-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathindex.html
More file actions
795 lines (751 loc) · 45.2 KB
/
Copy pathindex.html
File metadata and controls
795 lines (751 loc) · 45.2 KB
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
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="description" content="Seeing Beyond8Bits: Subjective and Objective Quality Assessment of HDR-UGC Videos. The largest crowdsourced HDR-UGC VQA dataset (~44K videos from 6,861 sources, >1.5M ratings) and HDR-Q, the first MLLM for HDR-UGC quality assessment, trained with HDR-Aware Policy Optimization (HAPO).">
<meta property="og:title" content="Seeing Beyond8Bits — HDR-UGC Quality Assessment (Beyond8Bits + HDR-Q)" />
<meta property="og:description" content="The largest HDR-UGC video quality dataset (~44K videos from 6,861 sources, >1.5M ratings) plus HDR-Q, the first MLLM for HDR-UGC VQA, trained with HAPO (HDR-Aware Policy Optimization)." />
<meta property="og:image" content="static/images/overview-fig1.png" />
<meta name="twitter:title" content="Seeing Beyond8Bits — HDR-UGC Quality Assessment (Beyond8Bits + HDR-Q)">
<meta name="twitter:description" content="Largest crowdsourced HDR-UGC dataset and the first MLLM (HDR-Q) for HDR-UGC VQA, trained with HAPO.">
<meta name="twitter:image" content="static/images/overview-fig1.png">
<meta name="twitter:card" content="summary_large_image">
<meta name="keywords" content="HDR, UGC, Video Quality Assessment, Dataset, MOS, VQA, MLLM, HDR-Q, Reinforcement Learning, HAPO, GRPO, LIVE-HDR, SFV+HDR">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Seeing Beyond8Bits — Subjective & Objective HDR-UGC Quality Assessment</title>
<link rel="icon" type="image/svg+xml" href="static/images/favicon.svg">
<link rel="icon" type="image/x-icon" href="static/images/favicon.ico">
<link rel="apple-touch-icon" sizes="180x180" href="static/images/apple-touch-icon.png">
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="static/css/bulma.min.css">
<link rel="stylesheet" href="static/css/fontawesome.all.min.css">
<link rel="stylesheet" href="static/css/index.css">
<script defer src="static/js/fontawesome.all.min.js"></script>
</head>
<body>
<!-- HERO -->
<section class="hero hero-main">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered hero-panel">
<span class="hero-eyebrow">Dataset & Method · CVPR 2026</span>
<h1 class="title is-1 publication-title">
Seeing <span class="title-accent">Beyond8Bits</span>:<br>
Subjective and Objective Quality Assessment of HDR-UGC Videos
</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">Shreshth Saini<sup>1</sup>, Bowen Chen<sup>1</sup>, Neil Birkbeck<sup>2</sup>, Yilin Wang<sup>2</sup>, Balu Adsumilli<sup>2</sup>, Alan C. Bovik<sup>1,3</sup></span>
</div>
<div class="is-size-6 publication-authors">
<span class="author-block"><sup>1</sup>Laboratory for Image and Video Engineering (LIVE), UT Austin <sup>2</sup>Google / YouTube <sup>3</sup>University of Colorado Boulder</span>
</div>
<div class="publication-authors">
<span class="venue-line">Dataset: <b>Beyond8Bits</b> · Method: <b>HDR-Q</b> (MLLM) with <b>HAPO</b> (HDR-Aware Policy Optimization)</span>
</div>
<div class="publication-links">
<span class="link-block">
<a href="static/pdfs/beyond8bits.pdf" target="_blank" class="external-link button is-normal is-rounded is-primary">
<span class="icon"><i class="fas fa-file-pdf"></i></span>
<span>Paper</span>
</a>
</span>
<span class="link-block">
<a href="static/pdfs/beyond8bits-supp.pdf" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon"><i class="fas fa-file-pdf"></i></span>
<span>Supplementary</span>
</a>
</span>
<span class="link-block">
<a href="https://github.com/shreshthsaini/Beyond8Bits" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon"><i class="fab fa-github"></i></span>
<span>GitHub</span>
</a>
</span>
<span class="link-block">
<a href="https://live.ece.utexas.edu/research/beyond8bits/index.html" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon"><i class="fas fa-database"></i></span>
<span>Dataset</span>
</a>
</span>
<span class="link-block">
<a href="https://live.ece.utexas.edu/" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon"><i class="fas fa-flask"></i></span>
<span>LIVE @ UT Austin</span>
</a>
</span>
</div>
