Research map / Video / Video-language understanding
LongVQUBench: Benchmarking Long-Term Video Quality Understanding of Vision-Language Models
Arpita Nema, Hanwei Zhu, Xi Zhang, Weisi Lin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase sampled frames from 8 to 256 (ablation) | GPT-5 · LongVQUBench val | overall accuracy: 70.5 -> 74.1 | Improved |
| increase sampled frames from 8 to 256 (ablation) | Gemini-3 · LongVQUBench val | overall accuracy: 68.1 -> 67.3 | No clear effect |
| increase sampled frames from 64 to 256 (ablation) | Qwen3-VL · LongVQUBench val | overall accuracy: 63.1 -> 51.8 | Hurt |
| increase sampled frames from 8 to 256 (ablation) | VQA2 · LongVQUBench val | overall accuracy: 59.4 -> 54.8 | Hurt |
| increase sampled frames in general (ablation) | most LVLMs · LongVQUBench val | overall accuracy | Mixed |
| move from local/cross-event to global quality questions | GPT-5 · LongVQUBench val | accuracy: 81.2 (CQR) -> 65.8 (GQU) | Hurt |
| move from cross-event to global questions | Gemini-3 · LongVQUBench val | accuracy: 76.5 (CQR) -> 59.2 (GQU) | Hurt |
| use agentic adaptive frame selection (DeepVideoDiscovery) instead of uniform sampling | agentic LVLM · LongVQUBench val | overall accuracy: 71.7 (DVD) vs 35.8 (VideoAgent) | Mixed |
| use video-quality-specialized model | VQA2 · LongVQUBench test | overall accuracy: 54.2 vs LLaVA-Video 54.5 | No clear effect |
| score open-ended answers for relevance versus completeness | all LVLMs · LongVQUBench test open-ended | score: relevance 88.9 vs completeness 45.8 (GPT-5) | Mixed |
Closest papers by meaning
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.92)
- Q-Bench-Video: Benchmark the Video Quality Understanding of LMMs (CVPR 2025 · similarity 0.91)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.91)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.90)
- MLVU: Benchmarking Multi-task Long Video Understanding (CVPR 2025 · similarity 0.90)
- RTV-Bench: Benchmarking MLLM Continuous Perception, Understanding and Reasoning through Real-Time Video (NeurIPS 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
