Research map / Video / Video-language understanding
QuoTA: Query-oriented Token Assignment via CoT Query Decouple for Long Video Comprehension
Yongdong Luo, Wang Chen, Weizhong Huang, Shukang Yin, Haojia Lin, Jinfa Huang, Chaoyou Fu, Jiayi Ji, Xiawu Zheng, Jiebo Luo
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 QuoTA query-oriented token assignment to LLaVA-Video-7B | LLaVA-Video-7B · Video-MME (w/o subtitles, overall) | accuracy: 63.3 -> 65.9 | Improved |
| add QuoTA to LLaVA-Video-7B | LLaVA-Video-7B · VNBench | accuracy: 54.4 -> 64.7 | Improved |
| add QuoTA to LLaVA-OV-7B | LLaVA-OV-7B · Video-MME (w/o subtitles, long videos) | accuracy: 47.7 -> 52.2 | Improved |
| add QuoTA to LLaVA-Video-7B | LLaVA-Video-7B · NeXT-QA | accuracy: 83.2 -> 83.9 | Improved |
| use bilinear interpolation rather than adaptive pooling for token assignment (ablation) | QuoTA · Video-MME (overall) | accuracy: 63.9 -> 65.9 | Improved |
| use bilinear interpolation rather than ToMe-style dynamic token merging (ablation) | QuoTA · Video-MME (overall) | accuracy: 64.5 -> 65.9 | Improved |
| add CoT entity query decoupling to LVLM frame scoring (ablation) | QuoTA · VNBench | accuracy: 49.5 -> 64.7 | Improved |
| add CoT entity query decoupling to LVLM frame scoring (ablation) | QuoTA · Video-MME (overall) | accuracy: 64.4 -> 65.9 | Improved |
| decouple query into events rather than entities (ablation) | QuoTA · Video-MME (overall) | accuracy: 65.9 -> 65.3 | Hurt |
| use CLIP instead of a scoring LVLM for frame scoring (ablation) | QuoTA · Video-MME (overall) | accuracy: 65.9 -> 64.0 | Hurt |
| add LVLM-based scoring on top of dynamic-length frame sampling (without CoT) (ablation) | QuoTA · VNBench | accuracy: 58.4 -> 48.6 | Hurt |
| add dynamic-length frame sampling alone (no scoring) (ablation) | LLaVA-Video-7B · Video-MME (overall) | accuracy: 64.0 -> 63.5 | No clear effect |
Closest papers by meaning
- Sink-Token-Aware Pruning for Fine-Grained Video Understanding in Efficient Video LLMs (ECCV 2026 · similarity 0.91)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.90)
- Rethinking Visual Token Reduction in LVLMs Under Cross-Modal Misalignment (AAAI 2026 · similarity 0.90)
- FlexSelect: Flexible Token Selection for Efficient Long Video Understanding (NeurIPS 2025 · similarity 0.89)
- Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding (NeurIPS 2025 · similarity 0.88)
- ApET: Approximation-Error Guided Token Compression for Efficient VLMs (CVPR 2026 · similarity 0.88)
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.
