Domain Generalisation with Bidirectional Encoder Representations from Vision Transformers

TL;DR

This paper explores the use of vision transformers, especially BEIT, for domain generalisation in out-of-distribution vision tasks, achieving significant accuracy improvements across multiple benchmarks.

Contribution

It demonstrates the effectiveness of BEIT architecture for domain generalisation and provides comprehensive evaluation on several OOD benchmarks.

Findings

BEIT outperforms other vision transformers in OOD settings

Significant accuracy improvements on PACS, Home-Office, and DomainNet

Implementation reduces gaps between in-distribution and OOD data performance

Abstract

Domain generalisation involves pooling knowledge from source domain(s) into a single model that can generalise to unseen target domain(s). Recent research in domain generalisation has faced challenges when using deep learning models as they interact with data distributions which differ from those they are trained on. Here we perform domain generalisation on out-of-distribution (OOD) vision benchmarks using vision transformers. Initially we examine four vision transformer architectures namely ViT, LeViT, DeiT, and BEIT on out-of-distribution data. As the bidirectional encoder representation from image transformers (BEIT) architecture performs best, we use it in further experiments on three benchmarks PACS, Home-Office and DomainNet. Our results show significant improvements in validation and test accuracy and our implementation significantly overcomes gaps between within-distribution and OOD data.