DR$^2$Seg: Decomposed Two-Stage Rollouts for Efficient Reasoning Segmentation in Multimodal Large Language Models
This work addresses a specific bottleneck in vision-language tasks for researchers and practitioners, offering an incremental improvement over existing methods.
The paper tackles the problem of overthinking in reasoning segmentation for multimodal large language models, which leads to verbose reasoning chains that interfere with object localization, and proposes DR^2Seg, a self-rewarding framework that improves reasoning efficiency and segmentation accuracy without extra supervision, as demonstrated through experiments on Qwen2.5-VL variants and SAM models.
Reasoning segmentation is an emerging vision-language task that requires reasoning over intricate text queries to precisely segment objects. However, existing methods typically suffer from overthinking, generating verbose reasoning chains that interfere with object localization in multimodal large language models (MLLMs). To address this issue, we propose DR$^2$Seg, a self-rewarding framework that improves both reasoning efficiency and segmentation accuracy without requiring extra thinking supervision. DR$^2$Seg employs a two-stage rollout strategy that decomposes reasoning segmentation into multimodal reasoning and referring segmentation. In the first stage, the model generates a self-contained description that explicitly specifies the target object. In the second stage, this description replaces the original complex query to verify its self-containment. Based on this design, two self-rewards are introduced to mitigate overthinking and the associated attention dispersion. Extensive experiments conducted on 3B and 7B variants of Qwen2.5-VL, as well as on both SAM2 and SAM3, demonstrate that DR$^2$Seg consistently improves reasoning efficiency and overall segmentation accuracy.