Equivariant Transformer Networks

Abstract

How can prior knowledge on the transformation invariances of a domain be incorporated into the architecture of a neural network? We propose Equivariant Transformers (ETs), a family of differentiable image-to-image mappings that improve the robustness of models towards pre-defined continuous transformation groups. Through the use of specially-derived canonical coordinate systems, ETs incorporate functions that are equivariant by construction with respect to these transformations. We show empirically that ETs can be flexibly composed to improve model robustness towards more complicated transformation groups in several parameters. On a real-world image classification task, ETs improve the sample efficiency of ResNet classifiers, achieving relative improvements in error rate of up to 15% in the limited data regime while increasing model parameter count by less than 1%.

ICML 2019

References in corpus (6)

• On the Convergence of Adam and Beyond
• Group Equivariant Convolutional Networks
• Rotation equivariant vector field networks
• Exploiting Cyclic Symmetry in Convolutional Neural Networks
• Deforming Autoencoders: Unsupervised Disentangling of Shape and Appearance
• Sparse Unsupervised Capsules Generalize Better

Cited by in corpus (10)

• Deep Learning for UAV-based Object Detection and Tracking: A Survey
• Choose a Transformer: Fourier or Galerkin
• Cortical surface registration using unsupervised learning
• Scale-Equivariant Steerable Networks
• B-Spline CNNs on Lie Groups
• Spatially Covariant Lesion Segmentation
• Group Equivariant Stand-Alone Self-Attention For Vision
• RoI Tanh-polar Transformer Network for Face Parsing in the Wild
• Deep Transformation-Invariant Clustering
• FILTRA: Rethinking Steerable CNN by Filter Transform