Reciprocal normalization for domain adaptation

Zhiyong Huang

doi:10.1016/j.patcog.2023.109533

Reciprocal normalization for domain adaptation

2023-08-01

5

OA

AI

DOI：10.1016/j.patcog.2023.109533

原文链接

查看原文

摘要

En 中文

Batch normalization (BN) is widely used in modern deep neural networks, which has been shown to represent the domain-related knowledge, and thus is ineffective for cross-domain tasks like unsuper-vised domain adaptation (UDA). Existing BN variant methods aggregate source and target domain knowl-edge in the same channel in normalization module. However, the misalignment between the features of corresponding channels across domains often leads to a sub-optimal transferability. In this paper, we exploit the cross-domain relation and propose a novel normalization method, Reciprocal Normal-ization (RN). Specifically, RN first presents a Reciprocal Compensation (RC) module to acquire the com-pensatory for each channel in both domains based on the cross-domain channel-wise correlation. Then RN develops a Reciprocal Aggregation (RA) module to adaptively aggregate the feature with its cross-domain compensatory components. As an alternative to BN, RN is more suitable for UDA problems and can be easily integrated into popular domain adaptation methods. Experiments show that the proposed RN outperforms existing normalization counterparts by a large margin and helps state-of-the-art adapta-tion approaches achieve better results. The source code is available on https://github.com/Openning07/ reciprocal- normalization- for-DA . (c) 2023 Published by Elsevier Ltd.

Keyword:

Domain adaptation

Feature normalization

Deep neural network

Reciprocal normalization for domain adaptation

摘要

AI总结

期刊

机构

引用论文