def _test_grad_different_shape(self, input_shape, output_shape):
with self.test_session():
test_image_shape = input_shape
test_image = np.random.randn(*test_image_shape)
test_image_tensor = constant_op.constant(test_image,
shape=test_image_shape)
test_transform = image_ops.angles_to_projective_transforms(
np.pi / 2, 4, 4)
if len(output_shape) == 2:
resize_shape = output_shape
elif len(output_shape) == 3:
resize_shape = output_shape[0:2]
elif len(output_shape) == 4:
resize_shape = output_shape[1:3]
output = image_ops.transform(images=test_image_tensor,
transforms=test_transform,
output_shape=resize_shape)
left_err = gradient_checker.compute_gradient_error(
test_image_tensor,
test_image_shape,
output,
output_shape,
x_init_value=test_image)
self.assertLess(left_err, 1e-10)
def _test_grad_different_shape(self, input_shape, output_shape):
with self.cached_session():
test_image_shape = input_shape
test_image = np.random.randn(*test_image_shape)
test_image_tensor = constant_op.constant(
test_image, shape=test_image_shape)
test_transform = image_ops.angles_to_projective_transforms(
np.pi / 2, 4, 4)
if len(output_shape) == 2:
resize_shape = output_shape
elif len(output_shape) == 3:
resize_shape = output_shape[0:2]
elif len(output_shape) == 4:
resize_shape = output_shape[1:3]
output = image_ops.transform(
images=test_image_tensor,
transforms=test_transform,
output_shape=resize_shape)
left_err = gradient_checker.compute_gradient_error(
test_image_tensor,
test_image_shape,
output,
output_shape,
x_init_value=test_image)
self.assertLess(left_err, 1e-10)
def rotate(images, angles, interpolation="NEAREST"):
"""A variant of tensorflow.contrib.image.python.ops.image_ops.rotate
that only works on 4D tensors, without checking image rank.
when the shape of images is unknown, len(image_or_images.get_shape()) will raise error.
Rotate image(s) by the passed angle(s) in radians.
Args:
images: A tensor of shape (num_images, num_rows, num_columns, num_channels)
(NHWC), (num_rows, num_columns, num_channels) (HWC), or
(num_rows, num_columns) (HW).
angles: A scalar angle to rotate all images by, or (if images has rank 4)
a vector of length num_images, with an angle for each image in the batch.
interpolation: Interpolation mode. Supported values: "NEAREST", "BILINEAR".
Returns:
Image(s) with the same type and shape as `images`, rotated by the given
angle(s). Empty space due to the rotation will be filled with zeros.
Raises:
TypeError: If `image` is an invalid type.
"""
images = ops.convert_to_tensor(images, name="images")
image_height = math_ops.cast(array_ops.shape(images)[1],
dtypes.float32)[None]
image_width = math_ops.cast(array_ops.shape(images)[2],
dtypes.float32)[None]
output = transform_4d(images,
angles_to_projective_transforms(
angles, image_height, image_width),
interpolation=interpolation)
return output
def _test_grad(self, shape_to_test):
with self.test_session():
test_image_shape = shape_to_test
test_image = np.random.randn(*test_image_shape)
test_image_tensor = constant_op.constant(test_image,
shape=test_image_shape)
test_transform = image_ops.angles_to_projective_transforms(np.pi / 2,
4,
4)
test_transform_shape = test_transform.shape
output_shape = test_image_shape
output = image_ops.transform(test_image_tensor, test_transform)
left_err = gradient_checker.compute_gradient_error(
test_image_tensor, test_image_shape, output, output_shape,
x_init_value=test_image)
self.assertLess(left_err, 1e-10)
def _test_grad(self, shape_to_test):
with self.test_session():
test_image_shape = shape_to_test
test_image = np.random.randn(*test_image_shape)
test_image_tensor = constant_op.constant(test_image,
shape=test_image_shape)
test_transform = image_ops.angles_to_projective_transforms(
np.pi / 2, 4, 4)
output_shape = test_image_shape
output = image_ops.transform(test_image_tensor, test_transform)
left_err = gradient_checker.compute_gradient_error(
test_image_tensor,
test_image_shape,
output,
output_shape,
x_init_value=test_image)
self.assertLess(left_err, 1e-10)
def test_compose(self):
for dtype in _DTYPES:
with self.test_session():
image = constant_op.constant(
[[1, 1, 1, 0], [1, 0, 0, 0], [1, 1, 1, 0], [0, 0, 0, 0]],
dtype=dtype)
# Rotate counter-clockwise by pi / 2.
rotation = image_ops.angles_to_projective_transforms(
np.pi / 2, 4, 4)
# Translate right by 1 (the transformation matrix is always inverted,
# hence the -1).
translation = constant_op.constant([1, 0, -1, 0, 1, 0, 0, 0],
dtype=dtypes.float32)
composed = image_ops.compose_transforms(rotation, translation)
image_transformed = image_ops.transform(image, composed)
self.assertAllEqual(
image_transformed.eval(),
[[0, 0, 0, 0], [0, 1, 0, 1], [0, 1, 0, 1], [0, 1, 1, 1]])
def test_compose(self):
for dtype in _DTYPES:
with self.cached_session():
image = constant_op.constant(
[[1, 1, 1, 0],
[1, 0, 0, 0],
[1, 1, 1, 0],
[0, 0, 0, 0]], dtype=dtype)
# Rotate counter-clockwise by pi / 2.
rotation = image_ops.angles_to_projective_transforms(np.pi / 2, 4, 4)
# Translate right by 1 (the transformation matrix is always inverted,
# hence the -1).
translation = constant_op.constant([1, 0, -1,
0, 1, 0,
0, 0],
dtype=dtypes.float32)
composed = image_ops.compose_transforms(rotation, translation)
image_transformed = image_ops.transform(image, composed)
self.assertAllEqual(image_transformed.eval(),
[[0, 0, 0, 0],
[0, 1, 0, 1],
[0, 1, 0, 1],
[0, 1, 1, 1]])
