期刊
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
卷 32, 期 6, 页码 3798-3809出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TCSVT.2021.3116210
关键词
Semantics; Image segmentation; Task analysis; Training; Annotations; Adaptation models; Feature extraction; Semantic segmentation; domain adaptation; partial adaptation
资金
- National Natural Science Foundation of China [12071458, 71731009]
The research of semantic segmentation based on unsupervised domain adaptation addresses the issue of knowledge transfer between domains by reducing the high cost of manual annotation. Previous domain alignment strategies suffer from class asymmetry in practical training. By introducing partial domain adaptive method, category-level knowledge can be selectively learned to avoid negative transfer.
The research of semantic segmentation based on unsupervised domain adaptation greatly alleviates the high-cost bottleneck of manual annotation in deep learning. Inevitably domain gap limits the ability of target domain to learn knowledge from source domain. Previous domain alignment strategies aim to explore maximum domain-invariant space to enlarge knowledge that target domain learns. They based on the assumption that images are symmetrical in categories covering all predefined categories. However, in practical training, there is class asymmetry in image samples inter domains. This will lead to outlier categories and negative transfer, because it lacks the guidance of the corresponding category of the source domain. To tackle this issue, we propose a partial domain adaptive method on semantic segmentation to guide target model to learn category-level knowledge selectively. PAM (Partial Adaptive Map) module is introduced to motivate the target model to obtain more knowledge from non-outliers and less knowledge from outliers to avoid negative transfer. We further analyze the effectiveness of this method from the perspective of JS divergence. Our method has achieved significant improvement without additional discriminators. Experiments on general datasets GTA5 and SYNTHIA can compare with SOTA methods.
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