app.py

import gradio as gr
from mtcnn import MTCNN
import cv2
import numpy as np
import time
import concurrent.futures

# loading haar
ff = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')
ff_alt = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_alt.xml')
ff_alt2 = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_alt2.xml')
pf = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_profileface.xml')

# loading mtcnn
mtcnn = MTCNN()

global_start = time.perf_counter()
haar_start = 0
mtcnn_start = 0

def get_unique_face_locations(all_face_locations):
    unique_detected_faces = []
    for (x1, y1, w1, h1) in all_face_locations:
        unique = True
        for (x2, y2, w2, h2) in unique_detected_faces:
            if abs(x1 - x2) < 50 and abs(y1 - y2) < 50:
                unique = False
                break
        if unique:
            unique_detected_faces.append((x1, y1, w1, h1))
    
    return unique_detected_faces

def detect_haar(gray):
    global haar_start 

    haar_start = time.perf_counter()
    ff_faces = ff.detectMultiScale(gray, scaleFactor=1.2, minNeighbors=10, minSize=(25, 25))
    ff_alt2_faces = ff_alt2.detectMultiScale(gray, scaleFactor=1.05, minNeighbors=10, minSize=(20, 20))
    pf_faces = pf.detectMultiScale(gray, scaleFactor=1.05, minNeighbors=5, minSize=(20, 20))

    return ff_faces, ff_alt2_faces, pf_faces

def detect_mtcnn(frame):
    global mtcnn_start

    mtcnn_start = time.perf_counter()
    faces = mtcnn.detect_faces(frame)
    mt_faces = [face['box'] for face in faces]

    return mt_faces

def detect_faces(image):
    frame = image
    gray = cv2.cvtColor(np.array(image), cv2.COLOR_BGR2GRAY)

    with concurrent.futures.ThreadPoolExecutor() as executor:
        haar_detections = executor.submit(detect_haar, gray)
        mtcnn_detections = executor.submit(detect_mtcnn, frame)
    
        ff_faces, ff_alt2_faces, pf_faces = haar_detections.result()
        mt_faces = mtcnn_detections.result()

        all_faces = [*ff_faces, *ff_alt2_faces, *pf_faces, *mt_faces]
        unique_detected_faces = get_unique_face_locations(all_faces)

        for (x, y, w, h) in unique_detected_faces:
            cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 3)
    
        frame = cv2.putText(frame, f"{len(unique_detected_faces)} Faces", (20, 650), cv2.FONT_HERSHEY_SIMPLEX, 1.6, (0, 0, 0), 5)

    print('\n\n')
    print(f"Haar Took - {time.perf_counter() - haar_start:.2f}s")
    print(f"MTCNN Took - {time.perf_counter() - mtcnn_start:.2f}s")
    print(f"Total Time - {time.perf_counter() - global_start:.2f}s")
    print('\n\n')

    return frame
# Create a Gradio interface
iface = gr.Interface(
    fn=detect_faces,
    inputs=gr.components.Image(sources="webcam"),
    outputs=[gr.components.Image(type="numpy", label="Processed Image")],
    live=True
)

# Launch the application
iface.launch()