milinddeore/mediapipe-face-detector-cpp

Mediapipe face detector tflite model running, without using mediapipe framework, c++ implementation.

mediapipe-face-detector-cpp:

Mediapipe face detector tflite model running, without using mediapipe framework, c++ implementation.

Build standalone Tensorflow Lite library for Linux OR MacOS:

Using the instruction mentioned in the following repo. Such that your tensorflow lite library is ready for your platform. I have tested it on MacOS, it should work perfectly on Linux too.

Build Face Detection project:

Standalone setup means you have all the included files at one place i.e. 3rdparty folder and libtensroflowlite.dylib(for MacOS) OR libtensroflowlite.so (for Linux) library ready for linking under libs folder. Run make.sh file to compile your project. make.sh file essentially, compile and link the project as:

I am also using opencv library so make sure all the includes and libraries are present on your machine: pkg-config --cflags opencv and pkg-config --libs opencv.

g++ -std=c++11 '-Wl,-rpath,$$ORIGIN/lib'  face_detect.cpp -o face_detector -L../3rdparty/libs -ltensorflowlite -I ../3rdparty/tensorflow/    
lite/tools/make/downloads/flatbuffers/include  -I ../3rdparty/tensorflow/lite/tools/make/downloads/absl  -I ../3rdparty/include -I ../       
3rdparty/ `pkg-config --cflags opencv` `pkg-config --libs opencv`  -lstdc++ -ldl -lpthread -lm -lz

Run the project:

./face_detector

mediapipe-face-detector-cpp's Issues

error C2070: “uint32_t [0]”: 非法的 sizeof 操作数

windows error when run code
error C2070: “uint32_t [0]”: 非法的 sizeof 操作数

Can I ask you what kind of tf model did you use?

using .tflite file in your repo, it takes almost 50~70ms for inference. I think it would be really helpful if I know the type or structure of the net you used for this project.
Thanks