Improving Diversity with Adversarially Learned Transformations for Domain Generalization

TL;DR

This paper introduces a novel adversarially learned transformation framework that generates diverse and challenging image transformations, significantly improving domain generalization performance over existing methods.

Contribution

It proposes a new ALT framework that models plausible, hard transformations adversarially, outperforming existing diversity augmentation techniques in domain generalization tasks.

Findings

Outperforms existing techniques on domain generalization benchmarks.

Achieves state-of-the-art performance when combined with existing diversity modules.

Generates transformations that are both diverse and challenging for classifiers.

Abstract

To be successful in single source domain generalization, maximizing diversity of synthesized domains has emerged as one of the most effective strategies. Many of the recent successes have come from methods that pre-specify the types of diversity that a model is exposed to during training, so that it can ultimately generalize well to new domains. However, na\"ive diversity based augmentations do not work effectively for domain generalization either because they cannot model large domain shift, or because the span of transforms that are pre-specified do not cover the types of shift commonly occurring in domain generalization. To address this issue, we present a novel framework that uses adversarially learned transformations (ALT) using a neural network to model plausible, yet hard image transformations that fool the classifier. This network is randomly initialized for each batch and trained for a fixed number of steps to maximize classification error. Further, we enforce consistency between the classifier's predictions on the clean and transformed images. With extensive empirical analysis, we find that this new form of adversarial transformations achieve both objectives of diversity and hardness simultaneously, outperforming all existing techniques on competitive benchmarks for single source domain generalization. We also show that ALT can naturally work with existing diversity modules to produce highly distinct, and large transformations of the source domain leading to state-of-the-art performance.