<div class="hero-stats">
<div class="stat-item">
<p class="stat-value">6,861</p>
<p class="stat-label">HDR-UGC Source Videos<br><span class="is-size-7">(2,253 Crowd + 4,608 Vimeo)</span></p>
</div>
<div class="stat-item">
<p class="stat-value">~44K</p>
<p class="stat-label">Transcoded / Distorted Videos<br><span class="is-size-7">bit-ladder 0.2–5 Mbps</span></p>
</div>
<div class="stat-item">
<p class="stat-value">>1.5 M</p>
<p class="stat-label">Crowd Subjective Ratings<br><span class="is-size-7">~35 ratings / video on AMT</span></p>
</div>
<div class="stat-item">
<p class="stat-value">HDR-Q</p>
<p class="stat-label">First MLLM for HDR-UGC VQA<br><span class="is-size-7">Ovis2.5 + LoRA, trained with HAPO</span></p>
</div>
</div>
<p class="is-size-7 has-text-grey mt-3" style="max-width:720px;margin:0 auto;">Publish-ready release: <b>5,917</b> sources (<b>2,153</b> Crowd + <b>3,764</b> Vimeo) and <b>41,419</b> transcoded clips in <code>data/Beyond8Bits_publish.csv</code>.</p>
</div>
</div>
</div>
</div>
</section>
<!-- Dataset Hero Figure -->
<section class="section section-soft">
<div class="container is-max-desktop">
<div class="column figure-frame">
<img src="static/images/overview-fig1.png" alt="Overview of Beyond8Bits dataset and performance evaluation" class="dataset-figure">
<p class="figure-caption"><strong>Overview of our dataset and performance evaluation.</strong> <em>Top:</em> Example comparisons between HDR and SDR frames, illustrating differences in brightness range, color depth, and visual detail across diverse scenes. <em>Bottom-left:</em> The distribution of video categories in the HDR-UGC-44K dataset, covering human-centered content, nature & outdoor scenes, and various other real-world scenarios. <em>Bottom-right:</em> Performance comparison between our proposed HDR-Q model and baseline methods on three datasets, where HDR-Q achieves significant improvements in PLCC.</p>
</div>
</div>
</section>
<!-- Abstract -->
<section class="section">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">Abstract</h2>
<div class="card abstract-card">
<div class="card-content">
<div class="content has-text-justified">
<p class="abstract-text">
<b>High Dynamic Range (HDR)</b> user-generated (UGC) videos are rapidly proliferating across social platforms, yet most perceptual video quality assessment (VQA) systems remain tailored to Standard Dynamic Range (SDR). HDR's higher bit depth, wide color gamut, and elevated luminance range expose distortions such as near-black crushing, highlight clipping, banding, and exposure flicker that amplify UGC artifacts and challenge SDR models. To catalyze progress, we curate <b><i>Beyond8Bits</i></b>, a large-scale subjective dataset of <b>~44K videos from 6,861 sources</b> with <b>>1.5M crowd ratings</b>, spanning diverse scenes, capture conditions, and compression settings. We further introduce <b>HDR-Q</b>, the first Multimodal Large Language Model (MLLM) for HDR-UGC VQA. We propose (i) a novel <b>HDR-aware vision encoder</b> (a SigLIP-2 adapter finetuned with an HDR–SDR dual-domain contrastive objective) to produce HDR-sensitive embeddings, and (ii) <b><i>HAPO</i></b> (HDR-Aware Policy Optimization), an RL fine-tuning framework that anchors reasoning to HDR cues. HAPO augments GRPO with an HDR–SDR contrastive KL that encourages token reliance on HDR inputs, dual-entropy regularization to prevent modality neglect, and a Gaussian-weighted regression reward for fine-grained MOS calibration. Across Beyond8Bits and public HDR-VQA benchmarks (LIVE-HDR, SFV+HDR), HDR-Q delivers state-of-the-art performance. Beyond8Bits subsumes our earlier <i>CHUG</i> (ICIP '25) and <i>BrightVQ</i> (WACV '26 Oral) releases.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- What's New -->
<section class="section section-soft">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">What's in Beyond8Bits</h2>
<p class="section-intro">
Beyond8Bits unifies two prior subjective studies (<b>CHUG</b>, <b>BrightVQ</b>) and adds a substantially larger Vimeo-sourced HDR-UGC partition, plus <b>HDR-Q</b> — the first MLLM tailored for HDR-UGC quality assessment. Each source video is transcoded across a multi-rung bit-ladder to expose realistic compression, up/down-scaling, and re-encoding distortions.
