arXiv 2602.074582026-05-13cs.CV

SpatialReward: 通过显式空间推理弥合在线强化学习中图像编辑的感知差距

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

Harbin Institute of Technology, Shenzhen（哈尔滨工业大学（深圳））；The Hong Kong University of Science and Technology（香港科学与技术大学）；Tsinghua Shenzhen International Graduate School, Tsinghua University（清华大学深圳国际研究生学院，清华大学）

Yancheng Long, Yankai Yang, Hongyang Wei, Wei Chen, Tianke Zhang, Haonan fan, Changyi Liu, Kaiyu Jiang, Jiankang Chen, Kaiyu Tang, Bin Wen, Fan Yang, Tingting G… 展开作者

Yancheng Long, Yankai Yang, Hongyang Wei, Wei Chen, Tianke Zhang, Haonan fan, Changyi Liu, Kaiyu Jiang, Jiankang Chen, Kaiyu Tang, Bin Wen, Fan Yang, Tingting Gao, Han Li, Shuo Yang

备注：: Accepted at the 43rd International Conference on Machine Learning (ICML 2026)

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出SpatialReward，通过显式空间推理提升在线强化学习中图像编辑的感知准确性，其在MMRB2和EditReward-Bench上表现优异，并在MultiEditReward-Bench上超越专用评估器。

摘要：

在线强化学习（RL）为复杂图像编辑提供了有前景的途径，但目前受限于可靠且细粒度的奖励信号稀缺。现有评估器常面临我们称为'注意力崩溃'的感知差距，即模型忽视跨图像比较和无法捕捉细粒度细节，导致感知不准确和评分失真。为解决这些限制，我们提出SpatialReward，一种通过显式空间推理进行精确验证的奖励模型。通过将推理锚定在预测的编辑区域，SpatialReward将语义判断基于像素级证据，显著提升评估准确性。在经过精心整理的260k个空间感知数据集上训练，我们的模型在MMRB2和EditReward-Bench上达到SOTA性能，并在我们提出的MultiEditReward-Bench上超越专用评估器。此外，SpatialReward在在线RL中提供稳健的信号，使OmniGen2在GEdit-Bench上提升+0.90，超过领先判别模型并使GPT-4.1的增益翻倍（+0.45）。这些结果表明，空间推理对于解锁图像编辑中的有效对齐至关重要。

英文摘要：

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.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。