search

seung-lab / znn-release

Multi-core CPU implementation of deep learning for 2D and 3D sliding window convolutional networks (ConvNets).
GNU General Public License v3.0
94 stars 33 forks source link

fix forward pass by adding layer specific "output map size" #44

Closed xiuliren closed 8 years ago

xiuliren commented 8 years ago
  1. the forward pass needs a output map size, which should be (1,1,1) in the last layer. The original variable name of layer_fov is misleading. The fov is equivalent of receptive field, which should be the whole network field of view in the last output layer. outmapsz could be a better variable name.
  2. add a default parameter value for malis_norm_type. The normal training do not use malis, so the parameter of malis normalization type should not be necessary.
xiuliren commented 8 years ago

since we already merged a forward pass fix, this update should be merged with master to solve conflicts first. Will create a pull request later.