Research map / Video / Video-language understanding
FOCUS: Efficient Keyframe Selection for Long Video Understanding
Zirui Zhu, Hailun Xu, Yang Luo, Yong Liu, Kanchan Sarkar, Zhenheng Yang, Yang You
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 |
|---|---|---|---|
| replace uniform frame sampling with FOCUS keyframe selection | Qwen2-VL-7B · LongVideoBench | accuracy: 55.6 -> 62.3 | Improved |
| replace uniform frame sampling with FOCUS keyframe selection | GPT-4o · Video-MME | accuracy: 61.8 -> 62.5 | Improved |
| replace uniform frame sampling with FOCUS keyframe selection | LLaVA-Video-7B · LongVideoBench long videos >20 min | accuracy: 51.8 -> 63.7 | Improved |
| replace Top-K selection with FOCUS | LLaVA-Video-7B · Video-MME overall | accuracy: 62.9 -> 65.4 | Improved |
| use Top-K keyframe selection instead of uniform sampling | LLaVA-Video-7B · Video-MME overall | accuracy: 64.4 -> 62.9 | Hurt |
| replace AKS with FOCUS | LLaVA-Video-7B · LongVideoBench overall | accuracy: 62.1 -> 63.5 | Improved |
| use FOCUS instead of AKS on short videos | LLaVA-Video-7B · LongVideoBench short | accuracy: 72.3 -> 72.3 | No clear effect |
| remove pre-filtering by using bandit exploration instead of 1 fps scoring | keyframe selection cost · LongVideoBench | GPU hours: 9.3 -> 5.5 | Improved |
| lower alpha from 0.25 to 0.1 (ablation) | FOCUS · LongVideoBench | accuracy: 63.5 -> 62.9 | Hurt |
| raise alpha from 0.25 to 0.5 (ablation) | FOCUS · LongVideoBench | accuracy: 63.5 -> 63.6 | No clear effect |
Closest papers by meaning
- Adaptive Keyframe Sampling for Long Video Understanding (CVPR 2025 · similarity 0.94)
- Efficient Frame Selection for Long Video Understanding via Reinforcement Learning (CVPR 2026 · similarity 0.92)
- Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding (NeurIPS 2025 · similarity 0.92)
- BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding (CVPR 2025 · similarity 0.91)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · similarity 0.91)
- M-LLM Based Video Frame Selection for Efficient Video Understanding (CVPR 2025 · similarity 0.89)
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.
