-
Notifications
You must be signed in to change notification settings - Fork 13
/
Copy pathindex.html
545 lines (506 loc) · 24.9 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
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
<!DOCTYPE html>
<html>
<head lang="en">
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-W7L2E12Y6D"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-W7L2E12Y6D');
</script>
<meta charset="UTF-8">
<meta http-equiv="x-ua-compatible" content="ie=edge">
<title>SMERF</title>
<meta name="description" content="SMERF: Streamable Memory Efficient Radiance Fields for Real-Time Large-Scene Exploration">
<meta name="viewport" content="width=device-width, initial-scale=1">
<!-- <base href="/"> -->
<!--FACEBOOK-->
<meta property="og:image" content="/assets/img/merf/social.jpg">
<meta property="og:image:type" content="image/png">
<meta property="og:image:width" content="1000">
<meta property="og:image:height" content="563">
<meta property="og:type" content="website" />
<meta property="og:url" content="https://smerf-3d.github.io" />
<meta property="og:title" content="SMERF" />
<meta property="og:description"
content="Project page for SMERF: Streamable Memory Efficient Radiance Fields for Real-Time Large-Scene Exploration." />
<!--TWITTER-->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="SMERF" />
<meta name="twitter:description"
content="Project page for SMERF: Streamable Memory Efficient Radiance Fields for Real-Time Large-Scene Exploration">
<meta name="twitter:image" content="/assets/img/merf/social.png" />
<!-- <link rel="apple-touch-icon" href="apple-touch-icon.png"> -->
<link rel="icon" href="/favicon.ico">
<!-- Place favicon.ico in the root directory -->
<link rel="stylesheet" href="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.5/css/bootstrap.min.css">
<link rel="stylesheet" href="https://maxcdn.bootstrapcdn.com/font-awesome/4.4.0/css/font-awesome.min.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/codemirror/5.8.0/codemirror.min.css">
<link rel="stylesheet" href="/assets/css/merf/app.css">
<link rel="stylesheet" href="/assets/css/merf/bootstrap.min.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/1.11.3/jquery.min.js"></script>
<script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.5/js/bootstrap.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/codemirror/5.8.0/codemirror.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/clipboard.js/1.5.3/clipboard.min.js"></script>
<script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
<script type="text/javascript" id="MathJax-script" async
src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml.js">
</script>
<link rel="stylesheet" href="/assets/css/merf/dics.min.css">
<script src="/assets/js/merf/dics.min.js"></script>
<script>
document.addEventListener('DOMContentLoaded', domReady);
function domReady() {
for (const e of document.querySelectorAll(".b-dics")) {
new Dics({
container: e,
textPosition: "top"
});
}
}
</script>
</head>
<body>
<div class="container" id="main">
<div class="row">
<h2 class="col-md-12 text-center">
<b>SMERF</b>: Streamable Memory Efficient Radiance Fields for Real-Time Large-Scene Exploration</br>
<!--small>
Under Review
</small-->
</h2>
</div>
<div class="row">
<div class="col-md-12 text-center">
<ul class="list-inline">
<li>
<a href="https://www.stronglyconvex.com/about.html">
Daniel Duckworth*<sup>1</sup>
</a>
</li>
<li>
<a href="https://phogzone.com/">
Peter Hedman*<sup>2</sup>
</a>
</li>
<li>
<a href="https://creiser.github.io/">
Christian Reiser<sup>2,4</sup>
</a>
</li>
<li>
<a href=https://www.linkedin.com/in/petr-zhizhin-9a5a6a75">
Peter Zhizhin<sup>2</sup>
</a>
</li>
<li>
<a href="https://linkedin.com/in/jean-francois-thibert-0682ba98">
Jean-François Thibert<sup>3</sup>
</a>
</li>
<li>
<a href="https://lucic.ai/">
Mario Lučić<sup>1</sup>
</a>
</li>
<li>
<a href="https://szeliski.org/RichardSzeliski.htm">
Richard Szeliski<sup>2</sup>
</a>
</li>
<li>
<a href="https://jonbarron.info/">
Jonathan T. Barron<sup>2</sup>
</a>
</li>
</ul>
<ul class="list-inline">
<li>
Google DeepMind<sup>1</sup>
</li>
<li>
Google Research<sup>2</sup>
</li>
<li>
Google Inc.<sup>3</sup>
</li>
<li>
Tübingen AI Center, University of Tübingen<sup>4</sup>
</li>
</ul>
<ul class="list-inline">
<li>
equal contribution*
</li>
</ul>
</div>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2 text-center">
<ul class="nav nav-pills nav-justified row">
<li>
<a href="https://arxiv.org/abs/2312.07541">
<img src="/assets/img/merf/paper.png" class="mini" alt="">
<h4><strong>Paper</strong></h4>
</a>
</li>
<li>
<a href="https://youtu.be/zhO8iUBpnCc">
<img src="/assets/img/merf/youtube_preview.jpg" class="mini" alt="">
<h4><strong>Video</strong></h4>
</a>
</li>
<li>
<a href="#demos">
<img src="/assets/img/merf/demo.png" class="mini" alt="">
<h4><strong>Demos</strong></h4>
</a>
</li>
<li>
<a href="https://github.com/google-research/google-research/tree/master/smerf">
<img src="/assets/img/merf/github_pad.png" height="120" alt="">
<h4><strong>Code</strong></h4>
</a>
</li>
<li>
<a href="#data">
<img src="/assets/img/merf/database.png" height="120" alt="">
<h4><strong>Data</strong></h4>
</a>
</li>
</ul>
</div>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2">
<div class="col-md-4">
<div class="aspect-ratio-container">
<video class="responsive-video" controls autoplay loop muted>
<source src="https://storage.googleapis.com/realtime-nerf-360/smerf/videos/mildenhall%20desktop.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<div class="col-md-4">
<div class="aspect-ratio-container">
<video class="responsive-video" controls autoplay loop muted>
<source src="https://storage.googleapis.com/realtime-nerf-360/smerf/videos/nottingham%20laptop.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<div class="col-md-4">
<div class="aspect-ratio-container">
<video class="responsive-video" controls autoplay loop muted>
<source src="https://storage.googleapis.com/realtime-nerf-360/smerf/videos/westcombe%20iphone.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</div>
</div>
<div class="row top-gutter">
<div class="col-md-8 col-md-offset-2">
<div class="aspect-ratio-container">
<video class="responsive-video" controls autoplay loop muted>
<source src="https://storage.googleapis.com/realtime-nerf-360/smerf/videos/duckworthd%20desktop%201080p.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2">
<h3>
Abstract
</h3>
<!-- <img src="/assets/img/merf/rays.jpg" class="img-responsive" alt="overview"><br> -->
<p class="text-justify">
Recent techniques for real-time view synthesis have rapidly advanced in
fidelity and speed, and modern methods are capable of rendering
near-photorealistic scenes at interactive frame rates. At the same time, a
tension has arisen between explicit scene representations amenable to
rasterization and neural fields built on ray marching, with state-of-the-art
instances of the latter surpassing the former in quality while being
prohibitively expensive for real-time applications. In this work, we introduce
SMERF, a view synthesis approach that achieves state-of-the-art accuracy among
real-time methods on large scenes with footprints <b>up to 300 m^2 at a volumetric
resolution of 3.5 mm^3</b>. Our method is built upon two primary contributions: a
hierarchical model partitioning scheme, which increases model capacity while
constraining compute and memory consumption, and a distillation training
strategy that simultaneously yields high fidelity and internal consistency. Our
approach enables full six degrees of freedom (6DOF) navigation within a web
browser and <b>renders in real-time on commodity smartphones and laptops</b>.
Extensive experiments show that our method exceeds the current state-of-the-art
in real-time novel view synthesis by <b>0.78 dB on standard benchmarks and 1.78 dB
on large scenes</b>, renders frames three orders of magnitude faster than
state-of-the-art radiance field models, and achieves real-time performance
across a wide variety of commodity devices, including smartphones.
