AI Paper Atlas

Research map / Video / Video-language understanding

Reason, Then Re-reason: Cross-view Revisiting Improves Spatial Reasoning

Chaofan Ma, Zhenjie Mao, Yuhuan Yang, Fanqin Zeng, Yue Shi, Yingjie Zhou, Xiaofeng Cao, Jiangchao Yao

ICML 2026 · Video › Video-language understanding · approach: Egocentric video understanding · 0 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 Re-reason phase with synthesized novel-view videoQwen3-VL-4B · VSI-Benchaverage accuracy: 30.7 -> 36.5Improved
Add ReRe to Qwen3-VL-2B on STI-Bench staticQwen3-VL-2B · STI-Bench static subsetaverage accuracy: 22.2 -> 30.2Improved
Add ReRe to InternVL2.5-4BInternVL2.5-4B · STI-Bench static subsetaverage accuracy: 30.5 -> 30.6No clear effect
Sequential revisiting instead of concatenating the two videos (ablation)Qwen2.5-VL-7B · VSI-Benchaverage accuracy: 25.4 -> 29.5Improved
Sequential revisiting instead of interleaved joint reasoning (ablation)Qwen2.5-VL-7B · VSI-Benchaverage accuracy: 25.6 -> 29.5Improved
Re-reason on the same original video instead of the novel view (ablation)Qwen2.5-VL-7B · VSI-Benchaverage accuracy: 24.8 -> 23.5Hurt
Use synthesized novel view alone for both phases instead of combining with egocentric video (ablation)Qwen2.5-VL-7B · VSI-Benchaverage accuracy: 29.5 -> 27.3Hurt
Use Oblique Sweep trajectory instead of Mid-level Traverse (ablation)ReRe · VSI-Benchaverage accuracy: 25.6 -> 29.5Improved
Use Oblique Sweep trajectory instead of Bird's-eye Orbit (ablation)ReRe · VSI-Benchaverage accuracy: 27.4 -> 29.5Improved
Reduce VGGT input frames from 100 to 20 (ablation)ReRe · VSI-Benchaverage accuracy: 35.8 -> 33.3Mixed
Add ReRe to spatial-specialized SpaceR-3BSpaceR-3B · VSI-Benchaverage accuracy: 34.69 -> 35.96Improved
Add ReRe to spatial-specialized SpatialLadder-3BSpatialLadder-3B · VSI-Benchaverage accuracy: 44.84 -> 45.58Improved

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