-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
623 lines (546 loc) · 23.3 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
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
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
<script src="http://www.google.com/jsapi" type="text/javascript"></script>
<script type="text/javascript">google.load("jquery", "1.3.2");</script>
<style type="text/css">
body {
font-family: "HelveticaNeue-Light", "Helvetica Neue Light", "Helvetica Neue", Helvetica, Arial, "Lucida Grande", sans-serif;
font-weight:300;
font-size:18px;
margin-left: auto;
margin-right: auto;
width: 1100px;
background: white;
}
h1 {
font-size:32px;
font-weight:300;
}
.disclaimerbox {
background-color: #eee;
border: 1px solid #eeeeee;
border-radius: 10px ;
-moz-border-radius: 10px ;
-webkit-border-radius: 10px ;
padding: 20px;
}
video.header-vid {
height: 140px;
border: 1px solid black;
border-radius: 10px ;
-moz-border-radius: 10px ;
-webkit-border-radius: 10px ;
}
img.header-img {
height: 140px;
border: 1px solid black;
border-radius: 10px ;
-moz-border-radius: 10px ;
-webkit-border-radius: 10px ;
}
img.rounded {
border: 1px solid #eeeeee;
border-radius: 10px ;
-moz-border-radius: 10px ;
-webkit-border-radius: 10px ;
}
a:link,a:visited
{
color: #1367a7;
text-decoration: none;
}
a:hover {
color: #208799;
}
td.dl-link {
height: 160px;
text-align: center;
font-size: 22px;
}
.layered-paper-big { /* modified from: http://css-tricks.com/snippets/css/layered-paper/ */
box-shadow:
0px 0px 1px 1px rgba(0,0,0,0.35), /* The top layer shadow */
5px 5px 0 0px #fff, /* The second layer */
5px 5px 1px 1px rgba(0,0,0,0.35), /* The second layer shadow */
10px 10px 0 0px #fff, /* The third layer */
10px 10px 1px 1px rgba(0,0,0,0.35), /* The third layer shadow */
15px 15px 0 0px #fff, /* The fourth layer */
15px 15px 1px 1px rgba(0,0,0,0.35), /* The fourth layer shadow */
20px 20px 0 0px #fff, /* The fifth layer */
20px 20px 1px 1px rgba(0,0,0,0.35), /* The fifth layer shadow */
25px 25px 0 0px #fff, /* The fifth layer */
25px 25px 1px 1px rgba(0,0,0,0.35); /* The fifth layer shadow */
margin-left: 10px;
margin-right: 45px;
}
.paper-big { /* modified from: http://css-tricks.com/snippets/css/layered-paper/ */
box-shadow:
0px 0px 1px 1px rgba(0,0,0,0.35); /* The top layer shadow */
margin-left: 10px;
margin-right: 45px;
}
.layered-paper { /* modified from: http://css-tricks.com/snippets/css/layered-paper/ */
box-shadow:
0px 0px 1px 1px rgba(0,0,0,0.35), /* The top layer shadow */
5px 5px 0 0px #fff, /* The second layer */
5px 5px 1px 1px rgba(0,0,0,0.35), /* The second layer shadow */
10px 10px 0 0px #fff, /* The third layer */
10px 10px 1px 1px rgba(0,0,0,0.35); /* The third layer shadow */
margin-top: 5px;
margin-left: 10px;
margin-right: 30px;
margin-bottom: 5px;
}
.vert-cent {
position: relative;
top: 50%;
transform: translateY(-50%);
}
hr
{
border: 0;
height: 1px;
background-image: linear-gradient(to right, rgba(0, 0, 0, 0), rgba(0, 0, 0, 0.75), rgba(0, 0, 0, 0));
}
.mycontainer {
display: flex;
align-items: center;
justify-content: center
}
myimg {
max-width: 100%;
max-height:100%;
}
.mytext {
font-size: 18px;
padding-left: 20px;
}
</style>
<html>
<head>
<title>Adapting MIMO video restoration networks to low latency constraints</title>
<meta property="og:image" content="./doc/Teaser.png"/> <!-- Facebook automatically scrapes this. Go to https://developers.facebook.com/tools/debug/ if you update and want to force Facebook to rescrape. -->
<meta property="og:title" content="Adapting MIMO video restoration networks to low latency constraints. Dewil et al." />
<meta property="og:description" content="Dewil et al. BMVC 2024" />
<link rel="icon" type="image/x-icon" href="./doc/Drone_dataset.png">
<!-- Get from Google Analytics -->
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src=""></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'UA-75863369-6');