</p>
</div>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2">
<h3>
Video
</h3>
<div class="text-center">
<div style="position:relative;padding-top:56.25%;">
<iframe src="https://www.youtube.com/embed/zhO8iUBpnCc?si=GxuaOyjUlZZcKJQj" allowfullscreen
style="position:absolute;top:0;left:0;width:100%;height:100%;"></iframe>
</div>
</div>
</div>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2" id="demos">
<h3>
Real-Time Interactive Viewer Demos
</h3>
</div>
</div>
<div class="row">
<div class="col-md-4 col-md-offset-2 text-center">
<ul class="nav nav-pills nav-justified">
<li>
<a href="/select_quality?scene=berlin" target="_blank">
<img src="/assets/img/merf/scene_preview/berlin_v2.png" width="160px"></img>
<h4><strong>Berlin</strong></h4>
</a>
</li>
<li>
<a href="/select_quality?scene=nyc" target="_blank">
<img src="/assets/img/merf/scene_preview/nyc.png" width="160px"></img>
<h4><strong>NYC</strong></h4>
</a>
</li>
<li>
<a href="/select_quality?scene=alameda" target="_blank">
<img src="/assets/img/merf/scene_preview/alameda.png" width="160px"></img>
<h4><strong>Alameda</strong></h4>
</a>
</li>
<li>
<a href="/select_quality?scene=london" target="_blank">
<img src="/assets/img/merf/scene_preview/london_v2.png" width="160px"></img>
<h4><strong>London</strong></h4>
</a>
</li>
<br>
<li>
<a href="/select_quality?scene=gardenvase" target="_blank">
<img src="/assets/img/merf/scene_preview/gardenvase.png" width="160px"></img>
<h4><strong>Gardenvase</strong></h4>
</a>
</li>
<li>
<a href="/select_quality?scene=bicycle" target="_blank">
<img src="/assets/img/merf/scene_preview/bicycle.png" width="160px"></img>
<h4><strong>Bicycle</strong></h4>
</a>
</li>
<li>
<a href="/select_quality?scene=kitchenlego" target="_blank">
<img src="/assets/img/merf/scene_preview/kitchenlego.png" width="160px"></img>
<h4><strong>Kitchen Lego</strong></h4>
</a>
</li>
<li>
<a href="/select_quality?scene=stump" target="_blank">
<img src="/assets/img/merf/scene_preview/stump.png" width="160px"></img>
<h4><strong>Stump</strong></h4>
</a>
</li>
<br>
<li>
<a href="/select_quality?scene=officebonsai" target="_blank">
<img src="/assets/img/merf/scene_preview/officebonsai.png" width="160px"></img>
<h4><strong>Office Bonsai</strong></h4>
</a>
</li>
<li>
<a href="/select_quality?scene=fulllivingroom" target="_blank">
<img src="/assets/img/merf/scene_preview/fulllivingroom.png" width="160px"></img>
<h4><strong>Full Living Room</strong></h4>
</a>
</li>
<li>
<a href="/select_quality?scene=kitchencounter" target="_blank">
<img src="/assets/img/merf/scene_preview/kitchencounter.png" width="160px"></img>
<h4><strong>Kitchen Counter</strong></h4>
</a>
</li>
<li>
<style>
.disabled-link {
pointer-events: none;
color: grey;
}
</style>
<a href="" class="disabled-link">
<img src="/assets/img/merf/scene_preview/flowertreehill.png" width="160px"></img>
<h4><strong>Treehill & Flower </strong></h4>
</a>
</li>
</ul>
</div>
</div>
<!--div class="row comp-margin">
<div class="col-md-8 col-md-offset-2">
<h3>
SMERF vs 3D Gaussian Splatting
</h3>
<div class="b-dics" style="width: 100%">
<img src="/assets/img/merf/comp/stump_snerg.png" alt="SNeRG++ (210 MB)" />
<img src="/assets/img/merf/comp/stump_merf.png" alt="MERF (220 MB)" />
</div>
</div>
</div>
<div class="row comp-margin">
<div class="col-md-8 col-md-offset-2">
<div class="b-dics" style="width: 100%">
<img src="/assets/img/merf/comp/kitchenlego_snerg.png" alt="SNeRG++ (213 MB)" />
<img src="/assets/img/merf/comp/kitchenlego_merf.png" alt="MERF (233 MB)" />
</div>
</div>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2">
<div class="b-dics" style="width: 100%">
<img src="/assets/img/merf/comp/gardenvase_snerg.png" alt="SNeRG++ (117 MB)" />
<img src="/assets/img/merf/comp/gardenvase_merf.png" alt="MERF (198 MB)" />
</div>
</div>
</div-->
<div class="row">
<div class="col-md-8 col-md-offset-2">
<h3>
How we boost representation power to handle large scenes
</h3>
<br>
<img src="/assets/img/merf/overview.png" style="width:100%;" class="img-responsive center-block"
alt="overview">
<br>
<p class="text-justify">
<b>(a)</b>: We model large multi-room scenes with a number of independent submodels, each of which is
assigned to a different region of the scene. During rendering the submodel is picked
based on camera origin. <b>(b)</b>: To model complex view-dependent effects, within each submodel
we additionally instantiate grid-aligned copies of deferred MLP parameters \(\theta\). These
parameters are trilinearly interpolated based on camera origin \(\mathbf{o}\).
