-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
525 lines (447 loc) · 18.1 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
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
---
redirect_from: saff/
---
<!DOCTYPE html>
<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
<title>Semantic Attention Flow Fields for Monocular Dynamic Scene Decomposition</title>
<style>
body {
padding: 1em;
color: rgb(0, 0, 0);
font-size: 14px;
font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Oxygen-Sans, Ubuntu, Cantarell, "Helvetica Neue", sans-serif;
background-color: rgb(255, 255, 255);
}
#container {
max-width: 800px;
text-align: justify;
margin: 0 auto;
}
h1 {
color: rgb(213, 28, 37);
text-align: center;
}
h2 {
color: rgb(213, 28, 37);
text-align: center;
}
h3 {
font-weight: bold;
margin-bottom: 0.5em;
}
li {
margin: 3px 0px;
width: 560px;
hyphens: auto;
text-align: justify;
}
PRE {
font-family: 'Monotype.com', Courier New, monospace;
font-size: 0.7em;
background-color: #f5f5f5;
border-radius: 4px;
}
a {
color: rgb(213, 28, 37);
text-decoration: none;
}
a:link {
color: rgb(213, 28, 37);
}
a:visited {
color: rgb(183, 0, 07);
}
a:hover {
color: rgb(53, 0, 0);
}
.dataset {
font-variant: small-caps;
}
.center {
margin-left: auto;
margin-right: auto;
}
.centerContent {
text-align: center;
}
.centerContent table {
margin-left: 50%;
transform: translateX(-50%);
-webkit-transform: translateX(-50%);
}
td p {
text-align: center;
margin-top: 0px;
margin-bottom: 5px;
}
</style>
<script>
function addVideoGrid(div, data) {
console.warn(data);
let str = "<div class='centerContent'><table><tbody>";
for (let i = 0; i < data.length; i++) {
str += "<tr>";
for (let j = 0; j < data[i].length; j++) {
if (data[i][j]) {
str += "<td><p>" + data[i][j].label + "</p><video loop controls muted playbackRate=0.5 width=300px><source src=" + data[i][j].path + " type='video/mp4'></video></td>";
} else {
str += "<td></td>"
}
}
str += "</tr>";
}
str += "</tbody></table></div>"
div.innerHTML = div.innerHTML + str;
// Videos are too fast, set playback to be half speed by default
justAddedVideos = div.getElementsByTagName("video")
setPlaybackRateOnVideos( justAddedVideos, 0.5 );
}
function addVideoControls(div, videoDivName) {
let str = "<div class='centerContent'>";
const playbackRates = [
{ rate: 0.05, label: '0.10x' },
{ rate: 0.125, label: '0.25x' },
{ rate: 0.25, label: '0.50x' },
{ rate: 0.5, label: '1.00x' },
]
str += "<button onclick='setPlayStateOnVideos( document.getElementById(\"" + videoDivName + "\").getElementsByTagName(\"video\" ), true)'>Play All</button>";
str += "<button onclick='setPlayStateOnVideos( document.getElementById(\"" + videoDivName + "\").getElementsByTagName( \"video\" ), false)'>Pause All</button>";
str += "<button onclick='setTimeOnVideos( document.getElementById(\"" + videoDivName + "\").getElementsByTagName( \"video\" ), 0)'>Reset All to t=0</button>";
playbackRates.forEach((pbr) => {
str += "<button onclick='setPlaybackRateOnVideos( document.getElementById(\"" + videoDivName + "\").getElementsByTagName( \"video\" ), " + pbr.rate + ")'>" + pbr.label + "</button>"
});
str += "</div>"
div.innerHTML += str;
}
function setPlaybackRateOnVideos(videos, rate) {
for (let i = 0; i < videos.length; i++)
videos[i].playbackRate = rate
}
function setPlayStateOnVideos(videos, playState) {
for (let i = 0; i < videos.length; i++) {
if (playState)
videos[i].play()
else
videos[i].pause()
}
}
function setTimeOnVideos(videos, time) {
for (let i = 0; i < videos.length; i++) {
videos[i].currentTime = time
}
}
function showVideoDiv(datasetName) {
dsDiv = document.getElementById(datasetName + '-videos');
videosInDiv = dsDiv.getElementsByTagName("video")
setPlayStateOnVideos(videosInDiv, false)
dsDiv.style.display = 'none'
dsDiv.style.display = 'block'
}
