oaid / facedetection

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

MTCNN C++ Implementation

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.

Build

• 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 CAFFE_ROOT, MXNET_ROOT or TENSORFLOW_ROOT to the right path in your machine. For example : CAFFE_ROOT=/usr/local/AID/Caffe-HRT/.

• make -j4

Run

If the basic work is ready (build caffe/Mxnet/Tensorflow sucessfully) followed by above steps. You can run the test now.

1. Test on single picture:

./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"

2. Test on camera (DL Framework is caffe)

./run.sh

Release History

Version 0.1.0 - 2018-2-11

• Modified readme file.
• Modified makefile.mk.
• Add run.sh script

Credit

MTCNN algorithm

https://github.com/kpzhang93/MTCNN_face_detection_alignment

MTCNN C++ on Caffe

https://github.com/wowo200/MTCNN

MTCNN python on Mxnet

https://github.com/pangyupo/mxnet_mtcnn_face_detection

MTCNN python on Tensorflow

FaceNet uses MTCNN to align face

https://github.com/davidsandberg/facenet

From this directory:

facenet/src/align

facedetection's Issues

Caffe is faster than tensorflow ?

The caffe runs faster than the tensorflow version, and caffe need less GPU memory. Is that right ?

make problem

i have disable the arm_compute, but met another problem whem make your project. I built the libmtcnn.so successfully. Link the test.a just print these problem below.

/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_DeleteTensor' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_DeleteStatus'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_DeleteSession' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_DeleteGraph'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_Message' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_TensorData'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_NewGraph' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_NewSessionOptions'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_SessionRun' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_NewSession'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_CloseSession' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_GraphOperationByName'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_GraphImportGraphDef' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_NewStatus'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_Dim' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_NewTensor'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_GetCode' /home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to TF_NewImportGraphDefOptions'
/home/sg/workspace/mtcnn/libmtcnn/libmtcnn.so: undefined reference to `TF_ImportGraphDefOptionsSetPrefix'
collect2: error: ld returned 1 exit status
Makefile:18: recipe for target 'test' failed
make: *** [test] Error 1

make problem

@xhbdahai i met a problem when i make your project,can you help me..
/usr/bin/ld: cannot find -larm_compute
collect2: error: ld returned 1 exit status
Makefile:17: recipe for target 'test' failed

Where is CAFFE_ROOT.

Hi,
A quick question. We have to edit CAFFE_ROOT in Makefile, but i am not able to see this parameter in Makefile. Also its not there in MXNet-HRT/Makefile.
Any suggestion?
Thanks

make error with tensorflow

I am working with tensorflow and success in bazel build.
I do have libtensorflow_framework.so libtensorflow.so in /usr/local/lib/
and I verify this with a toy .c file with

#include <stdio.h>
#include <tensorflow/c/c_api.h>

int main() {
  printf(“Hello from TensorFlow C library version %s\n”, TF_Version());
  return 0;
}

and gcc -I/usr/local/include -L/usr/local/lib hello_tf.c -ltensorflow
and it runs as expected

but in this project, I got these errors like:

