Visual Parallel Search: Learning to Search High-Resolution Images with Parallel Tile Inspection and Adaptive Zoom

2026-09-29

AI Summary

This study addresses the difficulty of acquiring evidence from local small objects in high-resolution visual question answering by proposing the VPS framework. This method introduces a novel search mechanism that combines parallel tile inspection with adaptive zooming, wherein a primary agent orchestrates sub-agents to retrieve image tiles in parallel while adaptively zooming to precisely localize critical regions. Furthermore, a prompt-free verification module and a role-specific GRPO supervision pipeline are designed to enable decoupled training of the controller and reader. Experimental results demonstrate that the proposed framework outperforms pure zoom-based search in 14 out of 15 comparisons, achieving improvements of up to 8.0 points. It also yields a 4.17-point gain on HR-Bench 4K while significantly reducing tool invocation overhead.

Abstract

High-resolution visual question answering often fails because a multimodal model does not acquire the small, spatially localized evidence needed to answer a question. Sequential zooming can recover detail, but it asks the main model to choose a region before obtaining a reliable overview. We introduce VPS, a visual parallel-search framework in which a main agent first invokes grid_search to inspect image tiles in parallel with question-conditioned sub-agents, and then adaptively invokes zoom_in PSisual Parallel Search improves mean accuracy over dedicated zoom-only search in 14 of 15 same-model comparisons, with gains up to 8.0 points and especially strong improvements for smaller main models. ZoomBench retains an approximately 3.2-point gain at every tested size. We further develop a supervision pipeline with hint-free verification and a paired role-specific GRPO surrogate for learning the controller and tile-reader roles. SFT improves observed accuracy on all five benchmark splits, including a 4.17-point gain on HR-Bench 4K. Role-specific RL further reshapes search behavior: main-only RL reduces mean tool use from 2.65 to 2.11 with similar pass@1 in an internal four-response evaluation, while external accuracy changes are mixed. Joint training reveals an asymmetry between local evidence reading and global search control. Together, these results support VPS as an effective inference-time scaffold and a trainable decomposition for visual evidence acquisition.

Problem

• Visual Question Answering
• High-Resolution Images
• Local Evidence Acquisition
• Sequential Zooming

Innovation

• Visual Parallel Search
• High-Resolution VQA
• Adaptive Zoom
• Role-specific GRPO
• Multi-agent Framework