CHUG is the first large-scale User-Generated HDR (UGC-HDR) video quality dataset, designed for perceptual video quality assessment and No-Reference HDR-VQA research.
✅ 5,992 videos from 856 UGC-HDR reference videos
✅ Authentic UGC-HDR distortions, including compression artifacts
✅ Bitrate ladder encoding, simulating real-world streaming scenarios
✅ 211,848 subjective ratings collected via Amazon Mechanical Turk (AMT)
✅ Balanced mix of portrait and landscape videos
This dataset serves as a benchmark for No-Reference (NR) UGC HDR-VQA models and HDR quality assessment research.
Each video is hosted on AWS S3 and can be accessed using:
https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/VIDEO.mp4
Replace VIDEO with a hashed video ID from chug.csv or chug-video.txt.
Example:
Museum: https://ugchdrmturk.s3.us-east-2.amazonaws.com/videos/9ae245a27cc5ea9d2f3fae9692250281.mp4
To download all videos:
cat chug-video.txt | while read video; do
aws s3 cp s3://ugchdrmturk/videos/${video}.mp4 ./CHUG_Videos/
doneTo download a single video:
aws s3 cp s3://ugchdrmturk/videos/VIDEO.mp4 ./CHUG_Dataset/To download selected videos, create a new text file with list of video IDs:
cat sample-video.txt | while read video; do
aws s3 cp s3://ugchdrmturk/videos/${video}.mp4 ./CHUG_Videos/
done- Higher resolutions & bitrates improve perceptual quality 📈
- UGC-HDR videos exhibit unique distortions, including banding and overexposure 🌈
- Landscape vs. Portrait orientation has minimal impact on MOS, though portrait is slightly favored 📱
- Compression artifacts degrade MOS significantly at low bitrates
⚠️
For a detailed analysis, check our paper and supplementary material. The accepted ICIP 2025 version is also available on IEEE Xplore.
Below, you can directly play some sample HDR videos from our dataset:
| Category | Video ID | MOS Score | Resolution | Link |
|---|---|---|---|---|
| Indoor Scene | 9ae245a27cc5ea9d2f3fae9692250281 |
33.46 | 1080p | ▶ Watch Video |
| Carousel | 273a5d8a3b8c2d0eb4d4c8ff5fcfe360 |
14.14 | 720p | ▶ Watch Video |
| Rodeo | 7b7c9033da9fdb1a5762527f19baf54d |
25.21 | 1080p | ▶ Watch Video |
| Nature | 482dc1789b58cd2a353408602e9cd903 |
51.28 | 1080p | ▶ Watch Video |
| Museum | 6ecb44305c0a4c421201b7bcfd369acb |
51.91 | 1080p | ▶ Watch Video |
| Mountains | 7435fdf9b5cda9a4299a7be5707ff911 |
53.37 | 1080p | ▶ Watch Video |
Please checkout the full dataset.
CHUG serves as a crucial benchmark for No-Reference UGC HDR Video Quality Assessment (NR-HDR-VQA) and real-world HDR streaming quality analysis. Key applications:
- CHUG captures banding, overexposure, luminance inconsistencies, making it an essential dataset for HDR distortion research.
- Streaming providers can leverage CHUG to evaluate bitrate-resolution trade-offs, improving HDR compression pipelines.
- CHUG enables refinement of HDR-specific VQA metrics such as HDR-VMAF, HDR-SSIM, and learning-based perceptual models.
CHUG is expected to guide industry standards and HDR-VQA research for years to come.
If you use CHUG in your research, please cite us:
@INPROCEEDINGS{11084488,
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},
volume={},
number={},
pages={2504-2509},
keywords={Visualization;Video on demand;User-generated content;Benchmark testing;Distortion;Quality assessment;High dynamic range;Web sites;Surges;Videos;Crowdsourced;High Dynamic Range (HDR);Video Quality Assessment;HDR VQA Dataset;User-Generated Content (UGC)},
doi={10.1109/ICIP55913.2025.11084488}
}CHUG is released under a Creative Commons Attribution-NonCommercial (CC BY-NC 4.0) License.
Last updated: 2024