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.
Geek Repo:Geek Repo
Github PK Tool:Github PK Tool