Beyond8Bits: Seeing Beyond 8 Bits

Subjective and objective quality assessment of HDR user-generated videos. CVPR 2026.

Shreshth Saini, Bowen Chen, Neil Birkbeck, Yilin Wang, Balu Adsumilli, Alan C. Bovik. LIVE, UT Austin; Google / YouTube; University of Colorado Boulder.

Paper (arXiv 2603.00938) · Project page · GitHub · LIVE dataset page · UT-Box full package

Beyond8Bits is the largest crowdsourced HDR-UGC video quality dataset we know of. The published release has 5,917 HDR source videos (2,153 crowd-contributed, 3,764 from Vimeo) transcoded along a bitrate ladder into 41,419 clips, rated by Amazon Mechanical Turk workers (about 1.46 M ratings, about 35 per clip) on a continuous 0 to 100 scale following ITU-R BT.500-14. MOS values come from SUREAL maximum-likelihood aggregation; median inter-subject SRCC is 0.90. The paper also introduces HDR-Q, a multimodal LLM for HDR-UGC VQA trained with HDR-aware policy optimization; code and weights are on the GitHub page.

What is in this repo

File	Rows	Contents
Beyond8Bits_publish.csv	41,419	Full release: every transcoded clip and source, with MOS and split
Beyond8Bits_publish.txt	41,419	Hashed video ids, one per line, for bulk download
Beyond8Bits_publish_crowd.csv	5,992	Crowd-only subset in the CHUG column layout
Beyond8Bits_publish_crowd.txt	5,992	Matching id list

The videos themselves are served from a public S3 mirror (see below). Per-rating raw CSVs are in the UT-Box package.

Columns (full config)

Column	Meaning
video_id	Hashed id, also the S3 object name
mos	Mean opinion score after SUREAL bias correction
sos	Standard deviation of opinion scores
type	Crowd or Vimeo
ref	1 for a source video, 0 for a transcode
resolution	360p, 720p, 1080p, or ref
bitrate	0.2M, 0.5M, 1M, 2M, 3M, or ref
orientation	Portrait or Landscape
framerate	Native playback frame rate
split	train, validation, test (70 / 10 / 20 by source identity: 28,987 / 4,151 / 8,281)
height, width	Frame size in pixels

The crowd config uses the CHUG layout: Video, mos_j, sos_j, ref, name, bitladder, resolution, bitrate, orientation, framerate, content_name, height, width.

Loading the scores

from datasets import load_dataset

full = load_dataset("shreshthsaini/Beyond8Bits", "full", split="train")   # all 41,419 rows; filter on the `split` column
crowd = load_dataset("shreshthsaini/Beyond8Bits", "crowd", split="train")

import pandas as pd
df = pd.read_csv("hf://datasets/shreshthsaini/Beyond8Bits/Beyond8Bits_publish.csv")
test = df[df.split == "test"]

Downloading the videos

Each clip is an HDR MP4 (10-bit PQ, BT.2020) at https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/<video_id>.mp4. Clips average about 30 MB, so the full set is roughly 1.2 TB.

import pandas as pd, urllib.request, pathlib
df = pd.read_csv("hf://datasets/shreshthsaini/Beyond8Bits/Beyond8Bits_publish.csv")
out = pathlib.Path("Beyond8Bits_Videos"); out.mkdir(exist_ok=True)
for vid in df.video_id:
    urllib.request.urlretrieve(f"https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/{vid}.mp4", out / f"{vid}.mp4")

Or with the AWS CLI:

while read v; do aws s3 cp "s3://ugchdrmturk/videos/${v}.mp4" ./Beyond8Bits_Videos/; done < Beyond8Bits_publish.txt

License

Metadata, manifests, and site code are CC BY 4.0. Videos keep their original terms: CC-licensed Vimeo content, and crowd-contributed videos released under a non-exclusive research redistribution agreement (non-commercial research use). For commercial use of the videos, contact the authors.

Citation

@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}
}

Beyond8Bits extends two earlier studies; cite them if you use the crowd subset or the BrightRate baseline:

@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}
}

Contact: Shreshth Saini, saini.2 at utexas.edu. Future data maintainer: Pragyadipta Adhya, pragyadipta.adhya at colorado.edu.