</p>
<div class="feature-grid">
<div class="feature-card">
<h4>Largest HDR-UGC study</h4>
<p>>1.5 M opinion scores across ~44K videos from 6,861 sources — an order of magnitude beyond prior HDR VQA datasets.</p>
</div>
<div class="feature-card">
<h4>Two complementary sources</h4>
<p>Consumer-device HDR captures (<b>Crowd</b>: iPhone / Pixel / Galaxy, 2,253 sources) plus CC-licensed Vimeo HDR uploads (<b>Vimeo</b>: 4,608 sources). Both share the same bit-ladder and AMT instrument.</p>
</div>
<div class="feature-card">
<h4>Full HDR signaling preserved</h4>
<p>Every clip retains <b>10-bit HEVC, PQ transfer, BT.2020</b> gamut through transcoding; clips are trimmed to ≤10 seconds.</p>
</div>
<div class="feature-card">
<h4>HDR-Q (first MLLM for HDR-UGC)</h4>
<p>An Ovis2.5-based MLLM with a SigLIP-2 HDR-aware vision encoder, trained with <b>HAPO</b> (extends GRPO with HDR–SDR contrastive KL, dual-entropy regularization, and high-entropy weighting).</p>
</div>
<div class="feature-card">
<h4>Rich metadata</h4>
<p>Per-video <b>MOS</b> / <b>SOS</b> (SUREAL-aggregated), <code>type</code>, <code>ref</code>, <code>resolution</code>, <code>bitrate</code>, <code>orientation</code>, <code>framerate</code>, <code>split</code>, and frame dimensions.</p>
</div>
<div class="feature-card">
<h4>Open & reusable</h4>
<p>Metadata released under <b>CC BY 4.0</b>; videos retain their original licenses for non-commercial research.</p>
</div>
</div>
</div>
</section>
<!-- Dataset Composition / Stats -->
<section class="section">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">Dataset Composition</h2>
<p class="section-intro">
Beyond8Bits is partitioned into a <b>Crowd</b> split (consumer-device HDR-UGC from a dedicated crowdsourcing campaign) and a <b>Vimeo</b> split (CC-licensed public HDR uploads). Both partitions share the same transcoding bit-ladder, AMT rating instrument, and subject pool, enabling apples-to-apples cross-domain analysis. The table below reports the <i>publish-ready release</i>; paper-reported totals are slightly larger and track ongoing license clearing.
</p>
<div class="split-grid">
<div class="split-card">
<h4>Crowd partition (includes CHUG)</h4>
<div class="sub-stat"><span>Source videos (released)</span><strong>2,153</strong></div>
<div class="sub-stat"><span>Transcoded videos (released)</span><strong>15,071</strong></div>
<div class="sub-stat"><span>Devices</span><strong>iPhone / Pixel / Galaxy (10-bit HEVC)</strong></div>
<div class="sub-stat"><span>Orientations</span><strong>Portrait 12,075 · Landscape 2,996</strong></div>
<div class="sub-stat"><span>Consent</span><strong>Non-exclusive research redistribution</strong></div>
</div>
<div class="split-card">
<h4>Vimeo partition</h4>
<div class="sub-stat"><span>Source videos (released)</span><strong>3,764</strong></div>
<div class="sub-stat"><span>Transcoded videos (released)</span><strong>26,348</strong></div>
<div class="sub-stat"><span>License</span><strong>Creative Commons (original Vimeo licenses retained)</strong></div>
<div class="sub-stat"><span>Orientations</span><strong>Landscape 22,974 · Portrait 3,374</strong></div>
<div class="sub-stat"><span>Common framerates</span><strong>24 / 25 / 30 / 60 fps</strong></div>
</div>
<div class="split-card">
<h4>Subjective study</h4>
<div class="sub-stat"><span>Platform</span><strong>Amazon Mechanical Turk (HDR-capable devices only)</strong></div>
<div class="sub-stat"><span>Rating instrument</span><strong>Continuous 0–100 Likert, ITU-R BT.500-14</strong></div>
<div class="sub-stat"><span>Ratings collected</span><strong>>1.5 M (valid, post-QC)</strong></div>
<div class="sub-stat"><span>Avg. ratings / video</span><strong>~35</strong></div>
<div class="sub-stat"><span>Screening</span><strong>HDR-display qualification quiz, training + calibration phase, golden-set + repeat videos, bit-depth / bandwidth checks</strong></div>
<div class="sub-stat"><span>MOS aggregation</span><strong>SUREAL MLE (median inter-subject SRCC 0.90)</strong></div>
</div>
</div>
<div class="split-grid" style="margin-top:1rem;">
<div class="split-card">
<h4>Transcoding grid (bit-ladder)</h4>
<div class="sub-stat"><span>Target resolutions</span><strong>360p · 720p · 1080p (+ source/ref)</strong></div>
<div class="sub-stat"><span>Bitrate rungs</span><strong>0.2 / 0.5 / 1 / 2 / 3 Mbps (paper up to 5 Mbps)</strong></div>
<div class="sub-stat"><span>Clip length</span><strong>≤ 10 seconds</strong></div>
<div class="sub-stat"><span>HDR signaling</span><strong>10-bit HEVC · PQ transfer · BT.2020 gamut</strong></div>
</div>
<div class="split-card">
