def load_and_process_example(example_string, mode,
image_size=224, preprocess=True):
"""To process records read from tfRecords file.
Args:
example_string: str, serialized string record.
mode: str, from tf.estimator.ModeKeys. Decides how the data is preprocessed.
image_size: int, for resizing the image.
preprocess: bool, if true resized to `image_size`.
Returns:
{'inputs': image, 'targets':label}
"""
data = tf.io.parse_single_example(
example_string,
features={
'image/encoded': tf.io.FixedLenFeature([], dtype=tf.string),
'image/class/label': tf.io.FixedLenFeature([], tf.int64)
})
image_string = data['image/encoded']
image_decoded = tf.image.decode_jpeg(image_string, channels=3)
if preprocess:
# The following does random crop/flip for training and center crop for test.
image_decoded = _do_scale(image_decoded, image_size + 32)
image_decoded = imagenet_preprocess_example({'inputs': image_decoded},
mode,
resize_size=(image_size,
image_size),
normalize=False)['inputs']
image_decoded = _keras_vgg16_preprocess(image_decoded)
return {'inputs': image_decoded, 'targets': data['image/class/label']}
def preprocess_example(self, example, mode, _):
return imagenet.imagenet_preprocess_example(example, mode)
def preprocess_example(self, example, mode, _): return imagenet.imagenet_preprocess_example(example, mode)
