def apply_augmentations(
self, augmentations: List[Union[Augmentation,
Transform]]) -> TransformList:
"""
Apply a list of Transform/Augmentation in-place and returned the applied transform.
Attributes of this class will be modified.
Returns:
TransformList:
returns transformed inputs and the list of transforms applied.
The TransformList can then be applied to other data associated with the inputs.
"""
tfms = []
for aug in augmentations:
if isinstance(aug, Augmentation):
args = []
for f in aug.input_args:
try:
args.append(getattr(self, f))
except AttributeError:
raise AttributeError(
f"Augmentation {aug} needs '{f}', which is not an attribute of {self}!"
)
tfm = aug.get_transform(*args)
assert isinstance(tfm, Transform), (
f"{type(aug)}.get_transform must return an instance of Transform! "
"Got {type(tfm)} instead.")
else:
tfm = aug
self.transform(tfm)
tfms.append(tfm)
return TransformList(tfms)
def apply_transform_gens(transform_gens, img):
"""
Apply a list of :class:`TransformGen` on the input image, and
returns the transformed image and a list of transforms.
We cannot simply create and return all transforms without
applying it to the image, because a subsequent transform may
need the output of the previous one.
Args:
transform_gens (list): list of :class:`TransformGen` instance to
be applied.
img (ndarray): uint8 or floating point images with 1 or 3 channels.
Returns:
ndarray: the transformed image
TransformList: contain the transforms that's used.
"""
for g in transform_gens:
assert isinstance(g, TransformGen), g
check_dtype(img)
tfms = []
for g in transform_gens:
tfm = g.get_transform(img)
assert isinstance(
tfm, Transform
), "TransformGen {} must return an instance of Transform! Got {} instead".format(
g, tfm)
img = tfm.apply_image(img)
tfms.append(tfm)
return img, TransformList(tfms)
def apply_augmentations(augmentations, img):
"""
Apply a list of :class:`Augmentation` or :class:`Transform` on the input image, and
returns the transformed image and a list of transforms.
We cannot simply create and return all transforms without
applying it to the image, because a subsequent transform may
need the output of the previous one.
Args:
augmentations (list): list of :class:`Augmentation` or :class:`Transform` instance to
be applied.
img (ndarray): uint8 or floating point images with 1 or 3 channels.
Returns:
ndarray: the transformed image
TransformList: contain the transforms that's used.
"""
for aug in augmentations:
assert isinstance(aug, (Transform, Augmentation)), aug
check_dtype(img)
tfms = []
for aug in augmentations:
tfm = aug.get_transform(img) if isinstance(aug, Augmentation) else aug
assert isinstance(
tfm, Transform
), f"Augmentation {aug} must return an instance of Transform! Got {tfm} instead."
img = tfm.apply_image(img)
tfms.append(tfm)
return img, TransformList(tfms)
def inverse(self):
if not self.expand:
raise NotImplementedError()
rotation = RotationTransform(self.bound_h, self.bound_w, -self.angle,
True, None, self.interp)
crop = CropTransform((rotation.bound_w - self.w) // 2,
(rotation.bound_h - self.h) // 2, self.w, self.h)
return TransformList([rotation, crop])
def __call__(self, aug_input) -> Transform:
tfms = []
for x in self.augs:
# 序列执行
tfm = x(aug_input)
# tfm is a instance of Transform
tfms.append(tfm)
# 返回Transorm 是可以用于处理User 需要处理的别的东西
return TransformList(tfms)
def inverse(self):
"""
The inverse is to rotate it back with expand, and crop to get the original shape.
"""
if not self.expand: # Not possible to inverse if a part of the image is lost
raise NotImplementedError()
rotation = RotationTransform(self.bound_h, self.bound_w, -self.angle,
True, None, self.interp)
crop = CropTransform((rotation.bound_w - self.w) // 2,
(rotation.bound_h - self.h) // 2, self.w, self.h)
return TransformList([rotation, crop])
def apply_transform_gens(transform_gens, img):
for g in transform_gens:
assert isinstance(g, (Transform, TransformGen)), g
check_dtype(img)
tfms = []
for g in transform_gens:
tfm = g.get_transform(img) if isinstance(g, TransformGen) else g
assert isinstance(tfm, Transform), \
f"TransformGen {g} must return an instance of Transform! Got {tfm} instead"
img = tfm.apply_image(img)
tfms.append(tfm)
return img, TransformList(tfms)
def get_transform(self, image: np.ndarray) -> TransformList:
# Compute the image scale and scaled size.
input_size = image.shape[:2]
output_size = self.crop_size
# Add random crop if the image is scaled up.
max_offset = np.subtract(input_size, output_size)
max_offset = np.maximum(max_offset, 0)
offset = np.multiply(max_offset, np.random.uniform(0.0, 1.0))
offset = np.round(offset).astype(int)
crop_transform = CropTransform(offset[1], offset[0], output_size[1],
output_size[0], input_size[1],
input_size[0])
# Add padding if the image is scaled down.
pad_size = np.subtract(output_size, input_size)
pad_size = np.maximum(pad_size, 0)
original_size = np.minimum(input_size, output_size)
pad_transform = PadTransform(0, 0, pad_size[1], pad_size[0],
original_size[1], original_size[0],
self.pad_value)
return TransformList([crop_transform, pad_transform])
def __call__(self, aug_input) -> Transform:
tfms = []
for x in self.augs:
tfm = x(aug_input)
tfms.append(tfm)
return TransformList(tfms)
def get_transform(self, image: np.ndarray) -> TransformList:
transforms = [self._get_crop(image)]
if self.pad:
transforms.append(self._get_pad(image))
return TransformList(transforms)
