模拟、聚焦与集成:面向域生成的注意力聚焦方案
摘要
域生成(DG)旨在从源域学习模型并应用于未见目标域,以处理外分布数据。在 CLIP 强大的语义概念编码能力吸引越来越多的关注后,CLIP 常在跨域间聚焦任务相关区域(即域不变区域),导致在未见目标域上表现不佳。为此,我们提出一种注意力聚焦方案,称为 Simulate, Refocus and Ensemble(SRE),通过注意力聚焦对齐 CLIP 中的注意力图来学习减少域迁移。SRE 首先通过对源数据进行增强来生成模拟目标域。随后,SRE 通过注意力聚焦学习在源域和模拟目标域之间减少域迁移。最后,SRE 使用集成学习来增强捕获源数据与模拟目标数据之间域不变注意力图的能力。广泛的实验结果表明,SRE 在多个数据集上通常优于当前方法。该代码已公开:https://github.com/bitPrincy/SRE-DG。
引用
@article{arxiv.2507.12851,
title = {Simulate, Refocus and Ensemble: An Attention-Refocusing Scheme for Domain Generalization},
author = {Ziyi Wang and Zhi Gao and Jin Chen and Qingjie Zhao and Xinxiao Wu and Jiebo Luo},
journal= {arXiv preprint arXiv:2507.12851},
year = {2025}
}
备注
\c{opyright} 20XX IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works