CHUG: Crowdsourced User-Generated HDR Video Quality Dataset

📌 Overview

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.

Key Features

✅ 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.

---

📂 Downloading Dataset

Direct download link for dataset: COMING SOON

For manual download, please see below.

---

🔗 Accessing Videos

1️⃣ Directly from AWS S3 (via Browser)

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

2️⃣ Downloading Videos Using AWS CLI

To download all videos:

cat chug-video.txt | while read video; do
    aws s3 cp s3://ugchdrmturk/videos/${video}.mp4 ./CHUG_Videos/
done

To 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

---

📊 Key Dataset Insights

• 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 ⚠️

📄 Paper and Supplementary Material

For a detailed analysis, check our paper and supplementary material.

---

🎬 Sample Videos (Direct Playback)

Below, you can directly play some sample HDR videos from our dataset:

Indoor Scene

Carousel

Rodeo

Nature

More sample are listed here in table:

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.

---

🏆 Use Cases and Future Impact

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:

✅ UGC-HDR Distortion Analysis

• CHUG captures banding, overexposure, luminance inconsistencies, making it an essential dataset for HDR distortion research.

✅ HDR Streaming Optimization

• Streaming providers can leverage CHUG to evaluate bitrate-resolution trade-offs, improving HDR compression pipelines.

✅ Advancing HDR Quality Metrics

• 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.

📜 Citation

If you use CHUG in your research, please cite:

COMING SOON

📜 License

CHUG is released under a Creative Commons Attribution-NonCommercial (CC BY-NC 4.0) License.

📬 Contact

For questions, please reach out: 📧 [Redacted for Blind Review]