Research map / Video / Video-language understanding
LvBench: A Benchmark for Long-form Video Understanding with Versatile Multi-modal Question Answering
Hongjie Zhang, Lu Dong, Yi Liu, Yifei Huang, Limin Wang, Limin Wang, Yu Qiao
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 |
|---|---|---|---|
| add dual pathway modeling (DPM) to SEVILA | SEVILA · LvBench test finetuned | accuracy: 35.0 -> 36.9 | Improved |
| add DPM to LongVA | LongVA 7B · LvBench test | accuracy: 44.4 -> 46.6 | Improved |
| add DPM to LLaVA-OV | LLaVA-OV 7B · LvBench test | accuracy: 46.5 -> 48.3 | Improved |
| add keyframe pathway to long context pathway (ablation) | DPM · LvBench test | accuracy: 36.0 -> 36.9 | Improved |
| add long context pathway to keyframe pathway (ablation) | DPM · LvBench test | accuracy: 36.3 -> 36.9 | Improved |
| increase input frames from 128 to 512 (ablation) | DPM · LvBench test | accuracy: 36.2 -> 36.9 | Improved |
| increase input frames from 10 to 384 (ablation) | GPT-4o · LvBench | accuracy: 52.8 -> 67.3 | Improved |
| increase input frames from 10 to 1000 (ablation) | Gemini-1.5 Pro · LvBench | accuracy: 57.2 -> 71.5 | Improved |
| use ChatGPT QA-based distractors instead of manual distractors (ablation) | FrozenBiLM · LvBench | accuracy: 31.5 -> 79.6 | Hurt |
| fine-tune on LvBench instead of zero-shot | SEVILA · LvBench | accuracy: 28.2 -> 35.0 | Improved |
| add subtitles and video to question for humans (ablation) | human · LvBench | accuracy: 40.2 -> 91.3 | Improved |
| scale LLaVA-OV from 7B to 72B | LLaVA-OV · LvBench zero-shot | accuracy: 40.4 -> 49.4 | Improved |
Closest papers by meaning
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.94)
- SVBench: A Benchmark with Temporal Multi-Turn Dialogues for Streaming Video Understanding (ICLR 2025 · similarity 0.92)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.92)
- CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding (ICLR 2025 · similarity 0.91)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.91)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.91)
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.
