Research map / Video / Video-language understanding
REVISOR: Beyond Textual Reflection, Towards Multimodal Introspective Reasoning in Long-Form Video Understanding
Jiaze Li, Hao Yin, Wenhui Tan, Jingyang Chen, Boshen Xu, Yuxun Qu, Yijing Chen, Jianzhong Ju, Zhenbo Luo, Jian Luan
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 Causal Segment Sufficiency Reward (lambda2=0.3) (ablation) | REVISOR GRPO training · VideoMME overall | accuracy: 62.2 -> 65.7 | Improved |
| add multimodal visual reflection instead of text-only reflection | Qwen2.5-VL-7B · VideoMME overall | accuracy: 63.4 -> 65.7 | Improved |
| add text-only reflection training | Qwen2.5-VL-7B · VideoMME overall | accuracy: 64.3 -> 63.4 | No clear effect |
| add REVISOR to base model | Qwen2.5-VL-7B · LVBench | accuracy: 40.2 -> 42 | Improved |
| add REVISOR to base model | Qwen2.5-VL-7B · LongVideoBench | accuracy: 56.5 -> 57.5 | Improved |
| set lambda2 larger than lambda1 (ablation) | REVISOR · MLVU | accuracy: 69.8 -> 68.7 | Hurt |
| set lambda2 larger than lambda1 (ablation) | REVISOR · NExT-GQA grounding | mIoU: 33.2 -> 33.9 | Mixed |
| force longer textual reflection by system prompt (ablation) | REVISOR · VideoMME overall | accuracy: 62.5 -> 60.3 | Hurt |
| add DADR during training (ablation) | REVISOR · LVBench key-segment review accuracy | accuracy: 36.9 -> 41.3 | Improved |
| train on more diverse data (add CG-Bench) (ablation) | REVISOR · MLVU | accuracy: 68.4 -> 69.8 | Improved |
| train on short video QA only (ablation) | REVISOR · VideoMME overall | accuracy: 64.3 -> 62.6 | Hurt |
| train on short video QA only (ablation) | REVISOR · NExT-GQA grounding | mIoU: 20.9 -> 25.0 | Improved |
Closest papers by meaning
- OneThinker: All-in-one Reasoning Model for Image and Video (CVPR 2026 · similarity 0.90)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.90)
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.89)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.89)
- A Benchmark for Omni-Modal Reasoning in Long Videos (NeurIPS 2026 · similarity 0.89)
- Mimic Human Cognition, Master Multi-Image Reasoning: A Meta-Action Framework for Enhanced Visual Understanding (CVPR 2026 · similarity 0.89)
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.
