| from typing import Tuple | |
| import cv2 | |
| import gradio as gr | |
| import numpy as np | |
| from mediapipe.python.solutions import drawing_utils | |
| from mediapipe.python.solutions.face_detection import FaceDetection | |
| from blurriness_estimation import BlurrinessEstimator | |
| from closed_eyes_detection import ClosedEyesDetector | |
| from closed_eyes_detection.eyes_extractor import EyesExtractor | |
| from config import InferenceConfig, BlurrinessConfig | |
| fd = FaceDetection() | |
| def draw_face_detection(img): | |
| """ | |
| Draws MediaPipe's face detection result on the input image. | |
| Args: | |
| img: input image, processed in place. | |
| """ | |
| results = fd.process(img) | |
| if results.detections: | |
| for detection in results.detections: | |
| drawing_utils.draw_detection(img, detection) | |
| def detect_closed_eyes(img: np.ndarray) -> Tuple[np.ndarray, str]: | |
| """ | |
| Runs the Closed Eyes Detector | |
| Args: | |
| img: input image | |
| Returns: | |
| str: Whether the eyes are opened or closed. | |
| """ | |
| if img is None: | |
| raise ValueError("No image provided") | |
| detector = ClosedEyesDetector(InferenceConfig()) | |
| extractor = EyesExtractor() | |
| is_closed = detector(img) | |
| if is_closed is not None: | |
| img = extractor.extract_eyes(img) | |
| b, h, w, c = img.shape | |
| vertical_separator = 255 * np.ones((h, 5, c)).astype(np.int8) | |
| img = np.hstack((img[1], vertical_separator, img[0])) | |
| s = "Eyes are closed." if is_closed else "Eyes are open." | |
| else: | |
| s = "No face detected" | |
| return img, s | |
| def detect_blur(img: np.ndarray) -> Tuple[np.ndarray, str]: | |
| """ | |
| Runs the Blurriness Detector | |
| Args: | |
| img: input image | |
| Returns: | |
| str: Whether the input image is blurry | |
| """ | |
| if img is None: | |
| raise ValueError("No image provided") | |
| cfg = BlurrinessConfig() | |
| be = BlurrinessEstimator(cfg) | |
| laplacian = np.absolute(cv2.Laplacian(img, cv2.CV_64F)).astype(np.int8) | |
| is_blurry, score = be(img) | |
| s = f"Sharpness threshold: {cfg.threshold}" | |
| s += f"\nScore: {score:.0f}\n" | |
| s += "Image is blurry (Score < threshold)" if is_blurry else "Image is sharp (score > threshold)." | |
| return laplacian, s | |
| demo = gr.Blocks() | |
| with demo: | |
| gr.Markdown("Detect closed eyes or estimate blurriness.") | |
| with gr.Tabs(): | |
| with gr.TabItem("Detect closed eyes"): | |
| with gr.Row(): | |
| with gr.Column(): | |
| eye_cam_input = gr.Image(sources="webcam", streaming=True) | |
| eyes_button = gr.Button("Detect") | |
| with gr.Column(): | |
| detection_output = gr.Image(label="Detection result (sorry for the creepiness)", ) | |
| detection_output_text = gr.Textbox(label='Result') | |
| with gr.TabItem("Estimate blur"): | |
| with gr.Row(): | |
| with gr.Column(): | |
| blur_cam_input = gr.Image(sources="webcam", streaming=True) | |
| blur_button = gr.Button("Estimate blur") | |
| with gr.Column(): | |
| laplacian_output = gr.Image(label="Laplacian filter view") | |
| blurriness_estimation_text = gr.Textbox(label='Result') | |
| eyes_button.click(detect_closed_eyes, inputs=[eye_cam_input], outputs=[detection_output, detection_output_text]) | |
| blur_button.click(detect_blur, inputs=[blur_cam_input], outputs=[laplacian_output, blurriness_estimation_text]) | |
| if __name__ == "__main__": | |
| demo.launch() | |