Research map / Video / Video-language understanding
QCA: Query- and Content-Aware Keyframe Selection for Long Video Understanding
Jun Peng, Baiyang Song, Jie Li, Hui Li, Yiyi Zhou, Rongrong Ji, Yonghong Tian
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 sampling with QCA keyframe selection | Qwen3-VL-8B · LongVideoBench (64 frames) | accuracy: 63.1 -> 66.9 | Improved |
| replace uniform sampling with QCA | Qwen3-VL-8B · LVBench (64 frames) | accuracy: 43.8 -> 51.8 | Improved |
| replace uniform sampling with QCA | LLaVA-Video-7B · Video-MME (64 frames) | accuracy: 64.4 -> 66.1 | Improved |
| replace uniform sampling with QCA | InternVL-3.5-8B · LVBench (64 frames) | accuracy: 42.8 -> 50.0 | Improved |
| use QCA instead of plain Top-k image-text matching selection | Qwen3-VL-8B · LongVideoBench (64 frames) | accuracy: 64.4 -> 66.9 | Improved |
| add semantic matching term M_s to segment allocation (ablation) | QCA · LongVideoBench | accuracy: 65.8 -> 66.9 | Improved |
| add content deviation term D_s to segment allocation (ablation) | QCA · Video-MME | accuracy: 70.4 -> 70.1 | Mixed |
| add semantic anchor instead of random initial frame (ablation) | QCA · LVBench | accuracy: 47.2 -> 52.4 | Improved |
| add candidate set restricting to high-relevance frames (ablation) | QCA · MLVU | accuracy: 72.4 -> 75.8 | Improved |
| add diversity-driven selection instead of top matching (ablation) | QCA · LVBench | accuracy: 49.2 -> 52.4 | Improved |
| increase frame budget from 16 to 128 (ablation) | QCA · LVBench | accuracy: 46.3 -> 53.2 | Improved |
| increase number of segments S beyond 16 (ablation) | QCA · LongVideoBench | accuracy: 66.9 at S=16, lower beyond | Hurt |
Closest papers by meaning
- Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding (NeurIPS 2025 · similarity 0.92)
- Adaptive Keyframe Sampling for Long Video Understanding (CVPR 2025 · similarity 0.90)
- AdaVideoRAG: Omni-Contextual Adaptive Retrieval-Augmented Efficient Long Video Understanding (NeurIPS 2025 · similarity 0.88)
- BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding (CVPR 2025 · similarity 0.88)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · similarity 0.88)
- FlexSelect: Flexible Token Selection for Efficient Long Video Understanding (NeurIPS 2025 · similarity 0.87)
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.
