arXiv 2605.108592026-05-11cs.CVcs.LG

你需要Masked Generative Transformer来进行图像编辑

Masked Generative Transformer Is What You Need for Image Editing

ByteDance（字节跳动）；National University of Singapore（新加坡国立大学）；Duke University（杜克大学）；Shanghai Jiao Tong University（上海交通大学）；HKUST(GZ)（香港科技大学（广州））

Wei Chow, Linfeng Li, Xian Sun, Lingdong Kong, Zefeng Li, Qi Xu, Hang Song, Tian Ye, Xian Wang, Jinbin Bai, Shilin Xu, Xiangtai Li, Junting Pan, Shaoteng Liu, R… 展开作者

Wei Chow, Linfeng Li, Xian Sun, Lingdong Kong, Zefeng Li, Qi Xu, Hang Song, Tian Ye, Xian Wang, Jinbin Bai, Shilin Xu, Xiangtai Li, Junting Pan, Shaoteng Liu, Ran Zhou, Tianshu Yang, Songhua Liu

备注：: CVPR 2026 HiGen Workshop; Project Page at https://weichow23.github.io/EditMGT/ GitHub at https://github.com/weichow23/EditMGT

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出EditMGT框架，利用Masked Generative Transformers实现局部化编辑，通过多层注意力整合和区域保持采样提升编辑精度，使用CrispEdit-2M数据集在多个基准上取得最佳性能。

摘要：

扩散模型主导图像编辑，但其全局去噪机制使修改区域与周围上下文纠缠，导致修改传播到应保持完整区域。我们提出基于Masked Generative Transformers（MGTs）的不同方法，其局部化token预测范式自然限制变化于目标区域。我们提出EditMGT框架，是首个此类框架。我们的方法采用多层注意力整合来聚合跨注意力图，生成精确的编辑定位信号，并通过区域保持采样防止非目标区域的token翻转。为支持训练，我们构建了CrispEdit-2M数据集，包含200万样本，覆盖七个类别。仅使用960M参数，EditMGT在多个基准上实现了最佳图像相似性，同时交付6倍更快的编辑速度，证明MGTs为基于扩散模型的编辑提供了有竞争力的替代方案。

英文摘要：

Diffusion models dominate image editing, yet their global denoising mechanism entangles edited regions with surrounding context, causing modifications to propagate into areas that should remain intact. We propose a fundamentally different approach by leveraging Masked Generative Transformers (MGTs), whose localized token-prediction paradigm naturally confines changes to intended regions. We present EditMGT, an MGT-based editing framework that is the first of its kind. Our approach employs multi-layer attention consolidation to aggregate cross-attention maps into precise edit localization signals, and region-hold sampling to explicitly prevent token flipping in non-target areas. To support training, we construct CrispEdit-2M, a 2M-sample high-resolution (>1024) editing dataset spanning seven categories. With only 960M parameters, EditMGT achieves state-of-the-art image similarity on multiple benchmarks while delivering 6x faster editing, demonstrating that MGTs offer a compelling alternative to diffusion-based 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。