wangzheng17 / CBDNet

Code for Composite Binary Decomposition Networks

Composite Binary Decomposition Networks

In this repository, we released code for Composite Binary Decomposition Networks (CBDNet).

Contents

1. Installation
2. Usage
3. Experiment Result
4. Reference

Installation

1. Clone the repository of Caffe and compile it

    git clone https://github.com/BVLC/caffe.git
    cd caffe
    # modify Makefile.config to the path of the library on your machine, please make sure the python interface is supported
    make -j8
    make pycaffe

2. Clone this repository

    https://github.com/wangzheng17/CBDNet.git

Usage

1. Download the original model files (.prototxt and .caffemodel) and move them to the directory of models

2. Command Line Usage To decompose a network, use the following command

    python main.py <arguments>
    arguments for CBDNet:
        -bottleneck           bottleneck ratio value (Numbers like 0.2,0.3,0.4,0.5,...)
        -j                    the number of overall channel except the sign channel(overall-fix channel, post-variable channel. Numbers like 5,6,7,8,...)
        -model MODEL          caffe prototxt file path
        -weight WEIGHT        caffemodel file path

For example, suppose the Resnet-18 network is in folder models/ and named as resnet-18.prototxt and resnet-18.caffemodel, you can decompose all layers with same number of channels:

    python main.py -model models/resnet-18.prototxt -weight models/resnet-18.caffemodel -bottleneck 0.5 -j 6

Note: please use same prefix name for prototxt and weights file, a floder will be created in "\models". btn_xx_a_list.npy will be stored for reusing, where xx is the bottleneck ratio. Results are saved under j_model sub-directory for same number of non-compressed channel setting and same number of overall channel setting. The parameter 'j' in the result indicates overall channels include the sign. Using model name compatiable with that in line 33-43, rank_a.py. These lines claim exclusive layers that don't require computing for specific model.

Reference

This work is based on our work Composite Binary Decomposition Networks (AAAI2019)[Paper]. If you think this is helpful for your research, please consider append following bibtex config in your latex file.