AI Paper Atlas

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

CVPR 2026 · Video › Video-language understanding · approach: Video LLMs with fine-grained understanding · 11 citations (OpenAlex)

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 changedWhereMeasuredEffect
add Causal Segment Sufficiency Reward (lambda2=0.3) (ablation)REVISOR GRPO training · VideoMME overallaccuracy: 62.2 -> 65.7Improved
add multimodal visual reflection instead of text-only reflectionQwen2.5-VL-7B · VideoMME overallaccuracy: 63.4 -> 65.7Improved
add text-only reflection trainingQwen2.5-VL-7B · VideoMME overallaccuracy: 64.3 -> 63.4No clear effect
add REVISOR to base modelQwen2.5-VL-7B · LVBenchaccuracy: 40.2 -> 42Improved
add REVISOR to base modelQwen2.5-VL-7B · LongVideoBenchaccuracy: 56.5 -> 57.5Improved
set lambda2 larger than lambda1 (ablation)REVISOR · MLVUaccuracy: 69.8 -> 68.7Hurt
set lambda2 larger than lambda1 (ablation)REVISOR · NExT-GQA groundingmIoU: 33.2 -> 33.9Mixed
force longer textual reflection by system prompt (ablation)REVISOR · VideoMME overallaccuracy: 62.5 -> 60.3Hurt
add DADR during training (ablation)REVISOR · LVBench key-segment review accuracyaccuracy: 36.9 -> 41.3Improved
train on more diverse data (add CG-Bench) (ablation)REVISOR · MLVUaccuracy: 68.4 -> 69.8Improved
train on short video QA only (ablation)REVISOR · VideoMME overallaccuracy: 64.3 -> 62.6Hurt
train on short video QA only (ablation)REVISOR · NExT-GQA groundingmIoU: 20.9 -> 25.0Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy