MetaFormer Is Actually What You Need for Vision

Abstract

Transformers have shown great potential in computer vision tasks. A common belief is their attention-based token mixer module contributes most to their competence. However, recent works show the attention-based module in Transformers can be replaced by spatial MLPs and the resulted models still perform quite well. Based on this observation, we hypothesize that the general architecture of the Transformers, instead of the specific token mixer module, is more essential to the model's performance. To verify this, we deliberately replace the attention module in Transformers with an embarrassingly simple spatial pooling operator to conduct only basic token mixing. Surprisingly, we observe that the derived model, termed as PoolFormer, achieves competitive performance on multiple computer vision tasks. For example, on ImageNet-1K, PoolFormer achieves 82.1% top-1 accuracy, surpassing well-tuned Vision Transformer/MLP-like baselines DeiT-B/ResMLP-B24 by 0.3%/1.1% accuracy with 35%/52% fewer parameters and 50%/62% fewer MACs. The effectiveness of PoolFormer verifies our hypothesis and urges us to initiate the concept of "MetaFormer", a general architecture abstracted from Transformers without specifying the token mixer. Based on the extensive experiments, we argue that MetaFormer is the key player in achieving superior results for recent Transformer and MLP-like models on vision tasks. This work calls for more future research dedicated to improving MetaFormer instead of focusing on the token mixer modules. Additionally, our proposed PoolFormer could serve as a starting baseline for future MetaFormer architecture design. Code is available at https://github.com/sail-sg/poolformer.

CVPR 2022 (Oral). Code: https://github.com/sail-sg/poolformer

References in corpus (16)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• MLP-Mixer: An all-MLP Architecture for Vision
• Transformer in Transformer
• MMDetection: Open MMLab Detection Toolbox and Benchmark
• Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
• CvT: Introducing Convolutions to Vision Transformers
• Do Vision Transformers See Like Convolutional Neural Networks?
• CycleMLP: A MLP-like Architecture for Dense Prediction
• Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth
• CMT: Convolutional Neural Networks Meet Vision Transformers
• Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions
• ResNet strikes back: An improved training procedure in timm
• Pay Attention to MLPs
• Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet
• Vision Permutator: A Permutable MLP-Like Architecture for Visual Recognition
• Global Filter Networks for Image Classification