-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
504 lines (453 loc) · 20.3 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
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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>GEM: A Generalizable Ego-Vision Multimodal World Model for Fine-Grained Ego-Motion, Object Dynamics, and Scene Composition Control</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<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">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<script defer src="./static/js/fontawesome.all.min.js"></script>
<style>
.video-container {
display: grid;
grid-template-columns: repeat(2, 1fr); /* Four equal columns for videos */
gap: 10px; /* Space between items */
margin: 20px 0; /* Add vertical spacing between sections */
align-items: stretch; /* Ensure items fill the grid cell vertically */
justify-items: stretch; /* Ensure items fill the grid cell horizontally */
}
.video-item {
display: flex;
flex-direction: column;
align-items: center; /* Center align descriptions and videos */
}
.video-description {
margin-bottom: 10px;
font-weight: bold; /* Make the description stand out */
}
.section-description {
margin-bottom: 30px;
font-size: 1.2em;
color: #555; /* Subtle color to differentiate from headings */
}
.content {
margin-bottom: 20px; /* Add space between sections (subtitles and videos) */
}
video, img {
width: 100%; /* Make both videos and GIFs fill their grid cells */
height: 100%; /* Fill the height of the grid cell */
border: 1px solid #ccc; /* Optional: Add a border */
border-radius: 5px; /* Optional: Add rounded corners */
object-fit: cover; /* Ensure content fills the container */
}
.pseudo-video-container {
display: flex; /* Use flexbox for layout */
flex-direction: column; /* Arrange items vertically */
justify-content: center; /* Center the videos vertically */
align-items: center; /* Center the videos horizontally */
gap: 20px; /* Add spacing between videos */
}
.pseudo-video-container video {
width: 100%; /* Each video takes 50% of the row */
height: 800; /* Set a large height for the videos */
max-height: 100%; /* Prevent overflow */
border: 1px solid #ccc; /* Optional: Add a border */
border-radius: 5px; /* Optional: Add rounded corners */
object-fit: cover; /* Ensures the video content scales nicely */
}
</style>
</head>
<body>
<!-- <div style="background-color: #e6ffe6; padding: 10px; text-align: center; color: green; font-weight: bold;">
We recommend using Chrome for the best experience. Videos might not work on Safari.
</div> -->
<section class="hero title-section" style="background-color: #f0f0f0;">
<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">
💎 GEM: A Generalizable Ego-Vision Multimodal World Model for Fine-Grained Ego-Motion, Object Dynamics, and Scene Composition Control
</h1>
<!-- Author List -->
<div class="author-list" style="margin-top: 20px; font-size: 1.1rem;">
Mariam Hassan<sup>★1</sup>, Sebastian Stapf<sup>★2</sup>, Ahmad Rahimi<sup>★1</sup>,
Pedro M B Rezende<sup>★2</sup>, Yasaman Haghighi<sup>♦1</sup>, <br>
David Brüggemann<sup>♦3</sup>, Isinsu Katircioglu<sup>♦3</sup>, Lin Zhang<sup>♦3</sup>,
Xiaoran Chen<sup>♦3</sup>, Suman Saha<sup>♦3</sup>, <br>
Marco Cannici<sup>♦4</sup>, Elie Aljalbout<sup>♦4</sup>, Botao Ye<sup>♦5</sup>,
Xi Wang<sup>♦5</sup>, Aram Davtyan<sup>2</sup>, <br>
Mathieu Salzmann<sup>1,3</sup>, Davide Scaramuzza<sup>4</sup>, Marc Pollefeys<sup>5</sup>,
Paolo Favaro<sup>2</sup>, Alexandre Alahi<sup>1</sup>
</div>
<!-- Affiliations -->
<div class="affiliations" style="margin-top: 20px; font-size: 1rem;">
<sup>1</sup>École Polytechnique Fédérale de Lausanne (EPFL),
<sup>2</sup>University of Bern, <br>
<sup>3</sup>Swiss Data Science Center,
<sup>4</sup>University of Zurich,
<sup>5</sup>ETH Zurich
</div>
<!-- Contribution Legend -->
<div class="contribution-legend" style="margin-top: 20px; font-size: 1rem; font-style: italic;">
<sup>★</sup> Main Contributors <sup>♦</sup> Data Contributors
</div>
<!-- arXiv Link -->
<span class="link-block" style="margin-left: 10px;">
<a href="https://arxiv.org/abs/2412.11198"
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>
<!-- Code Link -->
<span class="link-block" style="margin-left: 10px;">
<a href="https://github.com/vita-epfl/GEM"
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>
</div>
</div>
</section>
<!-- Other Sections -->
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">Unconditional Generations</h2>
<p class="section-description">
We show examples of unconditional generations from the model in diverse scenes with different driving dynamics.