function hideVideoDiv(datasetName) {
dsDiv = document.getElementById(datasetName + '-videos');
videosInDiv = dsDiv.getElementsByTagName("video")
dsDiv.style.display = 'none'
}
</script>
</head>
<body>
<div id="container">
<h1>Semantic Attention Flow Fields for<br>Monocular Dynamic Scene Decomposition</h1>
<h2>ICCV 2023</h2>
<table class="center">
<tr>
<td>
<h3 style="text-align: left";>
<a href="https://lynl7130.github.io">Yiqing Liang</a><br>
<a href="https://www.linkedin.com/in/eliot-laidlaw-472640197/">Eliot Laidlaw</a><br>
<a href="https://www.linkedin.com/in/ameyerow/">Alexander Meyerowitz</a> <br>
<a href="https://cs.brown.edu/people/ssrinath/">Srinath Sridhar</a><br>
<a href="https://jamestompkin.com/">James Tompkin</a>
</h3>
</td>
<td>
</td>
<td>
<img src="./images/BrownCSLogo.png" width="150px">
</td>
</tr>
</table>
<table class="center">
<tr>
<td><video src="umbrella_input.mp4" type="video/mp4" height="128" autoplay muted loop></td>
<td><video src="umbrella_nvs_spacetime.mp4" type="video/mp4" height="128" autoplay muted loop></td>
</tr>
<tr style="text-align:center">
<td>Input video</td>
<td>RGB (novel spacetime)</td>
</tr>
<tr>
<td><video src="umbrella_clusters_spacetime.mp4" type="video/mp4" height="128" autoplay muted loop></td>
<td><video src="umbrella_fg_spacetime.mp4" type="video/mp4" height="128" autoplay muted loop></td>
</tr>
</tr>
<tr style="text-align:center">
<td>Decomposition (novel spacetime)</td>
<td>Foreground (novel spacetime)</td>
</tr>
</table>
<h2>Abstract</h2>
<p class="center">
We present SAFF: a dynamic neural volume reconstruction of a casual monocular video that consists of time-varying color, density, scene flow, semantics, and attention information. The semantics and attention let us identify salient foreground objects separately from the background in arbitrary spacetime views. We add two network heads to represent the semantic and attention information. For optimization, we design semantic attention pyramids from DINO-ViT outputs that trade detail with whole-image context. After optimization, we perform a saliency-aware clustering to decompose the scene. For evaluation on real-world dynamic scene decomposition across spacetime, we annotate object masks in the NVIDIA Dynamic Scene Dataset. We demonstrate that SAFF can decompose dynamic scenes without affecting RGB or depth reconstruction quality, that volume-integrated SAFF outperforms 2D baselines, and that SAFF improves foreground/background segmentation over recent static/dynamic split methods.
</p>
<div style="text-align: center">
<img src="./images/teaser.svg" style="width:400px;">
</div>
<br>
<table class="center">
<tr>
<td>
<a href="https://arxiv.org/abs/2303.01526">
<img src="./images/arxiv.png" style="width:151px;border:2px solid #000;">
</a>
<br>
<p>
<a href="https://arxiv.org/abs/2303.01526">arXiv.org</a> | <a href="https://arxiv.org/pdf/2303.01526.pdf">PDF</a>
</p>
</td>
<td>
<a href="https://github.com/brownvc/saff/">
<img src="./images/github-mark.svg" style="width:199px;border:2px solid #000;">
</a>
<br>
<p>
<a href="https://github.com/brownvc/saff/">Code (Github)</a>
</p>
</td>
</tr>
</table>
<table class="center">
<tr>
<td>
<h3>Bibtex</h3>
<pre><code>
@inproceedings{Liang2023SAFF,
title={Semantic Attention Flow Fields for Monocular Dynamic Scene Decomposition},
author={Yiqing Liang and Eliot Laidlaw and Alexander Meyerowitz
and Srinath Sridhar and James Tompkin},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month = {October},
year={2023},
}</code>
</pre>
</td>
</tr>
</table>
<h2>Presentation</h2>
<div style="text-align: center">
<iframe src="https://onedrive.live.com/embed?cid=E10E204FB9E6F665&resid=E10E204FB9E6F665%21244211&authkey=ALNwvPtpgE6mzsM&em=2" width="540" height="330" frameborder="0" scrolling="no"></iframe>
</div>
<h2>Supplemental video results</h2>
