Dr. Seg: Revisiting GRPO Training for Visual Large Language Models through Perception-Oriented Design

TL;DR

Dr.~Seg is proposed, a simple, plug-and-play GRPO-based framework consisting of a Look-to-Confirm mechanism and a Distribution-Ranked Reward module, requiring no architectural modifications and integrating seamlessly with existing GRPO-based VLLMs, improving performance in complex visual scenarios while maintaining strong generalization.

Abstract

Following the success of Group Relative Policy Optimization (GRPO) in foundation LLMs, an increasing number of works have sought to adapt GRPO to Visual Large Language Models (VLLMs) for visual perception tasks (e.g., detection and segmentation). However, much of this line of research rests on a long-standing yet unexamined assumption: training paradigms developed for language reasoning can be transferred seamlessly to visual perception. Our experiments show that this assumption is not valid, revealing intrinsic differences between reasoning-oriented and perception-oriented settings. Using reasoning segmentation as a representative case, we surface two overlooked factors: (i) the need for a broader output space, and (ii) the importance of fine-grained, stable rewards. Building on these observations, we propose Dr.~Seg, a simple, plug-and-play GRPO-based framework consisting of a Look-to-Confirm mechanism and a Distribution-Ranked Reward module, requiring no architectural modifications and integrating seamlessly with existing GRPO-based VLLMs. Extensive experiments demonstrate that Dr.~Seg improves performance in complex visual scenarios while maintaining strong generalization. Code and models will be available at https://github.com/xVI-group-SCU/Dr-Seg.

Figures (14)

• Figure 1: Dr. Seg achieves new state-of-the-art results on 5 out of 6 benchmarks under both in-distribution (ID) and out-of-distribution (OOD) conditions, demonstrating strong generalization ability.
• Figure 2: Illustration of breadth-oriented exploration in perception-oriented tasks. Multiple level visual attributes can support different reasoning trajectories.
• Figure 3: Comparison of token-level entropy during training between baseline VisionReasonerliu2025visionreasoner model and baseline with a Look-to-Confirm strategy introduced in next section. The numbers in parentheses are the performance on ReasonSeg dataset.
• Figure 4: PCA visualization of last-token embeddings for baseline VisionReasonerliu2025visionreasoner model vs. baseline with Look-to-Confirm. Points are 2D projections; shaded ellipses denote $2\sigma$ covariance contours around each group mean.
• Figure 5: Overall training pipeline of Dr. Seg.