g++ -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/include pkg-config --cflags opencv -I/usr/local//include -c test.cpp -o test.o
make[1]: Entering directory /home/sensetime/miaospace/FaceDetection/libmtcnn' g++ -fpic -shared -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/include pkg-config --cflags opencv-I/usr/local//include -c mtcnn.cpp -o mtcnn.os g++ -fpic -shared -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/includepkg-config --cflags opencv-I/usr/local//include -c comm_lib.cpp -o comm_lib.os g++ -fpic -shared -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/includepkg-config --cflags opencv-I/usr/local//include -c utils.cpp -o utils.os g++ -fpic -shared -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/includepkg-config --cflags opencv-I/usr/local//include -c tensorflow_mtcnn.cpp -o tensorflow_mtcnn.os g++ -fpic -shared -o libmtcnn.so mtcnn.os comm_lib.os utils.os tensorflow_mtcnn.os g++ -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/includepkg-config --cflags opencv-I/usr/local//include -c mtcnn.cpp -o mtcnn.o g++ -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/includepkg-config --cflags opencv-I/usr/local//include -c comm_lib.cpp -o comm_lib.o g++ -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/includepkg-config --cflags opencv-I/usr/local//include -c utils.cpp -o utils.o g++ -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/includepkg-config --cflags opencv-I/usr/local//include -c tensorflow_mtcnn.cpp -o tensorflow_mtcnn.o ar -rcv libmtcnn.a mtcnn.o comm_lib.o utils.o tensorflow_mtcnn.o a - mtcnn.o a - comm_lib.o a - utils.o a - tensorflow_mtcnn.o make[1]: Leaving directory/home/sensetime/miaospace/FaceDetection/libmtcnn'
g++ -Wall -O2 -ggdb -std=c++11 -I/home/sensetime/miaospace/FaceDetection/include pkg-config --cflags opencv -I/usr/local//include test.o -o test -L/home/sensetime/miaospace/FaceDetection/libmtcnn -lmtcnn pkg-config --libs opencv pkg-config --libs computelibrary
Package computelibrary was not found in the pkg-config search path.
Perhaps you should add the directory containing computelibrary.pc' to the PKG_CONFIG_PATH environment variable No package 'computelibrary' found /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_ImportGraphDefOptionsSetPrefix'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_NewStatus' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_CloseSession'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_NewImportGraphDefOptions' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_TensorData'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_DeleteTensor' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_SessionRun'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_NewTensor' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_DeleteStatus'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_GetCode' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_Dim'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_NewSessionOptions' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_GraphImportGraphDef'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_Message' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_DeleteGraph'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_GraphOperationByName' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_NewGraph'
/home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_DeleteSession' /home/sensetime/miaospace/FaceDetection/libmtcnn/libmtcnn.so: undefined reference to TF_NewSession'
collect2: error: ld returned 1 exit status
make: *** [test] Error 1

I think there is something missing in linking tensorflow in test.o

Face alignment

Can I please know how to align face in c++ after detection as done in https://github.com/davidsandberg/facenet/blob/master/src/align/align_dataset_mtcnn.py

error while loading shared librarie

compile whit the "make problem #2" suggestions, the compilation was successful but when running test or camera I get:
"error while loading shared libraries: libmtcnn.so: cannot open shared object file: No such file or directory"

how can I fix this?

How can I use GPU+NEON to test mtcnn?

Hi, thank you sharing the code ,Now I can use CPU to run the mtcnn ,How can I use GPU+NEON to test mtcnn? THX

mac make issue

i'm using mac v10.14.4
i build tensorflow with the command shown on readme(bazel build --config=opt //tensorflow/tools/lib_package:libtensorflow) and when i do 'make -j4' i get
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
this error.

