期刊
2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021
卷 -, 期 -, 页码 10993-11002出版社
IEEE COMPUTER SOC
DOI: 10.1109/CVPR46437.2021.01085
关键词
-
资金
- NSF [IIS1924937, IIS-2041009]
- Amazon
This paper introduces dynamic transfer to address domain conflicts by adapting model parameters to samples, transforming multi-source domains into a single-source domain and simplifying alignment between source and target domains. Experimental results demonstrate that dynamic transfer outperforms the state-of-the-art method on the large multi-source domain adaptation dataset DomainNet.
Recent works of multi-source domain adaptation focus on learning a domain-agnostic model, of which the parameters are static. However, such a static model is difficult to handle conflicts across multiple domains, and suffers from a performance degradation in both source domains and target domain. In this paper, we present dynamic transfer to address domain conflicts, where the model parameters are adapted to samples. The key insight is that adapting model across domains is achieved via adapting model across samples. Thus, it breaks down source domain barriers and turns multi-source domains into a single-source domain. This also simplifies the alignment between source and target domains, as it only requires the target domain to be aligned with any part of the union of source domains. Furthermore, we find dynamic transfer can be simply modeled by aggregating residual matrices and a static convolution matrix. Experimental results show that, without using domain labels, our dynamic transfer outperforms the state-of-theart method by more than 3% on the large multi-source domain adaptation datasets - DomainNet.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据