anonymizer.py

def preprocess(path):
    img = cv2.imread(path)
    input_tensor = tf.convert_to_tensor(image_np)
    return input_tensor

def blur(img, bbox):
    
    imagette = img[bbox[0]:bbox[2], bbox[1]:bbox[3]]
    w, h = imagette.shape[:2]
    
    to_blur = imagette[0:int(0.7*w), :, :]
    blur = cv2.medianBlur(to_blur,55)
    
    imagette[0:int(0.7*w), :, :] = blur
    img[bbox[0]:bbox[2], bbox[1]:bbox[3]] = imagette
    
    return img

def anonymize(face_detector, imgpath, output_dir, threshold=0.5):
    
    im = cv2.imread(imgpath)
    input_image = preprocess(imgpath)
    input_tensor = tf.convert_to_tensor(im)
    input_tensor = input_tensor[tf.newaxis, ...]
    detections = face_detector(input_tensor)

    num_detections = int(detections.pop('num_detections'))
    detections = {key: value[0, :num_detections].numpy()
                   for key, value in detections.items()}
    
    detections['num_detections'] = num_detections
    detections['detection_classes'] = detections['detection_classes'].astype(np.int64)
    (w, h) = im.shape[:2]
    im1 = im.copy()
    for bbox, conf in list(zip(detections["detection_boxes"], detections["detection_scores"])):
        if conf > threshold:
            bbox = bbox * np.array([w, h, w, h])
            bbox = bbox.astype("int")
            im = blur(im, bbox)
            
    output_path = os.path.join(output_dir,imgpath.split('/')[-1])
    cv2.imwrite(os.path.join(output_dir, imgpath.split('/')[-1]), im)
    return