i have opencv v4.1.0 and tensorflow v1.12

exact output is ***********************

g++ -Wall -O2 -ggdb -std=c++11 -I/Users/angie/Desktop/FaceDetection/include pkg-config --cflags /usr/local/Cellar/opencv/4.1.0_1/lib/pkgconfig/opencv4.pc -I/Users/angie/Desktop/vision-2d/tf_standalone/include -c camera.cpp -o camera.o
g++ -fpic -shared -o libmtcnn.so mtcnn.os comm_lib.os utils.os tensorflow_mtcnn.os
g++ -Wall -O2 -ggdb -std=c++11 -I/Users/angie/Desktop/FaceDetection/include pkg-config --cflags /usr/local/Cellar/opencv/4.1.0_1/lib/pkgconfig/opencv4.pc -I/Users/angie/Desktop/vision-2d/tf_standalone/include -c mtcnn.cpp -o mtcnn.o
g++ -Wall -O2 -ggdb -std=c++11 -I/Users/angie/Desktop/FaceDetection/include pkg-config --cflags /usr/local/Cellar/opencv/4.1.0_1/lib/pkgconfig/opencv4.pc -I/Users/angie/Desktop/vision-2d/tf_standalone/include -c comm_lib.cpp -o comm_lib.o
Undefined symbols for architecture x86_64:
"_TF_CloseSession", referenced from:
tf_mtcnn::~tf_mtcnn() in tensorflow_mtcnn.os
"_TF_DeleteGraph", referenced from:
tf_mtcnn::~tf_mtcnn() in tensorflow_mtcnn.os
"_TF_DeleteSession", referenced from:
tf_mtcnn::~tf_mtcnn() in tensorflow_mtcnn.os
"_TF_DeleteStatus", referenced from:
tf_mtcnn::~tf_mtcnn() in tensorflow_mtcnn.os
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_DeleteTensor", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_Dim", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_GetCode", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_GraphImportGraphDef", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_GraphOperationByName", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_ImportGraphDefOptionsSetPrefix", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_Message", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewGraph", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewImportGraphDefOptions", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewSession", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewSessionOptions", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewStatus", referenced from:
tf_mtcnn::tf_mtcnn() in tensorflow_mtcnn.os
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_NewTensor", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_SessionRun", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_TensorData", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::copyMakeBorder(cv::_InputArray const&, cv::OutputArray const&, int, int, int, int, int, cv::Scalar const&)", referenced from:
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
"cv::Mat::deallocate()", referenced from:
set_input_buffer(std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >&, float*, int, int) in comm_lib.os
std::__1::__split_buffer<cv::Mat, std::__1::allocatorcv::Mat&>::__split_buffer() in comm_lib.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
cv::MatExpr::MatExpr() in tensorflow_mtcnn.os
"cv::Mat::updateContinuityFlag()", referenced from:
set_input_buffer(std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >&, float*, int, int) in comm_lib.os
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
"cv::Mat::copySize(cv::Mat const&)", referenced from:
set_input_buffer(std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >&, float*, int, int) in comm_lib.os
void std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >::__push_back_slow_path<cv::Mat const&>(cv::Mat const&&&) in comm_lib.os
std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >::__swap_out_circular_buffer(std::__1::__split_buffer<cv::Mat, std::__1::allocatorcv::Mat&>&) in comm_lib.os
"cv::Mat::Mat(cv::Mat const&, cv::Range const&, cv::Range const&)", referenced from:
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
"cv::error(int, std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&, char const*, char const*, int)", referenced from:
set_input_buffer(std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >&, float*, int, int) in comm_lib.os
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
"cv::resize(cv::_InputArray const&, cv::OutputArray const&, cv::Size, double, double, int)", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
"cv::cvtColor(cv::_InputArray const&, cv::_OutputArray const&, int, int)", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::fastFree(void*)", referenced from:
set_input_buffer(std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >&, float*, int, int) in comm_lib.os
std::__1::__split_buffer<cv::Mat, std::__1::allocatorcv::Mat&>::__split_buffer() in comm_lib.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::_1::allocator<face_box> >&) in tensorflow_mtcnn.os
cv::MatExpr::~MatExpr() in tensorflow_mtcnn.os
"cv::operator-(cv::Mat const&, cv::Scalar const&)", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::operator*(cv::MatExpr const&, double)", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::Mat::t() const", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::Mat::convertTo(cv::_OutputArray const&, int, double, double) const", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
make[1]: *** [libmtcnn.so] Error 1
make[1]: *** Waiting for unfinished jobs....
camera.cpp:101:71: warning: implicit conversion from 'double' to 'int' changes value from 1.8 to 1 [-Wliteral-conversion]
box.landmark.y[l]), 1, cv::Scalar(0, 0, 255), 1.8);
^~~
1 warning generated.
make: *** [libs] Error 2

where is tensorflow model comes from?