<h4>Train / Val / Test split (release)</h4>
<div class="sub-stat"><span>Split policy</span><strong>70 / 10 / 20 by source identity</strong></div>
<div class="sub-stat"><span>Train</span><strong>28,987</strong></div>
<div class="sub-stat"><span>Validation</span><strong>4,151</strong></div>
<div class="sub-stat"><span>Test</span><strong>8,281</strong></div>
<div class="sub-stat"><span>Split column</span><strong><code>split</code> in <code>Beyond8Bits_publish.csv</code></strong></div>
</div>
<div class="split-card">
<h4>Source vs. transcoded</h4>
<div class="sub-stat"><span>Reference sources (<code>ref=1</code>)</span><strong>5,917</strong></div>
<div class="sub-stat"><span>Transcoded clips (<code>ref=0</code>)</span><strong>35,502</strong></div>
<div class="sub-stat"><span>360p transcodes</span><strong>5,917</strong></div>
<div class="sub-stat"><span>720p transcodes</span><strong>11,834</strong></div>
<div class="sub-stat"><span>1080p transcodes</span><strong>17,751</strong></div>
</div>
</div>
</div>
</section>
<!-- Content Distribution Figure -->
<section class="section section-soft">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">Content & Score Distribution</h2>
<p class="section-intro">
Beyond8Bits covers a wide swath of perceptual space along spatial and temporal complexity, luminance/chrominance, and overall quality. The grid below samples representative frames across the dataset's content categories — portraits, group & event footage, and indoor/outdoor scenes — captured under diverse lighting and device conditions.
</p>
<div class="column figure-frame">
<img src="static/images/video-grid-supple.png" alt="Overview of the Beyond8Bits video dataset illustrated through sampled frames" class="dataset-figure">
<p class="figure-caption">Overview of our video dataset illustrated through sampled frames across portraits, group & events, and indoor / outdoor scenes.</p>
</div>
</div>
</section>
<!-- Carousel of Beyond8Bits source frames with crowd-sourced MOS -->
<section class="section">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">Sample Frames with Crowd-Sourced MOS</h2>
<p class="section-intro">
A sample of Beyond8Bits source frames spanning the full MOS range, with the corresponding crowd-sourced <b>Actual MOS</b>. Frames cover both Crowd and Vimeo partitions and both orientations.
</p>
<div class="carousel-container frame-carousel">
<div class="carousel-item carousel-shell" id="carouselItem">
<figure class="image carousel-media">
<img id="carouselImage" src="" alt="HDR-UGC sample frame">
</figure>
</div>
<div id="carouselCaption" class="has-text-centered mt-3 carousel-caption">
<p><strong>Actual MOS:</strong> <span id="actualMos"></span> · <span id="frameMeta" class="has-text-grey"></span></p>
</div>
</div>
<div class="buttons is-centered mt-4">
<button class="button is-dark is-rounded nav-btn" id="prevBtn">
<span class="icon"><i class="fas fa-chevron-left"></i></span>
<span>Previous</span>
</button>
<button class="button is-dark is-rounded nav-btn" id="nextBtn">
<span>Next</span>
<span class="icon"><i class="fas fa-chevron-right"></i></span>
</button>
</div>
</div>
</section>
<!-- AMT Interface -->
<section class="section">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">Subjective Study Interface</h2>
<p class="section-intro">
Ratings were collected on Amazon Mechanical Turk with a continuous single-stimulus rating bar, HDR-capable display attestation, and gold / self-consistency screening HITs interleaved throughout the session.
</p>
<div class="column figure-frame">
<img src="static/images/amt-interface.png" alt="Amazon Mechanical Turk rating interface" class="dataset-figure">
<p class="figure-caption">The web-based AMT rating interface used across the CHUG, BrightVQ and Vimeo sub-studies that together form Beyond8Bits.</p>
</div>
</div>
</section>
<!-- Dataset Access -->
<section class="section section-soft dataset-access" id="dataset-access">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">How to Access the Dataset</h2>
<p class="section-intro">
We release per-video metadata (<code>video_id</code>, <code>mos</code>, <code>sos</code>, <code>type</code>, <code>ref</code>, <code>resolution</code>, <code>bitrate</code>, <code>orientation</code>, <code>framerate</code>, <code>split</code>, <code>height</code>, <code>width</code>) for the <b>full publish-ready release</b> (41,419 clips across Crowd + Vimeo) and the corresponding video payloads via an S3-hosted mirror. The complete dataset package, including raw per-rating CSV files, is also available on UT-Box.
