Robust Image Classification Using A Low-Pass Activation Function and DCT Augmentation
This work addresses robustness issues in image classification for corrupted datasets, offering a novel method that improves SOTA performance, though it is incremental in combining existing ideas.
The paper tackles the problem of CNNs' performance disparity on clean versus corrupted images by introducing a low-pass activation function (LP-ReLU) and DCT augmentation to handle high- and low-frequency corruptions, achieving improvements of 5% on CIFAR-10-C and 7.3% on Tiny ImageNet-C compared to SOTA.
Convolutional Neural Network's (CNN's) performance disparity on clean and corrupted datasets has recently come under scrutiny. In this work, we analyse common corruptions in the frequency domain, i.e., High Frequency corruptions (HFc, e.g., noise) and Low Frequency corruptions (LFc, e.g., blur). Although a simple solution to HFc is low-pass filtering, ReLU -- a widely used Activation Function (AF), does not have any filtering mechanism. In this work, we instill low-pass filtering into the AF (LP-ReLU) to improve robustness against HFc. To deal with LFc, we complement LP-ReLU with Discrete Cosine Transform based augmentation. LP-ReLU, coupled with DCT augmentation, enables a deep network to tackle the entire spectrum of corruption. We use CIFAR-10-C and Tiny ImageNet-C for evaluation and demonstrate improvements of 5% and 7.3% in accuracy respectively, compared to the State-Of-The-Art (SOTA). We further evaluate our method's stability on a variety of perturbations in CIFAR-10-P and Tiny ImageNet-P, achieving new SOTA in these experiments as well. To further strengthen our understanding regarding CNN's lack of robustness, a decision space visualisation process is proposed and presented in this work.