Thanks for the code. I am wondering where this mtcnn_frozen_model.pb comes from? I need details of how is freeze and the layer names setup, cause I am training my own mtcnn tensorflow models.

Thanks again.

make problem

Hi everyone,

I want to tensorflow and have built the libtensorflow.so myself. I have the following error when make:

Undefined symbols for architecture x86_64:
"_TF_CloseSession", referenced from:
tf_mtcnn::~tf_mtcnn() in tensorflow_mtcnn.os
"_TF_DeleteGraph", referenced from:
tf_mtcnn::~tf_mtcnn() in tensorflow_mtcnn.os
"_TF_DeleteSession", referenced from:
tf_mtcnn::~tf_mtcnn() in tensorflow_mtcnn.os
"_TF_DeleteStatus", referenced from:
tf_mtcnn::~tf_mtcnn() in tensorflow_mtcnn.os
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_DeleteTensor", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_Dim", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_GetCode", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_GraphImportGraphDef", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_GraphOperationByName", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_ImportGraphDefOptionsSetPrefix", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_Message", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewGraph", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewImportGraphDefOptions", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewSession", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewSessionOptions", referenced from:
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
"_TF_NewStatus", referenced from:
tf_mtcnn::tf_mtcnn() in tensorflow_mtcnn.os
tf_mtcnn::load_model(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&) in tensorflow_mtcnn.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_NewTensor", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_SessionRun", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"_TF_TensorData", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_RNet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
tf_mtcnn::run_ONet(cv::Mat const&, std::__1::vector<face_box, std::__1::allocator<face_box> >&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::copyMakeBorder(cv::_InputArray const&, cv::OutputArray const&, int, int, int, int, int, cv::Scalar const&)", referenced from:
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
"cv::Mat::deallocate()", referenced from:
set_input_buffer(std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >&, float*, int, int) in comm_lib.os
std::__1::__split_buffer<cv::Mat, std::__1::allocatorcv::Mat&>::__split_buffer() in comm_lib.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
cv::MatExpr::MatExpr() in tensorflow_mtcnn.os
"cv::Mat::copySize(cv::Mat const&)", referenced from:
set_input_buffer(std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >&, float*, int, int) in comm_lib.os
void std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >::__push_back_slow_path<cv::Mat const&>(cv::Mat const&&&) in comm_lib.os
std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >::__swap_out_circular_buffer(std::__1::__split_buffer<cv::Mat, std::__1::allocatorcv::Mat&>&) in comm_lib.os
"cv::Mat::Mat(cv::Mat const&, cv::Range const&, cv::Range const&)", referenced from:
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
"cv::error(int, cv::String const&, char const*, char const*, int)", referenced from:
cv::Mat::Mat(int, int, int, void*, unsigned long) in comm_lib.os
cv::Mat::Mat(int, int, int, void*, unsigned long) in tensorflow_mtcnn.os
"cv::String::deallocate()", referenced from:
cvflann::anyimpl::big_any_policycv::String::static_delete(void**) in mtcnn.os
cvflann::anyimpl::big_any_policycv::String::move(void* const*, void**) in mtcnn.os
cv::Mat::Mat(int, int, int, void*, unsigned long) in comm_lib.os
cvflann::anyimpl::big_any_policycv::String::static_delete(void**) in comm_lib.os
cvflann::anyimpl::big_any_policycv::String::move(void* const*, void**) in comm_lib.os
cvflann::anyimpl::big_any_policycv::String::static_delete(void**) in utils.os
cvflann::anyimpl::big_any_policycv::String::move(void* const*, void**) in utils.os
...
"cv::String::allocate(unsigned long)", referenced from:
cv::Mat::Mat(int, int, int, void*, unsigned long) in comm_lib.os
cv::Mat::Mat(int, int, int, void*, unsigned long) in tensorflow_mtcnn.os
"cv::resize(cv::_InputArray const&, cv::OutputArray const&, cv::Size, double, double, int)", referenced from:
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
"cv::cvtColor(cv::_InputArray const&, cv::_OutputArray const&, int, int)", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::fastFree(void*)", referenced from:
set_input_buffer(std::__1::vector<cv::Mat, std::__1::allocatorcv::Mat >&, float*, int, int) in comm_lib.os
std::__1::__split_buffer<cv::Mat, std::__1::allocatorcv::Mat&>::__split_buffer() in comm_lib.os
tf_mtcnn::run_PNet(cv::Mat const&, scale_window&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
copy_one_patch(cv::Mat const&, face_box&, float*, int, int) in tensorflow_mtcnn.os
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::_1::allocator<face_box> >&) in tensorflow_mtcnn.os
cv::MatExpr::~MatExpr() in tensorflow_mtcnn.os
"cv::operator-(cv::Mat const&, cv::Scalar const&)", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::operator*(cv::MatExpr const&, double)", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::Mat::t() const", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
"cv::Mat::convertTo(cv::_OutputArray const&, int, double, double) const", referenced from:
tf_mtcnn::detect(cv::Mat&, std::__1::vector<face_box, std::__1::allocator<face_box> >&) in tensorflow_mtcnn.os
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
make[1]: *** [Makefile:32: libmtcnn.so] Error 1
make[1]: Leaving directory '/Users/jinjunjie/Desktop/Work/codes/combined-mtcnn-cpp/libmtcnn'
make: *** [Makefile:23: libs] Error 2

