Rotation Equivariant Vector Field Networks (2016-12-29T00:00:00.000000Z)

TL;DR

The Rotation Equivariant Vector Field Networks (RotEqNet), a Convolutional Neural Network architecture encoding rotation equivariance, invariance and covariance, is proposed and a modified convolution operator relying on this representation to obtain deep architectures is developed.

Abstract

In many computer vision tasks, we expect a particular behavior of the output with respect to rotations of the input image. If this relationship is explicitly encoded, instead of treated as any other variation, the complexity of the problem is decreased, leading to a reduction in the size of the required model. In this paper, we propose the Rotation Equivariant Vector Field Networks (RotEqNet), a Convolutional Neural Network (CNN) architecture encoding rotation equivariance, invariance and covariance. Each convolutional filter is applied at multiple orientations and returns a vector field representing magnitude and angle of the highest scoring orientation at every spatial location. We develop a modified convolution operator relying on this representation to obtain deep architectures. We test RotEqNet on several problems requiring different responses with respect to the inputs’ rotation: image classification, biomedical image segmentation, orientation estimation and patch matching. In all cases, we show that RotEqNet offers extremely compact models in terms of number of parameters and provides results in line to those of networks orders of magnitude larger.

Authors

D. Tuia

6 papers

N. Komodakis

8 papers

Diego Marcos

1 papers

TL;DR

Abstract

Authors

References37 items

Oriented Response Networks

Steerable CNNs

Harmonic Networks: Deep Translation and Rotation Equivariance

An Efficient Fusion Move Algorithm for the Minimum Cost Lifted Multicut Problem

Warped Convolutions: Efficient Invariance to Spatial Transformations

The Importance of Skip Connections in Biomedical Image Segmentation

Deep models for brain EM image segmentation: novel insights and improved performance

RIFD-CNN: Rotation-Invariant and Fisher Discriminative Convolutional Neural Networks for Object Detection

Learning rotation invariant convolutional filters for texture classification

TI-POOLING: Transformation-Invariant Pooling for Feature Learning in Convolutional Neural Networks

Group Equivariant Convolutional Networks

Deep Contextual Networks for Neuronal Structure Segmentation

Discriminative Learning of Deep Convolutional Feature Point Descriptors

Surpassing Humans in Boundary Detection using Deep Learning

Pushing the Boundaries of Boundary Detection using Deep Learning

Crowdsourcing the creation of image segmentation algorithms for connectomics

Spatial Transformer Networks

U-Net: Convolutional Networks for Biomedical Image Segmentation

Learning Deconvolution Network for Semantic Segmentation

Learning to compare image patches via convolutional neural networks

When Face Recognition Meets with Deep Learning: An Evaluation of Convolutional Neural Networks for Face Recognition

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

MatConvNet: Convolutional Neural Networks for MATLAB

Deep Symmetry Networks

Understanding Image Representations by Measuring Their Equivariance and Equivalence

Designing deep networks for surface normal estimation

Fully convolutional networks for semantic segmentation

Rotation, Scaling and Deformation Invariant Scattering for Texture Discrimination

Learning Invariant Representations with Local Transformations

Transformation Equivariant Boltzmann Machines

Tiled convolutional neural networks

An Integrated Micro- and Macroarchitectural Analysis of the Drosophila Brain by Computer-Assisted Serial Section Electron Microscopy

An empirical evaluation of deep architectures on problems with many factors of variation

Learning Local Image Descriptors

Best practices for convolutional neural networks applied to visual document analysis

Handwritten zip code recognition with multilayer networks

Dropout: a simple way to prevent neural networks from overfitting

Field of Study

Journal Information

Name

Page

Venue Information

Name

Type

URL

Alternate Names