timctho/caffe-fold-batchnorm

Folds batch normalisation and the following scale layer into a single scale layer for networks trained in Caffe. This can be done at inference time to reduce memory consumption.

Python