<b>(c)</b>: While each submodel represents the entire scene, only the submodel's assiociated grid cell
is modelled with high resolution, which is realized by contracting the submodel-specific local coordinates.
</p>
</div>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2">
<h3>
Getting the maximum out of our representation via distillation
</h3>
<img src="/assets/img/merf/distill.png" style="width:80%;" class="img-responsive center-block"
alt="contraction";</img>
<p class="text-justify">
We demonstrate that image fidelity can be greatly boosted via distillation. We first train a
state-of-the-art offline radiance field (Zip-NeRF). We then use the RGB color predictions \(\mathbf{c}\) of this teacher
model as supervision for our own model. Additionally, we access the volumetric density values \(\tau\) of the
pre-trained teacher by minimizing the discrepancy of volume rendering weights between teacher and student.
</p>
</p>
</div>
</div>
<div class="col-md-8 col-md-offset-2" id="data">
<h3>
Datasets & Teacher Checkpoints
</h3>
<p class="text-justify">
SMERF models are distilled from Zip-NeRF checkpoints trained on the Mip-NeRF 360 and Zip-NeRF scenes.
The checkpoints below are used to recreate the quantitative results in the published text.
Both datasets and checkpoints are released under CC-BY 4.0 license.
See text for additional details.
</p>
<table class="table">
<thead>
<tr>
<th class="col-md-2"></th>
<th class="col-md-2 text-center">Photos</th>
<th class="col-md-2 text-center">Teacher Checkpoints</th>
<th class="col-md-6 text-center">Notes</th>
</tr>
</thead>
<tbody>
<tr>
<td class="text-center">Mip-NeRF 360</td>
<td class="text-center"><a href="https://jonbarron.info/mipnerf360/">Project website</a></td>
<td class="text-center"><a href="https://storage.googleapis.com/gresearch/refraw360/360_checkpoints.zip">Download All</a></td>
<td>
For quantitative and qualitative results.
Zip-NeRF checkpoints are trained for 50,000 steps.
Checkpoints for treehill and flower available upon request.
</td>
</tr>
<tr>
<td class="text-center">Zip-NeRF (fisheye)</td>
<td class="text-center">
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf/alameda.zip">Alameda</a>,
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf/berlin.zip">Berlin</a>,
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf/london.zip">London</a>,
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf/nyc.zip">NYC</a>
</td>
<td class="text-center">
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf/checkpoints.zip">Download All</a>
</td>
<td>For qualitative results. Zip-NeRF checkpoints are trained for 100,000 step.</td>
</tr>
<tr>
<td class="text-center">Zip-NeRF (undistorted)</td>
<td class="text-center">
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf-undistorted/alameda.zip">Alameda</a>,
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf-undistorted/berlin.zip">Berlin</a>,
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf-undistorted/london.zip">London</a>,
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf-undistorted/nyc.zip">NYC</a>
</td>
<td class="text-center">
<a href="https://storage.googleapis.com/gresearch/refraw360/zipnerf-undistorted/checkpoints.zip">Download All</a>
</td>
<td>For quantitative results. Zip-NeRF checkpoints are trained for 100,000 steps.</td>
</tr>
</tbody>
</table>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2">
<h3>
Citation
</h3>
<p>
If you want to cite our work, please use:
</p>
<pre>
@misc{duckworth2023smerf,
title={SMERF: Streamable Memory Efficient Radiance Fields for Real-Time Large-Scene Exploration},
author={Daniel Duckworth and Peter Hedman and Christian Reiser and Peter Zhizhin and Jean-François Thibert and Mario Lučić and Richard Szeliski and Jonathan T. Barron},
year={2023},
eprint={2312.07541},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
</pre>
</div>
</div>
<div class="row">
<div class="col-md-8 col-md-offset-2">
<h3>
Acknowledgements
</h3>
<p class="text-justify">
The website template was borrowed from <a href="http://mgharbi.com/">Michaël
Gharbi</a>.
Image sliders are based on <a
href="https://github.com/abelcabezaroman/definitive-image-comparison-slider">dics</a>.
</p>
</div>
</div>
</div>
</body>
</html>