IRPO: Boosting Image Restoration via Post-training GRPO

TL;DR

IRPO introduces a novel post-training paradigm for image restoration that leverages data selection and reward modeling to improve performance and generalization, surpassing existing methods on multiple benchmarks.

Contribution

The paper proposes IRPO, a low-level GRPO-based post-training approach that systematically explores data formulation and reward modeling for enhanced image restoration.

Findings

Achieves state-of-the-art results on multiple benchmarks.

Surpasses baseline by 0.83 dB on in-domain tasks.

Surpasses baseline by 3.43 dB on out-of-domain tasks.

Abstract

Recent advances in post-training paradigms have achieved remarkable success in high-level generation tasks, yet their potential for low-level vision remains rarely explored. Existing image restoration (IR) methods rely on pixel-level hard-fitting to ground-truth images, struggling with over-smoothing and poor generalization. To address these limitations, we propose IRPO, a low-level GRPO-based post-training paradigm that systematically explores both data formulation and reward modeling. We first explore a data formulation principle for low-level post-training paradigm, in which selecting underperforming samples from the pre-training stage yields optimal performance and improved efficiency. Furthermore, we model a reward-level criteria system that balances objective accuracy and human perceptual preference through three complementary components: a General Reward for structural fidelity, an Expert Reward leveraging Qwen-VL for perceptual alignment, and a Restoration Reward for task-specific low-level quality. Comprehensive experiments on six in-domain and five out-of-domain (OOD) low-level benchmarks demonstrate that IRPO achieves state-of-the-art results across diverse degradation types, surpassing the AdaIR baseline by 0.83 dB on in-domain tasks and 3.43 dB on OOD settings. Our code can be shown in https://github.com/HaoxuanXU1024/IRPO.