partition_padded

dnois.utils.partition_padded(image: Tensor, n_patches: int | tuple[int, int], padding: int | tuple[int, int] = 0, mode: str = 'constant', value: float | int = 0, sequential: bool = False) → list[Tensor] | list[list[Tensor]]

Partition an image into patches, each of which is padded with pixels from neighbouring patches. Paddings of marginal patches depend on the parameters mode and value.

Parameters:

  • image (Tensor) – One or more images of shape … x H x W.

  • n_patches (int | tuple[int, int]) – Number of patches in vertical and horizontal direction.

  • padding (int | tuple[int, int]) – Padding width in vertical and horizontal direction.

  • mode (str) – See :func:torch.nn.functional.pad.

  • value (int | float) – See :func:torch.nn.functional.pad.

  • sequential (bool) – Whether to arrange all patches into one dimension.

Returns:

Image patches of shape patches[0] x patches[1] x … x PH x PW or (patches[0] x patches[1]) x … x PH x PW, depending on sequential wherein PH and PW are height and width of each patch, respectively.

Return type:

list[Tensor] if sequential is True, list[list[Tensor]] otherwise.