def preprocess(self, image: tf.Tensor,
label: tf.Tensor) -> Tuple[tf.Tensor, tf.Tensor]:
"""Apply image preprocessing and augmentation to the image and label."""
if self.is_training:
image = preprocessing.preprocess_for_train(
image,
image_size=self.image_size,
mean_subtract=self.config.mean_subtract,
standardize=self.config.standardize,
dtype=self.dtype,
augmenter=self.augmenter)
else:
image = preprocessing.preprocess_for_eval(
image,
image_size=self.image_size,
num_channels=self.num_channels,
mean_subtract=self.config.mean_subtract,
standardize=self.config.standardize,
dtype=self.dtype)
label = tf.cast(label, tf.int32)
if self.config.one_hot:
label = tf.one_hot(label, self.num_classes)
label = tf.reshape(label, [self.num_classes])
return image, label
def preprocess_data_imagenet(data, is_training):
"""ImageNet data preprocessing"""
if is_training:
image = preprocessing.preprocess_for_train(data['image'],
image_size=224)
else:
image = preprocessing.preprocess_for_eval(data['image'],
image_size=224)
return (image, data['label'])
def preprocess_data(data, is_training):
"""ImageNet data preprocessing"""
if is_training:
image = preprocessing.preprocess_for_train(
data['image'],
image_size=224,
augmenter=augment.AutoAugment() if FLAGS.use_autoaugment else None)
if FLAGS.distort_color:
image = color_distortion(image, s=1.0)
else:
image = preprocessing.preprocess_for_eval(data['image'], image_size=224)
return {'input_1': image, 'label': data['label']}