Spatial Transformer Networks with TPS transformation in Caffe

This code implemented the hot-off-the-press-paper Spatial Transformer Network in one of the most popular deep learning frameworks -- Caffe. This code implemented the Spatial Transformer Network with TPS transformation.

Oragnization of the code

• src/caffe/layers: contains the CPU implementation (two .cpp files) for Spatial Transformer with TPS Layer.I will update the GPU part as soon as possible. In that part, there will have two .cu files for the Layer.
• src/caffe/proto: contains my proto definition for Spatial Transformer Layer (Search for 'CtoParaParameter' and 'TPStransformParameter').
• include: contains two .hpp files for the layer.
• examples: contains the experiment code on EM3D and Cremi dataset. You can take a look as an example.

Reference

This code was implemented based on the Spatial Transformer Networks with Affine transformation, which has been implemented by Carey Mo:(https://github.com/daerduoCarey/SpatialTransformerLayer).

Acknowledge

The code is implemented at the first semester of my Master degree study. Since it is my first time implement an C++ application, the code might not be as effective as possible. But here, I still want to thank great help from my advisor Dr.Ji and Ph.D students T.Zeng, J.Li and Z.Wang.