Research map / Video / Video-language understanding
Select Less, Reason More: Prioritizing Evidence Purity for Video Reasoning
Xuchen Li, Xuzhao Li, Shiyu Hu, Kaiqi Huang
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 EARL reinforcement learning stage after SFT (ablation) | EARL SFT model · LongVideoBench | accuracy (%): 51.9% -> 59.8% | Improved |
| Add EARL reinforcement learning stage after SFT (ablation) | EARL SFT model · MVBench | accuracy (%): 63.8% -> 69.0% | Improved |
| Add EARL reinforcement learning stage after SFT (ablation) | EARL SFT model · VideoMME (Long) | accuracy (%): 51.8% -> 57.8% | Improved |
| Add IoU-based relevance reward (ablation) | EARL · LongVideoBench | accuracy (%): 56.8% -> 59.8% | Improved |
| Add IoU-based relevance reward (ablation) | EARL · MLVU | accuracy (%): 47.1% -> 49.3% | Improved |
| Add IoU constraint to the correctness reward (ablation) | EARL · LongVideoBench | accuracy (%): 57.8% -> 59.8% | Improved |
| Add IoU constraint to the correctness reward (ablation) | EARL · LVBench | accuracy (%): 44.7% -> 46.2% | Improved |
| Add IoU constraint to the correctness reward (ablation) | EARL · MLVU | accuracy (%): 47.9 -> 49.3 | Improved |
| Add dynamic adjustment of action and relevance reward scales (ablation) | EARL · LongVideoBench | accuracy (%): 58.4 -> 59.8 | Improved |
| Add dynamic adjustment of action and relevance reward scales (ablation) | EARL · MVBench | accuracy (%): 68.3 -> 69.0 | Improved |
| EARL-trained adaptive frame selection with localized re-sampling vs base Qwen2.5-VL-7B | Qwen2.5-VL-7B · VideoMME (Overall) | accuracy (%): 53.6% -> 64.9% | Improved |
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.
- LongVU — LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025) · better on 1 of 1 shared result
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
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.91)
- Beyond Perceptual Shortcuts: Causal-Inspired Debiasing Optimization for Generalizable Video Reasoning in Lightweight MLLMs (CVPR 2026 · similarity 0.90)
- Find, Fix, Reason: Context Repair for Video Reasoning (ICML 2026 · similarity 0.89)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.88)
- What Should a Streaming Video Model Remember? (NeurIPS 2026 · similarity 0.88)
- OneThinker: All-in-one Reasoning Model for Image and Video (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.