</p>
<div class="columns is-multiline">
<div class="column is-full">
<div class="card access-card">
<header class="card-header">
<p class="card-header-title">Clone the GitHub repository</p>
</header>
<div class="card-content">
<p>All metadata files, ID lists, rating CSVs and licensing info live in the <a href="https://github.com/shreshthsaini/Beyond8Bits" target="_blank">Beyond8Bits GitHub repository</a>. The paper supplementary contains detailed per-field documentation.</p>
<pre><code>git clone https://github.com/shreshthsaini/Beyond8Bits.git
cd Beyond8Bits</code></pre>
</div>
</div>
</div>
<div class="column is-full">
<div class="card access-card">
<header class="card-header">
<p class="card-header-title">Grab the video-ID manifest</p>
</header>
<div class="card-content">
<p>The full Beyond8Bits manifest (41,419 transcoded videos across Crowd + Vimeo partitions) is at <code>data/Beyond8Bits_publish.txt</code> (one hashed ID per line) with matched MOS / SOS and per-video metadata in <code>data/Beyond8Bits_publish.csv</code>. The CHUG-compatible crowd subset remains available at <code>data/Beyond8Bits_publish_crowd.csv</code> / <code>.txt</code>.</p>
</div>
</div>
</div>
<div class="column is-full">
<div class="card access-card">
<header class="card-header">
<p class="card-header-title">Download a single video with the AWS CLI</p>
</header>
<div class="card-content">
<p>Replace <code>VIDEO_ID</code> with any hashed ID from the manifest:</p>
<pre><code>aws s3 cp s3://ugchdrmturk/videos/VIDEO_ID.mp4 ./Beyond8Bits_Videos/</code></pre>
</div>
</div>
</div>
<div class="column is-full">
<div class="card access-card">
<header class="card-header">
<p class="card-header-title">Bulk download all videos</p>
</header>
<div class="card-content">
<p>To mirror the full published partition in one shot:</p>
<pre><code>cat data/Beyond8Bits_publish.txt | while read video; do
aws s3 cp s3://ugchdrmturk/videos/${video}.mp4 ./Beyond8Bits_Videos/
done</code></pre>
</div>
</div>
</div>
<div class="column is-full">
<div class="card access-card">
<header class="card-header">
<p class="card-header-title">Stream a video directly in the browser</p>
</header>
<div class="card-content">
<p>Replace <code>VIDEO_ID</code> below to play any video without the AWS CLI:</p>
<p><a href="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/VIDEO_ID.mp4" target="_blank">https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/VIDEO_ID.mp4</a></p>
<p class="mt-2"><strong>Example:</strong>
<a href="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/9ae245a27cc5ea9d2f3fae9692250281.mp4" target="_blank">9ae245a27cc5ea9d2f3fae9692250281.mp4</a>
</p>
</div>
</div>
</div>
<div class="column is-full">
<div class="card access-card">
<header class="card-header">
<p class="card-header-title">Load scores & metadata in Python</p>
</header>
<div class="card-content">
<pre><code>import pandas as pd
df = pd.read_csv("data/Beyond8Bits_publish.csv")
print(df.columns.tolist())
# ['video_id', 'mos', 'sos', 'type', 'ref', 'resolution',
# 'bitrate', 'orientation', 'framerate', 'split',
# 'height', 'width']
print(df["mos"].describe())
print(df["type"].value_counts()) # Crowd vs Vimeo
print(df["split"].value_counts()) # train / test</code></pre>
</div>
</div>
</div>
<div class="column is-full">
<div class="card access-card">
<header class="card-header">
<p class="card-header-title">Full dataset package on UT-Box</p>
</header>
<div class="card-content">
<p>The complete Beyond8Bits release — including per-rating raw CSVs and the full metadata bundle — is mirrored on UT-Box:</p>
<p><a href="https://utexas.box.com/s/pvz8zpmpogvpy62pqpar2e54c6ovyd5z" target="_blank">https://utexas.box.com/s/pvz8zpmpogvpy62pqpar2e54c6ovyd5z</a></p>
</div>
</div>
</div>
</div>
<div class="placeholder-note">
<strong>Note:</strong> <code>Beyond8Bits_publish.csv</code> / <code>.txt</code> now covers all 41,419 transcoded videos across Crowd + Vimeo partitions.
