Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Uncertainty-guided Contrastive Learning for Single Source Domain Generalisation

Anastasios Arsenos, Dimitrios Kollias, Evangelos Petrongonas, Christos Skliros, Stefanos Kollias

TL;DR

A novel model referred to as Contrastive Uncertainty Domain Generalisation Network (CUDGNet) is introduced, which surpasses the state-of-the-art single-DG methods by up to 7.08%.

Abstract

In the context of single domain generalisation, the objective is for models that have been exclusively trained on data from a single domain to demonstrate strong performance when confronted with various unfamiliar domains. In this paper, we introduce a novel model referred to as Contrastive Uncertainty Domain Generalisation Network (CUDGNet). The key idea is to augment the source capacity in both input and label spaces through the fictitious domain generator and jointly learn the domain invariant representation of each class through contrastive learning. Extensive experiments on two Single Source Domain Generalisation (SSDG) datasets demonstrate the effectiveness of our approach, which surpasses the state-of-the-art single-DG methods by up to $7.08\%$. Our method also provides efficient uncertainty estimation at inference time from a single forward pass through the generator subnetwork.

Uncertainty-guided Contrastive Learning for Single Source Domain Generalisation

TL;DR

A novel model referred to as Contrastive Uncertainty Domain Generalisation Network (CUDGNet) is introduced, which surpasses the state-of-the-art single-DG methods by up to 7.08%.

Abstract

In the context of single domain generalisation, the objective is for models that have been exclusively trained on data from a single domain to demonstrate strong performance when confronted with various unfamiliar domains. In this paper, we introduce a novel model referred to as Contrastive Uncertainty Domain Generalisation Network (CUDGNet). The key idea is to augment the source capacity in both input and label spaces through the fictitious domain generator and jointly learn the domain invariant representation of each class through contrastive learning. Extensive experiments on two Single Source Domain Generalisation (SSDG) datasets demonstrate the effectiveness of our approach, which surpasses the state-of-the-art single-DG methods by up to . Our method also provides efficient uncertainty estimation at inference time from a single forward pass through the generator subnetwork.
Paper Structure (4 sections, 9 equations, 2 figures, 3 tables)

This paper contains 4 sections, 9 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Materials and Method
  3. Experimental Results
  4. CONCLUSION

Figures (2)

  • Figure 1: The overall framework of the proposed CUDGNet. The Task Model M and the domain augmentation Generator G are jointly trained while the transformation component TC and style mixing (EFDMix) further enrich the augmentation capacity. The contrastive loss will guide semantically similar samples from different domains to be closer in the embedding space.
  • Figure 2: Estimation of Uncertainty CIFAR-10-C. Our domain uncertainty prediction aligns with Bayesian uncertainty, while our approach is significantly faster.