app.py

#!pip install streamlit
from keras_vggface.utils import preprocess_input
from keras_vggface.vggface import VGGFace
import pickle
from sklearn.metrics.pairwise import cosine_similarity
import streamlit as st
from PIL import Image
import os
import cv2
from mtcnn import MTCNN
import numpy as np

detector = MTCNN()
model = VGGFace(model='resnet50', include_top=False,
                input_shape=(224, 224, 3), pooling='avg')
feature_list = pickle.load(open('embedding.pkl', 'rb'))
filenames = pickle.load(open('filenames.pkl', 'rb'))


def extract_features(img_path, model, detector):
    img = cv2.imread(img_path)
    results = detector.detect_faces(img)

    x, y, width, height = results[0]['box']

    face = img[y:y + height, x:x + width]

    #  extract its features
    image = Image.fromarray(face)
    image = image.resize((224, 224))

    face_array = np.asarray(image)

    face_array = face_array.astype('float32')

    expanded_img = np.expand_dims(face_array, axis=0)
    preprocessed_img = preprocess_input(expanded_img)
    result = model.predict(preprocessed_img).flatten()
    return result


def recommend(feature_list, features):
    similarity = []
    for i in range(len(feature_list)):
        similarity.append(cosine_similarity(features.reshape(
            1, -1), feature_list[i].reshape(1, -1))[0][0])

    result = sorted(list(enumerate(similarity)),
                    reverse=True, key=lambda x: x[1])[0]
    index_pos = result[0]
    percentage = result[1]
    return index_pos, percentage


def save_uploaded_image(uploaded_image):
    try:
        with open(os.path.join('uploads', uploaded_image.name), 'wb') as f:
            f.write(uploaded_image.getbuffer())
        return True
    except:
        return False


st.title('Which bollywood celebrity do you look like?')

uploaded_image = st.file_uploader('Upload an image')

if uploaded_image is not None:
    # save the image in a directory
    if save_uploaded_image(uploaded_image):
        # load the image
        display_image = Image.open(uploaded_image)
        resized_display_img = display_image.resize((260, 320), Image.ANTIALIAS)

        # extract the features
        features = extract_features(os.path.join(
            'uploads', uploaded_image.name), model, detector)
        # recommend
        img_path, percentage = recommend(feature_list, features)
        actor_path = filenames[img_path]

        predicted_actor = " ".join(
            actor_path.split('\\')[1].split('_'))

        display_pred_image = Image.open(actor_path)
        resized_pred_img = display_pred_image.resize(
            (260, 320), Image.ANTIALIAS)
        # display
        st.header(
            f'You look like {predicted_actor} with {np.round(percentage*100,0)}% similarity')
        col1, col2 = st.columns(2)

        with col1:
            st.header('Your uploaded image')
            st.image(resized_display_img)
        with col2:
            st.header('Your celeb look-alike')
            st.image(resized_pred_img)