Search papers, labs, and topics across Lattice.
This paper investigates visual-text fusion in MLLMs through layer-wise masking and attention analysis, revealing non-uniform fusion across layers and a late-stage visual signal reactivation. The authors identify persistent high-attention noise on irrelevant regions and increasing attention on text-aligned areas during processing. Based on these insights, they propose a training-free contrastive attention framework that models attention shifts between early fusion and final layers, enhancing multimodal reasoning.
MLLMs don't fuse vision and language uniformly: targeted interventions guided by layer-wise attention analysis can significantly boost multimodal reasoning without retraining.
Multimodal Large Language Models (MLLMs) have achieved remarkable progress in vision-language understanding, yet how they internally integrate visual and textual information remains poorly understood. To bridge this gap, we perform a systematic layer-wise masking analysis across multiple architectures, revealing how visual-text fusion evolves within MLLMs. The results show that fusion emerges at several specific layers rather than being uniformly distributed across the network, and certain models exhibit a late-stage"review"phenomenon where visual signals are reactivated before output generation. Besides, we further analyze layer-wise attention evolution and observe persistent high-attention noise on irrelevant regions, along with gradually increasing attention on text-aligned areas. Guided by these insights, we introduce a training-free contrastive attention framework that models the transformation between early fusion and final layers to highlight meaningful attention shifts. Extensive experiments across various MLLMs and benchmarks validate our analysis and demonstrate that the proposed approach improves multimodal reasoning performance. Code will be released.
