Skip to content

Latest commit

 

History

History
72 lines (55 loc) · 2.17 KB

tf2caffe_tf2_mkldnn.README.md

File metadata and controls

72 lines (55 loc) · 2.17 KB
Raw

Usage:

install caffe

  • donwload and checkout

    • git clone https://github.com/intel/caffe.git
    • cd caffe && git checkout 1.0.6

  • modify caffe source

    • move PATCH/caffe.patch, PATCH/clip.patch to caffe folder
    • run git apply caffe.patch to modify caffe

  • install caffe and pycaffe

    • mkdir build && cd build && cmake -DCPU_ONLY=1 -DCMAKE_BUILD_TYPE=Release ..
    • make all -j$(nproc)

  • set caffe and pycaffe env

    • change user to root
    • CAFFE_ROOT=$(YOUR CAFFE FOLDER)
    • cd $CAFFE_ROOT and run:
for req in $(cat python/requirements.txt) pydot; do pip install  $req; done
export PYCAFFE_ROOT=$CAFFE_ROOT/python
export PYTHONPATH=$PYCAFFE_ROOT:$PYTHONPATH
export PATH=$CAFFE_ROOT/build/tools:$PYCAFFE_ROOT:$PATH
echo "$CAFFE_ROOT/build/lib" >> /etc/ld.so.conf.d/caffe.conf && ldconfig

install mkl-dnn

  • download

    • git clone https://github.com/intel/mkl-dnn

  • modify mkl-dnn source

    • move clip.patch to mkl-dnn folder
    • cd mkl-dnn folder
    • run git apply clip.patch

  • install mkl-dnn

    • cd scripts && ./prepare_mkl.sh && cd ..
    • mkdir -p build && cd build && cmake .. && make
    • make test
    • sudo make install

convert model

  • now you should have installed caffe, pycaffe, and mkl-dnn.

  • make sure you have tensorflow and pycaffe installed, can check by import tensorflow and import caffe in python.

  • prepare pb format tensorflow model.

  • create a folder to save caffe model.

  • cd to the script folder and run, eg:

# Transform the model to internal representation
python tf2topo.py --input_model_filename=./mobilenet.pb  --weights_file=save_model/weights.bin\
             --pkl_file=save_model/weights.pkl --topo_file=save_model/topo.txt

# Transform to caffe model
python topo2caffe.py --topo_file=save_model/topo.txt --pkl_file=save_model/weights.pkl\
    --batch_size=1 --save_folder=save_model

# Transform to C++ code which is based on MKLDNN
python make_main.py --topo=save_model/topo.txt
  • run net of mkl-dnn as below:
    • cd run_mkl_dnn
    • vi build.sh and replace ${path of mkl-dnn} to '/home/enxiang/mkl-dnn' after 'MKLDNN_ROOT='
    • sh build.sh

done