</script>
</head>
<body>
<br>
<center>
<span style="font-size:36px">Adapting MIMO video restoration networks to low latency constraints</span> <br>
<br>
<table align=center width=940px>
<table align=center width=1025px>
<tr>
<td align=center width=90px>
<center>
<span style="font-size:20px"><a href="https://centreborelli.ens-paris-saclay.fr/fr/annuaire-des-personnes/valery-dewil/">Valéry Dewil</a></span>
</center>
</td>
<td align=center width=80px>
<center>
<span style="font-size:20px"><a href="https://openreview.net/profile?id=~Zhe_Zheng3/">Zhe Zheng</a></span>
</center>
</td>
<td align=center width=100px>
<center>
<span style="font-size:20px"><a href="https://centreborelli.ens-paris-saclay.fr/fr/annuaire-des-personnes/arnaud-barral/">Arnaud Barral</a></span>
</center>
</td>
<td align=center width=80px>
<center>
<span style="font-size:20px"><a href="https://dev.ipol.im/~lraad/">Lara Raad</a></span>
</center>
</td>
<td align=center width=90px>
<center>
<span style="font-size:20px">Nao Nicolas</span>
</center>
</td>
<td align=center width=100px>
<center>
<span style="font-size:20px">Ioannis Cassagne</span>
</center>
</td>
<td align=center width=110px>
<center>
<span style="font-size:20px"><a href="https://centreborelli.ens-paris-saclay.fr/fr/annuaire-des-personnes/jean-michel-morel">Jean-Michel Morel</a></span>
</center>
</td>
<td align=center width=100px>
<center>
<span style="font-size:20px"><a href="http://gfacciol.github.io/">Gabriele Facciolo</a></span>
</center>
</td>
<td align=center width=100px>
<center>
<span style="font-size:20px"><a href="https://github.com/bgalerne/">Bruno Galerne</a></span>
</center>
</td>
<td align=center width=100px>
<center>
<span style="font-size:20px"><a href="http://dev.ipol.im/~pariasm/">Pablo<br>Arias</a></span>
</center>
</td>
</table>
<br>
<div class="container">
<a href="https://bmvc2024.org/"><img src="./doc/BMVC_logo (copie).png" height="100" /></a>
<a href="https://centreborelli.ens-paris-saclay.fr/fr"><img src="./doc/Centre_Borelli_logo (copie).png" height="100" /></a>
<a href="https://ens-paris-saclay.fr/"><img src="./doc/ENS_logo (copie).png" height="100" /></a>
</div>
<br>
<table align=center width=550px>
<tr>
<td align=center width=120px>
<center>
<span style="font-size:20px"><a href='https://www.arxiv.org/pdf/2408.12439'>[Paper]</a></span>
</center>
</td>
<td align=center width=120px>
<center>
<span style="font-size:20px"><a href='https://github.com/centreborelli/video-MIMO-ROSO/'>[GitHub]</a></span><br>
</center>
</td>
<td align=center width=120px>
<center>
<span style="font-size:20px"><a href='https://xxxx'>[Dataset]</a></span><br>
</center>
</td>
<td align=center width=280px>
<center>
<span style="font-size:20px"><a href='https://xxxx'>[Supplementary material]</a></span><br>
</center>
</td>
</tr>
</table>
</table>
</center>
<br>
<center>
<div style="text-align: center;">
<img width="1050" src="./doc/Teaser.png">
</div>
<br>
<table align=center width=1070px>
<tr>
<td>
Recently, MIMO (Multiple Input, Multiple Output) architectures has been proposed for the video restoration task. They have a better performance/running time trade-off than still single output architectures (MISO: Multiple Input, Multiple Output). In this paper, we focus on the three main state-of-the-art architectures for denoising in the low-latency setting (limited number of frames in the output stack, typically 5 or 7 frames). We show that the PSNR of MIMO networks is not uniform within the output stack and propose to use recurrence across stacks (RAS) and output stack overlap (OSO) to smooth
this non uniformity. See Figure (a). <br><br>
We also show that MIMO networks are temporally consistent within output stacks. However, they have very strong and unwanted changes at stack transitions. See Figure (b). <br><br>
The proposed contributions (shortened to ROSO when applied together) significantly reduce these changes.
