From Drop-off to Recovery: A Mechanistic Analysis of Segmentation in MLLMs
This provides a mechanistic analysis of MLLMs for segmentation, informing future model design, but it is incremental as it builds on existing understanding of attention mechanisms.
The study investigated the spatial understanding of Multimodal Large Language Models (MLLMs) for segmentation tasks, finding that the adapter causes a drop-off in segmentation representation, but LLM layers recover it through attention-mediated refinement, with bidirectional attention among image tokens improving spatial consistency.
Multimodal Large Language Models (MLLMs) are increasingly applied to pixel-level vision tasks, yet their intrinsic capacity for spatial understanding remains poorly understood. We investigate segmentation capacity through a layerwise linear probing evaluation across the entire MLLM pipeline: vision encoder, adapter, and LLM. We further conduct an intervention based attention knockout analysis to test whether cross-token attention progressively refines visual representations, and an evaluation of bidirectional attention among image tokens on spatial consistency. Our analysis reveals that the adapter introduces a segmentation representation drop-off, but LLM layers progressively recover through attention-mediated refinement, where correctly classified tokens steer misclassified neighbors toward the correct label. At early image token positions, this recovery is bounded by causal attention, which bidirectional attention among image tokens alleviates. These findings provide a mechanistic account of how MLLMs process visual information for segmentation, informing the design of future segmentation-capable models.