face_compare.py

from facenet_pytorch import MTCNN, InceptionResnetV1
import torchvision.transforms as transforms
from PIL import Image
import torch

mtcnn = MTCNN(keep_all=False, device='cpu')

model = InceptionResnetV1(pretrained='vggface2').eval()

def preprocess_face(base64_img):
        img = Image.open(base64_img).convert('RGB')

        preprocess = transforms.Compose([
            transforms.Resize((160, 160)),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
        ])

        # Detect face and get the bounding box
        box, _ = mtcnn.detect(img)

        if box is not None:
            # Crop the image using the bounding box
            img_cropped = img.crop(box[0])  # box[0] contains (x1, y1, x2, y2)

            # Apply the rest of preprocessing (resize, tensor, normalize)
            img_preprocessed = preprocess(img_cropped).unsqueeze(0)
            return img_preprocessed
        else:
            print("No face detected")
            return None

def compare_faces(base64_img1, base64_img2):
    # Load and detect face, then crop the image

    # Preprocess both images
    img1 = preprocess_face(base64_img1)  # Replace with your image path
    img2 = preprocess_face(base64_img2)  # Replace with your image path

    # Check if faces were detected in both images
    if img1 is not None and img2 is not None:
        # Get the embeddings from the FaceNet model
        with torch.no_grad():
            emb1 = model(img1)
            emb2 = model(img2)

        # Calculate cosine similarity
        cosine_similarity = torch.nn.functional.cosine_similarity(emb1, emb2)
        return cosine_similarity.item()
    else:
        print("Face detection failed on one or both images.")
        return None