</td>
</tr>
</table>
<table align=center width=1070px>
<tr>
<td>
<br><br>
<i> Project developed at the <a href='https://centreborelli.ens-paris-saclay.fr/fr'>ENS Paris-Saclay, Centre Borelli</a> and accepted at <a href='https://bmvc2024.org/'>BMVC 2024</a>.<i/><br>
</td>
</tr>
</table>
</center>
<hr>
<table align=center width=850px>
<center><h1>Abstract</h1></center>
<tr>
<td>
MIMO (multiple input, multiple output) approaches are a recent trend in neural network architectures for video restoration problems, where each network evaluation produces multiple output frames. The video is split into non-overlapping stacks of frames that are processed independently, resulting in a very appealing trade-off between output quality and computational cost. In this work we focus on the low-latency setting by limiting the number of available future frames. We find that MIMO architectures suffer from problems that have received little attention so far, namely (1) the performance drops significantly due to the reduced temporal receptive field, particularly for frames at the boundaries of the stack, (2) there are strong temporal discontinuities at stack transitions which induce a step-wise motion artifact. We propose two simple solutions to alleviate these problems: recurrence across MIMO stacks to boost the output quality by implicitly increasing the temporal receptive field, and overlapping of the output stacks to smooth the temporal discontinuity at stack transitions. These modifications can be applied to any MIMO architecture. We test them on three state-of-the-art video denoising networks with different computational cost. The proposed contributions result in a new state-of-the-art for low-latency networks, both in terms of reconstruction error and temporal consistency. As an additional contribution, we introduce a new benchmark consisting of drone footage that highlights temporal consistency issues that are not apparent in the standard benchmarks.
</td>
</tr>
</table>
<br>
<!-- <hr>
<center><h1>Talk</h1></center>
<p align="center">
<iframe width="660" height="395" src="https://www.youtube.com/embed/dQw4w9WgXcQ" frameborder="0" allow="autoplay; encrypted-media" allowfullscreen align="center"></iframe>
</p>
<table align=center width=800px>
<br>
<tr>
<center>
<span style="font-size:28px"><a href=''>[Slides]</a>
</span>
</center>
</tr>
</table>
-->
<hr>
<center><h1>Proposed framework</h1></center>
<table align=center width=420px>
<center>
<tr>
<td>
</td>
</tr>
</center>
</table>
<div style="text-align: center;">
<img width="1000" src="./doc/Proposed_framework.png">
</div>
<table align=center width=1000px>
<center>
<tr>
<td>
<strong>Our proposed ROSO mechanism.</strong> We propose two contributions to drastically alleviate the problems mentionned above: Output Stack Overlapping (OSO) and Recurrence Across Stack (RAS). In the three figures, the red dashed lines indicates the RAS. <br>
(a) Baseline MIMO network applied to two non-overlapping frame stacks (b) OSO: the output in the overlapped frames is computed as a weighted averafe of the two denoised versions. (c) OSO with RAS with a specific overlapped recurrence using the output overlapped frames from the previous stack.
</td>
</tr>
</center>
</table>
<br>
<hr>
<center><h1>Visual results</h1></center>
<div style="text-align: center;">
<img width="1000" src="./doc/Visual_results.png">
</div>
<table align=center width=1000px>
<center>
<tr>
<td>
<strong>Visual results at stack transition.</strong> We display here the results of the baseline network and with our proposed RAS & OSO (abbreviated to ROSO). The network architecture is M2Mnet. We show the frames at stack transition <i> t </i>. We show the two last frames of one stack and the first frame of the next stack. Between them, we display the warping error. The contrast has been enhanced for better visualization. <br>
The strong transition is clearly visible with the baseline network. Using RAS slightly alleviate the transition, while ROSO does not show larger transition inside a stack (see here at time <i> t-2 </i> to <i> t-1 </i>) than acros two stacks (transition <i> t-1 </i> to <i> t </i>).
</td>
</tr>
</center>
</table>
<br>
<video controls autoplay preload="auto" id="idvideo" >
<source src="./doc/drone-d160-40.mp4" type="video/mp4">
</video>
<table align=center width=1000px>
<center>
<tr>
<td>
<strong>Video results of M2Mnet, BasicVSR++ and ReMoNet.</strong> In this video, we present in a split view: the clean ground-truth, the noisy sequence (AWGN σ = 40), the results of the baseline networks and the results of ours baseline+ROSO. The video is played twice, the second time with enhanced contrast to highlight the differences between the methods. The sequence is a crop from one of the videos of the proposed drone benchmark containing an urban scene. The step-wise motion effect induced by the stack transitions can be noticed in low contrast textures (e.g. building facades), and also in the motion of the vehicles. With the extensions proposed in the paper (+ROSO), the textures are much more stable and the cars on the road follow a more fluid motion.
