Do Vision Transformers See Like Convolutional Neural Networks?

Abstract

Convolutional neural networks (CNNs) have so far been the de-facto model for visual data. Recent work has shown that (Vision) Transformer models (ViT) can achieve comparable or even superior performance on image classification tasks. This raises a central question: how are Vision Transformers solving these tasks? Are they acting like convolutional networks, or learning entirely different visual representations? Analyzing the internal representation structure of ViTs and CNNs on image classification benchmarks, we find striking differences between the two architectures, such as ViT having more uniform representations across all layers. We explore how these differences arise, finding crucial roles played by self-attention, which enables early aggregation of global information, and ViT residual connections, which strongly propagate features from lower to higher layers. We study the ramifications for spatial localization, demonstrating ViTs successfully preserve input spatial information, with noticeable effects from different classification methods. Finally, we study the effect of (pretraining) dataset scale on intermediate features and transfer learning, and conclude with a discussion on connections to new architectures such as the MLP-Mixer.

References in corpus (7)

• MLP-Mixer: An all-MLP Architecture for Vision
• Convolutional Neural Networks as a Model of the Visual System: Past, Present, and Future
• Similarity of Neural Network Representations Revisited
• Intriguing Properties of Vision Transformers
• Stand-Alone Self-Attention in Vision Models
• Are Pre-trained Convolutions Better than Pre-trained Transformers?
• Supervised Transfer Learning at Scale for Medical Imaging

Cited by in corpus (11)

• Universal Paralinguistic Speech Representations Using Self-Supervised Conformers
• MetaFormer Is Actually What You Need for Vision
• Comparison Analysis of Traditional Machine Learning and Deep Learning Techniques for Data and Image Classification
• Similarity of Neural Network Models: A Survey of Functional and Representational Measures
• Pruning Self-attentions into Convolutional Layers in Single Path
• Multi-Scale High-Resolution Vision Transformer for Semantic Segmentation
• TransMix: Attend to Mix for Vision Transformers
• Self-supervised pseudo-colorizing of masked cells
• Do Self-Supervised and Supervised Methods Learn Similar Visual Representations?
• Hybrid BYOL-ViT: Efficient approach to deal with small datasets
• Physics-Augmented Learning: A New Paradigm Beyond Physics-Informed Learning