[2603.04676] Decoding the Pulse of Reasoning VLMs in Multi-Image Understanding Tasks

Computer Science > Computer Vision and Pattern Recognition arXiv:2603.04676 (cs) [Submitted on 4 Mar 2026] Title:Decoding the Pulse of Reasoning VLMs in Multi-Image Understanding Tasks Authors:Chenjun Li View a PDF of the paper titled Decoding the Pulse of Reasoning VLMs in Multi-Image Understanding Tasks, by Chenjun Li View PDF HTML (experimental) Abstract:Multi-image reasoning remains a significant challenge for vision-language models (VLMs). We investigate a previously overlooked phenomenon: during chain-of-thought (CoT) generation, the text-to-image (T2I) attention of reasoning VLMs exhibits diffuse "pulses": sporadic and unfocused attention patterns that fail to concentrate on task-relevant images. We further reveal a systematic positional bias in attention allocation across images. Motivated by these observations, we propose PulseFocus, a training-free, inference-time method that structures CoT reasoning into interleaved plan/focus blocks with soft attention gating. By forcing the model to explicitly plan which image to examine and then gating decode-time attention to the referenced image, PulseFocus sharpens attention focus and yields consistent improvements on multi-image benchmarks like BLINK benchmark (+3.7%) and MuirBench (+1.07%). Comments: Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) Cite as: arXiv:2603.04676 [cs.CV]   (or arXiv:2603.04676v1 [cs.CV] for this version)   https://doi.org/10.48550/arXiv.2603.04676 Focu...