Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations

TL;DR

This paper introduces an extended anti-aliasing method for CNNs that effectively addresses aliasing in both downsampling and non-linear layers, achieving true shift invariance including fractional translations, and improving robustness to adversarial shifts.

Contribution

The paper presents a novel anti-aliasing approach that ensures CNNs are invariant to both integer and fractional translations, surpassing previous methods in shift robustness.

Findings

Model is invariant to integer and fractional translations.

Outperforms existing shift-invariant methods in robustness.

Effectively prevents aliasing in non-linear layers.

Abstract

Although CNNs are believed to be invariant to translations, recent works have shown this is not the case, due to aliasing effects that stem from downsampling layers. The existing architectural solutions to prevent aliasing are partial since they do not solve these effects, that originate in non-linearities. We propose an extended anti-aliasing method that tackles both downsampling and non-linear layers, thus creating truly alias-free, shift-invariant CNNs. We show that the presented model is invariant to integer as well as fractional (i.e., sub-pixel) translations, thus outperforming other shift-invariant methods in terms of robustness to adversarial translations.