def get_x_from_path(self, path: str) -> np.ndarray:
img = cv2.imread(path)
x = normalize_img(img,
with_aug=self.with_aug,
width=self.img_w,
height=self.img_h)
x = np.moveaxis(np.array(x), 2, 0)
return x
def preprocess(self, imgs):
xs = []
for img in imgs:
x = normalize_img(img, width=self.width, height=self.height)
xs.append(x)
xs = np.moveaxis(np.array(xs), 3, 1)
xs = torch.tensor(xs)
xs = xs.to(device_torch)
return xs
def predict_with_confidence(self, imgs: List[np.ndarray]) -> Tuple:
"""
TODO: describe method
"""
xs = [normalize_img(img) for img in imgs]
if not bool(xs):
return [], []
predicted = self._predict(xs)
confidences, orientations = self.unzip_predicted(predicted)
return orientations, confidences, predicted
def predict_with_confidence(self, imgs: List[np.ndarray or List]) -> Tuple:
"""
Predict options(region, count lines) with confidence by numberplate images
"""
xs = [normalize_img(img) for img in imgs]
if not bool(xs):
return [], [], [], []
predicted = self._predict(xs)
confidences, region_ids, count_lines = self.unzip_predicted(predicted)
count_lines = self.custom_count_lines_id_to_all_count_lines(
count_lines)
return region_ids, count_lines, confidences, predicted
def predict_with_confidence(self, imgs: List[np.ndarray]) -> Tuple:
"""
Predict options(region, count lines) with confidence by numberplate images
"""
xs = [normalize_img(img) for img in imgs]
predicted = [[], []]
if bool(xs):
xs = np.moveaxis(np.array(xs), 3, 1)
predicted = self.ort_session.run(None, {self.input_name: xs})
confidences, region_ids, count_lines = self.unzip_predicted(predicted)
count_lines = self.get_count_lines_labels(count_lines)
return region_ids, count_lines, confidences, predicted
def get_x_from_path(self, path: str, bbox: List, angle: int) -> np.ndarray:
img = cv2.imread(path)
rotated_img, rotated_bbox = rotate_image_and_bboxes(
img, np.array([bbox]), int(angle))
rotated_bbox = rotated_bbox[0]
rotated_bbox[0] = rotated_bbox[0] if rotated_bbox[0] > 0 else 0
rotated_bbox[1] = rotated_bbox[1] if rotated_bbox[1] > 0 else 0
min_x = rotated_bbox[0]
max_x = rotated_bbox[2]
min_y = rotated_bbox[1]
max_y = rotated_bbox[3]
img_part = rotated_img[min_y:max_y, min_x:max_x]
img_part = normalize_img(img_part, width=self.img_w, height=self.img_h)
img_part = np.moveaxis(np.array(img_part), 2, 0)
return img_part
def run_iteration(self, with_aug=False):
ys = []
xs = []
paths = []
for _ in np.arange(self.batch_size):
x, y = self.next_sample()
paths.append(x)
img = cv2.imread(x)
x = normalize_img(img,
with_aug=with_aug,
width=self.img_w,
height=self.img_h)
xs.append(x)
ys.append(y)
xs = np.moveaxis(np.array(xs), 3, 1)
return paths, xs, ys
def preprocess(images):
x = convert_cv_zones_rgb_to_bgr(images)
x = [normalize_img(img) for img in x]
x = np.moveaxis(np.array(x), 3, 1)
return x