PatchShuffle Regularization

TL;DR

PatchShuffle is a novel regularization technique for CNNs that shuffles pixels within local patches to enhance generalization, robustness, and reduce overfitting, especially effective with limited data.

Contribution

Introduces PatchShuffle, a simple and effective regularization method that can be integrated into any CNN training process to improve performance and robustness.

Findings

Improves CNN generalization on multiple datasets.

Enhances robustness to noise and local image changes.

Effective especially with scarce training data.

Abstract

This paper focuses on regularizing the training of the convolutional neural network (CNN). We propose a new regularization approach named ``PatchShuffle`` that can be adopted in any classification-oriented CNN models. It is easy to implement: in each mini-batch, images or feature maps are randomly chosen to undergo a transformation such that pixels within each local patch are shuffled. Through generating images and feature maps with interior orderless patches, PatchShuffle creates rich local variations, reduces the risk of network overfitting, and can be viewed as a beneficial supplement to various kinds of training regularization techniques, such as weight decay, model ensemble and dropout. Experiments on four representative classification datasets show that PatchShuffle improves the generalization ability of CNN especially when the data is scarce. Moreover, we empirically illustrate that CNN models trained with PatchShuffle are more robust to noise and local changes in an image.