Skip to content

Let LinearTransformation return datapoints instead of tensors #7244

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Feb 14, 2023
Merged

Let LinearTransformation return datapoints instead of tensors #7244

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
59b301c
Let LinearTransformation return datapoints instead of tensors
NicolasHug Feb 14, 2023
243a5d0
Any
NicolasHug Feb 14, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 10 additions & 11 deletions torchvision/prototype/transforms/_misc.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,12 +76,7 @@ def _check_inputs(self, sample: Any) -> Any:
if has_any(sample, PIL.Image.Image):
raise TypeError("LinearTransformation does not work on PIL Images")

def _transform(
self, inpt: Union[datapoints.TensorImageType, datapoints.TensorVideoType], params: Dict[str, Any]
) -> torch.Tensor:
# Image instance after linear transformation is not Image anymore due to unknown data range
# Thus we will return Tensor for input Image

def _transform(self, inpt: Any, params: Dict[str, Any]) -> Any:
shape = inpt.shape
n = shape[-3] * shape[-2] * shape[-1]
if n != self.transformation_matrix.shape[0]:
Expand All @@ -97,11 +92,15 @@ def _transform(
f"Got {inpt.device} vs {self.mean_vector.device}"
)

flat_tensor = inpt.reshape(-1, n) - self.mean_vector
flat_inpt = inpt.reshape(-1, n) - self.mean_vector

transformation_matrix = self.transformation_matrix.to(flat_inpt.dtype)
output = torch.mm(flat_inpt, transformation_matrix)
output = output.reshape(shape)

transformation_matrix = self.transformation_matrix.to(flat_tensor.dtype)
transformed_tensor = torch.mm(flat_tensor, transformation_matrix)
return transformed_tensor.reshape(shape)
if isinstance(inpt, (datapoints.Image, datapoints.Video)):
output = type(inpt).wrap_like(inpt, output) # type: ignore[arg-type]
return output


class Normalize(Transform):
Expand All @@ -120,7 +119,7 @@ def _check_inputs(self, sample: Any) -> Any:

def _transform(
self, inpt: Union[datapoints.TensorImageType, datapoints.TensorVideoType], params: Dict[str, Any]
) -> torch.Tensor:
) -> Any:
return F.normalize(inpt, mean=self.mean, std=self.std, inplace=self.inplace)


Expand Down