Batch Renormalization module for PyTorch
Batch Renormalization implementation based on paper https://arxiv.org/abs/1702.03275 . Currently there is only 2D implementation for conv nets.
Usage is the same as for PyTorch BatchNorm2D class: You intialize an instance by providing number of features (number of channels in conv layer). Input is expected to be in shape (N, C, H, W), where: (N - number of samples per batch, C - number of channels, H - height, W - width)