Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations
arXiv:2303.08085 · doi:10.1109/CVPR52729.2023.01567
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.
arXiv commentThe paper was accepted to CVPR 2023. Our code is available at https://github.com/hmichaeli/alias_free_convnets/
References in corpus (13)
- Geometric deep learning: going beyond Euclidean data
- Alias-Free Generative Adversarial Networks
- Joint 2D-3D-Semantic Data for Indoor Scene Understanding
- Making Convolutional Networks Shift-Invariant Again
- Exploring the Landscape of Spatial Robustness
- Attentive Group Equivariant Convolutional Networks
- An Effective Anti-Aliasing Approach for Residual Networks
- Delving Deeper into Anti-aliasing in ConvNets
- Shift Invariance Can Reduce Adversarial Robustness
- Anti-aliasing Deep Image Classifiers using Novel Depth Adaptive Blurring and Activation Function
- Equivariance Discovery by Learned Parameter-Sharing
- Generalization to translation shifts: a study in architectures and augmentations
- Learnable Polyphase Sampling for Shift Invariant and Equivariant Convolutional Networks