Abstract

Deep Convolution Neural Networks (DCNNs) are capable of learning unprecedentedly effective image representations. However, their ability in handling signifificant 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 fifilter bank containing the fifilter itself and its multiple unmaterialised rotated versions. During backpropagation, 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 classifification 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 fifilters with the proposed ARFs leads to signifificant reduction in the number of network parameters and improvement in classifification performance. We report the best results on several commonly used benchmarks