SimpleDG: Simple Domain Generalization Baseline without Bells and Whistles

TL;DR

This paper introduces SimpleDG, a straightforward domain generalization baseline that outperforms many complex methods, demonstrating that simple ERM-based approaches can be highly effective for domain generalization tasks.

Contribution

The paper proposes SimpleDG, a simple yet effective domain generalization method that achieves competitive results, emphasizing the strength of basic ERM strategies.

Findings

SimpleDG achieved second place in NICO CHALLENGE 2022

ERM is a strong baseline compared to complex methods

Simple designs can significantly boost generalization performance

Abstract

We present a simple domain generalization baseline, which wins second place in both the common context generalization track and the hybrid context generalization track respectively in NICO CHALLENGE 2022. We verify the founding in recent literature, domainbed, that ERM is a strong baseline compared to recent state-of-the-art domain generalization methods and propose SimpleDG which includes several simple yet effective designs that further boost generalization performance. Code is available at https://github.com/megvii-research/SimpleDG