Making Convolutional Networks Shift-Invariant Again

TL;DR

This paper demonstrates that integrating anti-aliasing filters into convolutional networks restores shift-invariance, improves accuracy, and enhances robustness across various architectures on ImageNet.

Contribution

It shows how to correctly incorporate anti-aliasing into deep networks, leading to better performance and robustness, addressing a longstanding issue in CNN shift-invariance.

Findings

Increased accuracy on ImageNet classification.

Enhanced robustness to input corruptions.

Compatibility of anti-aliasing with existing architectures.

Abstract

Modern convolutional networks are not shift-invariant, as small input shifts or translations can cause drastic changes in the output. Commonly used downsampling methods, such as max-pooling, strided-convolution, and average-pooling, ignore the sampling theorem. The well-known signal processing fix is anti-aliasing by low-pass filtering before downsampling. However, simply inserting this module into deep networks degrades performance; as a result, it is seldomly used today. We show that when integrated correctly, it is compatible with existing architectural components, such as max-pooling and strided-convolution. We observe \textit{increased accuracy} in ImageNet classification, across several commonly-used architectures, such as ResNet, DenseNet, and MobileNet, indicating effective regularization. Furthermore, we observe \textit{better generalization}, in terms of stability and robustness to input corruptions. Our results demonstrate that this classical signal processing technique has been undeservingly overlooked in modern deep networks. Code and anti-aliased versions of popular networks are available at https://richzhang.github.io/antialiased-cnns/ .