Perceive-to-Reason: Decoupling Perception and Reasoning for Fine-Grained Visual Reasoning

2026-07-01

AI Summary

This work addresses the challenge that vision-language models struggle to capture fine-grained visual cues in high-resolution images. To this end, the authors propose Perceive-to-Reason (P2R), a perception-reasoning decoupled framework that separates fine-grained visual reasoning into two stages: visual evidence localization (perception) and answer generation (reasoning). They introduce a novel role-aware alternating reinforcement learning algorithm, PRA-GRPO, which enables joint optimization of both stages using only final-answer supervision. Built upon the Qwen3-VL-Instruct model series, the resulting P2R-4B achieves 93.2% on V-Star, and 81.9% and 80.5% on HR-Bench-4K and HR-Bench-8K, respectively, substantially outperforming existing methods while demonstrating strong generalization in multimodal reasoning.

Abstract

Fine-grained visual reasoning remains challenging for vision-language models, especially when small but critical visual cues are buried in high-resolution images. Existing approaches rely on repeated cropping or test-time visual search to introduce local evidence, but they typically do not explicitly distinguish perception from reasoning. In this paper, we propose Perceive-to-Reason (P2R), a unified framework that formulates fine-grained visual reasoning as a two-stage process: the model first localizes question-relevant evidence as a Perceiver, and then answers the question as a Reasoner based on the annotated image and cropped regions. To better align training with this decoupled formulation, we further introduce Perception-Reasoning Alternating GRPO (PRA-GRPO), a role-aware reinforcement learning strategy that alternates between perception-focused and reasoning-focused updates using only final-answer supervision. Built on top of Qwen3-VL-Instruct-2B/4B/8B, P2R consistently improves performance across model scales. In particular, P2R-4B achieves 93.2% on V-Star, 81.9% on HR-Bench-4K, and 80.5% on HR-Bench-8K, substantially outperforming its corresponding backbone. Further experiments show that the benefits of P2R extend beyond high-resolution benchmarks to broader multimodal reasoning tasks. These results suggest that explicitly decoupling perception from reasoning provides an effective framework for fine-grained visual reasoning.

Problem

• fine-grained visual reasoning
• vision-language models
• perception-reasoning decoupling
• high-resolution images
• visual cues

Innovation

• fine-grained visual reasoning
• perception-reasoning decoupling
• Perceive-to-Reason
• reinforcement learning
• vision-language models