AI Paper Atlas

Research map / Video / Video-language understanding

Reinforcing Video Reasoning with Focused Thinking

Jisheng Dang, Jingze Wu, Teng Wang, Xuanhui Lin, Nannan Zhu, Hongbo Chen, WEISHI ZHENG, Meng Wang, Tat-Seng Chua

ECCV 2026 · Video › Video-language understanding · approach: Long video understanding

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
train on multi-choice instead of single-choice questions (ablation)GRPO · CLEVRER counterfactualaccuracy (all): 32.6 -> 41.1Improved
train on multi-choice instead of single-choice questions (ablation)GRPO · NExT-GQA sourceaccuracy (all): 30.7 -> 36.7Improved
train on multi-choice instead of single-choice questions (ablation)GRPO · STAR sourceaccuracy (all): 29.3 -> 29.1No clear effect
train on multi-choice instead of single-choice (ablation)TW-GRPO · CLEVRER counterfactualaccuracy (all): 38.9 -> 50.4Improved
use soft reward instead of fixed reward (ablation)TW-GRPO · CLEVRER counterfactualsoft accuracy (all): 58.7 -> 62.9Improved
add token weighting (alpha=0.7) (ablation)GRPO with soft reward · CLEVRER counterfactualsoft accuracy (multiple choice): 57.6 -> 64.4Improved
add token weighting (ablation)GRPO with fixed reward · CLEVRER counterfactualaccuracy (all): 41.1 -> 42.6Improved
use nonzero token weight alpha=0.7 instead of alpha=0 (ablation)TW-GRPO · CLEVRER counterfactualaccuracy (all): 48.6 -> 50.4Improved
raise sampling number from 4 to 8 (ablation)TW-GRPO · CLEVRER counterfactualaccuracy (all): 38.9 -> 50.4Improved
raise sampling number from 8 to 12 (ablation)TW-GRPO · CLEVRER counterfactualaccuracy (all): 50.4 -> 49.1Hurt
temperature 1.0 instead of 0.5 (ablation)TW-GRPO · CLEVRER counterfactualaccuracy (all): 46.3 -> 50.4Improved
temperature 1.0 instead of 1.5 (ablation)TW-GRPO · CLEVRER counterfactualaccuracy (all): 44.3 -> 50.4Improved

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