</div>
</div>
</section>
<!-- Sample Videos -->
<section class="section samples-section">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">Sample Videos</h2>
<p class="section-intro">Representative clips drawn from the CHUG and BrightVQ sub-studies (all part of Beyond8Bits). Best viewed on an HDR10/HLG-capable display.</p>
<h3 class="title is-4 sample-group-title">Portraits</h3>
<div class="columns is-multiline">
<div class="column is-one-third">
<div class="card sample-card crowd portrait">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/b882a56f87be722f24206e007c31ee4c.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Crowd · Portrait</p>
<p class="is-size-7 has-text-grey">1080p · 30 fps · MOS 86.0</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card crowd portrait">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/166f24971e567ed278e8b2a3ec711446.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Crowd · Portrait</p>
<p class="is-size-7 has-text-grey">1080p · 30 fps · MOS 79.8</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card crowd portrait">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/3db0ff4a6485e5f549e8d0bbf07b475b.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Crowd · Portrait</p>
<p class="is-size-7 has-text-grey">1080p · 29.97 fps · MOS 78.8</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card vimeo portrait">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/93cd38af54ae7762d8c3c44f50130a49.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Vimeo · Portrait</p>
<p class="is-size-7 has-text-grey">1080p · 30 fps · MOS 80.1</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card vimeo portrait">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/6e09adb286b5a926f54999b67cfa0fd5.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Vimeo · Portrait</p>
<p class="is-size-7 has-text-grey">1080p · 30 fps · MOS 79.4</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card vimeo portrait">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/1e17f5bde97ff2b7e24514c99a436d56.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Vimeo · Portrait</p>
<p class="is-size-7 has-text-grey">1080p · 30 fps · MOS 79.3</p>
</div>
</div>
</div>
</div>
<h3 class="title is-4 mt-6 sample-group-title">Landscapes</h3>
<div class="columns is-multiline">
<div class="column is-one-third">
<div class="card sample-card crowd landscape">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/bf44eaf6fff562e1f3fd061617c8392f.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Crowd · Landscape</p>
<p class="is-size-7 has-text-grey">1080p · 29.97 fps · MOS 78.4</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card crowd landscape">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/9dbe220b1d07df571e6a207cf6c126eb.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Crowd · Landscape</p>
<p class="is-size-7 has-text-grey">1080p · 30 fps · MOS 77.1</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card crowd landscape">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/e5d086082992215d5e18024df6fd0f0c.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Crowd · Landscape</p>
<p class="is-size-7 has-text-grey">1080p · 59.94 fps · MOS 76.9</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card vimeo landscape">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/3b77fab10ee478d0b98fa27e04e24601.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Vimeo · Landscape</p>
<p class="is-size-7 has-text-grey">1080p · 30 fps · MOS 81.1</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card vimeo landscape">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/7dc195a8b4da9caf63de2fb14b9dddba.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Vimeo · Landscape</p>
<p class="is-size-7 has-text-grey">1080p · 58.69 fps · MOS 80.9</p>
</div>
</div>
</div>
<div class="column is-one-third">
<div class="card sample-card vimeo landscape">
<div class="card-image">
<video class="video-card" controls preload="metadata">
<source src="https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/4b5bb3bc6b46b275c957f1dd512b89e8.mp4" type="video/mp4">
</video>
</div>
<div class="card-content has-text-centered">
<p class="subtitle is-6">Vimeo · Landscape</p>
<p class="is-size-7 has-text-grey">1080p · 58.1 fps · MOS 81.8</p>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Related Sub-studies -->
<section class="section">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">Part of a Family of HDR-UGC Releases</h2>
<p class="section-intro">
Beyond8Bits consolidates and extends two earlier releases from our lab. If you use Beyond8Bits, please also consider citing its predecessors.
</p>
<div class="related-row">
<a class="related-card" href="https://live.ece.utexas.edu/research/chug/index.html" target="_blank">
<span class="tag-row">ICIP 2025 · Dataset</span>
<h4>CHUG</h4>
<p>Crowdsourced HDR-UGC Quality Dataset — 856 sources, 5,992 videos, 211,848 ratings. The crowd-captured seed study that Beyond8Bits extends.</p>
</a>
<a class="related-card" href="https://shreshthsaini.github.io/BrightVQ/" target="_blank">
<span class="tag-row">WACV 2026 Oral · Dataset + Method</span>
<h4>BrightRate / BrightVQ</h4>
<p>HDR-UGC VQA dataset (300 src / 2,100 videos / 73,794 ratings) and a classical feature-based quality model. Subset of the Beyond8Bits corpus.</p>
</a>
<a class="related-card" href="https://live.ece.utexas.edu/research/LIVE_UGC_HDR/index.html" target="_blank">
<span class="tag-row">LIVE @ UT Austin</span>
<h4>LIVE HDR UGC Program</h4>
<p>Long-running program at LIVE on HDR capture, tone-mapping, inverse tone-mapping, and perceptual quality assessment.</p>
</a>
</div>
</div>
</section>
<!-- Method: HDR-Q + HAPO -->
<section class="section">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">Method: HDR-Q + HAPO</h2>
<p class="section-intro">
<b>HDR-Q</b> is the first multimodal large language model for HDR-UGC quality assessment. It integrates (i) an <b>HDR-aware vision encoder</b> that produces HDR-sensitive embeddings while retaining semantic alignment, and (ii) <b>HDR-Aware Policy Optimization (HAPO)</b>, a reinforcement-learning objective that extends GRPO with HDR-specific grounding and stability terms.
</p>
<div class="column figure-frame">
<img src="static/images/hdrq-hapo-method.png" alt="Overview of HDR-Q with HAPO" class="dataset-figure">
<p class="figure-caption"><strong>Overview of HDR-Q with HAPO.</strong> <em>Left:</em> HAPO compares rollouts under HDR inputs (text + SDR + HDR tokens) versus an HDR-deprived pathway (text + SDR only), maximizing their KL divergence to enforce HDR grounding and applying dual-entropy regularization to prevent reward hacking. Group-wise rewards include MOS/attribute accuracy, reasoning quality, and self-rewarding. <em>Right:</em> a LoRA-tuned LLM decodes the HDR-aware reasoning; visual inputs originate from both a standard encoder and our HDR-aware adapter.</p>
</div>
<div class="feature-grid">
<div class="feature-card">
<h4>HDR-aware vision encoder</h4>
<p>A SigLIP-2 encoder adapted with HDR–SDR dual-domain contrastive supervision. Captions are generated by Qwen2.5-VL-72B; HDR embeddings are pushed closer to their caption than the tone-mapped SDR counterpart, preserving 10-bit PQ / BT.2020 cues without collapsing onto SDR representations.</p>
</div>
<div class="feature-card">
<h4>HDR–SDR contrastive KL</h4>
<p>Maximizes <code>D<sub>KL</sub>(π<sup>HDR</sup> ‖ π<sup>SDR</sup>)</code> between rollouts with and without HDR tokens, preventing modality neglect (policies that read textual priors while ignoring the HDR signal).</p>
</div>
<div class="feature-card">
<h4>Dual-entropy regularization</h4>
<p>Per-pathway entropy penalties on both HDR and SDR rollouts prevent the trivial "entropy inflation" solution to the contrastive KL while keeping HDR-grounded distributions sharp.</p>
</div>
<div class="feature-card">
<h4>High-entropy weighting (HEW)</h4>
<p>Rescales GRPO's group-normalized advantage with per-token entropy, concentrating the learning signal on informative reasoning tokens (e.g., banding, highlight clipping, near-black crushing).</p>
</div>
<div class="feature-card">
<h4>Gaussian MOS reward + self-reward</h4>
<p>A Gaussian-weighted regression reward <code>R<sub>sc</sub></code> calibrates fine-grained MOS; a group-level self-reward <code>R<sub>self</sub></code> consolidates within-group consensus for reasoning stability.</p>
</div>
<div class="feature-card">
<h4>Two-stage RL training</h4>
<p><b>Stage 1 — Modality Alignment:</b> short HAPO runs align HDR tokens and projection layers. <b>Stage 2 — Full-RFT:</b> complete HAPO on the HDR-UGC corpus. Base: Ovis2.5 + rank-4 LoRA; 8 uniformly-sampled frames at native 10-bit PQ; trained on 4× NVIDIA H200 GPUs.</p>
</div>
</div>
</div>
</section>
<!-- License -->
<section class="section section-soft">
<div class="container is-max-desktop">
<h2 class="title is-3 section-title">License & Terms of Use</h2>
<div class="card abstract-card"><div class="card-content"><div class="content">
<p>
Beyond8Bits metadata (CSV / TXT manifests, rating aggregates) is released under <b>Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)</b>. Video payloads retain their original licenses (CC-licensed Vimeo clips; user-contributed crowd videos for which we have non-exclusive research redistribution rights). The dataset is intended strictly for <b>non-commercial research</b> in HDR video quality assessment, tone-mapping, inverse tone-mapping, generative modeling evaluation, and related perceptual studies.
</p>
<p>
By downloading any portion of the dataset you agree not to (a) redistribute the raw video payloads outside the approved S3 mirror without written consent, (b) attempt to de-anonymize contributing workers, or (c) use the data to train models for deployment in a commercial product without a separate license from UT Austin / YouTube. Please refer to <code>LICENSE</code> in the GitHub repo for the full terms.
