This is a C++ project to implement MTCNN, a perfect face detect algorithm, on different DL frameworks.
The most popular frameworks: caffe/mxnet/tensorflow, are all suppported now.
Bulid caffe, mxnet or tensorflow first Please edit makefile.mk (set xxx_ON flags to enable corresponding dp framework) to select one or more to be supported
Build Caffe-HRT, refer to Caffe-HRT Release notes
Build MXNet-HRT, refer to MXNet-HRT release notes
Build tensorflow, to generate libtensorflow.so, please use:
bazel build --config=opt //tensorflow/tools/lib_package:libtensorflow
the tarball, bazel-bin/tensorflow/tools/lib_package/libtensorflow.tar.gz, includes the libtensorflow.so and c header files
Edit Makefile to set
TENSORFLOW_ROOT to the right path in your machine. For example : CAFFE_ROOT=/usr/local/AID/Caffe-HRT/.
If the basic work is ready (build caffe/Mxnet/Tensorflow sucessfully) followed by above steps. You can run the test now.
./test -f photo_fname [ -t DL_type] [-s] -f photo_fname picture to be detected -t DL_type DL frame: "caffe" , "mxnet"(default) or "tensorflow" -s Save face chop into jpg files
The new picture, which boxed face and 5 landmark points will be created and saved as "new.jpg"
FaceNet uses MTCNN to align face
From this directory: