From Failure to Feedback: Group Revision Unlocks Hard Cases in Object-Level Grounding
This work addresses the problem of sparse reward signals in reinforcement learning-based fine-tuning of vision-language models for object-level grounding, which is a known bottleneck for hard cases.
The paper proposes a group-revision optimization paradigm for fine-tuning Large Vision-Language Models on object-level grounding tasks, which converts sparse response-level rewards into informative shaping signals by comparing revised candidates to an initial response. The method achieves consistent gains across referring segmentation, reasoning segmentation, referring expression comprehension, and counting benchmarks compared to prior GRPO-based models.
Finetuning Large Vision-Language Models with reinforcement learning has emerged as a promising approach to enhance their capability in object-level grounding. However, existing methods, mainly based on GRPO, assign rewards at the response level. Such sparse reward, often criterion-induced, leads to minimal learning signals when all candidate responses fail in challenging scenarios. In this work, we propose a group-revision optimisation paradigm that enhances learning on hard cases. It begins with a sampled initial response and generates a set of revised candidates to explore improved grounding outcomes. Inspired by reward shaping, we introduce a consolidation process that quantifies each candidate's improvement over the initial attempt and converts it into informative shaping signals. These signals are used to both refine the reward and modulate the advantage, amplifying the influence of high-quality revisions. Our method achieves consistent gains across referring and reasoning segmentation, REC, and counting benchmarks compared with prior GRPO-based models. Our code is available at https://github.com/yyliu01/GroupRevision.