From Failure to Feedback: Group Revision Unlocks Hard Cases in Object-Level Grounding

2026-05-15

AI Summary

This work addresses a key limitation of existing GRPO-based reinforcement fine-tuning approaches for object-level grounding tasks: their reliance on sparse response-level rewards, which fails to provide effective learning signals when all candidate responses are incorrect on hard examples. To overcome this, the authors propose a Group Revision Optimization (GRO) paradigm that generates multiple revised candidates and constructs a reward-shaping signal based on their relative improvements over the initial response. This signal modulates the advantage function to enhance learning on challenging instances. Extensive experiments demonstrate that the proposed method significantly outperforms current GRPO baselines across multiple benchmarks, including referring expression segmentation, reasoning-based segmentation, referring expression comprehension (REC), and counting tasks, thereby validating its effectiveness and generalization capability.

Abstract

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.

Problem

• object-level grounding
• sparse reward
• hard cases
• reinforcement learning
• vision-language models

Innovation

• group revision
• reward shaping
• object-level grounding
• reinforcement learning
• vision-language models