-
Notifications
You must be signed in to change notification settings - Fork 1
/
trnerf.html
427 lines (416 loc) · 19.3 KB
/
trnerf.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
<!DOCTYPE HTML>
<html lang="en-US">
<head>
<meta charset="UTF-8">
<title>TRNeRF: Restoring Blurry, Rolling Shutter, and Noisy Thermal Images with Neural Radiance Fields</title>
<meta name="description" content="TRNeRF">
<style>
.title{
font-family: Arial, Helvetica, sans-serif;
text-align:center;
margin-top:30pt;
margin-bottom:0pt;
margin-left: auto;
margin-right: auto;
color:#303030;
font-size:35pt;
font-weight:800;
letter-spacing:0pt;
max-width:1200px;
}
.subtitle{
font-family: Arial, Helvetica, sans-serif;
text-align:center;
margin-top:8pt;
margin-bottom:0;
margin-left: auto;
margin-right: auto;
color:#303030;
font-size:25pt;
font-weight:500;
letter-spacing:0pt;
max-width:1200px;
}
.section_title{
font-family: Arial, Helvetica, sans-serif;
text-align:left;
padding-left:25pt;
padding-right:25pt;
color:#181818;
font-size:25pt;
font-weight:500;
letter-spacing:0pt;
}
.body{
font-family: Arial, Helvetica, sans-serif;
color:#000000;
padding-right:25pt;
padding-left:25pt;
padding-bottom:25pt;
text-align:justify;
font-size:14pt;
font-weight:100;
letter-spacing:0pt;
line-height: 20pt;
}
.col{
font-family: Arial, Helvetica, sans-serif;
text-align: center;
float: left;
color:#000000;
font-weight:100;
letter-spacing:0pt;
height: 25pt;
}
.row{
margin-left: auto;
margin-right: auto;
}
.row:after{
content: "";
display: table;
clear: both;
}
.affiliation{
font-family: Arial, Helvetica, sans-serif;
text-align: center;
color:#000000;
font-size:10pt;
font-weight:100;
letter-spacing:0pt;
}
.container{
margin-left:auto;
margin-right:auto;
margin-top: 5pt;
max-width:1200px
}
/*Adapted from source: https://www.w3schools.com/howto/howto_js_slideshow.asp */
.mySlides {display: none}
.slideshow-container {
max-width: 1200px;
position: relative;
margin: auto;
}
.prev, .next {
cursor: pointer;
position: absolute;
top: 54%;
width: auto;
padding: 16px;
margin-top: -22px;
color: white;
font-weight: bold;
font-size: 18px;
transition: 0.6s ease;
border-radius: 0 3px 3px 0;
user-select: none;
}
.next {
right: 0;
border-radius: 3px 0 0 3px;
}
.prev:hover, .next:hover {
background-color: rgba(0,0,0,0.8);
}
.caption {
font-family: Arial, Helvetica, sans-serif;
color: #ffffff;
font-size: 15pt;
position: absolute;
font-weight: 600;
}
.dot {
cursor: pointer;
height: 15px;
width: 15px;
margin: 0 2px;
background-color: #bbb;
border-radius: 50%;
display: inline-block;
transition: background-color 0.6s ease;
}
.active, .dot:hover {
background-color: #717171;
}
</style>
</head>
<body style="background-color:#ffffff;">
<div class="section">
<h1 class="title">TRNeRF</h1>
<h1 class="subtitle">Restoring Blurry, Rolling Shutter, and Noisy Thermal Images with Neural Radiance Fields</h1>
<!-- Author names first row -->
<div class="row" style="max-width:800px;">
<div class="col" style="font-size:12pt; font-weight: 200; height: auto; margin-bottom: -20pt; width: 25%;">
<p>Spencer Carmichael<sup>1</sup></p>
</div>
<div class="col" style="font-size:12pt; font-weight: 200; height: auto; margin-bottom: -20pt; width: 25%;">
<p>Manohar Bhat<sup>1</sup></p>
</div>
<div class="col" style="font-size:12pt; font-weight: 200; height: auto; margin-bottom: -20pt; width: 25%;">
<p>Mani Ramanagopal<sup>3</sup></p>
</div>
<div class="col" style="font-size:12pt; font-weight: 200; height: auto; margin-bottom: -20pt; width: 25%;">
<p>Austin Buchan<sup>1</sup></p>
</div>
</div>
<!-- Mailbox first row -->
<div class="row" style="max-width:800px;">
<div class="col", style="font-size:10pt; width: 25%;">
<p>[email protected]</p>
</div>
<div class="col", style="font-size:10pt; width: 25%;">
<p>[email protected]</p>
</div>
<div class="col", style="font-size:10pt; width: 25%;">
<p>[email protected]</p>
</div>
