Scale Steerable Filters for Locally Scale-Invariant Convolutional Neural Networks
This work addresses the challenge of scale invariance in CNNs for computer vision tasks, offering a domain-specific improvement over existing methods.
The paper tackled the problem of achieving local scale invariance in convolutional neural networks by proposing scale-steerable filters based on log-radial harmonics, resulting in significant performance improvements on MNIST-Scale and FMNIST-Scale datasets and on-par generalization with global affine transformation methods like Spatial Transformers.
Augmenting transformation knowledge onto a convolutional neural network's weights has often yielded significant improvements in performance. For rotational transformation augmentation, an important element to recent approaches has been the use of a steerable basis i.e. the circular harmonics. Here, we propose a scale-steerable filter basis for the locally scale-invariant CNN, denoted as log-radial harmonics. By replacing the kernels in the locally scale-invariant CNN \cite{lsi_cnn} with scale-steered kernels, significant improvements in performance can be observed on the MNIST-Scale and FMNIST-Scale datasets. Training with a scale-steerable basis results in filters which show meaningful structure, and feature maps demonstrate which demonstrate visibly higher spatial-structure preservation of input. Furthermore, the proposed scale-steerable CNN shows on-par generalization to global affine transformation estimation methods such as Spatial Transformers, in response to test-time data distortions.