Impact of Aliasing on Generalization in Deep Convolutional Networks
This addresses generalization issues in image classification for researchers and practitioners, offering a simple architectural fix that is incremental but effective.
The paper tackles the problem of aliasing in deep convolutional networks, showing that it harms generalization, and demonstrates that inserting non-trainable low-pass filters improves performance, achieving state-of-the-art results on ImageNet-C and Meta-Dataset without extra parameters.
We investigate the impact of aliasing on generalization in Deep Convolutional Networks and show that data augmentation schemes alone are unable to prevent it due to structural limitations in widely used architectures. Drawing insights from frequency analysis theory, we take a closer look at ResNet and EfficientNet architectures and review the trade-off between aliasing and information loss in each of their major components. We show how to mitigate aliasing by inserting non-trainable low-pass filters at key locations, particularly where networks lack the capacity to learn them. These simple architectural changes lead to substantial improvements in generalization on i.i.d. and even more on out-of-distribution conditions, such as image classification under natural corruptions on ImageNet-C [11] and few-shot learning on Meta-Dataset [26]. State-of-the art results are achieved on both datasets without introducing additional trainable parameters and using the default hyper-parameters of open source codebases.