Domain Adaptation and Image Classification via Deep Conditional Adaptation Network

Abstract

Unsupervised domain adaptation aims to generalize the supervised model trained on a source domain to an unlabeled target domain. Marginal distribution alignment of feature spaces is widely used to reduce the domain discrepancy between the source and target domains. However, it assumes that the source and target domains share the same label distribution, which limits their application scope. In this paper, we consider a more general application scenario where the label distributions of the source and target domains are not the same. In this scenario, marginal distribution alignment-based methods will be vulnerable to negative transfer. To address this issue, we propose a novel unsupervised domain adaptation method, Deep Conditional Adaptation Network (DCAN), based on conditional distribution alignment of feature spaces. To be specific, we reduce the domain discrepancy by minimizing the Conditional Maximum Mean Discrepancy between the conditional distributions of deep features on the source and target domains, and extract the discriminant information from target domain by maximizing the mutual information between samples and the prediction labels. In addition, DCAN can be used to address a special scenario, Partial unsupervised domain adaptation, where the target domain category is a subset of the source domain category. Experiments on both unsupervised domain adaptation and Partial unsupervised domain adaptation show that DCAN achieves superior classification performance over state-of-the-art methods. In particular, DCAN achieves great improvement in the tasks with large difference in label distributions (6.1% on SVHN→MNIST, 5.4% in UDA tasks on Office-Home and 4.5% in Partial UDA tasks on Office-Home).

Strong-Weak Integrated Semi-Supervision for Unsupervised Domain Adaptation

“…For quantitative assessment of the proposed strong-weak integrated semi-supervision (SWISS) method, we evaluate it on two popular benchmarks Office-Home [23], DomainNet [2]; and, we compare our proposed method with state-of-the-art methods, namely, DADA [20], CDAN+BSP [16], MCC [22], DCAN [17], SHOT-IM [12], and the most recent FixBi [18], SCDA [19], AANet [4]. We employ the pre-trained ResNet-50 or ResNet-101 as the feature generator G as was done in [12,24,17,21]. For the classifier F, we use a linear layer with the input and output dimensions being (1024, k), where k is the number of classes.…”

“…Baselines. For single-target domain adaptation, we compare with several classical and state-of-the-art methods, namely, the classical DANN [39], MSTN [40], ADDA [41], MCD [42], RTN [43], JAN [44], CDAN+BSP [45], CAN [46], MDD [47], DCAN [48], SHOT-IM [29], MIMTFL [49], and the most recent FixBi [23], ATDOC [22], SCDA [50], AANet [7]. For multi-target domain adaptation, we compare with three recent works MT-MTDA [12], D-CGCT [14], and HGAN [9].…”

“…Implementation Details We employ the pre-trained ResNet-50 or ResNet-101 as the feature generator G like that in [20], [29], [48], [51], [52]. For the classifier F, we use a linear layer with the input and output dimensions being (1024, k), where k is the number of classes.…”

Cell Detection in Domain Shift Problem Using Pseudo-Cell-Position Heatmap

“…To address the domain shift problem, unsupervised domain adaptation (UDA) methods, which use only data on the source domain without data on the target domain, have been proposed [9,10,12,13,15,[25][26][27]. Most are designed for classification tasks but not for detection tasks.…”