</p>
</div></div></div>
</div>
</section>
<!-- BibTeX -->
<section class="section bibtex-section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title is-3 section-title">BibTeX</h2>
<p class="section-intro">If you find Beyond8Bits useful in your research, please cite:</p>
<pre class="bibtex-block"><code>@InProceedings{Saini_2026_CVPR,
author = {Saini, Shreshth and Chen, Bowen and Wang, Yilin and Birkbeck, Neil and Adsumilli, Balu and Bovik, Alan C.},
title = {Seeing Beyond 8bits: Subjective and Objective Quality Assessment of HDR-UGC Videos},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2026},
pages = {15538-15549}
}</code></pre>
<p class="section-intro" style="margin-top:1.2rem;">Please also consider citing the two sub-studies that Beyond8Bits extends:</p>
<pre class="bibtex-block"><code>@INPROCEEDINGS{Saini_2025_ICIP_CHUG,
author = {Saini, Shreshth and Bovik, Alan C. and Birkbeck, Neil and Wang, Yilin and Adsumilli, Balu},
booktitle = {2025 IEEE International Conference on Image Processing (ICIP)},
title = {CHUG: Crowdsourced User-Generated HDR Video Quality Dataset},
year = {2025},
pages = {2504-2509},
doi = {10.1109/ICIP55913.2025.11084488}
}
@InProceedings{Saini_2026_WACV_BrightRate,
author = {Saini, Shreshth and Chen, Bowen and Wang, Yilin and Birkbeck, Neil and Adsumilli, Balu and Bovik, Alan C.},
title = {BrightRate: Quality Assessment for User-Generated HDR Videos},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
month = {March},
year = {2026},
pages = {1522-1532}
}</code></pre>
</div>
</section>
<!-- Footer -->
<footer class="footer site-footer">
<div class="container">
<div class="columns is-centered">
<div class="column is-8">
<div class="content has-text-centered">
<p>
© 2026 Shreshth Saini · Laboratory for Image and Video Engineering (<a href="https://live.ece.utexas.edu/" target="_blank">LIVE</a>), The University of Texas at Austin.<br>
Page adapted from the <a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">Academic Project Page Template</a> (originally from <a href="https://nerfies.github.io" target="_blank">Nerfies</a>).
This website is licensed under a <a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/" target="_blank">Creative Commons Attribution-ShareAlike 4.0 International License</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
<!-- Carousel script -->
<script>
let imagesData = [];
let currentIndex = 0;
const imageElement = document.getElementById("carouselImage");
const mosElement = document.getElementById("actualMos");
const metaElement = document.getElementById("frameMeta");
const captionElement = document.getElementById("carouselCaption");
async function fetchJson(url) {
const response = await fetch(url, { cache: "no-store" });
if (!response.ok) {
throw new Error("Failed to fetch " + url + " (" + response.status + ")");
}
return response.json();
}
async function loadImagesData() {
const dataSources = [
"images_data.json",
"./images_data.json",
"https://raw.githubusercontent.com/shreshthsaini/Beyond8Bits/main/images_data.json",
"https://raw.githubusercontent.com/shreshthsaini/Beyond8Bits/master/images_data.json"
];
let lastError = null;
for (const source of dataSources) {
try {
const data = await fetchJson(source);
if (Array.isArray(data) && data.length > 0) return data;
} catch (error) {
lastError = error;
}
}
throw lastError || new Error("Unable to load images_data.json from all sources.");
}
function showCarouselError(message) {
if (captionElement) {
captionElement.innerHTML = "<p><strong>Sample frames unavailable:</strong> " + message + "</p>";
}
}
function updateCarousel() {
if (imagesData.length === 0) return;
const frame = imagesData[currentIndex];
const mos = Number.parseFloat(frame.mos);
imageElement.style.opacity = "0";
imageElement.onload = () => { imageElement.style.opacity = "1"; };
imageElement.onerror = () => {
imageElement.style.opacity = "1";
showCarouselError("Could not load frame image at " + frame.src);
};
imageElement.src = frame.src;
imageElement.alt = `HDR-UGC sample frame ${currentIndex + 1}`;
mosElement.textContent = Number.isFinite(mos) ? mos.toFixed(2) : "N/A";
const metaBits = [frame.type, frame.orientation, frame.resolution, frame.framerate ? `${frame.framerate} fps` : null].filter(Boolean);
metaElement.textContent = metaBits.join(" · ");
}
document.getElementById("prevBtn").addEventListener("click", () => {
if (imagesData.length === 0) return;
currentIndex = (currentIndex - 1 + imagesData.length) % imagesData.length;
updateCarousel();
});
document.getElementById("nextBtn").addEventListener("click", () => {
if (imagesData.length === 0) return;
currentIndex = (currentIndex + 1) % imagesData.length;
updateCarousel();
});
loadImagesData()
.then(images => {
imagesData = images;
updateCarousel();
})
.catch(error => {
console.error("Error loading images:", error);
showCarouselError("Failed to load frame metadata. Ensure images_data.json is available.");
});
</script>
</body>
</html>