Exploring Robustness of Unsupervised Domain Adaptation in Semantic Segmentation

Yang, Jinyu; Li, Chunyuan; An, Weizhi; Ma, Hehuan; Guo, Yuzhi; Rong, Yu; Zhao, Peilin; Huang, Junzhou

doi:10.1109/iccv48922.2021.00906

Cited by 24 publications

(12 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The white-box FGSM [11] attack with perturbation budget of = 3 is used for both AT and testing. Following the practice of [19], [20], we assume that attackers have the labels of the target dataset to generate adversarial examples. The rationale behind these settings is that (i) most existing UDA approaches [6], [8] are based on DANN's key idea; (ii) the white-box threat model has been considered a standard evaluation protocol for defenses [15], [17], [18], [23].…”

Section: Exploring At For Udamentioning

confidence: 99%

“…A naive way is to produce pseudo labels y t using an external pre-trained UDA model. ASSUDA [20] resorts to this idea and applies it to the UDA semantic segmentation problem. Note that ASSUDA only evaluates black-box robustness.…”

Section: A Conventional At On Udamentioning

confidence: 99%

“…However, its performance is sensitive to the teacher models' perturbation budget, architecture, etc., which limits the flexibility in a wide range of uses. ASSUDA [20] uses an external pre-trained UDA model to produce pseudo labels for AT on target data. Unfortunately, we show that it suffers from suboptimal accuracy and robustness under white-box attacks.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Exploring Adversarially Robust Training for Unsupervised Domain Adaptation

Lo¹,

Patel²

2022

Preprint

View full text Add to dashboard Cite

Unsupervised Domain Adaptation (UDA) methods aim to transfer knowledge from a labeled source domain to an unlabeled target domain. UDA has been extensively studied in the computer vision literature. Deep networks have been shown to be vulnerable to adversarial attacks. However, very little focus is devoted to improving the adversarial robustness of deep UDA models, causing serious concerns about model reliability.Adversarial Training (AT) has been considered to be the most successful adversarial defense approach. Nevertheless, conventional AT requires ground-truth labels to generate adversarial examples and train models, which limits its effectiveness in the unlabeled target domain. In this paper, we aim to explore AT to robustify UDA models: How to enhance the unlabeled data robustness via AT while learning domain-invariant features for UDA? To answer this, we provide a systematic study into multiple AT variants that potentially apply to UDA. Moreover, we propose a novel Adversarially Robust Training method for UDA accordingly, referred to as ARTUDA. Extensive experiments on multiple attacks and benchmarks show that ARTUDA consistently improves the adversarial robustness of UDA models.

show abstract

Section: Exploring At For Udamentioning

confidence: 99%

Section: A Conventional At On Udamentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Exploring Adversarially Robust Training for Unsupervised Domain Adaptation

Lo¹,

Patel²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…, the tendency of the classifier to collapse into the classes that are more represented, in contrast with the long tail of the most underrepresented ones) and conduct adversarial training on the segmentation network to improve its robustness. Yang et al [219] study the adversarial vulnerability of existing DA-SiS methods and propose the adversarial self-supervision UDA, where the objective is to maximize the proximity between clean images and their adversarial counterparts in the output space -by using a contrastive loss. Huang et al [78] propose a Fourier adversarial training method, where the pipeline is (i) generating adversarial samples by perturbing certain high frequency components that do not carry significant semantic information and (ii) using them to train the model.…”

Section: Entropy Minimization Of Target Predictions (Tem)mentioning

confidence: 99%

“…The model is trained end-to-end using co-evolving pseudo-labels -using a momentum network (a copy of the original model that evolves slowly) -and maintaining an exponentially moving class prior, which is used to discount the confidence thresholds for classes with few samples, in order to increase their relative contribution to the training loss. Also, Yang et al [219] -as mentioned in the previous paragraph -exploit self-supervision in DASiS by minimizing the distance between clean and adversarial samples in the output space via a contrastive loss.…”

Section: Entropy Minimization Of Target Predictions (Tem)mentioning

confidence: 99%

Unsupervised Domain Adaptation for Semantic Image Segmentation: a Comprehensive Survey

Volpi¹,

Chidlovskii²

2021

Preprint

View full text Add to dashboard Cite

Semantic segmentation plays a fundamental role in a broad variety of computer vision applications, providing key information for the global understanding of an image. Yet, the state-of-the-art models rely on large amount of annotated samples, which are more expensive to obtain than in tasks such as image classification. Since unlabelled data is instead significantly cheaper to obtain, it is not surprising that Unsupervised Domain Adaptation reached a broad success within the semantic segmentation community.This survey is an effort to summarize five years of this incredibly rapidly growing field, which embraces the importance of semantic segmentation itself and a critical need of adapting segmentation models to new environments. We present the most important semantic segmentation methods; we provide a comprehensive survey on domain adaptation techniques for semantic segmentation; we unveil newer trends such as multi-domain learning, domain generalization, test-time adaptation or source-free domain adaptation; we conclude this survey by describing datasets and benchmarks most widely used in semantic segmentation research. We hope that this survey will provide researchers across academia and industry with a comprehensive reference guide and will help them in fostering new research directions in the field.1 See as example the https://github.com/errolPereira/ Semi-Automatic-Image-Annotation-Tool 2 We refer to the dedicated page https://github.com/ heartexlabs/awesome-data-labeling for an up-to-date collection of annotation tools.

show abstract

DecoupleNet: Decoupled Network for Domain Adaptive Semantic Segmentation

Lai

Tian

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Exploring Robustness of Unsupervised Domain Adaptation in Semantic Segmentation

Cited by 24 publications

References 28 publications

Exploring Adversarially Robust Training for Unsupervised Domain Adaptation

Exploring Adversarially Robust Training for Unsupervised Domain Adaptation

Unsupervised Domain Adaptation for Semantic Image Segmentation: a Comprehensive Survey

DecoupleNet: Decoupled Network for Domain Adaptive Semantic Segmentation

Contact Info

Product

Resources

About