Scale Steerable Filters for Locally Scale-Invariant Convolutional Neural Networks
arXiv:1906.03861
Abstract
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.
Accepted as a Spotlight talk to ICML Workshop on Theoretical Physics for Deep Learning, 2019
References in corpus (3)
Cited by in corpus (10)
- Incorporating Symmetry into Deep Dynamics Models for Improved Generalization
- Scale-Equivariant Steerable Networks
- A General Theory of Equivariant CNNs on Homogeneous Spaces
- Scale-covariant and scale-invariant Gaussian derivative networks
- Scale-invariant scale-channel networks: Deep networks that generalise to previously unseen scales
- Scaling-Translation-Equivariant Networks with Decomposed Convolutional Filters
- Coordinate Independent Convolutional Networks -- Isometry and Gauge Equivariant Convolutions on Riemannian Manifolds
- On the Realization and Analysis of Circular Harmonic Transforms for Feature Detection
- Implicit Equivariance in Convolutional Networks
- Scale Equivariant Neural Networks with Morphological Scale-Spaces