Caffe to TensorFlow

Convert Caffe models to TensorFlow.

Usage with standalone model inside docker container

1 - Install caffe-tensorflow

git clone https://github.com/psaboia/caffe-tensorflow
# This fork has only be tested with Python 2.7 and Tensorflow CPU 

2 - Build docker image

cd caffe-tensorflow
sudo docker build -t caffe-tensorflow .

3 - Run docker container

docker run -ti caffe-tensorflow bash

4 - Convert your model

cd caffe-tensorflow
python2.7 convert.py --caffemodel ./model.caffemodel ./model.prototxt --standalone-output-path ./standalone.pb --code-output-path ./output.py

5- Use the standalone.pb file

It contains the weights and the architecture of the network.

Usage

Run convert.py to convert an existing Caffe model to TensorFlow.

The input consists of an existing Caffe model.

1. --caffemodel ./model.caffemodel ./model.prototxt. Make sure you're using the latest Caffe format.

The output consists of the files:

1. --standalone-output-path ./standalone.pb, a standalone.pb GraphDef model file containing the model's graph and learned parameters (to be loaded on Tensorflow)
2. --code-output-path ./output.py, a Python class that constructs the model's graph.
3. --data-output-path ./output.mat, a data file (in NumPy's native format) containing the model's learned parameters.

Example: Converting googlenet model from caffe to tensorflow

# get Caffe model
wget http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel
wget https://raw.githubusercontent.com/BVLC/caffe/master/models/bvlc_googlenet/deploy.prototxt

# convert 
python2.7 convert.py --caffemodel ./bvlc_googlenet.caffemodel ./deploy.prototxt --standalone-output-path ./standalone.pb --code-output-path ./output.py