Universal Domain Adaptation through Self Supervision

Saito, Kuniaki; Kim, Donghyun; Sclaroff, Stan; Saenko, Kate

doi:10.48550/arxiv.2002.07953

Cited by 25 publications

(52 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A surge of research interest has been recently attracted to unsupervised learning [43]. Early efforts dedicate to designing pretext tasks [17,19,30,70], which are proven beneficial for UDA when utilized as auxiliary tasks on target data [48,54,67]. The gap with supervised learning is considerably closed by a few prominent works [10,24] that build on contrastive learning.…”

Section: Unsupervised Representation Learningmentioning

confidence: 99%

Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation

Zhang

et al. 2021

Preprint

View full text Add to dashboard Cite

Self-training is a competitive approach in domain adaptive segmentation, which trains the network with the pseudo labels on the target domain. However inevitably, the pseudo labels are noisy and the target features are dispersed due to the discrepancy between source and target domains. In this paper, we rely on representative prototypes, the feature centroids of classes, to address the two issues for unsupervised domain adaptation. In particular, we take one step further and exploit the feature distances from prototypes that provide richer information than mere prototypes. Specifically, we use it to estimate the likelihood of pseudo labels to facilitate online correction in the course of training. Meanwhile, we align the prototypical assignments based on relative feature distances for two different views of the same target, producing a more compact target feature space. Moreover, we find that distilling the already learned knowledge to a self-supervised pretrained model further boosts the performance. Our method shows tremendous performance advantage over state-of-the-art methods. We will make the code publicly available.

show abstract

Section: Unsupervised Representation Learningmentioning

confidence: 99%

Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation

Zhang

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In domain adaptation some recent approaches [5,36,41] have used selfsupervision tasks as an auxiliary objective to regularize their model. Saito et al [30], used a self-supervised feature space clustering objective for universal domain adaptation. PAC differs from these approaches in that we use rotation prediction to pretrain our feature extractor.…”

Section: Background and Related Workmentioning

confidence: 99%

Surprisingly Simple Semi-Supervised Domain Adaptation with Pretraining and Consistency

Mishra¹,

Saenko²,

Saligrama³

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Visual domain adaptation involves learning to classify images from a target visual domain using labels available in a different source domain. A range of prior work uses adversarial domain alignment to try and learn a domain invariant feature space, where a good source classifier can perform well on target data. This however, can lead to errors where class A features in the target domain get aligned to class B features in source. We show that in the presence of a few target labels, simple techniques like selfsupervision (via rotation prediction) and consistency regularization can be effective without any adversarial alignment to learn a good target classifier. Our Pretraining and Consistency (PAC) approach, can achieve state of the art accuracy on this semi-supervised domain adaptation task, surpassing multiple adversarial domain alignment methods, across multiple datasets. Notably, it outperforms all recent approaches by 3-5% on the large and challenging DomainNet benchmark, showing the strength of these simple techniques in fixing errors made by adversarial alignment.

show abstract

“…Prior domain adaptation methods first extract discriminative features on the source domain guided by source supervision. Then, they align the target features with the source features by: minimizing maximum mean discrepancy [19], minimizing maximum discrepancy of domain distributions [33,42], feature-level or pixel-level adversarial domain classifier based learning [7,14,18,36], entropy optimization [32,18,31], and finding matching pairs across domains based on optimal transport [1,4,34,38] or nearest neighbors [24,10]. Some semi-supervised learning techniques such as entropy minimization [9], pseudo-labeling [16], and Virtual Adversarial Training (VAT) [21] have been often used in domain adaptation (e.g., [17,31,44]).…”

Section: Related Workmentioning

confidence: 99%

“…Instance Discrimination [39] learns an embedding which maps visually similar images closer to each other and far from dissimilar images by classifying an image as its own unique class. Other methods propose to cluster local neighborhoods [3,15,32,43] within the same domain.…”

Section: Related Workmentioning

confidence: 99%

Cross-domain Self-supervised Learning for Domain Adaptation with Few Source Labels

Kim,

Saito,

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Existing unsupervised domain adaptation methods aim to transfer knowledge from a label-rich source domain to an unlabeled target domain. However, obtaining labels for some source domains may be very expensive, making complete labeling as used in prior work impractical. In this work, we investigate a new domain adaptation scenario with sparsely labeled source data, where only a few examples in the source domain have been labeled, while the target domain is unlabeled. We show that when labeled source examples are limited, existing methods often fail to learn discriminative features applicable for both source and target domains. We propose a novel Cross-Domain Self-supervised (CDS) learning approach for domain adaptation, which learns features that are not only domain-invariant but also class-discriminative. Our self-supervised learning method captures apparent visual similarity with in-domain selfsupervision in a domain adaptive manner and performs cross-domain feature matching with across-domain self-supervision. In extensive experiments with three standard benchmark datasets, our method significantly boosts performance of target accuracy in the new target domain with few source labels and is even helpful on classical domain adaptation scenarios.

show abstract

Universal Domain Adaptation through Self Supervision

Cited by 25 publications

References 0 publications

Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation

Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation

Surprisingly Simple Semi-Supervised Domain Adaptation with Pretraining and Consistency

Cross-domain Self-supervised Learning for Domain Adaptation with Few Source Labels

Contact Info

Product

Resources

About