SpatialReward: Bridging the Perception Gap in Online RL for Image Editing via Explicit Spatial Reasoning

TL;DR

SpatialReward introduces an explicit spatial reasoning reward model that improves evaluation accuracy and enhances online RL performance in complex image editing tasks by addressing the perception gap.

Contribution

The paper presents SpatialReward, a novel reward model leveraging spatial reasoning to improve evaluation and RL in image editing, trained on a large spatial-aware dataset.

Findings

SpatialReward achieves state-of-the-art results on multiple benchmarks.

It outperforms proprietary evaluators and enhances RL agent performance.

Spatial reasoning is shown to be crucial for effective image editing alignment.

Abstract

Online Reinforcement Learning (RL) offers a promising avenue for complex image editing but is currently constrained by the scarcity of reliable and fine-grained reward signals. Existing evaluators frequently struggle with a critical perception gap we term "Attention Collapse," where models neglect cross-image comparisons and fail to capture fine-grained details, resulting in inaccurate perception and miscalibrated scores. To address these limitations, we propose SpatialReward, a reward model that enforces precise verification via explicit spatial reasoning. By anchoring reasoning to predicted edit regions, SpatialReward grounds semantic judgments in pixel-level evidence, significantly enhancing evaluative accuracy. Trained on a curated 260k spatial-aware dataset, our model achieves state-of-the-art performance on MMRB2 and EditReward-Bench, and outperforms proprietary evaluators on our proposed MultiEditReward-Bench. Furthermore, SpatialReward serves as a robust signal in online RL, boosting OmniGen2 by +0.90 on GEdit-Bench--surpassing the leading discriminative model and doubling the gain of GPT-4.1 (+0.45). These results demonstrate that spatial reasoning is essential for unlocking effective alignment in image editing.