Uncertainty-guided Model Generalization to Unseen Domains

Qiao, Fengchun; Peng, Xi

doi:10.1109/cvpr46437.2021.00672

Cited by 56 publications

(15 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[22], [23], [31], [34], [72], [73], [95], [96], [164], [165], [166] Data Augmentation ( § 3.3) -Hand-Engineered Transformations [32], [33], [65], [86], [167], [168], [169], [170] -Gradient-Based Augmentation [24], [40], [45], [171] -Model-Based Augmentation [19], [25], [26], [137], [172], [173], [174], [175], [176] -Feature-Based Augmentation [27], [86], [177], [178] Ensemble Learning ( § 3.4) -Exemplar-SVMs [47], [50], [179] -Domain-Specific Neural Networks [104], [138], [180], [181], [182] -Domain-Specific Batch Normalization [183], [184], [185], [186] -Weight Averaging [187] Network Architecture Design ( § 3.5) -Exploiting Instance Normalization [39], [57], [58], [59], [188] -Problem-Specific Modules ...…”

Section: Action Recognitionmentioning

confidence: 99%

“…Combining feature-level augmentation with image-level augmentation has been studied in [86] for zeroshot DG, where Mixup [214] is applied to mixing instances from different domains at both the image and feature space. In [178], a neural network is learned via a worstcase loss [171] to generate feature-level perturbations, and a learnable Mixup is proposed to mix the perturbed and original instances in both feature and label space.…”

Section: Feature-based Augmentationmentioning

confidence: 99%

See 1 more Smart Citation

Domain Generalization: A Survey

Zhou¹,

Liu²,

Qiao³

et al. 2021

Preprint

View full text Add to dashboard Cite

Generalization to out-of-distribution (OOD) data is a capability natural to humans yet challenging for machines to reproduce. This is because most statistical learning algorithms strongly rely on the i.i.d. assumption on source/target data, while in practice domain shift between source and target is common. Domain generalization (DG) aims to achieve OOD generalization by using only source data for model learning. Since first introduced in 2011, research in DG has made great progresses. In particular, intensive research in this topic has led to a broad spectrum of methodologies, e.g., those based on domain alignment, meta-learning, data augmentation, or ensemble learning, just to name a few; and has covered various applications such as object recognition, segmentation, action recognition, and person re-identification. In this paper, for the first time, a comprehensive literature review is provided to summarize the developments in DG in the past decade. Specifically, we first cover the background by formally defining DG and relating it to other research fields like domain adaptation and transfer learning. Second, we conduct a thorough review into existing methods and present a categorization based on their methodologies and motivations. Finally, we conclude this survey with insights and discussions on future research directions.

show abstract

Section: Action Recognitionmentioning

confidence: 99%

Section: Feature-based Augmentationmentioning

confidence: 99%

Domain Generalization: A Survey

Zhou¹,

Liu²,

Qiao³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…disentangle domain-specific and domain-invariant information [2,3,16,28], or align feature distributions of different domains while preserving their semantics [1,17,22,43]. Typical approaches for single DG simulate the presence of new domains with data augmentation either through adversarial strategies [9,18,25,26,33,41] or direct input transformation [32]. For instance, [33] performs adversarial data augmentation under a worst case formulation, assuming samples of unseen domains to be close to the training distribution.…”

Section: Related Workmentioning

confidence: 99%

Attention Consistency on Visual Corruptions for Single-Source Domain Generalization

Cugu¹,

Mancini²,

Chen³

et al. 2022

Preprint

View full text Add to dashboard Cite

Generalizing visual recognition models trained on a single distribution to unseen input distributions (i.e. domains) requires making them robust to superfluous correlations in the training set. In this work, we achieve this goal by altering the training images to simulate new domains and imposing consistent visual attention across the different views of the same sample. We discover that the first objective can be simply and effectively met through visual corruptions. Specifically, we alter the content of the training images using the nineteen corruptions of the ImageNet-C benchmark and three additional transformations based on Fourier transform. Since these corruptions preserve object locations, we propose an attention consistency loss to ensure that class activation maps across original and corrupted versions of the same training sample are aligned. We name our model Attention Consistency on Visual Corruptions (ACVC). We show that ACVC consistently achieves the state of the art on three single-source domain generalization benchmarks, PACS, COCO, and the large-scale DomainNet 1 .

show abstract

“…For example, DDAIG developed by Zhou et al [48] learns a neural network to transform images' appearance such that a domain classifier cannot identify their source domain labels. The last group transitions from pixel-to feature-level augmentation by, e.g., mixing feature statistics [49] or learning feature perturbation networks [28]. Most existing DG methods cannot handle unlabeled data except those based on self-supervised learning [6,39].…”

Section: Ablation Study and Analysismentioning

confidence: 99%

Semi-Supervised Domain Generalization with Stochastic StyleMatch

Zhou¹,

Loy²,

Liu

2021

Preprint

View full text Add to dashboard Cite

Most existing research on domain generalization assumes source data gathered from multiple domains are fully annotated. However, in real-world applications, we might have only a few labels available from each source domain due to high annotation cost, along with abundant unlabeled data that are much easier to obtain. In this work, we investigate semi-supervised domain generalization (SSDG), a more realistic and practical setting. Our proposed approach, StyleMatch, is inspired by FixMatch, a state-of-the-art semi-supervised learning method based on pseudolabeling, with several new ingredients tailored to solve SSDG. Specifically, 1) to mitigate overfitting in the scarce labeled source data while improving robustness against noisy pseudo labels, we introduce stochastic modeling to the classifier's weights, seen as class prototypes, with Gaussian distributions. 2) To enhance generalization under domain shift, we upgrade FixMatch's two-view consistency learning paradigm based on weak and strong augmentations to a multi-view version with style augmentation as the third complementary view. To provide a comprehensive study and evaluation, we establish two SSDG benchmarks, which cover a wide range of strong baseline methods developed in relevant areas including domain generalization and semi-supervised learning. Extensive experiments demonstrate that StyleMatch achieves the best out-of-distribution generalization performance in the low-data regime. We hope our approach and benchmarks can pave the way for future research on data-efficient and generalizable learning systems.Preprint. Under review.

show abstract

Uncertainty-guided Model Generalization to Unseen Domains

Cited by 56 publications

References 22 publications

Domain Generalization: A Survey

Domain Generalization: A Survey

Attention Consistency on Visual Corruptions for Single-Source Domain Generalization

Semi-Supervised Domain Generalization with Stochastic StyleMatch

Contact Info

Product

Resources

About