Spatial Transformer Networks

TL;DR

Spatial Transformer Networks introduce a learnable module that enables neural networks to actively manipulate spatial data, improving invariance to transformations like translation, scale, and rotation, and achieving state-of-the-art results.

Contribution

The paper presents the Spatial Transformer module, a novel differentiable component that can be integrated into CNNs to enhance spatial invariance without extra supervision.

Findings

Improved invariance to translation, scale, and rotation.

Achieved state-of-the-art performance on multiple benchmarks.

Enables neural networks to learn spatial transformations automatically.

Abstract

Convolutional Neural Networks define an exceptionally powerful class of models, but are still limited by the lack of ability to be spatially invariant to the input data in a computationally and parameter efficient manner. In this work we introduce a new learnable module, the Spatial Transformer, which explicitly allows the spatial manipulation of data within the network. This differentiable module can be inserted into existing convolutional architectures, giving neural networks the ability to actively spatially transform feature maps, conditional on the feature map itself, without any extra training supervision or modification to the optimisation process. We show that the use of spatial transformers results in models which learn invariance to translation, scale, rotation and more generic warping, resulting in state-of-the-art performance on several benchmarks, and for a number of classes of transformations.