def complex_center_crop(data_list, shape, offset=1, contiguous=False):
"""
Apply a center crop to the input data, or to a list of complex images
Parameters
----------
data_list : List[torch.Tensor] or torch.Tensor
The complex input tensor to be center cropped. It should have at least 3 dimensions
and the cropping is applied along dimensions didx and didx+1 and the last dimensions should have a size of 2.
shape : Tuple[int, int]
The output shape. The shape should be smaller than the corresponding dimensions of data.
If one value is None, this is filled in by the image shape.
offset : int
Starting dimension for cropping.
contiguous : bool
Return as a contiguous array. Useful for fast reshaping or viewing.
Returns
-------
torch.Tensor or list[torch.Tensor]: The center cropped input_image
"""
data_list = ensure_list(data_list)
assert_same_shape(data_list)
image_shape = list(data_list[0].shape)
ndim = data_list[0].ndim
bbox = [0] * ndim + image_shape
# Allow for False in crop directions
shape = [
_ if _ else image_shape[idx + offset] for idx, _ in enumerate(shape)
]
for idx in range(len(shape)):
bbox[idx + offset] = (image_shape[idx + offset] - shape[idx]) // 2
bbox[len(image_shape) + idx + offset] = shape[idx]
if not all([_ >= 0 for _ in bbox[:ndim]]):
raise ValueError(
f"Bounding box requested has negative values, "
f"this is likely to data size being smaller than the crop size. Got {bbox} with image_shape {image_shape} "
f"and requested shape {shape}.")
output = [crop_to_bbox(data, bbox) for data in data_list]
if contiguous:
output = [_.contiguous() for _ in output]
if len(output) == 1: # Only one element:
output = output[0]
return output
def complex_center_crop_previous(data_list, shape, didx=-3, contiguous=False):
"""
Apply a center crop to the input data, or to a list of complex images
Parameters_o
----------
data_list : List[torch.Tensor] or torch.Tensor
The complex input tensor to be center cropped. It should have at least 3 dimensions
and the cropping is applied along dimensions didx and didx+1 and the last dimensions should have a size of 2.
shape : Tuple[int, int]
The output shape. The shape should be smaller than the corresponding dimensions of data.
didx : int
Starting dimension for cropping.
contiguous : bool
Return as a contiguous array. Useful for fast reshaping or viewing.
Returns
-------
torch.Tensor or list[torch.Tensor]: The center cropped input_image
# TODO(jt): We can use crop_to_bbox here.
"""
data_list = ensure_list(data_list)
for data in data_list:
assert didx in [
-3, -2
], "Cropping needs to be done in the spatial dimensions."
assert 0 < shape[0] <= data.shape[didx]
assert 0 < shape[1] <= data.shape[didx + 1]
w_from = (data_list[0].shape[didx] - shape[0]) // 2
h_from = (data_list[0].shape[didx + 1] - shape[1]) // 2
w_to = w_from + shape[0]
h_to = h_from + shape[1]
if didx == -3:
output = [data[..., w_from:w_to, h_from:h_to, :] for data in data_list]
else:
output = [data[..., w_from:w_to, h_from:h_to] for data in data_list]
if contiguous:
output = [_.contiguous() for _ in output]
if len(output) == 1: # Only one element:
output = output[0]
return output
def complex_random_crop(data_list, shape, contiguous=False):
"""
Apply a random crop to the input data tensor or a list of complex.
Parameters
----------
data_list : Union[List[torch.Tensor], torch.Tensor]
The complex input tensor to be center cropped. It should have at least 3 dimensions and the cropping is applied
along dimensions -3 and -2 and the last dimensions should have a size of 2.
shape : Tuple[int, int]
The output shape. The shape should be smaller than the corresponding dimensions of data.
contiguous : bool
Return as a contiguous array. Useful for fast reshaping or viewing.
Returns
-------
torch.Tensor: The center cropped input tensor or list of tensors
"""
data_list = ensure_list(data_list)
# TODO: Check if all have same shape
for data in data_list:
assert 0 < shape[0] <= data.shape[-3]
assert 0 < shape[1] <= data.shape[-2]
w_from = np.random.randint(0, data_list[0].shape[-3] - shape[0] + 1)
h_from = np.random.randint(0, data_list[0].shape[-2] - shape[1] + 1)
w_to = w_from + shape[0]
h_to = h_from + shape[1]
output = [data[..., w_from:w_to, h_from:h_to, :] for data in data_list]
if contiguous:
output = [_.contiguous() for _ in output]
if len(output) == 1:
return output[0]
return output
def complex_random_crop(
data_list,
crop_shape,
offset: int = 1,
contiguous: bool = False,
sampler: str = "uniform",
sigma: bool = None,
):
"""
Apply a random crop to the input data tensor or a list of complex.
Parameters
----------
data_list : Union[List[torch.Tensor], torch.Tensor]
The complex input tensor to be center cropped. It should have at least 3 dimensions and the cropping is applied
along dimensions -3 and -2 and the last dimensions should have a size of 2.
crop_shape : Tuple[int, ...]
The output shape. The shape should be smaller than the corresponding dimensions of data.
offset : int
Starting dimension for cropping.
contiguous : bool
Return as a contiguous array. Useful for fast reshaping or viewing.
sampler : str
Select the random indices from either a `uniform` or `gaussian` distribution (around the center)
sigma : float or list of float
Standard variance of the gaussian when sampler is `gaussian`. If not set will take 1/3th of image shape
Returns
-------
torch.Tensor: The center cropped input tensor or list of tensors
"""
if sampler == "uniform" and sigma is not None:
raise ValueError(
f"sampler `uniform` is incompatible with sigma {sigma}, has to be None."
)
data_list = ensure_list(data_list)
assert_same_shape(data_list)
image_shape = list(data_list[0].shape)
ndim = data_list[0].ndim
bbox = [0] * ndim + image_shape
crop_shape = [
_ if _ else image_shape[idx + offset]
for idx, _ in enumerate(crop_shape)
]
crop_shape = np.asarray(crop_shape)
limits = []
for idx in range(len(crop_shape)):
limits.append(image_shape[offset + idx] - crop_shape[idx])
limits = np.asarray(limits)
if not all([_ >= 0 for _ in limits]):
raise ValueError(
f"Bounding box limits have negative values, "
f"this is likely to data size being smaller than the crop size. Got {limits}"
)
if sampler == "uniform":
lower_point = np.random.randint(0, limits + 1).tolist()
elif sampler == "gaussian":
data_shape = np.asarray(image_shape[offset:offset + len(crop_shape)])
if not sigma:
sigma = data_shape / 6 # w, h
if len(sigma) != 1 and len(sigma) != len(crop_shape):
raise ValueError(
f"Either one sigma has to be set or same as the length of the bounding box. Got {sigma}."
)
lower_point = (np.random.normal(
loc=data_shape / 2, scale=sigma, size=len(data_shape)) -
crop_shape / 2).astype(int)
lower_point = np.clip(lower_point, 0, limits)
else:
raise ValueError(
f"Sampler is either `uniform` or `gaussian`. Got {sampler}.")
for idx in range(len(crop_shape)):
bbox[offset + idx] = lower_point[idx]
bbox[offset + ndim + idx] = crop_shape[idx]
output = [crop_to_bbox(data, bbox) for data in data_list]
if contiguous:
output = [_.contiguous() for _ in output]
if len(output) == 1:
return output[0]
return output