</td>
</tr>
</center>
</table>
<br>
<hr>
<center><h1>Landscapes of low-latency video denoising networks </h1></center>
In the following two plots, we show several several video denoising networks mapped in a measure of performance (PSNR) or measure of temporal consistency at stack transition vs per-frame running time. The landcapes were computed on the drone benchmark and the running times measured on a Nvidia A100 GPU. The values are obtained on the drone benchmark and averaged over noise levels σ = 10, 20, 30, 40, 50? The methods plotted with a star are variants proposed in this paper.
<br><br>
Using this visualization, we can identify a Pareto frontier showing the optimal trade-off between computational cost and PSNR or temporal consistency at stack transition. Methods in the gray region are sub-optimal in the sense that it is possible to achieve better results at lower or equal computational cost. The proposed strategy (<i>ROSO</i>) define a new <i>Pareto</i> frontier by improving both the PSNR and the temporal consistency at stack transition, in spite of the incease in the running tme, expected to be of a factor 1.4.
<br><br>
<div style="text-align: center;">
<img width="1000" src="./doc/Landscapes.png">
</div>
<table align=center width=1000px>
<center>
<tr>
<td>
<strong><i>(left)</i> PSNR vs running time landscape.</strong> The proposed ROSO improves over the current <i>Pareto</i> frontier by a significant increase in the PSNR
<br><br>
<strong><i>(right></i> inter-TC vs running time landscape.</strong> We used a metric based on the warping error designed to measure the temporal consistency at output stack transitions. This metric is called the <i> inter-tc </i>. Our contribution largely improves the inter-tc among low-latency networks.
</td>
</tr>
</center>
</table>
<br>
<hr>
<center><h1>Dataset</h1></center>
<div style="text-align: center;">
<a href='https:/xxxx'><img width="1000" src="./doc/Drone_dataset.png"></a>
</div>
<table align=center width=1000px>
<center>
<tr>
<td>
<strong>Video drone dataset. </strong> Temporal consistency issues are masked in existing benchmark datasets. Consequently, we introduce a new evaluation dataset of 14 stabilized videos taken with drone-mounted cameras. This dataset features smooth motions to highlight temporal consistency issues. We hope that this will encourage research towards better restoration on stabilized video, which is a very relevant use case. For that, we make this validation dataset publicly available.
</td>
</tr>
</center>
</table>
<table align=center width=800px>
<br>
<tr><center>
<span style="font-size:28px"> <a href='https://xxxx'>[Download our drone dataset]</a>
</span>
</center>
</table>
<br>
<hr>
<center><h1>Code</h1></center>
<div class="mycontainer">
<div class="image">
<a href="https://github.com/centreborelli/video-MIMO-ROSO/"><img width="300" src="./doc/GitHub_logo.png"></a>
</div>
<div class="mytext">
In the following <a href="https://github.com/centreborelli/video-MIMO-ROSO/">github repository</a>, we provide training and testing code for the three state-of-sthe-art video denoising network: BasicVSR++<span style="font-size:9pt"> [1] </span>, M2Mnet<span style="font-size:9pt"> [2] </span> and ReMoNet<span style="font-size:9pt"> [3] </span>. For BasicVSR++, we adpated the <a href="https://ckkelvinchan.github.io/projects/BasicVSR++/">code</a> publicly available provided by the authors. For M2Mnet and ReMoNet, we implemented ourselves the corresponding architecture accordingly to the paper. <br>
In particular, we provide a recurrent implementation of BasicVSR++ and M2Mnet.
<br><br>
We also let available the code for evaluation, in particular our temporal consistency metric.
<br><br>
Our code uses Python3 and PyTorch.
</div>
</div>
<br>
<br>
<span style="font-size:11pt">[1]``On the Generalization of BasicVSR++ to Video Deblurring and Denoising'', Chan <i> et al </i>, 2022 </span> <br>
<span style="font-size:11pt">[2] ``Multiframe-to-Multiframe Network for Video Denoising'', Chen <i> et al </i>, 2021 </span> <br>
<span style="font-size:11pt">[3] ``ReMoNet: Recurrent Multi-Output Network for Efficient Video Denoising'', Xiang <i> et al </i>, 2022 </span> <br>
<br>
<hr>
<!--
<center><h1>Sentinel-2 sensor layout and L1BSR dataset</h1></center>
<table align=center width=400px>
<tr>
<td align=center width=400px>
<center>
<td><img class="round" style="width:550px" src="./resources/S2_sensor_layout.png"/></td>
</center>
</td>
</tr>
</table>
<table align=center width=850px>
<center>
<tr>
<td>
<strong>Sensor layout of the Sentinel-2 MSI.</strong> The Sentinel-2 MSI carries 12 CMOS detectors for the VNIR bands, with adjacent detectors having overlapping fields of view that result in overlapping regions in level-1B (L1B) images. The push-broom acquisition is done in the vertical direction.
