def compute_transformation(self, input: Tensor, params: Dict[str, Tensor]) -> Tensor:
if self.flags["cropping_mode"] == "resample":
transform: Tensor = get_perspective_transform(params["src"].to(input), params["dst"].to(input))
return transform
if self.flags["cropping_mode"] == "slice": # Skip the computation for slicing.
return self.identity_matrix(input)
raise NotImplementedError(f"Not supported type: {self.flags['cropping_mode']}.")
def compute_transformation(self, input: Tensor,
params: Dict[str, Tensor]) -> Tensor:
if params["output_size"] == input.shape[-2:]:
return eye_like(3, input)
transform: Tensor = get_perspective_transform(params["src"],
params["dst"])
transform = transform.expand(input.shape[0], -1, -1)
return transform
def apply_perspective(input: torch.Tensor,
params: Dict[str, torch.Tensor],
return_transform: bool = False) -> UnionType:
r"""Perform perspective transform of the given torch.Tensor or batch of tensors.
Args:
input (torch.Tensor): Tensor to be transformed with shape BxCxHxW.
start_points (torch.Tensor): Tensor containing [top-left, top-right, bottom-right,
bottom-left] of the orignal image with shape Bx4x2.
end_points (torch.Tensor): Tensor containing [top-left, top-right, bottom-right,
bottom-left] of the transformed image with shape Bx4x2.
return_transform (bool): if ``True`` return the matrix describing the transformation
applied to each. Default: False.
Returns:
torch.Tensor: Perspectively transformed tensor.
"""
if not torch.is_tensor(input):
raise TypeError(f"Input type is not a torch.Tensor. Got {type(input)}")
device: torch.device = input.device
dtype: torch.dtype = input.dtype
# arrange input data
x_data: torch.Tensor = input.view(-1, *input.shape[-3:])
batch_size, _, height, width = x_data.shape
# compute the homography between the input points
transform: torch.Tensor = get_perspective_transform(
params['start_points'], params['end_points']).to(device, dtype)
out_data: torch.Tensor = x_data.clone()
# process valid samples
mask = params['batch_prob'].to(device)
# TODO: look for a workaround for this hack. In CUDA it fails when no elements found.
if bool(mask.sum() > 0):
# apply the computed transform
height, width = x_data.shape[-2:]
out_data[mask] = warp_perspective(x_data[mask], transform[mask],
(height, width))
if return_transform:
return out_data.view_as(input), transform
return out_data.view_as(input)
def compute_transformation(self, input: Tensor,
params: Dict[str, Tensor]) -> Tensor:
return get_perspective_transform(params["start_points"].to(input),
params["end_points"].to(input))