Active Adversarial Domain Adaptation

Jong-Chyi Su, Yi-Hsuan Tsai, Kihyuk Sohn, Buyu Liu, Subhransu Maji, Manmohan Chandraker; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 1-4


The covariate shift problem is common in many practical computer vision applications, where the training and test data are drawn from different distribution, e.g., the seasonal distribution of natural species may change in a camera trap dataset. Many domain adaptation (DA) methods have been proposed to address this issue [3, 10, 19, 17, 11, 5, 18] by matching the marginal distributions of source and target domain. While domain adaptation provides a good starting point, the performance of unsupervised DA methods often fall far behind their supervised counterparts [16, 1]. In such cases, some labeled data from the target domain can bring in performance benefits. However, obtaining ground-truth annotations can be laborious and naively collecting annotated data could be inefficient. In this work, we aim to answer the following questions: 1) how to select data to label from the target domain effectively, and 2) how to perform adaptation given these labeled data from the target domain.

Related Material

[pdf] [dataset]
author = {Su, Jong-Chyi and Tsai, Yi-Hsuan and Sohn, Kihyuk and Liu, Buyu and Maji, Subhransu and Chandraker, Manmohan},
title = {Active Adversarial Domain Adaptation},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}