CVSearch: Empowering Multimodal LLMs with Cognitive Visual Search for High-Resolution Image Perception

High-resolution (HR) image perception presents a key bottleneck for multimodal large language models (MLLMs). While visual search offers a promising solution, existing methods struggle with the trade-off between coverage and efficiency. Visual expert-assisted search is efficient but prone to blind spots when proposals fail, whereas scan-based search guarantees coverage at the cost of computational redundancy and semantic fragmentation. To address this dilemma, we introduce CVSearch, a training-free adaptive framework that dynamically schedules search strategies via an Assess-then-Search workflow. Specifically, CVSearch first invokes expert-assisted search when global information is insufficient, and only triggers a novel semantic-aware scanning mechanism upon failure. Distinct from rigid grid partitioning, this efficient scanning paradigm incorporates Semantic Guided Adaptive Patching to decompose images into semantically consistent regions, effectively mitigating object fragmentation. Furthermore, we devise a Dynamic Bottom-Up Search strategy driven by a Visual Complexity prior to enable efficient and precise iterative exploration of local details. Extensive experiments on HR benchmarks demonstrate that CVSearch achieves state-of-the-art accuracy while substantially improving search efficiency. Code is released at https://github.com/liliupeng28/ICML26-CVSearch.

Current multimodal AI models struggle with high-resolution images because they either aggressively downscale the picture and lose crucial details, or blindly chop it into rigid tiles that fracture objects and slow computation. To overcome this, we introduce CVSearch, a framework that mimics human visual search. It first takes a global glance at the scene before rapidly zooming into promising regions using a lightweight object detector. For complex areas, CVSearch applies a context-aware scan that groups related pixels to preserve object integrity and targets visually rich zones while skipping empty backgrounds entirely. This approach empowers AI to resolve fine details like small text or distant objects with state-of-the-art accuracy. By operating three times faster than previous methods, CVSearch makes efficient and scalable high-resolution AI perception highly practical for real-world applications.