def transform(img, scale, angle, bbox_center, output_shape):
tx = bbox_center[0] - output_shape[1] * scale / 2
ty = bbox_center[1] - output_shape[0] * scale / 2
# for offsetting translations caused by rotation:
# https://docs.opencv.org/2.4/modules/imgproc/doc/geometric_transformations.html
rx = (1 - tf.cos(angle)) * output_shape[1] * scale / 2 - tf.sin(angle) * output_shape[0] * scale / 2
ry = tf.sin(angle) * output_shape[1] * scale / 2 + (1 - tf.cos(angle)) * output_shape[0] * scale / 2
transform = [scale * tf.cos(angle), scale * tf.sin(angle), rx + tx,
-scale * tf.sin(angle), scale * tf.cos(angle), ry + ty,
0., 0.]
img = image_ops.transform(tf.expand_dims(img, axis=0),
tf.expand_dims(transform, axis=0),
fill_mode='constant',
output_shape=output_shape[:2])
img = tf.squeeze(img)
# transform for keypoints
alpha = 1 / scale * tf.cos(-angle)
beta = 1 / scale * tf.sin(-angle)
rx_xy = (1 - alpha) * bbox_center[0] - beta * bbox_center[1]
ry_xy = beta * bbox_center[0] + (1 - alpha) * bbox_center[1]
transform_xy = [[alpha, beta],
[-beta, alpha]]
tx_xy = bbox_center[0] - output_shape[1] / 2
ty_xy = bbox_center[1] - output_shape[0] / 2
M = tf.concat([transform_xy, [[rx_xy - tx_xy], [ry_xy - ty_xy]]], axis=1)
return img, M
def transform(image: tf.Tensor, transforms) -> tf.Tensor:
"""Prepares input data for `image_ops.transform`."""
original_ndims = tf.rank(image)
transforms = tf.convert_to_tensor(transforms, dtype=tf.float32)
if transforms.shape.rank == 1:
transforms = transforms[None]
image = to_4d(image)
image = image_ops.transform(
images=image, transforms=transforms, interpolation='nearest')
return from_4d(image, original_ndims)
def _run_random_transform_with_mock(self,
transform_matrix,
expected_output,
mode,
interpolation='bilinear'):
inp = np.arange(15).reshape((1, 5, 3, 1)).astype(np.float32)
with self.cached_session(use_gpu=True):
output = image_preprocessing.transform(
inp, transform_matrix, fill_mode=mode, interpolation=interpolation)
self.assertAllClose(expected_output, output)
def rotate(inputs):
"""rotate image."""
inputs_shape = array_ops.shape(inputs)
batch_size = inputs_shape[0]
img_hd = math_ops.cast(inputs_shape[1], dtypes.float32)
img_wd = math_ops.cast(inputs_shape[2], dtypes.float32)
min_angle = LOWER * 2. * np.pi
max_angle = UPPER * 2. * np.pi
angles = random_ops.random_uniform(
shape=[batch_size], minval=min_angle, maxval=max_angle)
return image_preprocessing.transform(
inputs, image_preprocessing.get_rotation_matrix(angles, img_hd, img_wd))
def zoom(inputs):
"""zoom image."""
inputs_shape = array_ops.shape(inputs)
batch_size = inputs_shape[0]
img_hd = math_ops.cast(inputs_shape[1], dtypes.float32)
img_wd = math_ops.cast(inputs_shape[2], dtypes.float32)
height_zoom = random_ops.random_uniform(
shape=[batch_size, 1], minval=1. + LOWER, maxval=1. + UPPER)
width_zoom = random_ops.random_uniform(
shape=[batch_size, 1], minval=1. + LOWER, maxval=1. + UPPER)
zooms = math_ops.cast(
array_ops.concat([width_zoom, height_zoom], axis=1), dtype=dtypes.float32)
return image_preprocessing.transform(
inputs, image_preprocessing.get_zoom_matrix(zooms, img_hd, img_wd))
