import cv2 import numpy as np from scipy.spatial import Delaunay import os import urllib.request import time import tempfile import shutil # Get the directory where this script is located script_dir = os.path.dirname(os.path.abspath(__file__)) # Download the face landmarker model if not present model_path_src = os.path.join(script_dir, "face_landmarker.task") if not os.path.exists(model_path_src): print("Downloading MediaPipe Face Landmarker model...") model_url = "https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/1/face_landmarker.task" urllib.request.urlretrieve(model_url, model_path_src) print("Model downloaded successfully!") # Copy to temp dir (MediaPipe cannot handle Unicode in paths) model_path = shutil.copy2(model_path_src, tempfile.mkdtemp()) # Import MediaPipe import mediapipe as mp from mediapipe.tasks import python from mediapipe.tasks.python import vision def create_face_detector(model_path): """Create a MediaPipe Face Landmarker for video processing.""" base_options = python.BaseOptions(model_asset_path=model_path) options = vision.FaceLandmarkerOptions( base_options=base_options, running_mode=vision.RunningMode.VIDEO, num_faces=1 ) return vision.FaceLandmarker.create_from_options(options) def detect_landmarks_video(detector, frame_rgb, timestamp_ms): """Detect facial landmarks in a video frame.""" mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=frame_rgb) result = detector.detect_for_video(mp_image, timestamp_ms) if not result.face_landmarks: return None height, width = frame_rgb.shape[:2] landmarks = [] for lm in result.face_landmarks[0]: x = lm.x * width y = lm.y * height landmarks.append([x, y]) return np.array(landmarks) def create_lowpoly_frame(frame_rgb, landmarks, show_edges=False): """Create a low-poly version of a video frame.""" height, width = frame_rgb.shape[:2] # Add boundary points for full coverage corners = np.array([ [0, 0], [width, 0], [width, height], [0, height], [width/2, 0], [width, height/2], [width/2, height], [0, height/2] ]) points = np.vstack([landmarks, corners]) # Compute Delaunay triangulation try: tri = Delaunay(points) except Exception: return frame_rgb # Return original if triangulation fails # Create output image output = np.zeros_like(frame_rgb) # Draw each triangle for simplex in tri.simplices: triangle = points[simplex].astype(np.int32) # Calculate centroid for color sampling centroid = np.mean(triangle, axis=0).astype(int) cx = np.clip(centroid[0], 0, width - 1) cy = np.clip(centroid[1], 0, height - 1) # Get color from original image color = frame_rgb[cy, cx].tolist() # Draw filled triangle cv2.fillPoly(output, [triangle], color) # Optionally draw edges if show_edges: cv2.polylines(output, [triangle], True, (255, 255, 255), 1) return output def main(): print("=" * 50) print("Real-Time Low-Poly Face Art") print("=" * 50) print("\nControls:") print(" 'q' - Quit") print(" 's' - Save current frame") print(" 'e' - Toggle edges") print(" 'f' - Toggle FPS display") print("\nStarting webcam...") # Open webcam cap = cv2.VideoCapture(0) if not cap.isOpened(): print("Error: Could not open webcam.") print("Please check that your webcam is connected and not in use.") return # Set resolution (lower for better performance) cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640) cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 480) # Create face detector detector = create_face_detector(model_path) # State variables show_edges = False show_fps = True frame_count = 0 start_time = time.time() fps = 0 print("Webcam opened successfully!") print("Position your face in front of the camera.") while True: ret, frame = cap.read() if not ret: print("Error: Could not read frame from webcam.") break # Mirror the frame for more intuitive interaction frame = cv2.flip(frame, 1) # Convert BGR to RGB for MediaPipe frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # Get timestamp for video mode timestamp_ms = int(cap.get(cv2.CAP_PROP_POS_MSEC)) if timestamp_ms == 0: timestamp_ms = int((time.time() - start_time) * 1000) # Detect landmarks landmarks = detect_landmarks_video(detector, frame_rgb, timestamp_ms) if landmarks is not None: # Create low-poly version output_rgb = create_lowpoly_frame(frame_rgb, landmarks, show_edges) # Convert back to BGR for display output = cv2.cvtColor(output_rgb, cv2.COLOR_RGB2BGR) else: # Show original with "No face detected" message output = frame.copy() cv2.putText(output, "No face detected", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2) # Calculate and display FPS frame_count += 1 elapsed = time.time() - start_time if elapsed > 0: fps = frame_count / elapsed if show_fps: cv2.putText(output, f"FPS: {fps:.1f}", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2) # Display edge status edge_status = "ON" if show_edges else "OFF" cv2.putText(output, f"Edges: {edge_status}", (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1) # Show the frame cv2.imshow(' Real-Time Low-Poly Face Triangulation (Press Q to quit)', output) # Handle key presses key = cv2.waitKey(1) & 0xFF if key == ord('q'): print("\nQuitting...") break elif key == ord('s'): # Save current frame save_path = os.path.join(script_dir, "realtime_lowpoly_screenshot.png") cv2.imwrite(save_path, output) print(f"Screenshot saved to: {save_path}") elif key == ord('e'): show_edges = not show_edges print(f"Edges: {'ON' if show_edges else 'OFF'}") elif key == ord('f'): show_fps = not show_fps print(f"FPS display: {'ON' if show_fps else 'OFF'}") # Clean up cap.release() cv2.destroyAllWindows() detector.close() print("Done!") if __name__ == '__main__': main()