<p><em>Note:</em> This page has 100+ videos. Each set of results will show/hide through a button to ease the burden on the browser.</p>
<h3>Index</h3>
<a id="datasetindex"></a>
<div id="datasetindexappend">0. <a href="#contributions">Key aspects of our method</a><br></div>
<h2>Key aspects of our method <a href='#top'>🔝</a></h2>
<a id="contributions"></a>
<div id="pyramidDiv">
<h3>Pyramid feature extraction</h3>
<p>Semantic and saliency quality improves with our pyramid feature extraction approach.</p>
<p><span class='dataset'>Balloon NBoard</span> sequence below shows low resolution raw semantics and saliency from DINO-ViT and our higher resolution versions on the right. <em>Note:</em> Features have been projected via PCA and so some color variation is expected; colors are not directly comparable.</p>
</div>
<script>
addVideoControls(document.getElementById("pyramidDiv"), "pyramidDiv");
addVideoGrid(document.getElementById("pyramidDiv"), [
[
{ path: 'pyramid/rgb_input.mp4', label: 'Input sequence' },
{ path: 'pyramid/feat_input.mp4', label: 'Raw semantics' },
{ path: 'pyramid/pyr_feat_input.mp4', label: 'Semantics after our pyramid extraction' }
],
[
null,
{ path: 'pyramid/sal_input.mp4', label: 'Raw saliency' },
{ path: 'pyramid/pyr_sal_input.mp4', label: 'Saliency after our pyramid extraction' }
]
]);
</script>
<div id="volumeDiv">
<h3>Feature volume integration</h3>
<p>Semantic and saliency quality improves further through volume integration.</p>
<p><span class='dataset'>Balloon NBoard</span> sequence below shows the semantic and saliency input to the
optimization process with the rendered output after optimization.</p>
</div>
<script>
addVideoControls(document.getElementById("volumeDiv"), "volumeDiv");
addVideoGrid(document.getElementById("volumeDiv"), [
[
{ path: 'volume/rgb_input.mp4', label: 'Input sequence' },
{ path: 'volume/pyr_feat_input.mp4', label: 'Pyramid semantics' },
{ path: 'volume/feat_output.mp4', label: 'Volume semantics (input views)' },
{ path: 'volume/feat_output_novel.mp4', label: 'Volume semantics (novel view fix time)' }
],
[
null,
{ path: 'volume/pyr_sal_input.mp4', label: 'Pyramid saliency' },
{ path: 'volume/sal_output.mp4', label: 'Volume saliency (input views)' },
{ path: 'volume/sal_output_novel.mp4', label: 'Volume saliency (novel view fix time)' }
]
]);
</script>
<div id="clusterDiv">
<h3>Saliency-aware clustering</h3>
<p>Saliency-aware clustering improves decomposition.</p>
<p><span class='dataset'>DynamicFace</span> sequence below shows the clustering results before saliency voting
and before cluster merging.</p>
</div>
<script>
addVideoControls(document.getElementById("clusterDiv"), "clusterDiv");
addVideoGrid(document.getElementById("clusterDiv"), [
[
{ path: 'clustering/novote.mp4', label: 'Before saliency voting (input views)' },
{ path: 'clustering/nomerge.mp4', label: 'Before merging (input views)' },
{ path: 'clustering/aftermerge.mp4', label: 'Final clustering (input views)' },
],
[
{ path: 'clustering/novote_novel.mp4', label: 'Before saliency voting (novel views)' },
{ path: 'clustering/nomerge_novel.mp4', label: 'Before merging (novel views)' },
{ path: 'clustering/aftermerge_novel.mp4', label: 'Final clustering (novel views)' },
]
]);
</script>
<h2>Datasets</h2>
<a id="datasets"></a>
<script>
const datasets = [
{
filenamePrefix: 'Balloon1-2',
name: 'balloon1',
displayName: 'Balloon NBoard',
},
{
filenamePrefix: 'Umbrella',
name: 'umbrella',
displayName: 'Umbrella',
},
{
filenamePrefix: 'Balloon2-2',
name: 'balloon2',
displayName: 'Balloon Wall',
},
{
filenamePrefix: 'Jumping',
name: 'jumping',
displayName: 'Jumping',
},
{
filenamePrefix: 'Skating',
name: 'skating',
