How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers

TL;DR

This paper systematically studies how data, augmentation, and regularization affect Vision Transformers (ViT), revealing that combining increased compute and AugReg can match or surpass larger datasets' performance.

Contribution

It provides a comprehensive empirical analysis of training strategies for ViT, demonstrating effective methods to improve performance without massive datasets.

Findings

Increased compute and AugReg can match larger datasets' performance.

ViT models trained on ImageNet-21k can outperform those trained on JFT-300M.

Systematic analysis of data, augmentation, and regularization effects.

Abstract

Vision Transformers (ViT) have been shown to attain highly competitive performance for a wide range of vision applications, such as image classification, object detection and semantic image segmentation. In comparison to convolutional neural networks, the Vision Transformer's weaker inductive bias is generally found to cause an increased reliance on model regularization or data augmentation ("AugReg" for short) when training on smaller training datasets. We conduct a systematic empirical study in order to better understand the interplay between the amount of training data, AugReg, model size and compute budget. As one result of this study we find that the combination of increased compute and AugReg can yield models with the same performance as models trained on an order of magnitude more training data: we train ViT models of various sizes on the public ImageNet-21k dataset which either match or outperform their counterparts trained on the larger, but not publicly available JFT-300M dataset.