Understanding Robustness of Transformers for Image Classification

by   Srinadh Bhojanapalli, et al.

Deep Convolutional Neural Networks (CNNs) have long been the architecture of choice for computer vision tasks. Recently, Transformer-based architectures like Vision Transformer (ViT) have matched or even surpassed ResNets for image classification. However, details of the Transformer architecture – such as the use of non-overlapping patches – lead one to wonder whether these networks are as robust. In this paper, we perform an extensive study of a variety of different measures of robustness of ViT models and compare the findings to ResNet baselines. We investigate robustness to input perturbations as well as robustness to model perturbations. We find that when pre-trained with a sufficient amount of data, ViT models are at least as robust as the ResNet counterparts on a broad range of perturbations. We also find that Transformers are robust to the removal of almost any single layer, and that while activations from later layers are highly correlated with each other, they nevertheless play an important role in classification.


page 1

page 6

page 7

page 20

page 21

page 22


MLP-Mixer: An all-MLP Architecture for Vision

Convolutional Neural Networks (CNNs) are the go-to model for computer vi...

Are Vision Transformers Robust to Spurious Correlations?

Deep neural networks may be susceptible to learning spurious correlation...

Equivariant Transformer Networks

How can prior knowledge on the transformation invariances of a domain be...

Vision Checklist: Towards Testable Error Analysis of Image Models to Help System Designers Interrogate Model Capabilities

Using large pre-trained models for image recognition tasks is becoming i...

Exploring Corruption Robustness: Inductive Biases in Vision Transformers and MLP-Mixers

Recently, vision transformers and MLP-based models have been developed i...

Visual Transformer for Soil Classification

Our food security is built on the foundation of soil. Farmers would be u...

On the interplay of adversarial robustness and architecture components: patches, convolution and attention

In recent years novel architecture components for image classification h...