Research map / Video / Video-language understanding
Wavelet-based Frame Selection by Detecting Semantic Boundary for Long Video Understanding
Wang Chen, Yuhui Zeng, Yongdong Luo, Tianyu Xie, Luojun Lin, Jiayi Ji, Yan Zhang, Xiawu Zheng
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 WFS-SB frame selection instead of uniform sampling | LLaVA-Video-7B · VideoMME (8 frames) | accuracy: 56.2 -> 61.7 | Improved |
| add WFS-SB frame selection instead of uniform sampling | LLaVA-Video-7B · MLVU (8 frames) | accuracy: 57.4 -> 66.9 | Improved |
| add WFS-SB frame selection instead of uniform sampling | LLaVA-Video-7B · LongVideoBench (8 frames) | accuracy: 54.9 -> 61.1 | Improved |
| add WFS-SB frame selection instead of uniform sampling | Qwen2.5-VL-7B · MLVU (32 frames) | accuracy: 59.7 -> 70.4 | Improved |
| add WFS-SB frame selection instead of uniform sampling | InternVL3-8B · VideoMME (32 frames) | accuracy: 65.6 -> 67.4 | Improved |
| use WFS-SB instead of A.I.R. iterative selection | Qwen2.5-VL-7B · VideoMME (32 frames) | accuracy: 65.0 vs 64.4 | Mixed |
| use wavelet boundaries instead of raw local minima (ablation) | WFS-SB · MLVU (K=16) | accuracy: 64.6 -> 67.9 | Improved |
| use wavelet boundaries instead of raw gradient (ablation) | WFS-SB · MLVU (K=16) | accuracy: 66.8 -> 67.9 | Improved |
| use wavelet boundaries instead of raw local minima (ablation) | WFS-SB · VideoMME (K=16) | accuracy: 60.8 -> 61.9 | Improved |
| add adaptive budget allocation instead of average (ablation) | WFS-SB · MLVU (K=16) | accuracy: 67.4 -> 67.9 | Improved |
| use MMR instead of top-K selection (ablation) | WFS-SB · VideoMME (K=16) | accuracy: 60.9 -> 61.9 | Improved |
| use MMR instead of uniform selection within segments (ablation) | WFS-SB · MLVU (K=16) | accuracy: 62.7 -> 67.9 | Improved |
Closest papers by meaning
- Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding (NeurIPS 2025 · similarity 0.88)
- BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding (CVPR 2025 · similarity 0.87)
- Adaptive Keyframe Sampling for Long Video Understanding (CVPR 2025 · similarity 0.86)
- M-LLM Based Video Frame Selection for Efficient Video Understanding (CVPR 2025 · similarity 0.86)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.86)
- LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025 · similarity 0.86)
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.
