SelfReg: Self-supervised Contrastive Regularization for Domain Generalization

Kim, Dae–Hee; Park, Seung-Hyun; Kim, Jinkyu; Lee, Jaekoo

doi:10.1109/iccv48922.2021.00948

Cited by 128 publications

(56 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…and to the baseline trained directly on RGB natural images. SelfReg [46] performs poorly as expected; it is intended for multi-source domain generalization. For L2D [51] which is designed specifically for the single-source task, we run the provided code and ensure that optimal learning rate according to validation performance on the source task is used; the reported score is reproduced.…”

Section: A Ablation Study On Sketchymentioning

confidence: 55%

“…c) Domain generalization: The most common approach for domain generalization is invariant feature learning, based on the theoretical results of Ben-David et al [42]. Representative approaches include kernel-based invariant feature learning by minimizing domain dissimilarity [43], multi-task autoencoders that transform the original image to other related domains, domain classifiers as adversaries to match the source domain distributions in the feature space [44], [45], and crossdomain non-contrastive learning as regularization [46]. Some methods specialize for single-source domain generalization.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Edge Augmentation for Large-Scale Sketch Recognition without Sketches

Efthymiadis¹,

Tolias²,

Chum³

2022

Preprint

View full text Add to dashboard Cite

This work addresses scaling up the sketch classification task into a large number of categories. Collecting sketches for training is a slow and tedious process that has so far precluded any attempts to large-scale sketch recognition. We overcome the lack of training sketch data by exploiting labeled collections of natural images that are easier to obtain. To bridge the domain gap we present a novel augmentation technique that is tailored to the task of learning sketch recognition from a training set of natural images. Randomization is introduced in the parameters of edge detection and edge selection. Natural images are translated to a pseudo-novel domain called "randomized Binary Thin Edges" (rBTE), which is used as a training domain instead of natural images. The ability to scale up is demonstrated by training CNN-based sketch recognition of more than 2.5 times larger number of categories than used previously. For this purpose, a dataset of natural images from 874 categories is constructed by combining a number of popular computer vision datasets. The categories are selected to be suitable for sketch recognition. To estimate the performance, a subset of 393 categories with sketches is also collected.

show abstract

Section: A Ablation Study On Sketchymentioning

confidence: 55%

Section: Introductionmentioning

confidence: 99%

Edge Augmentation for Large-Scale Sketch Recognition without Sketches

Efthymiadis¹,

Tolias²,

Chum³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…To alleviate this challenge, Pandey et al [48] used a metric transformation to keep the source samples clustering near the corresponding categories in the feature space. Kim et al [49] used a self-supervised contrastive loss to make the representations of the positive pair samples close. However, these interesting works are designed for the task of domain generalization, in which the target domain is unaccessible during training, thus, they rarely consider transferring discriminative information for the target domain.…”

Section: B Pseudo Target Domain: Construction and Alignmentmentioning

confidence: 99%

Multi-Source Unsupervised Domain Adaptation via Pseudo Target Domain

Ren

Liu

Zhang

et al. 2022

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

Multi-source domain adaptation (MDA) aims to transfer knowledge from multiple source domains to an unlabeled target domain. MDA is a challenging task due to the severe domain shift, which not only exists between target and source but also exists among diverse sources. Prior studies on MDA either estimate a mixed distribution of source domains or combine multiple single-source models, but few of them delve into the relevant information among diverse source domains. For this reason, we propose a novel MDA approach, termed Pseudo Target for MDA (PTMDA). Specifically, PTMDA maps each group of source and target domains into a group-specific subspace using adversarial learning with a metric constraint, and constructs a series of pseudo target domains correspondingly. Then we align the remainder source domains with the pseudo target domain in the subspace efficiently, which allows to exploit additional structured source information through the training on pseudo target domain and improves the performance on the real target domain. Besides, to improve the transferability of deep neural networks (DNNs), we replace the traditional batch normalization layer with an effective matching normalization layer, which enforces alignments in latent layers of DNNs and thus gains further promotion. We give theoretical analysis showing that PTMDA as a whole can reduce the target error bound and leads to a better approximation of the target risk in MDA settings. Extensive experiments demonstrate PTMDA's effectiveness on MDA tasks, as it outperforms state-of-the-art methods in most experimental settings.

show abstract

“…Domain Generalization (DG) task also tackles the OOD generalization problem, but requires sufficient domain samples and full ImageNet pretraining. In this paper, we select three SOTA DG approaches (SD [70], SelfReg [50] and SagNet [63]) for comparison. These methods do not require domain labels which share the same setting as ours.…”

Section: Implementation Detailsmentioning

confidence: 99%

Equivariance and Invariance Inductive Bias for Learning from Insufficient Data

Wang¹,

Sun²,

Pranata³

et al. 2022

Preprint

View full text Add to dashboard Cite

We are interested in learning robust models from insufficient data, without the need for any externally pre-trained checkpoints. First, compared to sufficient data, we show why insufficient data renders the model more easily biased to the limited training environments that are usually different from testing. For example, if all the training swan samples are "white", the model may wrongly use the "white" environment to represent the intrinsic class swan. Then, we justify that equivariance inductive bias can retain the class feature while invariance inductive bias can remove the environmental feature, leaving the class feature that generalizes to any environmental changes in testing. To impose them on learning, for equivariance, we demonstrate that any off-theshelf contrastive-based self-supervised feature learning method can be deployed; for invariance, we propose a class-wise invariant risk minimization (IRM) that efficiently tackles the challenge of missing environmental annotation in conventional IRM. State-of-the-art experimental results on real-world benchmarks (VIPriors, ImageNet100 and NICO) validate the great potential of equivariance and invariance in data-efficient learning. The code is available at https://github.com/Wangt-CN/EqInv.

show abstract

SelfReg: Self-supervised Contrastive Regularization for Domain Generalization

Cited by 128 publications

References 18 publications

Edge Augmentation for Large-Scale Sketch Recognition without Sketches

Edge Augmentation for Large-Scale Sketch Recognition without Sketches

Multi-Source Unsupervised Domain Adaptation via Pseudo Target Domain

Equivariance and Invariance Inductive Bias for Learning from Insufficient Data

Contact Info

Product

Resources

About