dcjones / gcnn.jl Goto Github PK

Implementation of a Group CNN (G-CNN) layer for Flux, which is kind of 2d convolution operation invariant under (a discretized set of) rotations. It's proposed in these papers.

Bekkers,E.J., Lafarge,M.W., Veta,M., Eppenhof,K.A.J., Pluim,J.P.W. and Duits,R. (2018) Roto-Translation Covariant Convolutional Networks for Medical Image Analysis. In Medical Image Computing and Computer Assisted Intervention – MICCAI 2018. Springer International Publishing, pp. 440–448.

Lafarge,M.W., Bekkers,E.J., Pluim,J.P.W., Duits,R. and Veta,M. (2021) Roto-translation equivariant convolutional networks: Application to histopathology image analysis. Med. Image Anal., 68, 101849.

The package exports RotGroupConv, which is a convolution operator applied to either a 4-dimensional input with shape [width, height, nchannels, nbatches], or applied to 5-dimensional input with [width, height, nchannels, nbatches, nrotations]. The former the paper calls a "lifting layer" and the latter a "group convolution layer".

Not really tested, use at your own risk.