlying causes of this issue and propose solutions tailored to DiT architectures. Our study reveals that camera control performance depends heavily on the choice of conditioning methods rather than camera pose representations that is commonly believed. To address the persistent motion degradation in DiT, we introduce Camera Motion Guidance (CMG), based on classifier-free guidance, which boosts camera control by over 400%. Additionally, we present a sparse camera control pipeline, significantly simplifying the process of specifying camera poses for long videos. Our method universally applies to both U-Net and DiT models, offering improved camera control for video generation tasks.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- End paper abstract -->
<section class="hero is-small">
<div class="hero-body">
<div class="container is-centered has-text-centered">
<h2 class="title is-3">Effective Camera Control for DiT</h2>
<p style="font-size: 20px;"> SOTA camera control methods for U-Net : MotionCtrl and CameraCtrl lose controllability and motion in DiT (diffusion transformer) architecture. Our method restores controllability and boosting motion.</p>
<div class="row is-centered">
<div class="col3 is-centered">
<p>Conditioning camera poses.</p>
</div>
<div class="col3 ">
<img src="supplementary/1_Method_Comparison_files/000.png" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/001.png" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/002.png" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/003.png" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/004.png" width="256"/>
</div>
</div>
<div class="row">
<div class="col">
<p>MotionCtrl method in DiT has uncontrollable motion.</p>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/050.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/053.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/056.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/059.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/062.gif" width="256"/>
</div>
</div>
<div class="row">
<div class="col">
<p>CameraCtrl method in DiT has limited motion</p>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/051.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/054.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/057.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/060.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/063.gif" width="256"/>
</div>
</div>
<div class="row">
<div class="col">
<p>Our method restores camera controllability with boosted motion.</p>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/052.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/055.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/043.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/061.gif" width="256"/>
</div>
<div class="col3">
<img src="supplementary/1_Method_Comparison_files/064.gif" width="256"/>
</div>
</div>
<br><br>
</div>
</div>
</section>
<section class="hero is-small">
<div class="hero-body">
<div class="container is-centered has-text-centered">
<h2 class="title is-3">Camera Motion Guidance (CMG) boosts camera motion</h2>
<p style="text-align:left; font-size: 20px;"> Increasing guidance scale boost camera motion. Our method has good disentanglement from text guidance which controls the video content and appearance.</p>
<br><br>
<div class="row">
<div class="col2">
<p> Camera Pose </p>
<img src="supplementary/2_Ablation_files/000.png" width="256"/>
</div>
<div class="col2">
<p> CMG scale = 0 </p>
<img src="supplementary/2_Ablation_files/001.gif" width="256"/>
</div>
<div class="col2">
<p> CMG scale = 1 </p>
<img src="supplementary/2_Ablation_files/002.gif" width="256"/>
</div>
<div class="col2">
<p> CMG scale = 2 </p>
<img src="supplementary/2_Ablation_files/003.gif" width="256"/>
</div>
<div class="col2">
<p> CMG scale = 3 </p>
<img src="supplementary/2_Ablation_files/004.gif" width="256"/>
</div>
<div class="col2">
<p> CMG scale = 4 </p>
<img src="supplementary/2_Ablation_files/005.gif" width="256"/>
</div>
<div class="col2">
<p> CMG scale = 5 </p>
<img src="supplementary/2_Ablation_files/006.gif" width="256"/>
</div>
<div class="col2">
<p> CMG scale = 6 </p>
<img src="supplementary/2_Ablation_files/007.gif" width="256"/>
</div>
<div class="col2">
<p> CMG scale = 7 </p>
<img src="supplementary/2_Ablation_files/008.gif" width="256"/>
</div>
</div>
<div class="row">
<div class="col2">
<img src="supplementary/2_Ablation_files/009.png" width="256"/>
</div>
<div class="col2">
<img src="supplementary/2_Ablation_files/010.gif" width="256"/>
</div>
<div class="col2">
<img src="supplementary/2_Ablation_files/011.gif" width="256"/>
</div>
<div class="col2">
<img src="supplementary/2_Ablation_files/012.gif" width="256"/>
</div>
<div class="col2">
<img src="supplementary/2_Ablation_files/013.gif" width="256"/>
</div>
<div class="col2">
<img src="supplementary/2_Ablation_files/014.gif" width="256"/>
</div>
<div class="col2">
<img src="supplementary/2_Ablation_files/015.gif" width="256"/>
</div>
<div class="col2">
<img src="supplementary/2_Ablation_files/016.gif" width="256"/>
</div>
<div class="col2">
<img src="supplementary/2_Ablation_files/017.gif" width="256"/>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-small">
<div class="hero-body">
<div class="container is-centered has-text-centered">
<h2 class="title is-3">Sparse Camera Control</h2>
<p style="font-size: 20px;"> Our data augmentation method enable sparse camera control. Videos with sparse camera control (b-e) resemble motion with dense control (a).</p>
<div class="row is-centered">
<div class="col is-centered">
<img src="supplementary/3_SparseControl_files/010.png" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/011.png" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/012.png" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/013.png" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/014.png" width="256"/>
</div>
</div>
<div class="row">
<div class="col">
<img src="supplementary/3_SparseControl_files/015.gif" width="256"/>
<p>(a)</p>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/016.gif" width="256"/>
<p>(b)</p>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/017.gif" width="256"/>
<p>(c)</p>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/018.gif" width="256"/>
<p>(d)</p>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/019.gif" width="256"/>
<p>(e)</p>
</div>
</div>
<p style="font-size: 20px;"> (i) The first and last camera pose is facing straight, thus the generated motion (rightmost) contains only translation with no rotation.</p>
<br><br>
<div class="row">
<div class="col">
<img src="supplementary/3_SparseControl_files/000.png" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/001.png" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/002.png" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/003.png" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/004.png" width="256"/>
</div>
</div>
<div class="row">
<div class="col">
<img src="supplementary/3_SparseControl_files/005.gif" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/006.gif" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/007.gif" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/008.gif" width="256"/>
</div>
<div class="col">
<img src="supplementary/3_SparseControl_files/009.gif" width="256"/>
</div>
</div>
<p style="font-size: 20px;">(ii) Now by adding rotation to the last frame in (e), our model 'interpolate' the missing poses to create smooth trajectory with rotation. </p>
</div>
</div>
</section>
<!--BibTex citation -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>
@article{cheong2024cmg,
author = {Cheong, Soon Yau and Mustafa, Armin and Ceylan, Duygu and Gilbert, Andrew and Huang, Chun-hao Paul},
title = {Boosting Camera Motion Control for Video Diffusion Transformers},
journal = {Arxiv Preprint 2410.10802},
month = {October},
year = {2024}}</code></pre>
</div>
</section>
<!--End BibTex citation -->
<footer class="footer">
<div class="container">
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p>
This page was built using the <a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">Academic Project Page Template</a> which was adopted from the <a href="https://nerfies.github.io" target="_blank">Nerfies</a> project page.
</p>
</div>
</div>
</div>
</div>
</footer>
<!-- Statcounter tracking code -->
<!-- You can add a tracker to track page visits by creating an account at statcounter.com -->
<!-- End of Statcounter Code -->
</body>
</html>