Feature-level augmentation to improve robustness of deep neural networks to affine transformations
This work addresses robustness issues in deep neural networks for image classification, offering an incremental improvement over existing stabilization techniques.
The paper tackled the problem of convolutional neural networks' poor generalization to small image transformations like rotations and translations by proposing feature-level data augmentation, which improved robustness across three image classification benchmarks and two architectures, achieving the best trade-off between accuracy and mean flip rate compared to state-of-the-art methods.
Recent studies revealed that convolutional neural networks do not generalize well to small image transformations, e.g. rotations by a few degrees or translations of a few pixels. To improve the robustness to such transformations, we propose to introduce data augmentation at intermediate layers of the neural architecture, in addition to the common data augmentation applied on the input images. By introducing small perturbations to activation maps (features) at various levels, we develop the capacity of the neural network to cope with such transformations. We conduct experiments on three image classification benchmarks (Tiny ImageNet, Caltech-256 and Food-101), considering two different convolutional architectures (ResNet-18 and DenseNet-121). When compared with two state-of-the-art stabilization methods, the empirical results show that our approach consistently attains the best trade-off between accuracy and mean flip rate.