</p>
<div class="video-container">
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/OpenDV_002133.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/OpenDV_000265.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/OpenDV_000322.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/OpenDV_000637.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/OpenDV_001165.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/OpenDV_001591.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/OpenDV_002209.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/OpenDV_002847.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/NUSCENES_000071.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/generation/NUSCENES_000964.mp4" type="video/mp4">
</video>
</div>
</div>
</section>
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">Ego Control</h2>
<p class="section-description">
We show examples of ego-motion controllability. All videos are generated by GEM using the same starting frame but with different trajectory control input. <br>
We observe that the model follows the control signals and generates realistic scenes.
</p>
<div class="video-container">
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/ego/OpenDV_001365-1.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/ego/OpenDV_001365-2.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/ego/OpenDV_001161-left.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/ego/OpenDV_001162.mp4" type="video/mp4">
</video>
</div>
</div>
</section>
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">Object Manipulation</h2>
<p class="section-description">
GEM can move objects in the scene using DINO features. <br>
In the following examples, we show an unconditional generation by GEM and the same generation with motion control. <br>
The green box indidcates the source DINO features and the blue ones indicate the target position tokens used. <br>
We observe that the object moves from the green box to the blue box.
</p>
<div class="video-container">
<div class="video-item">
<p class="video-description">Unconditional Generation</p>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/move/demo6_uc.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<p class="video-description">Object Motion Control</p>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/move/demo6_1_c.mp4" type="video/mp4">
</video>
</div>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/move/bike_uc.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/move/bike_c.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/move/demo8_uc.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/move/demo8_c.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">Object Insertion</h2>
<!-- First Row of Videos -->
<div class="video-container">
<div class="video-item">
<p class="video-description">Unconditional Generation</p>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/insert/insert1_uc.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<p class="video-description">Insertion Control</p>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/insert/insert1.mp4" type="video/mp4">
</video>
</div>
</div>
<!-- Text between rows -->
<p style="margin: 10px 0; font-size: 1.0rem;">
In the following example, we insert a car on the left and control the motion of another car on the right.
</p>
<!-- Second Row of Videos -->
<div class="video-container">
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/insert/demo1_uc.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/insert/demo1_c.mp4" type="video/mp4">
</video>
</div>
</div>
</section>
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">Human Pose Manipulation</h2>
<p class="section-description">
GEM can use human poses to control the motion of pedestrians in the scene. <br>
In this examples, the pedesterians are crossing the street or stopping based on the human poses controls.
</p>
<div class="video-container">
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/human/OpenDV_000000_walk.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/human/OpenDV_000000.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/human/OpenDV_000001_walk.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/human/OpenDV_000001.mp4" type="video/mp4">
</video>
<!-- <video autoplay muted loop playsinline preload="metadata">
<source src="path/to/human/video3.mp4" type="video/mp4">
</video> -->
</div>
</div>
</section>
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">Long Generation</h2>
<p class="section-description">
We compare our long generation with the only world model trained on OpenDV capable of generating long sequences.<br>
We observe that our generations have higher ego motion temporal consistency and more realistic dynamics.
