Argus: Vision-Centric Reasoning with Grounded Chain-of-Thought
This work addresses a specific limitation in multimodal AI for researchers and practitioners, representing an incremental improvement by enhancing visual attention mechanisms in existing models.
The paper tackles the problem of multimodal large language models struggling with vision-centric scenarios requiring precise visual focus, and introduces Argus, which uses object-centric grounding as visual chain-of-thought signals to improve performance, achieving excellence in multimodal reasoning and referring object grounding tasks as validated on diverse benchmarks.
Recent advances in multimodal large language models (MLLMs) have demonstrated remarkable capabilities in vision-language tasks, yet they often struggle with vision-centric scenarios where precise visual focus is needed for accurate reasoning. In this paper, we introduce Argus to address these limitations with a new visual attention grounding mechanism. Our approach employs object-centric grounding as visual chain-of-thought signals, enabling more effective goal-conditioned visual attention during multimodal reasoning tasks. Evaluations on diverse benchmarks demonstrate that Argus excels in both multimodal reasoning tasks and referring object grounding tasks. Extensive analysis further validates various design choices of Argus, and reveals the effectiveness of explicit language-guided visual region-of-interest engagement in MLLMs, highlighting the importance of advancing multimodal intelligence from a visual-centric perspective. Project page: https://yunzeman.github.io/argus/