DomainDrop: Suppressing Domain-Sensitive Channels for Domain Generalization

TL;DR

DomainDrop is a novel framework that improves domain generalization by identifying and suppressing unstable channels in neural networks, leading to better performance on unseen domains.

Contribution

We introduce a new approach that enhances channel robustness by dropping unstable channels during training, backed by theoretical generalization bounds and state-of-the-art experimental results.

Findings

Achieves state-of-the-art performance on multiple benchmarks.

Theoretically proven to lower the generalization bound.

Effectively suppresses domain-specific channels.

Abstract

Deep Neural Networks have exhibited considerable success in various visual tasks. However, when applied to unseen test datasets, state-of-the-art models often suffer performance degradation due to domain shifts. In this paper, we introduce a novel approach for domain generalization from a novel perspective of enhancing the robustness of channels in feature maps to domain shifts. We observe that models trained on source domains contain a substantial number of channels that exhibit unstable activations across different domains, which are inclined to capture domain-specific features and behave abnormally when exposed to unseen target domains. To address the issue, we propose a DomainDrop framework to continuously enhance the channel robustness to domain shifts, where a domain discriminator is used to identify and drop unstable channels in feature maps of each network layer during forward propagation. We theoretically prove that our framework could effectively lower the generalization bound. Extensive experiments on several benchmarks indicate that our framework achieves state-of-the-art performance compared to other competing methods. Our code is available at https://github.com/lingeringlight/DomainDrop.