</p>
<div class="video-container">
<!-- Video Item 1 -->
<div class="video-item">
<p class="video-description">GEM</p>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/long/OpenDV-250_000291.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<p class="video-description">Vista</p>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/long/OpenDV_000291_LONG_vista.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<!-- <p class="video-description">GEM</p> -->
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/long/OpenDV_000276_LONG.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<!-- <p class="video-description">Vista</p> -->
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/long/OpenDV_000276_LONG_vista.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<!-- <p class="video-description">GEM</p> -->
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/long/OpenDV_000047_LONG.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<!-- <p class="video-description">Vista</p> -->
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/long/OpenDV_000047_LONG_vista.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<!-- <p class="video-description">GEM</p> -->
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/long/OpenDV_000395_LONG.mp4" type="video/mp4">
</video>
</div>
<div class="video-item">
<!-- <p class="video-description">Vista</p> -->
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/long/OpenDV_000395_LONG_vista.mp4" type="video/mp4">
</video>
</div>
</div>
</section>
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">Interesting Observations</h2>
<!-- Section Description -->
<p class="section-description">
We show interesting behaviors observed in the generated videos. <br> These behaviors do not necessarily exist in the ground truth videos, but emerge from the model's learned dynamics.
</p>
<!-- Video Grid -->
<div class="video-container">
<!-- Video Item 1 -->
<div class="video-item">
<p class="video-description">Break Lights go off before moving</p>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/interesting/OpenDV-250_000291.mp4" type="video/mp4">
</video>
</div>
<!-- Video Item 2 -->
<div class="video-item">
<p class="video-description">Smooth takeover dynamics on a long generation </p>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/interesting/OpenDV-250_000287.mp4" type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</section>
<!-- MultiModal Section -->
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">MultiModal</h2>
<p class="section-description">
GEM generates two modalities simultaneously: RGB and Depth. We show examples of multimodal generations.
</p>
<!-- Images Row -->
<div class="video-container">
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/multimodal/OpenDV_000226.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/multimodal/OpenDV_000226_depth.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/multimodal/OpenDV_000195.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/multimodal/OpenDV_000195_depth.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/multimodal/OpenDV_001132.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/multimodal/OpenDV_001132_depth.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/multimodal/OpenDV_000221.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/multimodal/OpenDV_000221_depth.mp4" type="video/mp4">
</video>
</div>
</div>
</section>
<!-- MultiDomain Section -->
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">MultiDomain</h2>
<p class="section-description">
GEM is finetuned on two other ego centric domains and we observe it quickly adapts to these new domains.
</p>
<!-- Drone Flights -->
<div class="content">
<h3 class="subtitle is-3">1-Drone Flights</h3>
<div class="video-container">
<img src="./static/videos/multidomain/1.gif" alt="Drone Flights GIF">
<img src="./static/videos/multidomain/2.gif" alt="Drone Flights GIF">
<img src="./static/videos/multidomain/3.gif" alt="Drone Flights GIF">
<img src="./static/videos/multidomain/4.gif" alt="Drone Flights GIF">
</div>
</div>
<!-- Human EgoCentric -->
<div class="content">
<h3 class="subtitle is-3">2-Human EgoCentric</h3>
<div class="video-container">
<img src="./static/videos/multidomain/5.gif" alt="Human EgoCentric GIF">
<img src="./static/videos/multidomain/6.gif" alt="Human EgoCentric GIF">
<img src="./static/videos/multidomain/7.gif" alt="Human EgoCentric GIF">
<img src="./static/videos/multidomain/8.gif" alt="Human EgoCentric GIF">
</div>
</div>
</div>
</section>
<!-- Pseudo labels Section -->
<section class="section">
<div class="container has-text-centered">
<h2 class="title is-2">Pseudo-Labelling</h2>
<p class="section-description">
Some visualisations of the outputs of our pseudo-labeling pipeline.
</p>
<div class="pseudo-video-container">
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/pseudo/pseudo_label.mp4" type="video/mp4">
</video>
<video autoplay muted loop playsinline preload="metadata">
<source src="./static/videos/pseudo/pseudo_label_1.mp4" type="video/mp4">
</video>
</div>
</div>
</section>
</body>
</html>