My make version is:
GNU Make 4.2.1
Built for x86_64-apple-darwin16.5.0

Thanks a ton!

when make -j4, occure an error "fatal error: caffe/caffe.hpp : No such file or directory".But the caffe.hpp was already existed in CaffeOnACL

It seems that modifying the code doesn't work

int caffe_mtcnn::load_model(const std::string &proto_model_dir)
{

	Caffe::set_mode(Caffe::GPU); //use GPU mode

	//load pre_trained model
	//first param is file path, second represent the net is used for test
	PNet_=new Net<float>(( "./models/det111.prototxt"), caffe::TEST);
	PNet_->CopyTrainedLayersFrom(proto_model_dir + "/det1.caffemodel");


	RNet_=new Net<float>((proto_model_dir + "/det2.prototxt"), caffe::TEST);
	RNet_->CopyTrainedLayersFrom(proto_model_dir + "/det2.caffemodel");


	ONet_=new Net<float>((proto_model_dir + "/det3.prototxt"), caffe::TEST);
	ONet_->CopyTrainedLayersFrom(proto_model_dir + "/det3.caffemodel");

	return 0;
}

I have modified the caffe_mtcnn::load_model() method like above, you can see i changed the mode but after recompiling the project, it still run in cpu mode.
And i modified the model's file name to det111.prototxt , but the log info still shows that the Net load the model lwhich is called det1.prototxt.

I wish you can help me! Thank you!

Here are part of the console log info

mtcnn git:(master) ✗ ./test      
1
WARNING: Logging before InitGoogleLogging() is written to STDERR
I1123 20:26:12.290751 14968 upgrade_proto.cpp:67] Attempting to upgrade input file specified using deprecated input fields: ./models/det1.prototxt
I1123 20:26:12.290843 14968 upgrade_proto.cpp:70] Successfully upgraded file specified using deprecated input fields.
W1123 20:26:12.290848 14968 upgrade_proto.cpp:72] Note that future Caffe releases will only support input layers and not input fields.
I1123 20:26:12.290976 14968 net.cpp:51] Initializing net from parameters: 
name: "PNet"
state {
  phase: TEST
  level: 0
}
layer {
  name: "input"
  type: "Input"
  top: "data"
  input_param {
    shape {
      dim: 1
      dim: 3
      dim: 12
      dim: 12
    }