<div class="col", style="font-size:10pt; width: 25%;">
<p>[email protected]</p>
</div>
</div>
<!-- Author names second row -->
<div class="row" style="max-width:800px;">
<div class="col" style="font-size:12pt; font-weight: 200; height: auto; margin-bottom: -20pt; width: 50%;">
<p>Ram Vasudevan<sup>1,2</sup></p>
</div>
<div class="col" style="font-size:12pt; font-weight: 200; height: auto; margin-bottom: -20pt; width: 50%;">
<p>Katherine A. Skinner<sup>1</sup></p>
</div>
</div>
<!-- Mailbox second row -->
<div class="row" style="max-width:800px;">
<div class="col", style="font-size:10pt; width: 50%;">
<p>[email protected]</p>
</div>
<div class="col", style="font-size:10pt; width: 50%;">
<p>[email protected]</p>
</div>
</div>
<!-- Affiliation -->
<div class="affiliation" style="margin-bottom: -10pt;">
<p><sup>1</sup>Robotics Department, University of Michigan, Ann Arbor</p>
</div>
<div class="affiliation" style="margin-bottom: -10pt;">
<p><sup>2</sup>Department of Mechanical Engineering, University of Michigan, Ann Arbor</p>
</div>
<div class="affiliation">
<p><sup>3</sup>Robotics Institute, Carnegie Mellon University, Pittsburgh</p>
</div>
<!-- Links -->
<div class="row" style="margin-left:auto; margin-right:auto; margin-top:10pt;max-width:800px;">
<div class="col" style="width: 33%; height: 27pt;">
<a href="https://umautobots.github.io/trnerf">
<img src="./web_elements/paper.png" height="40pt"/>
</a>
</div>
<div class="col" style="width: 33%; height: 27pt;">
<a href="https://umautobots.github.io/trnerf">
<img src="./web_elements/github.png" height="40pt"/>
</a>
</div>
<div class="col" style="width: 33%; height: 27pt;">
<a href="https://umautobots.github.io/trnerf">
<img src="./web_elements/database.png" height="40pt"/>
</a>
</div>
</div>
<!-- Link names -->
<div class="row" style="margin-left:auto; margin-right:auto; max-width:800px; line-height: 10pt;">
<div class="col" style="width: 33%; font-size:12pt; font-weight: 600;">
<p>Preprint (coming soon)</p>
</div>
<div class="col" style="width: 33%; font-size:12pt; font-weight: 600;">
<p>Code (coming soon)</p>
</div>
<div class="col" style="width: 33%; font-size:12pt; font-weight: 600;">
<p>Dataset (coming soon)</p>
</div>
</div>
</div>
<div class="container" style="margin-top: 0pt; margin-bottom: -30pt;">
<h2 class="section_title">Example Results</h2>
</div>
<!-- Example results slideshow -->
<div class="slideshow-container">
<div class="mySlides">
<div class="col", style="font-size:18pt; font-weight: 100; width: 100%; margin-bottom: 20pt; margin-top: 10pt;">
<p>(1/4) <b>Fast</b> camera movement (playback 1/4x speed) <b>Indoor</b> scene</p>
</div>
<video width="100%" id="fast_indoor" autoplay muted loop controls>
<source src="media/trnerf/fast_indoor_raw_vs_trnerf_quarter_speed_first_15_sec.mp4">
</video>
<div class="caption" style="top: 17%; left: 2%"> Raw (Undistorted) </div>
<div class="caption" style="top: 17%; left: 52%"> TRNeRF Restored </div>
</div>
<div class="mySlides">
<div class="col", style="font-size:18pt; font-weight: 100; width: 100%; margin-bottom: 20pt; margin-top: 10pt;">
<p>(2/4) <b>Fast</b> camera movement (playback 1/4x speed) <b>Outdoor</b> scene</p>
</div>
<video width="100%" id="fast_outdoor" autoplay muted loop controls>
<source src="media/trnerf/fast_outdoor_raw_vs_trnerf_quarter_speed_first_15_sec.mp4">
</video>
<div class="caption" style="top: 17%; left: 2%"> Raw (Undistorted) </div>
<div class="caption" style="top: 17%; left: 52%"> TRNeRF Restored </div>
</div>
<div class="mySlides">
<div class="col", style="font-size:18pt; font-weight: 100; width: 100%; margin-bottom: 20pt; margin-top: 10pt;">
<p>(3/4) <b>Medium</b> camera movement (playback 1x speed) <b>Indoor</b> scene</p>
</div>
<video width="100%" id="medium_indoor" autoplay muted loop controls>
<source src="media/trnerf/medium_indoor_raw_vs_trnerf_first_15_sec.mp4">
</video>
