D3GU: Multi-target Active Domain Adaptation via Enhancing Domain Alignment

Lin Zhang1      Linghan Xu1      Saman Motamed1
      Shayok Chakraborty2      Fernando de la Torre1
1Carnegie Mellon University      2Florida State University     

WACV 2024

Paper Code

The proposed Multi-target Active Domain Adaptation (MT-ADA) framework. In each stage, sampled are sampled in the union pool composed of unlabeled target images.

Abstract

Unsupervised domain adaptation (UDA) for image classification has made remarkable progress in transferring classification knowledge from a labeled source domain to an unlabeled target domain, thanks to effective domain alignment techniques. Recently, in order to further improve performance on a target domain, many Single-Target Active Domain Adaptation (ST-ADA) methods have been proposed to identify and annotate the salient and exemplar target samples. However, it requires one model to be trained and deployed for each target domain and the domain label associated with each test sample. This largely restricts its application in the ubiquitous scenarios with multiple target domains. Therefore, we propose a Multi-Target Active Domain Adaptation (MT-ADA) framework for image classification, named D3GU, to simultaneously align different domains and actively select samples from them for annotation. This is the first research effort in this field to our best knowledge. D3GU applies Decomposed Domain Discrimination (D3) during training to achieve both source-target and target-target domain alignments. Then during active sampling, a Gradient Utility (GU) score is designed to weight every unlabeled target image by its contribution towards classification and domain alignment tasks, and is further combined with KMeans clustering to form GU-KMeans for diverse image sampling. Extensive experiments on three benchmark datasets, Office31, OfficeHome, and DomainNet, have been conducted to validate consistently superior performance of D3GU for MT-ADA.

Performance

MT-ADA accuracies with total budget 120 and 400 on Office31 and OfficeHome, respectively. We conducted 4 active learning stages with equal budgets. Pretraining stage and active learning training stages apply the same domain discrimination, which is indicated by “(a)”(all-way) and “(b)”(binary) postfixes in “Method” column.

MT-ADA accuracies on DomainNet with total budget =10, 000 in 4 active learning stages.

BibTex

@inproceedings{zhang2024d3gu,
      title={D3GU: Multi-target Active Domain Adaptation via Enhancing Domain Alignment},
      author={Zhang, Lin and Xu, Linghan and Motamed, Saman and Chakraborty, Shayok and De la Torre, Fernando},
      booktitle = {WACV},
      year={2024}
    }
}