Hello, Please be noted : This is few quick steps taken for running in new version(R3). You might do in better way All policies/rules/terms/conditions etc will remain same as it is with original contributer: https://github.com/MekkaSiekka/Face-Recognition-with-OpenVino-Toolkit Thanks! Todo: 0. In cmake file, you should be able to comment out "server_plugin" executable. This is used for an outdated Intel video server and may cause issue during compilation. 1. install intel's OpenVino computational vision sdk from https://software.intel.com/en-us/openvino-toolkit Follow installation instructions at https://software.intel.com/en-us/articles/OpenVINO-Install-Linux, make sure you can run "Verify the Installation Using the Demo Scripts" Currently, only Ubuntu operation system is supported. ***terminal command to setup openvino/opencv environment: source /opt/intel/computer_vision_sdk/bin/setupvars.sh 2. In order to run the pre-trained Tensorflow model in Intel's Inference Engine for face recognition, a conversion using Model Optimizer is needed: To download the tensorflow model and have some understanding of face recognition, please refer to FaceNet's github page: https://github.com/davidsandberg/facenet This is a general view of model optimizer w. TF support https://software.intel.com/en-us/articles/OpenVINO-Using-TensorFlow For this specific project, run the following commands, and you need to specify the path to tensorflow .pb model cd /opt/intel/computer_vision_sdk/deployment_tools/model_optimizer python3 ./mo_tf.py --input_model path_to_model/model_name.pb --freeze_placeholder_with_value "phase_train->False" --reverse_input_channels Copy the converted .bin and .xml files to /model directory of *this project*, or you can specify absolute paths to them by modifying the code. -----or----- For your convenience, we have prepared necessary models to be downloaded: https://drive.google.com/drive/folders/1XS1PAZ8XdyhbPVXl2gcKDEt02zVOnsko?usp=sharing 3. For server plugin, you must change path from relative to absolute i.e: FaceRecognition.initialize("/home/xiaojiang/Desktop/webrtc-mcu-analytics-sdk/plugin/samples/face_recognition_plugin/model/20180402-114759.xml"); <------ vector<pair<string, vector<float> > > read_text(){ cout<< "Reading vectors of saved person"<<endl; vector<pair<string, vector<float> > > data; string file="/home/xiaojiang/Desktop/webrtc-mcu-analytics-sdk/plugin/samples/face_recognition_plugin/vectors.txt"; <------ 4. This project supports out-of-tree build, given that you have installed the OpenVino toolkit using sudo Building : cd build //under build directory rm -rf * (optional, if something bad happened you need to retry the build) cmake .. make ***Executables & .so are located under build/intel64/Release for Ubuntu OS** Description of each program : preprocess tool: processes "/raw_photos", and save to "/output_photos", raw_photos should have sub-directories as following example. Names of pictures do not affect execution, but directory names do matter. -- Jianlin |----01-08-2018-15-59-13.jpg |----01-08-2018-15-59-20.jpg |----01-08-2018-15-59-19.jpg |----01-08-2018-15-59-18.jpg |----01-08-2018-15-59-14.jpg -- Jay |----Jay_Chou_0001.jpg |----Jay_Chou_0003.jpg |----Jay_Chou_0004.jpg |----Jay_Chou_0002.jpg -- Bush |----George_W_Bush_0003.jpg |----George_W_Bush_0001.jpg |----George_W_Bush_0002.jpg proprocess_tool also outputs the features represented by 512-D vectors for each photo and saves to "vectors.txt" local_recognizer: Reads saved pictures from "vectors.txt", reads each frame from a camera, do infer of face detection/recognition for each frame, and compare with data contained in vectors.txt You need to provide