Ported from https://github.com/qassemoquab/stnbhwd according to pytorch tutorial. Ported from https://github.com/fxia22/stn.pytorch only for personal use now support 3D spatial transformation, but have removed the GPU part.

cd script
./make.sh
python build.py
python test.py

There is a demo in test_stn.py

STN is the spatial transformer module, it takes a B*C*H*W*D tensor and a B*C*H*W*3 grid normalized to [-1,1] as an input and do bilinear sampling.

AffineGridGen takes a B*3*4 matrix and generate an affine transformation grid.