AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Learning Active Perception for Pixel-Space Reasoning via Visual-Intent Stratified GRPO

Mingkang Zhu, Xi Chen, Senqiao Yang, Bei Yu, Hengshuang ZHAO, Jiaya Jia

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Reinforcement learning for visual reasoning · 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
replace GRPO with VIS-GRPO (depth and intent stratification)Qwen3-VL-8B · average of four benchmarksaccuracy: 74.2 -> 77.1Improved
replace GRPO with VIS-GRPOQwen3-VL-8B · HR-Bench 8Kaccuracy: 77.6 -> 81.6Improved
replace GRPO with VIS-GRPOQwen3-VL-8B · MME-RealWorldaccuracy: 52.7 -> 56.3Improved
add visual-intent stratification on top of depth-only stratification (ablation)VIS-GRPO · average of four benchmarksaccuracy: 74.8 -> 77.1Improved
add depth-only stratification to GRPO (ablation)GRPO · MME-RealWorldaccuracy: 52.7 -> 55.7Improved
add depth-only stratification to GRPO (ablation)GRPO · V* Benchaccuracy: 87.4 -> 84.8Hurt
add depth-only stratification to GRPO (ablation)GRPO · average of four benchmarksaccuracy: 74.2 -> 74.8Improved
use full VIS-GRPO instead of Pixel Reasoner zoom-reward heuristicsQwen3-VL-8B · average of four benchmarksaccuracy: 74.4 -> 77.1Improved
apply Pixel Reasoner reward heuristics instead of GRPOQwen3-VL-8B · V* Benchaccuracy: 87.4 -> 83.8Hurt
apply Pixel Reasoner reward heuristics instead of GRPOQwen3-VL-8B · HR-Bench 8Kaccuracy: 77.6 -> 79.0Improved
apply GRPO RL training to base Qwen3-VL-8BQwen3-VL-8B · average of four benchmarksaccuracy: 71.5 -> 74.2Improved
add visual-intent stratification (VIS-GRPO) over depth-only on intent balance (ablation)Depth-Stratified GRPO · MME-RealWorld multi-zoomexplore:verify ratio: 9:1 -> 7:3Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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