You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
The LongVQUBench dataset contains identifiers to videos originally sourced from MLVU, LongVideoBench, and LongVideo-Reason-Eval for data collection purposes. LongVQUBench does not own the content linked within this dataset; all rights and copyright belong to the respective dataset owners and original channel creators. Ensuring compliance with platform terms and conditions is the responsibility of these source datasets. By accessing this dataset, you acknowledge and agree to the following terms:
Log in or Sign Up to review the conditions and access this dataset content.
LongVQUBench
Benchmarking Long-Term Video Quality Understanding of Vision-Language Models
📄 Paper | 🌐 Project Page | 💻 GitHub
Dataset Description
LongVQUBench is a comprehensive benchmark for evaluating long-term video quality understanding in large vision-language models (LVLMs), featuring 1,200 videos and 1,500 QA pairs across three hierarchical evaluation levels.
Dataset Structure
- LQU — Local Event Quality Understanding
- CQR — Cross-Event Quality Reasoning
- GQU — Global Quality Understanding
Stats
| Property | Value |
|---|---|
| Videos | 1,200+ |
| QA Pairs | 1,500 |
| Avg. Duration | ~12 min |
| LVLMs Evaluated | 14 |
- Downloads last month
- 7