Observations on K-image Expansion of Image-Mixing Augmentation for Classification

TL;DR

This paper introduces a novel K-image mixing augmentation based on a probabilistic stick-breaking process, demonstrating improved robustness, generalization, and efficiency in image classification tasks over traditional two-image methods.

Contribution

It proposes a new K-image mixing augmentation method using Dirichlet prior, enhancing classifier robustness, loss landscape, and adversarial resistance, with applications in uncertainty measurement and architecture search.

Findings

Outperforms traditional two-image mixing methods in accuracy and robustness.

Provides a probabilistic model for sample-wise uncertainty estimation.

Reduces network architecture search time by 7-fold.

Abstract

Image-mixing augmentations (e.g., Mixup and CutMix), which typically involve mixing two images, have become the de-facto training techniques for image classification. Despite their huge success in image classification, the number of images to be mixed has not been elucidated in the literature: only the naive K-image expansion has been shown to lead to performance degradation. This study derives a new K-image mixing augmentation based on the stick-breaking process under Dirichlet prior distribution. We demonstrate the superiority of our K-image expansion augmentation over conventional two-image mixing augmentation methods through extensive experiments and analyses: (1) more robust and generalized classifiers; (2) a more desirable loss landscape shape; (3) better adversarial robustness. Moreover, we show that our probabilistic model can measure the sample-wise uncertainty and boost the efficiency for network architecture search by achieving a 7-fold reduction in the search time. Code will be available at https://github.com/yjyoo3312/DCutMix-PyTorch.git.