def testDieOnTargetSizeGreaterThanImageSize(self):
image = np.dstack([[[5, 6], [9, 0]], [[4, 3], [3, 5]]])
with self.test_session():
image_placeholder = tf.placeholder(tf.float32)
padded_image = utils.pad_to_bounding_box(image_placeholder, 0, 0,
2, 1, 255)
with self.assertRaisesWithPredicateMatch(
errors.InvalidArgumentError,
'target_width must be >= width'):
padded_image.eval(feed_dict={image_placeholder: image})
padded_image = utils.pad_to_bounding_box(image_placeholder, 0, 0,
1, 2, 255)
with self.assertRaisesWithPredicateMatch(
errors.InvalidArgumentError,
'target_height must be >= height'):
padded_image.eval(feed_dict={image_placeholder: image})
def testDieIfImageTensorRankIsNotThree(self):
image = np.vstack([[5, 6], [9, 0]])
with self.test_session():
image_placeholder = tf.placeholder(tf.float32)
padded_image = utils.pad_to_bounding_box(image_placeholder, 0, 0,
2, 2, 255)
with self.assertRaisesWithPredicateMatch(
errors.InvalidArgumentError, 'Wrong image tensor rank'):
padded_image.eval(feed_dict={image_placeholder: image})
def testReturnOriginalImageWhenTargetSizeIsEqualToImageSize(self):
image = np.dstack([[[5, 6], [9, 0]], [[4, 3], [3, 5]]])
with self.test_session():
image_placeholder = tf.placeholder(tf.float32)
padded_image = utils.pad_to_bounding_box(image_placeholder, 0, 0,
2, 2, 255)
self.assertAllClose(
padded_image.eval(feed_dict={image_placeholder: image}), image)
def testDieIfTargetSizeNotPossibleWithGivenOffset(self):
image = np.dstack([[[5, 6], [9, 0]], [[4, 3], [3, 5]]])
with self.test_session():
image_placeholder = tf.placeholder(tf.float32)
padded_image = utils.pad_to_bounding_box(image_placeholder, 3, 0,
4, 4, 255)
with self.assertRaisesWithPredicateMatch(
errors.InvalidArgumentError,
'target size not possible with the given target offsets'):
padded_image.eval(feed_dict={image_placeholder: image})
def testReturnPaddedImageWithNonZeroPadValue(self):
for dtype in [np.int32, np.int64, np.float32, np.float64]:
image = np.dstack([[[5, 6], [9, 0]], [[4, 3], [3,
5]]]).astype(dtype)
expected_image = np.dstack([[[255, 255, 255, 255, 255],
[255, 255, 255, 255, 255],
[255, 5, 6, 255, 255],
[255, 9, 0, 255, 255],
[255, 255, 255, 255, 255]],
[[255, 255, 255, 255, 255],
[255, 255, 255, 255, 255],
[255, 4, 3, 255, 255],
[255, 3, 5, 255, 255],
[255, 255, 255, 255,
255]]]).astype(dtype)
with self.test_session():
image_placeholder = tf.placeholder(tf.float32)
padded_image = utils.pad_to_bounding_box(
image_placeholder, 2, 1, 5, 5, 255)
self.assertAllClose(
padded_image.eval(feed_dict={image_placeholder: image}),
expected_image)
def preprocess_image_and_label(image,
label,
model_input_height,
model_input_width,
min_scale_factor=1.,
max_scale_factor=1.,
scale_factor_step_size=0,
ignore_label=255,
is_training=True):
"""Preprocesses the image and label.
Args:
image: Input image.
label: Ground truth annotation label.
model_input_height: The height of the input feed to model extractor.
model_input_width: The width of the input feed to model extractor.
min_scale_factor: Minimum scale factor value.
max_scale_factor: Maximum scale factor value.
scale_factor_step_size: The step size from min scale factor to max scale
factor. The input is randomly scaled based on the value of
(min_scale_factor, max_scale_factor, scale_factor_step_size).
ignore_label: The label value which will be ignored for training and
evaluation.
is_training: If the preprocessing is used for training or not.
Returns:
original_image: Original image (could be resized).
processed_image: Preprocessed image.
label: Preprocessed ground truth segmentation label.
Raises:
ValueError: Ground truth label not provided during training.
"""
if is_training and label is None:
raise ValueError('During training, label must be provided.')
processed_image = tf.cast(image, tf.float32)
if label is not None:
label = tf.cast(label, tf.int32)
# Data augmentation by randomly scaling the inputs.
if is_training:
scale = utils.get_random_scale(min_scale_factor, max_scale_factor,
scale_factor_step_size)
processed_image, label = utils.randomly_scale_image_and_label(
processed_image, label, scale)
processed_image.set_shape([None, None, 3])
if not is_training:
processed_image, label = utils.resize_to_target(
processed_image, label, model_input_height, model_input_width)
# Pad image and label to have
# dimensions >= [model_input_height, model_input_width]
image_shape = tf.shape(processed_image)
image_height = image_shape[0]
image_width = image_shape[1]
target_height = image_height +\
tf.maximum(model_input_height - image_height, 0)
target_width = image_width +\
tf.maximum(model_input_width - image_width, 0)
# Pad image with mean pixel value.
mean_pixel = tf.reshape(MEAN_UINT_PIXEL, [1, 1, 3])
processed_image = utils.pad_to_bounding_box(processed_image, 0, 0,
target_height, target_width,
mean_pixel)
if label is not None:
label = utils.pad_to_bounding_box(label, 0, 0, target_height,
target_width, ignore_label)
# Randomly crop the image and label.
if is_training and label is not None:
processed_image, label = utils.random_crop([processed_image, label],
model_input_height,
model_input_width)
processed_image.set_shape([model_input_height, model_input_width, 3])
if label is not None:
label.set_shape([model_input_height, model_input_width, 1])
if is_training:
# Randomly left-right flip the image and label.
processed_image, label, _ = utils.flip_dim([processed_image, label],
_PROB_OF_FLIP,
dim=1)
processed_image = _preprocess_zero_mean_unit_range(processed_image)
return processed_image, label