AI Paper Atlas

Research map / Video / Video-language understanding

VidBridge-R1: Bridging QA and Captioning for RL-based Video Understanding Models with Intermediate Proxy Tasks

Xinlong Chen, Yuanxing Zhang, Yushuo Guan, Weihong Lin, Zekun Moore Wang, Bohan Zeng, Yang Shi, Sihan Yang, Qiang Liu, Pengfei Wan, Liang Wang

ICLR 2026 · Video › Video-language understanding · approach: Video question answering

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
naively mix VideoQA and Caption data and rewards in RL (ablation)Qwen2.5-VL-7B GRPO · Video-MMEaccuracy: 63.2 -> 54.8Hurt
naively mix VideoQA and Caption in RL, effect on captioning (ablation)Qwen2.5-VL-7B GRPO · DREAM-1KAutoDQ F1: 34.8 -> 30.6Hurt
add DarkEventInfer proxy task to VideoQA (ablation)RL training mix · DarkEventInfer-Test / MixVidQA-Testjudge score / accuracy: 60.0 -> 113.0 / 24.0 -> 41.0Improved
add MixVidQA proxy task on top of VideoQA + DarkEventInfer (ablation)RL training mix · MixVidQA-Testaccuracy: 41.0 -> 54.0Improved
add Caption task on top of VideoQA + both proxy tasks (ablation)RL training mix · DREAM-1KAutoDQ F1: 32.2 -> 35.2Improved
add Caption task on top of VideoQA + both proxy tasks, effect on reasoning (ablation)RL training mix · DarkEventInfer-Testjudge score: 121.0 -> 117.0Hurt
scale up VideoQA and Caption data volume in naive mix (ablation)RL training mix · Video-MMEaccuracy: 54.8 -> 61.5Mixed
add KL divergence regularization to GRPO (coefficient 0.10) (ablation)VidBridge-R1 · Video-MMEaccuracy: 64.3 -> 62.6Hurt
use implicit format reward instead of explicit or none (ablation)VidBridge-R1 · MVBenchaccuracy: 58.2 -> 61.9Improved
set precision weight alpha=1.0 without keywords reward in caption reward (ablation)caption reward · VidCapBenchaccuracy: 13.0 -> 10.3Hurt
add keywords reward (beta=0.2) to caption reward (ablation)caption reward · DREAM-1KAutoDQ F1: 32.9 -> 34.8Improved
use SFT cold start with reasoning chains before RL (ablation)Qwen2.5-VL-7B · Video-MMEaccuracy: 57.0 vs 64.3 for direct RLHurt

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