Title	Face Recognition
Team	Dang Quoc Vu - Vu. qvdang.96@gmail.com
Predicting	Build a model for Vietnamese football players face recognition. The data is collected from google and captured from videos. Need to build a model that predicts people in the large image.
Data	capture images from videos (https://www.youtube.com), images from https://www.google.com/search. Link:https://drive.google.com/file/d/1P4Vf5wsrsWZc8AHH7jLbceiG5KS3580Y/view?usp=sharing
Features	image: continuous label: discrete
Models	MMOD is used for human face detection. Link: http://dlib.net/files/mmod_human_face_detector.dat.bz2 Use VGG_Face_net as the model to get face emmbedding. It outputs 2622 embedding for each face image then we take this 2622 embeddings for later classification of image. Link:https://drive.google.com/uc?id=1CPSeum3HpopfomUEK1gybeuIVoeJT_Eo
Future	impove the speed and accuracy in detecting and recognizing the human faces
References	[1] https://arxiv.org/pdf/1502.00046.pdf [2] https://medium.com/analytics-vidhya/face-recognition-with-vgg-face-in-keras-96e6bc1951d5 [3] https://www.pyimagesearch.com/2018/01/22/install-dlib-easy-complete-guide/ [4] https://sefiks.com/2018/08/06/deep-face-recognition-with-keras/
Results	Score: > 0.96

Directory structure

|cnn_model/
|  |-- mmod_face_detector.dat
|  |-- vgg_face_net_weight.h5
|dataset/
|  |--data_images/
|     |--video/
|        |--(video)
|     |--video_frames/
|        |--(frames of video)
|  |--test_image/ 
|     |--(images)
|  |--train_image/ 
|     |--Football-player[1]/
|        |--(images)
|     |--Football-player[2]/
|        |--(images)
|     |--Football-player[3]/
|        |--(images)
|     |--Football-player[4]/
|        |--(images)
|other_file/
|  |--(processed data files)
|src/
|  |--(files .py)
|test_result/
|  |--(images)
|trained_model/
|  |--(model file .h5)

Command Line

Training:

python3 face_train.py --p <bool>
- True  : Process data before training 
- False : Use current processed data to train

Testing:

python3 face_test.py --image <image_name>
- Paste the test-image into /dataset/test_image/
- The result image is saved in test_result folder with the same name

Result:

The picture below shows the trained model is able to classify 11 football players in Vietnam national team successfully.