CoAtNet: Marrying Convolution and Attention for All Data Sizes

TL;DR

CoAtNet introduces a hybrid model combining convolution and attention mechanisms, achieving state-of-the-art image classification accuracy across various data scales by effectively unifying the strengths of both architectures.

Contribution

This work presents a novel hybrid architecture, CoAtNet, that unifies depthwise convolution and self-attention, improving generalization, capacity, and efficiency in vision models.

Findings

Achieves 86.0% ImageNet top-1 accuracy without extra data.

Matches ViT-huge performance with significantly less pre-training data.

Sets new state-of-the-art with 90.88% accuracy on ImageNet using JFT-3B.

Abstract

Transformers have attracted increasing interests in computer vision, but they still fall behind state-of-the-art convolutional networks. In this work, we show that while Transformers tend to have larger model capacity, their generalization can be worse than convolutional networks due to the lack of the right inductive bias. To effectively combine the strengths from both architectures, we present CoAtNets(pronounced "coat" nets), a family of hybrid models built from two key insights: (1) depthwise Convolution and self-Attention can be naturally unified via simple relative attention; (2) vertically stacking convolution layers and attention layers in a principled way is surprisingly effective in improving generalization, capacity and efficiency. Experiments show that our CoAtNets achieve state-of-the-art performance under different resource constraints across various datasets: Without extra data, CoAtNet achieves 86.0% ImageNet top-1 accuracy; When pre-trained with 13M images from ImageNet-21K, our CoAtNet achieves 88.56% top-1 accuracy, matching ViT-huge pre-trained with 300M images from JFT-300M while using 23x less data; Notably, when we further scale up CoAtNet with JFT-3B, it achieves 90.88% top-1 accuracy on ImageNet, establishing a new state-of-the-art result.