<div class="caption" style="top: 17%; left: 2%"> Raw (Undistorted) </div>
<div class="caption" style="top: 17%; left: 52%"> TRNeRF Restored </div>
</div>
<div class="mySlides">
<div class="col", style="font-size:18pt; font-weight: 100; width: 100%; margin-bottom: 20pt; margin-top: 10pt;">
<p>(4/4) <b>Medium</b> camera movement (playback 1x speed) <b>Outdoor</b> scene</p>
</div>
<video width="100%" id="medium_outdoor" autoplay muted loop controls>
<source src="media/trnerf/medium_outdoor_raw_vs_trnerf_first_15_sec.mp4">
</video>
<div class="caption" style="top: 17%; left: 2%"> Raw (Undistorted) </div>
<div class="caption" style="top: 17%; left: 52%"> TRNeRF Restored </div>
</div>
<a class="prev" onclick="plusSlides(-1)">❮</a>
<a class="next" onclick="plusSlides(1)">❯</a>
</div>
<br>
<div style="text-align:center">
<span class="dot" onclick="currentSlide(1)"></span>
<span class="dot" onclick="currentSlide(2)"></span>
<span class="dot" onclick="currentSlide(3)"></span>
<span class="dot" onclick="currentSlide(4)"></span>
</div>
<!-- Abstract -->
<div class="container">
<h2 class="section_title">Abstract</h2>
<p class="body"> Thermal cameras offer unique detection capabilities in building inspections, search and rescue
operations, and autonomous vehicle perception. Of the different types of thermal cameras, uncooled
microbolometers are often chosen due to their relative affordability, small size, and low power consumption.
However, microbolometers suffer from motion blur, rolling shutter distortions, and fixed pattern noise,
which limit the conditions of their use. Nearly all prior methods for microbolometer image restoration
account for only one of these degradations, and current techniques addressing microbolometer blur and
rolling shutter are limited. This paper presents TRNeRF, a thermal image restoration method that jointly
addresses all three degradations by incorporating the microbolometer image formation model with Neural
Radiance Fields (NeRFs). To evaluate TRNeRF, this paper introduces a new real-world dataset that is uniquely
designed to support two novel quantitative evaluation strategies for thermal image restoration. Experiments
demonstrate that TRNeRF is able to recover sharp, global shutter, and clear thermal images, even under
extremely aggressive camera motion that causes existing methods to fail.
</p>
</div>
<!-- Rendering Pipeline -->
<div class="container">
<h2 class="section_title">Rendering Pipeline</h2>
<p class="body"> We train an accurate representation of the thermal scene by incorporating motion blur, rolling
shutter distortions, and fixed pattern noise into the NeRF rendering pipeline such that the rendered pixels
can be supervised by the real captured images subject to degradation. The left side of this figure depicts
the learned scene and the views traversed by two pixels read out at different times. The remaining sections
show how a single degraded pixel value is estimated and contributes to the loss. At the inference stage, the
augmentations to the rendering pipeline are removed and the trained model can be used to directly render
restored images.
</p>
<div style="text-align: center;">
<img src="images/trnerf/method.png", width="94%">
</div>
</div>
<!-- Dataset -->
<div class="container">
<h2 class="section_title">Dataset</h2>
<p class="body"> To collect our dataset, we designed a rig with a backpack mounted computer and hand-held sensor
platform including two microbolometer thermal cameras, two monochrome cameras, and an IMU (included for future
work). The thermal cameras are placed side-by-side to support the comparison of different camera settings. The
dataset includes a sunny outdoor scene with high thermal contrast, and an air-conditioned indoor scene with low
thermal contrast. For each scene there are three recorded sequences, denoted slow, medium, and fast, with
increasingly aggressive six degree-of-freedom camera motion.
