torch_staintools.functional.utility package

Submodules

torch_staintools.functional.utility.implementation module

torch_staintools.functional.utility.implementation.default_device(device: device | None = None) → device | None

Default device if device is not given.

Parameters:

device – input device.

Returns:

torch.device(‘cpu’) if None, otherwise the input device itself.

torch_staintools.functional.utility.implementation.default_rng(rng: Generator | int | None, device: device | None) → Generator | None

Helper function to get the default random number generator (torch.Generator)

Parameters:

  • rng – Optional. int seed or torch.Generator. If not set (None) then return None. Identity mapping if input is already a generator. Create a new generator and specify the seed if an int seed is given.

  • device – device of the rng

Returns:

torch.Generator

torch_staintools.functional.utility.implementation.img_from_concentration(concentration: Tensor, stain_matrix: Tensor, img_shape: Tuple[int, ...], out_range: Tuple[float, float] = (0, 1))

reconstruct image from concentration and stain matrix to RGB

Parameters:

  • concentration – B x (HW) x num_stain

  • stain_matrix – B x num_stain x input channel

  • img_shape –

  • out_range –

Returns:

torch_staintools.functional.utility.implementation.nanstd(data: Tensor, dim: int | tuple | None = None, correction: float = 1) → Tensor

Compute the standard deviation while ignoring NaNs.

Always keep the dim.

Parameters:

  • data – Input tensor.

  • dim – The dimension or dimensions to reduce. If None (default), reduces all dimensions.

  • correction – Difference between the sample size and sample degrees of freedom. Defaults 1 (Bessel’s).

Returns:

Standard deviation with NaNs ignored. If dim is provided, it reduces along that dimension(s), otherwise reduces all dimensions.

Return type:

torch.Tensor

torch_staintools.functional.utility.implementation.transpose_trailing(mat: Tensor)

Helper function to transpose the trailing dimension, since data is batchified.

Parameters:

mat – input tensor ixjxk

Returns:

output with flipped dimension from ixjxk –> ixkxj

Module contents