Oriented Response Networks

TL;DR

This paper introduces Active Rotating Filters (ARFs) for deep neural networks, enabling explicit encoding of orientation information, leading to improved rotation-invariant features and classification accuracy across various architectures.

Contribution

The paper proposes ARFs that actively rotate during convolution, allowing networks to learn rotation-invariant features and improve performance on classification and orientation estimation tasks.

Findings

ARFs improve rotation invariance in deep networks

Replacing regular filters with ARFs reduces parameters and boosts accuracy

Consistent performance gains across multiple architectures and benchmarks

Abstract

Deep Convolution Neural Networks (DCNNs) are capable of learning unprecedentedly effective image representations. However, their ability in handling significant local and global image rotations remains limited. In this paper, we propose Active Rotating Filters (ARFs) that actively rotate during convolution and produce feature maps with location and orientation explicitly encoded. An ARF acts as a virtual filter bank containing the filter itself and its multiple unmaterialised rotated versions. During back-propagation, an ARF is collectively updated using errors from all its rotated versions. DCNNs using ARFs, referred to as Oriented Response Networks (ORNs), can produce within-class rotation-invariant deep features while maintaining inter-class discrimination for classification tasks. The oriented response produced by ORNs can also be used for image and object orientation estimation tasks. Over multiple state-of-the-art DCNN architectures, such as VGG, ResNet, and STN, we consistently observe that replacing regular filters with the proposed ARFs leads to significant reduction in the number of network parameters and improvement in classification performance. We report the best results on several commonly used benchmarks.