Research map / Video / Video-language understanding
Flow4Agent: Long-form Video Understanding via Motion Prior from Optical Flow
Ruyang Liu, Shangkun Sun, Haoran Tang, Wei Gao, Ge Li
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 dynamic event split (DES) (ablation) | LLaVA-Video-7B · VideoMME (overall, no subs) | accuracy: 62.6 -> 63.2 | Improved |
| add event-center cross-modal query (ECQ) (ablation) | LLaVA-Video-7B · VideoMME (overall, no subs) | accuracy: 62.6 -> 63.4 | Improved |
| add motion token pruning (MTP) (ablation) | LLaVA-Video-7B · VideoMME (long, no subs) | accuracy: 50.6 -> 52.4 | Improved |
| add ECQ on top of DES (ablation) | Flow4Agent TGO · VideoMME (overall, no subs) | accuracy: 63.2 -> 64.0 | Improved |
| add MTP on top of DES+ECQ (full model) (ablation) | Flow4Agent · VideoMME (overall, no subs) | accuracy: 64.0 -> 64.7 | Improved |
| apply Flow4Agent | LLaVA-NeXT-7B · VideoMME (overall, no subs) | accuracy: 44.9 -> 47.0 | Improved |
| apply Flow4Agent | LLaVA-OneVision-7B · VideoMME (overall, no subs) | accuracy: 58.2 -> 59.9 | Improved |
| apply Flow4Agent | Qwen2-VL-7B · VideoMME (overall, no subs) | accuracy: 61.7 -> 63.9 | Improved |
| apply Flow4Agent | LLaVA-Video-72B · VideoMME (overall, no subs) | accuracy: 67.1 -> 69.0 | Improved |
| use Sea-RAFT instead of NeuFlow (ablation) | Flow4Agent · VideoMME (overall, no subs) | accuracy: 64.1 -> 64.7 | Improved |
| use Sea-RAFT instead of StreamFlow (ablation) | Flow4Agent · VideoMME (overall, no subs) | accuracy: 64.5 -> 64.7 | Improved |
| raise MTP flow iterations from 4 to 12 (ablation) | Flow4Agent · VideoMME (overall, no subs) | accuracy: 64.2 -> 64.7 | Improved |
Closest papers by meaning
- LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025 · similarity 0.89)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.89)
- FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.88)
- HIMMEL: Hierarchical Interleaved Multi-stream Motion Encoding for Long Video Understanding (NeurIPS 2026 · similarity 0.88)
- V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models (NeurIPS 2026 · similarity 0.88)
- Video-XL: Extra-Long Vision Language Model for Hour-Scale Video Understanding (CVPR 2025 · 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.
