face.py

from deepface import DeepFace
import logging
import json
import numpy as np

 # set level of logging
logger = logging.getLogger()
logger.setLevel(logging.INFO)

handler = logging.FileHandler('logfile.log')
handler.setLevel(logging.INFO)

# set format of logs
formatter = logging.Formatter("%(asctime)s - %(levelname)s - %(message)s")
handler.setFormatter(formatter)
logger.addHandler(handler)


# creating embeddings
def create_embedding(login: str, is_login: bool):
    if is_login:
        img_path = f'tmp/{login}/img.png'
        emb_path = f'tmp/{login}/emb.txt'
    else:
        img_path = f'db/{login}/img.png'
        emb_path = f'db/{login}/emb.txt'

    result = DeepFace.represent(img_path=img_path, anti_spoofing=True)
    logger.info(f'Created embedding of user {login}')
    
    with open(emb_path, 'w') as f:
        f.write(str(result[0]['embedding']))

    logger.info(f'Saved embedding of user {login} in {emb_path}')


# to calculate similarity
def verify_face(login: str):
    try:
        emb1_path = f'db/{login}/emb.txt'
        emb2_path = f'tmp/{login}/emb.txt'

        with open(emb1_path, 'r') as f:
            emb1 = f.read()
            emb1 = emb1.replace("'", '"')

        with open(emb2_path, 'r') as f:
            emb2 = f.read()
            emb2 = emb2.replace("'", '"')

        emb1 = json.loads(emb1)
        emb1 = np.array(emb1)

        emb2 = json.loads(emb2)
        emb2 = np.array(emb2)

        dot_product = np.dot(emb1, emb2)
        source_norm = np.linalg.norm(emb1)
        test_norm = np.linalg.norm(emb2)
        distances = 1 - dot_product / (source_norm * test_norm)

        logger.info(f'Successfully calculated cosine distance of user {login}')
        return distances < 0.68
    except:
        logger.error('Error while calculating cosine distance')

    return False


# do testowania 
if __name__ == '__main__':
    emb1 = DeepFace.represent('db/Linas/img.png')
    emb2 = DeepFace.represent('tmp/Linas/img.png')

    with open('db/Linas/test_emb.txt', 'w') as f:
        f.write(str(emb1[0]))

    with open('tmp/Linas/test_emb.txt', 'w') as f:
        f.write(str(emb2[0]))

    with open('db/Linas/test_emb.txt', 'r') as f:
        test1 = f.read()
        test1 = test1.replace("'", '"')

    with open('tmp/Linas/test_emb.txt', 'r') as f:
        test2 = f.read()
        test2 = test2.replace("'", '"')

    test1 = json.loads(test1)
    test1 = np.array(test1)

    test2 = json.loads(test2)
    test2 = np.array(test2)

    dot_product = np.dot(test1, test2)
    source_norm = np.linalg.norm(test1)
    test_norm = np.linalg.norm(test2)
    distances = 1 - dot_product / (source_norm * test_norm)

    print(distances)