ResMLP: Feedforward networks for image classification with data-efficient training

TL;DR

ResMLP is a simple, fully feedforward residual architecture for image classification that achieves competitive accuracy with efficient training, and extends to self-supervised learning and machine translation.

Contribution

It introduces ResMLP, a novel architecture based solely on multi-layer perceptrons for image classification, demonstrating strong performance and versatility.

Findings

Achieves good accuracy/complexity trade-offs on ImageNet

Effective in self-supervised training setups

Adapts well to machine translation tasks

Abstract

We present ResMLP, an architecture built entirely upon multi-layer perceptrons for image classification. It is a simple residual network that alternates (i) a linear layer in which image patches interact, independently and identically across channels, and (ii) a two-layer feed-forward network in which channels interact independently per patch. When trained with a modern training strategy using heavy data-augmentation and optionally distillation, it attains surprisingly good accuracy/complexity trade-offs on ImageNet. We also train ResMLP models in a self-supervised setup, to further remove priors from employing a labelled dataset. Finally, by adapting our model to machine translation we achieve surprisingly good results. We share pre-trained models and our code based on the Timm library.