Training data-efficient image transformers & distillation through attention

TL;DR

This paper introduces a data-efficient vision transformer trained solely on ImageNet within three days, utilizing a novel attention-based distillation method that achieves competitive accuracy without external data.

Contribution

The work presents a convolution-free transformer trained on ImageNet in limited time and introduces a teacher-student distillation strategy using attention tokens for improved learning.

Findings

Achieved 83.1% top-1 accuracy on ImageNet with 86M parameters.

Introduced a token-based distillation method for transformers.

Reported competitive results with convnets on ImageNet and transfer tasks.

Abstract

Recently, neural networks purely based on attention were shown to address image understanding tasks such as image classification. However, these visual transformers are pre-trained with hundreds of millions of images using an expensive infrastructure, thereby limiting their adoption. In this work, we produce a competitive convolution-free transformer by training on Imagenet only. We train them on a single computer in less than 3 days. Our reference vision transformer (86M parameters) achieves top-1 accuracy of 83.1% (single-crop evaluation) on ImageNet with no external data. More importantly, we introduce a teacher-student strategy specific to transformers. It relies on a distillation token ensuring that the student learns from the teacher through attention. We show the interest of this token-based distillation, especially when using a convnet as a teacher. This leads us to report results competitive with convnets for both Imagenet (where we obtain up to 85.2% accuracy) and when transferring to other tasks. We share our code and models.