A Caffe Re-Implementation of Network Slimming

This repository is the caffe re-implementation for the following paper
Learning Efficient Convolutional Networks Through Network Slimming (ICCV 2017).

I have applied this method to prune MobileNet v2 in Pruned-MobileNet_v2.

Usage

Merge the two folders into your $(CAFFE_HOME).

Attention

• You'd better back up your own files first. And after merging the files, you should manually modify the code in $(CAFFE_HOME)/src/caffe/proto/caffe.proto as follows.

message ScaleParameter {
  optional int32 axis = 1 [default = 1];
  optional int32 num_axes = 2 [default = 1];
  optional FillerParameter filler = 3;
  optional bool bias_term = 4 [default = false];
  optional FillerParameter bias_filler = 5;
  optional float l1_lambda = 6 [default = 1e-9];  // The code that you need to add manually.
 }

I didn't directly upload the caffe.proto, because the file is likely to contain your own modifications, and you need to be careful to replace it.

• After that, run make -j4 under your $(CAFFE_HOME).

Application example

layer {
  name: "conv1/scale"
  type: "Scale"
  bottom: "conv1/bn"
  top: "conv1/bn"
  scale_param {
    filler {
      value: 0.5
    }
    bias_term: true
    bias_filler {
      value: 0
    }
    l1_lambda: 0.0001  // The L1 regularization coefficient for gamma
  }
}

Contact

If you have any problems, please feel free to contact me via [email protected].