Exploring Self-attention for Image Recognition

2020/04/28 by Hengshuang Zhao, Jiaya Jia, Zhao, Hengshuang +3 · 5 citations
Computer Science · #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences

paper · pdf · doi:10.48550/arxiv.2004.13621

openalex publication_date 2020/04/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Recent work has shown that self-attention can serve as a basic building block for image recognition models. We explore variations of self-attention and assess their effectiveness for image recognition. We consider two forms of self-attention. One is pairwise self-attention, which generalizes standard dot-product attention and is fundamentally a set operator. The other is patchwise self-attention, which is strictly more powerful than convolution. Our pairwise self-attention networks match or outperform their convolutional counterparts, and the patchwise models substantially outperform the convolutional baselines. We also conduct experiments that probe the robustness of learned representations and conclude that self-attention networks may have significant benefits in terms of robustness and generalization.

Citations

• ImageNet Large Scale Visual Recognition Challenge
• Backpropagation Applied to Handwritten Zip Code Recognition
• ImageNet classification with deep convolutional neural networks
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal\n Networks
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
• Large Scale GAN Training for High Fidelity Natural Image Synthesis
• Pay Less Attention with Lightweight and Dynamic Convolutions
• An Intriguing Failing of Convolutional Neural Networks and the CoordConv Solution

Cited by

• SDGraph: Multi-Level Sketch Representation Learning by Sparse-Dense Graph Architecture
• Bidirectional Feature-aligned Motion Transformation for Efficient Dynamic Point Cloud Compression
• Hierarchical MLANet: Multi-level Attention for 3D Face Reconstruction From Single Images
• CoSwin: Convolution Enhanced Hierarchical Shifted Window Attention For Small-Scale Vision
• On the Integration of Self-Attention and Convolution

Related

• Attention Augmented Convolutional Networks
• Stand-Alone Self-Attention in Vision Models
• Scaling Local Self-Attention for Parameter Efficient Visual Backbones
• Locally Enhanced Self-Attention: Combining Self-Attention and Convolution as Local and Context Terms
• On the Relationship between Self-Attention and Convolutional Layers