Research map / Video / Video-language understanding
VideoMiner: Iteratively Grounding Key Frames of Hour-Long Videos via Tree-based Group Relative Policy Optimization
Xinye Cao, Hongcan Guo, Jiawen Qian, Guoshun Nan, Chao Wang, Yuqi Pan, Tianhao Hou, Xiaojuan Wang, Yutong Gao
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 |
|---|---|---|---|
| plug VideoMiner key-frame tree into Qwen2-VL-7B | Qwen2-VL · EgoSchema | accuracy: 50.0 -> 66.2 | Improved |
| plug VideoMiner key-frame tree into Qwen2-VL-7B | Qwen2-VL · LongVideoBench (900-3600s) | accuracy: 38.0 -> 49.3 | Improved |
| plug VideoMiner key-frame tree into Qwen2-VL-7B | Qwen2-VL · Video-MME (long) | accuracy: 45.8 -> 52.2 | Improved |
| plug VideoMiner key-frame tree into Qwen2-VL-7B | Qwen2-VL · MLVU M-Avg | accuracy: 60.1 -> 65.1 | Improved |
| plug VideoMiner into Qwen2-VL-7B on short videos | Qwen2-VL · LongVideoBench (8-15s) | accuracy: 68.8 -> 65.1 | Hurt |
| add event clustering instead of frame clustering (ablation) | VideoMiner · long-video benchmarks | accuracy | Improved |
| add clustering instead of no clustering (ablation) | VideoMiner · long-video benchmarks | accuracy | Improved |
| train policy with T-GRPO instead of other RL methods (ablation) | VideoMiner · three long-video benchmarks | accuracy | Improved |
| add RL training to the policy model (ablation) | VideoMiner · long-video benchmarks | accuracy | Improved |
| increase response length from 20 to 400 tokens (ablation) | VideoMiner T-GRPO · three long-video benchmarks | accuracy: 20 -> 400 tokens, accuracy improved by over 10% | Improved |
| set tree growth rate below 1 (ablation) | VideoMiner T-GRPO · long-video benchmarks | accuracy | Hurt |
| lower growth rate toward more continue actions (ablation) | VideoMiner T-GRPO · long-video benchmarks | accuracy | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VideoTree — VideoTree: Adaptive Tree-based Video Representation for LLM Reasoning on Long Videos (CVPR 2025) · better on 9 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TSPO: Temporal Sampling Policy Optimization for Long-form Video Language Understanding (AAAI 2026 · similarity 0.90)
- FrameThinker: Learning to Think with Long Videos via Multi-Turn Frame Spotlighting (ICLR 2026 · similarity 0.89)
- Efficient Frame Selection for Long Video Understanding via Reinforcement Learning (CVPR 2026 · similarity 0.89)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.88)
- DeepVideo-R1: Video Reinforcement Fine-Tuning via Difficulty-aware Regressive GRPO (NeurIPS 2025 · similarity 0.88)
- Reinforcing Structured Chain-of-Thought for Video Understanding (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.
