PixelEyes: Decoupling Perception and Reasoning for Pinpoint Visual Evidence Seeking

2026-06-30

AI Summary

This work addresses the limitations of current large vision-language models in multi-turn visual reasoning, where tightly coupled perception and reasoning often lead to target localization failures and redundant reasoning trajectories. To overcome this, the authors propose PixelEyes, a framework that explicitly decouples reasoning from perception: a reasoning module determines *what* to look for, while a dedicated perception tool precisely returns the corresponding location. The approach integrates mask-guided visual search with a semantic-region breadth-first strategy and leverages a referring expression segmentation model for mask-level localization. To support evaluation, the authors introduce the PixelEyes-6K dataset and Pinpoint-Bench, a zero-shot benchmark for precise visual grounding. Experiments demonstrate that PixelEyes significantly outperforms existing methods, achieving higher localization accuracy, reduced redundancy, and improved reasoning efficiency.

Abstract

This paper explores multi-turn visual reasoning and observes that MLLMs repeatedly fail to localize the target, leading to long, redundant trajectories. We attribute this failure to the entanglement of reasoning and perception within a single model, the MLLM reasons and localizes simultaneously, and inaccurate localization triggers additional reasoning turns that bloat the trajectory. To solve this problem, we propose PixelEyes, a multi-turn visual reasoning agent that explicitly decouples reasoning from perception, i.e., the reasoner decides what to look for, while a specialized perception tool answers where it is. Specifically, PixelEyes introduces 1) Mask-guided Visual Search. A referring segmentation model is invoked to provide mask-precise localization, freeing the reasoner from the need to compensate for imprecise grounding. 2) Semantic-region Breadth-first Search (BFS). To eliminate redundant loops caused by repeatedly cropping incorrect sub-regions, we organize exploration as a breadth-first search over semantic regions. To internalize these capabilities, we construct the PixelEyes-6K dataset by resynthesizing expert trajectories from existing data. This explicitly embeds our mask-guided search and BFS logic into the model. We further introduce Pinpoint-Bench, a zero-hint visual search benchmark, i.e., no location cues are provided in the question, with instance-level masks and bounding boxes that separate localization failures from reasoning failures, enabling fine-grained analysis of failure modes such as inattentional blindness. Recent state-of-the-art MLLMs and visual reasoning agents leave large headroom on Pinpoint-Bench, demonstrating its quality and difficulty. Code and models are open-sourced.

Problem

• visual reasoning
• localization
• perception-reasoning entanglement
• multi-turn interaction
• inattentional blindness

Innovation

• decoupled perception and reasoning
• mask-guided visual search
• semantic-region BFS
• Pinpoint-Bench
• referring segmentation