Note

This works very well. I've not been not maintaining it for a while. I'll give you gals and guys some pointers.

• OpenCV Version
• Tensorflow Version (This is a major problem. This code was written for a really really old version. I don't even remeber it. I did it in a trance!)

Facenet C++ Classifier

• This is C++ implementation of the Facenet classifier for the Facenet project by David Sandberg https://github.com/davidsandberg/facenet
• Part of the Tensor Flow code has been taken from davidsandberg/facenet#357 and tensorflow/tensorflow#8033
• Preprocessing of Images is done using code posted by knighthappy on davidsandberg/facenet#357
• It has been compiled & run using OpenCV 3.3.0 and Tensor Flow 1.4.0 on Fedora 26 x86_64 and Ubuntu 16.04 x86_64
• It has been also adapted for OpenCV 4

Build

cd src && mkdir build && cd build
cmake ..
make

Usage

• ./run <TRAIN|CLASSIFY> <Path/To/TensorFlowModel> <Path/To/Image/Directory/Structure> <Path/To/Classifier/Model>
• Directory structure should be <class_id>/<image_files>
• Face images should be 160x160, otherwise they will be resized

Issues

• More Testing Needed
• Results are good with more than one input image given to the pre-trained model.

Instructions to compile Tensor Flow C++ shared library

• https://www.tensorflow.org/install/install_sources to get libtensorflow_framework.so in /usr/lib/python2.7/site-packages/tensorflow/
• http://tuatini.me/building-tensorflow-as-a-standalone-project/
• https://github.com/FloopCZ/tensorflow_cc <- this one was used here