def _load_iam_paragraphs():
print('Loading IAM paragraph crops and ground truth from image files...')
images = []
gt_images = []
ids = []
for filename in CROPS_DIRNAME.glob('*.jpg'):
id_ = filename.stem
image = util.read_image(filename, grayscale=True)
image = 1. - image / 255
gt_filename = GT_DIRNAME / f'{id_}.png'
gt_image = util.read_image(gt_filename, grayscale=True)
images.append(image)
gt_images.append(gt_image)
ids.append(id_)
images = np.array(images).astype(np.float32)
gt_images = util.to_categorical(np.array(gt_images), 3).astype(np.uint8)
return images, gt_images, np.array(ids)
def y_test(self):
return to_categorical(self.y_test_int, self.num_classes)
def y_train(self):
return to_categorical(self.y_train_int, self.num_classes)
def convert_strings_to_categorical_labels(labels, mapping):
return np.array([
to_categorical([mapping[c] for c in label], num_classes=len(mapping))
for label in labels
])
def y_test(self):
"""Return y_test"""
return util.to_categorical(self.y_test_int, self.num_classes)
def y_train(self):
"""Return y_train"""
return util.to_categorical(self.y_train_int, self.num_classes)