</p>
<div style="text-align: center;">
<img src="images/trnerf/rig_fig_website.png", width="94%">
</div>
</div>
<!-- Evaluation Methods -->
<div class="container">
<h2 class="section_title">Evaluation Methods</h2>
<p class="body"> We introduce two novel quantitative methods for evaluating thermal
image restoration. The first method utilizes a multi-spectral Aprilgrid board that is placed in each scene. The
board can be detected in both the sharp visible spectrum images and the restored thermal images. We use the
the detections in the monochrome images to project the board's corners into the thermal images. We then compute the
percentage of successfully detected corners in the thermal images to assess how well areas of the board were
restored. The second method involves rendering pseudo ground truth images using a standard NeRF model
trained on the relatively undegraded slow sequence thermal images (following the removal of fixed pattern noise
through a two point non-uniformity correction). We compute the LPIPS metric between the restored and pseudo
ground truth images.
</p>
<div class="col", style="font-size:18pt; font-weight: 100; width: 100%; margin-bottom: 20pt; margin-top: -35pt">
<p>Detection Based Evaluation Method</p>
</div>
<video style="margin-left:25pt;" width="94%" id="detection" autoplay muted loop controls>
<source src="media/trnerf/medium_indoor_detection_sixth_speed_first_15_sec.mp4">
</video>
<div style="text-align: center;">
<img src="images/trnerf/detection_legend.png", width="350pt">
</div>
<div class="col", style="font-size:18pt; font-weight: 100; width: 100%; margin-bottom: 20pt;">
<p>Pseudo Ground Truth Based Evaluation Method</p>
</div>
<video style="margin-left:25pt;" width="94%" id="lpips" autoplay muted loop controls>
<source src="media/trnerf/fast_indoor_lpips_fifteenth_speed.mp4">
</video>
</div>
<!-- Abalation Study -->
<div class="container">
<h2 class="section_title">Ablation Study</h2>
<p class="body"> Our ablation study demonstrates that it is necessary to account for
all three degradations and that seemingly subtle details can have a substantial impact on restoration
performance. In particular, onboard image filters, which were enabled by default in our thermal cameras, can
produce significant artifacts in the restored images.
</p>
<video style="margin-left:25pt;" width="94%" id="ablations" autoplay muted loop controls>
<source src="media/trnerf/fast_indoor_ablation_quarter_speed.mp4">
</video>
</div>
<!-- Citation -->
<div class="container">
<h2 class="section_title">Citation</h2>
<ul class="body" style="font-size: 10pt; line-height: 15pt;">
<ul style="padding-left:0"><div style="background-color:#e6e6e6; padding-bottom: 10pt; padding-top: 10pt">
<p class="body" style="padding-left:0pt; font-size: 10pt;padding-bottom:0;text-indent:-3rem;margin-left:4rem;font-family:monospace;line-height: 1.2rem;">
@misc{trnerf2024, <br>
title={TRNeRF: Restoring Blurry, Rolling Shutter, and Noisy Thermal Images with Neural Radiance Fields}, <br>
author={Spencer Carmichael and Manohar Bhat and Mani Ramanagopal and Austin Buchan and Ram Vasudevan and Katherine A. Skinner}, <br>
year={2024}, <br>
eprint={TODO:PLACEHOLDER}, <br>
archivePrefix={arXiv}, <br>
primaryClass={cs.CV}}
</div></li>
</ul>
</div>
<script>
// From source: https://www.w3schools.com/howto/howto_js_slideshow.asp
let slideIndex = 1;
showSlides(slideIndex);
function plusSlides(n) {
showSlides(slideIndex += n);
}
function currentSlide(n) {
showSlides(slideIndex = n);
}
function showSlides(n) {
let i;
let slides = document.getElementsByClassName("mySlides");
let dots = document.getElementsByClassName("dot");
if (n > slides.length) {slideIndex = 1}
if (n < 1) {slideIndex = slides.length}
for (i = 0; i < slides.length; i++) {
slides[i].style.display = "none";
}
for (i = 0; i < dots.length; i++) {
dots[i].className = dots[i].className.replace(" active", "");
}
slides[slideIndex-1].style.display = "block";
dots[slideIndex-1].className += " active";
}
</script>
</body>
</html>