</td>
</tr>
</center>
</table>
<table align=center width=850px>
<center>
<tr>
<td>
The <strong>L1BSR dataset</strong> includes 3740 pairs of overlapping image crops extracted from two L1B products. Each crop has a height of around 400 pixels and a variable width that depends on the overlap width between detectors for RGBN bands, typically around 120-200 pixels. In addition to detector parallax, there is also cross-band parallax for each detector, resulting in shifts between bands. Pre-registration is performed for both cross-band and cross-detector parallax, with a precision of up to a few pixels (typically less than 10 pixels).
</td>
</tr>
</center>
</table>
<div class="image-container">
<div class="image-pair">
<img src="resources/1_left.png" class="image1">
<img src="resources/1_right.png" class="image2">
</div>
<div class="image-pair">
<img src="resources/2_left.png" class="image1">
<img src="resources/2_right.png" class="image2">
</div>
<div class="image-pair">
<img src="resources/4_left.png" class="image1">
<img src="resources/4_right.png" class="image2">
</div>
<div class="image-pair">
<img src="resources/5_left.png" class="image1">
<img src="resources/5_right.png" class="image2">
</div>
</div>
<style>
.image-container {
display: flex;
justify-content: center;
align-items: center;
}
.image-pair {
position: relative;
width: 176px;
height: 176px;
margin: 10px;
overflow: hidden;
}
.image1, .image2 {
position: absolute;
top: 0;
left: 0;
width: 100%;
height: 100%;
opacity: 0;
}
.image1 {
opacity: 1;
z-index: 2;
animation: show-and-switch 4s infinite;
}
.image2 {
z-index: 1;
animation: show-and-switch 4s infinite;
animation-delay: 2s;
}
@keyframes show-and-switch {
0%, 48% {
opacity: 1;
}
48.01%, 50%, 98% {
opacity: 0;
}
98.01%, 100% {
opacity: 1;
}
}
</style>
<table align=center width=850px>
<center>
<tr>
<td>
Examples of overlapping L1B crops from the L1BSR dataset.
</td>
</tr>
</center>
</table>
<table align=center width=800px>
<br>
<tr><center>
<span style="font-size:28px"> <a href='https://zenodo.org/record/7826696'>[Download link. L1BSR dataset]</a>
</span>
</center>
</table>
<br>
<hr>
-->
<table align=center width=450px>
<center><h1>Paper</h1></center>
<tr>
<td><a href="https://arxiv.org/pdf/2408.12439"><img class="layered-paper-big" style="height:175px" src="./doc/Paper.png"/></a></td>
<td><span style="font-size:14pt">V. Dewil, Z. Zheng, A. Barral, L. Raad, N. Nicolas, I. Cassagne J-M. Morel, G. Facciolo B. Galerne and P. Arias<br>
<b>Adapting MIMO video restoration networks to low latency constraints.</b><br>
In <a href='https://bmvc2024.org/'>BMVC 2024</a>.<br>
(hosted on <a href="https://arxiv.org/pdf/2408.12439">ArXiv</a>)<br>
<!-- (<a href="./resources/camera-ready.pdf">camera ready</a>)<br> -->
<span style="font-size:4pt"><a href=""><br></a>
</span>
</td>
</tr>
</table>
<br>
<div class="disclaimerbox">
<span style="color:#646464">
<center><span style="font-size:28px"><b><a href="./doc/bibtex.txt"><span style="color:#646464">To cite us</span></a></b></span></center>
<br>
<i> @article{dewil2024adapting,<br>
title={Adapting MIMO video restoration networks to low latency constraints},<br>
author={Dewil, Val{\'e}ry and Zheng, Zhe and Barral, Arnaud and Raad, Lara and Nicolas, Nao and Cassagne, Ioannis and Morel, Jean-michel and Facciolo, Gabriele and Galerne, Bruno and Arias, Pablo},<br>
journal={arXiv preprint arXiv:2408.12439},<br>
year={2024}<br>
}
</i>
<br>
</span>
</div>
<hr>
<br>
<table align=center width=900px>
<tr>
<td width=400px>
<left>
<center><h1>Acknowledgements</h1></center>
This template was originally made by <a href="http://web.mit.edu/phillipi/">Phillip Isola</a> and <a href="http://richzhang.github.io/">Richard Zhang</a> for a <a href="http://richzhang.github.io/colorization/">colorful</a> ECCV project; the code can be found <a href="https://github.com/richzhang/webpage-template">here</a>.
</left>
</td>
</tr>
</table>
<br>
</body>
</html>