Learning Target-Domain-Specific Classifier for Partial Domain Adaptation

Abstract

Unsupervised domain adaptation (UDA) aims at reducing the distribution discrepancy when transferring knowledge from a labeled source domain to an unlabeled target domain. Previous UDA methods assume that the source and target domains share an identical label space, which is unrealistic in practice since the label information of the target domain is agnostic. This paper focuses on a more realistic UDA scenario, i.e. partial domain adaptation (PDA), where the target label space is subsumed to the source label space. In the PDA scenario, the source outliers that are absent in the target domain may be wrongly matched to the target domain (technically named negative transfer), leading to performance degradation of UDA methods. This paper proposes a novel Target Domain Specific Classifier Learning-based Domain Adaptation (TSCDA) method. TSCDA presents a soft-weighed maximum mean discrepancy criterion to partially align feature distributions and alleviate negative transfer. Also, it learns a target-specific classifier for the target domain with pseudo-labels and multiple auxiliary classifiers, to further address classifier shift. A module named Peers Assisted Learning is used to minimize the prediction difference between multiple target-specific classifiers, which makes the classifiers more discriminant for the target domain. Extensive experiments conducted on three PDA benchmark datasets show that TSCDA outperforms other state-of-the-art methods with a large margin, e.g. 4% and 5.6% averagely on Office-31 and Office-Home, respectively.

Source Data-absent Unsupervised Domain Adaptation through Hypothesis Transfer and Labeling Transfer

“…In particular, there are totally 25 classes (the first 25 in the alphabetical order) out of 65 classes in the target domain for Office-Home, while the first 6 classes in the alphabetical order out of 12 classes are included in the target domain for VISDA-C. Results of our methods and previous state-of-the-art PDA methods [105], [106], [107], [108] are shown in Table 7. As explained in Section 3.6, β = 0 is utilized in all of our methods here.…”

“…For vanilla unsupervised DA in digit recognition, we compare SHOT with ADDA [5], ADR [56], CDAN [42], CyCADA [8], CAT [96], SWD [97] and STAR [98]; for object recognition, we compare ours with DANN [24], DAN [4], SAFN [82], BSP [99], MDD [100], TransNorm [55], DSBN [25], BNM [101] and GVB-GD [102]. For partial-set DA tasks, we compare ours with IWAN [103], SAN [89], ETN [104], DRCN [105], RTNet adv [106], BA 3 US [107], and TSCDA [108]. For multi-source UDA, we compare ours with DCTN [109], MCD [6], WBN [110], M 3 SDA-β [18], CMSS [111], and CMSS [111].…”

Cross-Site Severity Assessment of COVID-19 From CT Images via Domain Adaptation

“…DA methods aim to find a latent space for source and target domains, so that the discrimination information in the source domain can be efficiently transferred to the recognition task in the target domain [22]- [25]. Generally, DA methods can be divided into three categories: unsupervised DA methods, weakly-supervised DA methods, and semi-supervised DA methods.…”

SPNet: A novel deep neural network for retinal vessel segmentation based on shared decoder and pyramid-like loss

“…In practice, visual discrepancies among the clinical datasets could be large [39,40,41,42]. For example, DRIVE dataset contains 33 images without any sign of diabetic retinopathy and 7 images with signs of mild early diabetic retinopathy.…”