note: 'Sometimes wanted salient objects are static, e.g., the human in <span class="dataset">Balloon NBoard</span>. Othertimes, salient static objects may be unwanted, like in this sequence. Should an application know that wanted objects are dynamic, then the SAFF representation allows us to isolate static salient clusters easily via the scene flow (see main paper discussion). We show all salient objects here regardless of dynamics.',
displayName: 'Skating',
},
{
filenamePrefix: 'Playground',
name: 'playground',
note: 'Sometimes wanted salient objects are static, e.g., the human in <span class="dataset">Balloon NBoard</span>. Othertimes, salient static objects may be unwanted, like in this sequence. Should an application know that wanted objects are dynamic, then the SAFF representation allows us to isolate static salient clusters easily via the scene flow (see main paper discussion). We show all salient objects here regardless of dynamics.',
displayName: 'Playground',
},
]
const methods = [
{
name: 'SAFF',
displayName: 'SAFF (Ours)',
rgb: true,
fg_blend: true,
depth: true,
cluster_blend: true,
},
{
name: 'NSFF',
displayName: 'NSFF',
rgb: true,
fg_blend: true,
depth: true,
cluster_blend: false,
},
{
name: 'DinoBaseline',
displayName: 'DINO-ViT Baseline',
rgb: false,
fg_blend: true,
depth: false,
cluster_blend: true,
},
{
name: 'ProposeReduce',
displayName: 'ProposeReduce',
rgb: false,
fg_blend: true,
depth: false,
cluster_blend: true,
},
{
name: 'D2NeRF',
displayName: 'D<sup>2</sup>NeRF',
rgb: true,
fg_blend: true,
depth: true,
cluster_blend: false,
}
]
channels = [
{ name: 'rgb', displayName: 'RGB' },
{ name: 'depth', displayName: 'depth' },
{ name: 'cluster_blend', displayName: 'clustering' },
{ name: 'fg_blend', displayName: 'fg/bg' },
]
datasets.forEach((d, i) => {
const index = document.getElementById("datasetindexappend")
index.innerHTML += (i + 1) + ". <a href='#" + d.name + "-results'>Results on the <span class='dataset'>" + d.displayName + "</span> dataset<br>"
const target = document.getElementById("datasets")
const dsDiv = document.createElement("div")
dsDiv.setAttribute("id", d.name + '-results')
target.append(dsDiv)
dsDiv.innerHTML += "<h2>" + (i + 1) + ". Results on the <span class='dataset'>" + d.displayName + "</span> dataset <a href='#top'>🔝</a></h2>"
dsDiv.innerHTML = dsDiv.innerHTML + `<button onclick='
showVideoDiv(\"` + d.name + `\");
setPlayStateOnVideos( document.getElementById(\"` + d.name + `-results\").getElementsByTagName( \"video\" ), true ` + `);
window.location = \"#Dataset` + d.name + `\";'
>
Show Videos
</button>`
dsDiv.innerHTML = dsDiv.innerHTML + `<button onclick='hideVideoDiv(\"` + d.name + `\")'>Hide Videos</button><br><br>`
const videoGridData = methods.map((method) => (channels.map((channel) => (
method[channel.name] ?
{
path: method.name + '/' + channel.name + '/' + d.filenamePrefix + '_train_' + channel.name + '.mp4',
label: method.displayName + ' ' + channel.displayName
} : null
))))
const videoDiv = document.createElement("div")
if( d.note != undefined && d.note != "" )
{
videoDiv.innerHTML += "<p><em>Note:</em> " + d.note + "</p>"
}
videoDiv.setAttribute("id", d.name + '-videos')
dsDiv.append(videoDiv);
addVideoControls(videoDiv, d.name + '-videos')
addVideoGrid(videoDiv, videoGridData);
hideVideoDiv(d.name);
});
</script>
<h2>Acknowledgement</h2>
The authors thank the computer vision community in New England for feedback, and acknowledge funding from NSF CNS-2038897 and an Amazon Research Award. Eliot Laidlaw was supported by a Randy F. Pausch '82 Computer Science Undergraduate Summer Research Award at Brown University.
<br>
<table style="border-spacing: 10px 0;">
<tr>
<td><img src="./images/nsf-logo-desktop.svg" width="200px"></td>
<td><img src="./images/amazon-science-logo.svg" width="200px"></td>
</tr>
</table>
<!--
The work was presented orally at <a href="https://necv2022.github.io/">NECV 2022 workshop</a>.
-->
</div>
</body>
</html>