Improving Deep Learning using Generic Data Augmentation

TL;DR

This paper benchmarks various generic data augmentation techniques for CNNs, demonstrating that cropping significantly improves task performance, thereby guiding researchers in selecting effective augmentation methods.

Contribution

It provides a comparative analysis of popular data augmentation schemes, highlighting the effectiveness of cropping in geometric augmentation for CNN training.

Findings

Cropping in geometric augmentation improves CNN accuracy.

Different augmentation schemes have varying impacts on performance.

Benchmark results guide optimal augmentation choices for CNNs.

Abstract

Deep artificial neural networks require a large corpus of training data in order to effectively learn, where collection of such training data is often expensive and laborious. Data augmentation overcomes this issue by artificially inflating the training set with label preserving transformations. Recently there has been extensive use of generic data augmentation to improve Convolutional Neural Network (CNN) task performance. This study benchmarks various popular data augmentation schemes to allow researchers to make informed decisions as to which training methods are most appropriate for their data sets. Various geometric and photometric schemes are evaluated on a coarse-grained data set using a relatively simple CNN. Experimental results, run using 4-fold cross-validation and reported in terms of Top-1 and Top-5 accuracy, indicate that cropping in geometric augmentation significantly increases CNN task performance.