HyperNets and their application to learning spatial transformations
This work addresses the challenge of modeling spatial transformations in image processing, which is important for computer vision applications, but it appears incremental as it builds on existing neural network concepts.
The paper tackles the problem of learning groups of spatial transformations, such as rotation and affine transformations, by proposing HyperNets, a conceptual framework for higher-order artificial neural networks. The result shows that HyperNets can generalize these transformations and compensate for image rotation to bring images into canonical forms.
In this paper we propose a conceptual framework for higher-order artificial neural networks. The idea of higher-order networks arises naturally when a model is required to learn some group of transformations, every element of which is well-approximated by a traditional feedforward network. Thus the group as a whole can be represented as a hyper network. One of typical examples of such groups is spatial transformations. We show that the proposed framework, which we call HyperNets, is able to deal with at least two basic spatial transformations of images: rotation and affine transformation. We show that HyperNets are able not only to generalize rotation and affine transformation, but also to compensate the rotation of images bringing them into canonical forms.