sets of raw photos and then use preprocess_tool to organize /output_photos Photo_taker: Opens a camera stream if any. Do not click outside the terminal after you execute the program Press "Esc" to take a photo, which will be saved to "/camera photos" You can later organize these photos to form your raw_photos, and then use preprocess tool to make them ready for face recognition Common work flow: Build the repo. 1. Use photo_taker to take photos, and then manually move photos from /camera_photo to organize photo sets in /raw_photos ----or----- Upload your own photos and build photo sets in /raw_photos 2. Use pre_process_tool to crop photos; this will do inference on cropped photos so 512-D vectors are stored in vectors.txt It is recommended to delete all files under /output_photos before running ./preprocess_tool, since the program will output to this directory in response to photo stored under /raw_photos 3. Run programes 4. If you want to add more people to be recognized, repeat step 1-2 Tips: 1. Thresholds of FaceRecognitionClass to decide whether a face is "Unknown" or mostly close to a known photo can be changed in public function "recognize()." A threshold of 0.6 is accurate but relatively hard to achieve, while 0.7 is less demanding, but easier to recognize with allowance of uncertainty 2. Clear, frontal photos greatly enhance recognition accuracy Appendix : Tree of the directory: . ├── build ├── camera_photos │ ├── 01-08-2018-15-42-27.jpg │ ├── 01-08-2018-15-42-29.jpg │ ├── 01-08-2018-15-42-31.jpg │ ├── 01-08-2018-15-42-32.jpg │ ├── 01-08-2018-15-42-33.jpg │ ├── 01-08-2018-15-59-13.jpg │ ├── 01-08-2018-15-59-14.jpg │ ├── 01-08-2018-15-59-18.jpg │ ├── 01-08-2018-15-59-19.jpg │ ├── 01-08-2018-15-59-20.jpg │ ├── 01-08-2018-16-02-38.jpg │ ├── 01-08-2018-16-02-39.jpg │ ├── 02-08-2018-08-46-15.jpg │ ├── 02-08-2018-08-46-16.jpg │ ├── 02-08-2018-08-46-19.jpg │ ├── 02-08-2018-08-46-21.jpg │ ├── 02-08-2018-09-36-50.jpg │ ├── 02-08-2018-09-36-54.jpg │ ├── 02-08-2018-09-38-14.jpg │ ├── 02-08-2018-09-38-15.jpg │ ├── 02-08-2018-09-38-16.jpg │ ├── 02-08-2018-09-38-17.jpg │ ├── 02-08-2018-14-08-57.jpg │ ├── 02-08-2018-14-08-58.jpg │ ├── 03-08-2018-16-29-29.jpg │ ├── 03-08-2018-16-29-30.jpg │ ├── 03-08-2018-16-29-32.jpg │ ├── 03-08-2018-16-29-33.jpg │ ├── 03-08-2018-16-30-05.jpg │ ├── 03-08-2018-16-30-06.jpg │ ├── 09-08-2018-15-15-24.jpg │ ├── 09-08-2018-15-15-26.jpg │ ├── 09-08-2018-15-15-28.jpg │ ├── 09-08-2018-15-52-21.jpg │ ├── 09-08-2018-15-52-22.jpg │ ├── 09-08-2018-15-52-24.jpg │ ├── 09-08-2018-15-52-27.jpg │ └── 09-08-2018-15-52-28.jpg ├── CMakeLists.txt ├── local_recognizer │ ├── backup │ ├── CMakeLists.txt │ └── main.cpp ├── local_recognizer_async │ ├── CMakeLists.txt │ ├── main.cpp │ └── thread_func.h ├── model │ ├── 20180402-114759.bin │ ├── 20180402-114759.mapping │ └── 20180402-114759.xml ├── output_photos ├── photo_taker │ ├── CMakeLists.txt │ └── main.cpp ├── preprocess_tool │ ├── CMakeLists.txt │ └── main.cpp ├── raw_photos │ ├── Foo │ ├── Foo2 │ └── WuYangzu │ ├── 01-08-2018-16-02-38.jpg │ ├── 01-08-2018-16-02-39.jpg │ ├── 02-08-2018-08-46-15.jpg │ ├── 02-08-2018-08-46-16.jpg │ ├── 02-08-2018-14-08-57.jpg │ ├── 02-08-2018-14-08-58.jpg │ ├── 03-08-2018-16-30-05.jpg │ ├── 03-08-2018-16-30-06.jpg │ ├── 09-08-2018-15-15-24.jpg │ ├── 09-08-2018-15-15-26.jpg │ └── 09-08-2018-15-15-28.jpg ├── Readme ├── server_plugin │ ├── CMakeLists.txt │ ├── main.cpp │ └── myplugin.h ├── src │ ├── common_ie.hpp │ ├── emotion_recognition.cpp │ ├── emotion_recognition.h │ ├── face_detection.cpp │ ├── face_detection.h │ ├── face_recognition.cpp │ ├── face_recognition.h │ ├── person_detection.cpp │ └── person_detection.h └── vectors.txt