Face Recognition

A light weight face recognition implementation using a pre-trained facenet model. Most of the code is taken from David Sandberg’s facenet repository.

Steps to follow:

1. Create a dataset of faces for each person and arrange them in below order

root folder  
│
└───Person 1
│   │───IMG1
│   │───IMG2
│   │   ....
└───Person 2
|   │───IMG1
|   │───IMG2
|   |   ....
 

2. Create embeddings of this dataset by running python script/create_embeddings.py --input_dir = $(/path/to/your/image/dataset) a pickle file embeddings.pickle will be created in your folder
3. script/face_recognition.py contains most functionalities for face recognition task. Input a face image of a person in the dataset, face_recognition.py
   calculates the distances between embedding of input image and embeddings in the dataset, and return the one with minimum distance
   i.e.

_, face, bboxes = align_face(image_path='path/to/image',
							 image_size = 160,
							 margin=44,
							 gpu_memory_fraction=1.0							 

emb = create_embeddings(
 
face, model_dir='/path/to/facenet/model')
embs = load_embeddings(emb_filepath='path/to/embedding/lists')
result, _ = identify_person(emb, embs)

align_face

detect face in an image and returns cropped image and resize it to 160x160
Args:

• image_path: path to image
• image_size: image size to be aligned
• margin: margin for the crop around the bounding box (height, width) in pixels.
• gpu_memory_fraction: percentage of gpu memory to be allocated

Returns:

• True if face is detected, false otherwise
• aligned face
• bounding box