spyrit.core.torch.walsh2_torch

spyrit.core.torch.walsh2_torch(img, H=None)[source]

Return 2D Walsh-ordered Hadamard transform of an image

This applies the 1D transform \(H \in \mathbb{R}^{n \times n}\) to the rows and to the columns of batches of images \(X\in \mathbb{R}^{n \times n}\)

\[Y = H X H^T.\]

Args:

img (torch.tensor): Batch of images \(X\) with shape \((*,n,n)\).

H (torch.tensor, optional): 1D Walsh-ordered Hadamard matrix with shape \((n,n)\).

Returns:

torch.tensor: Transformed image \(Y\) with shape \((*, n, n)\) where \(*\) is the same number as for img.

See Also:

fwht_2d() implements the same transform with a different algorithm.

Example:

Example 1: Basic example

>>> img = torch.randn(256, 1, 64, 64)
>>> had = walsh2_torch(img)

Example 2: Same on CPU

>>> img = torch.randn(256, 1, 64, 64)
>>> img = img.to(device='cpu')
>>> had = walsh2_torch(img)
>>> print(had.device)
cpu

Example 3: On GPU using torch.float64

>>> img = torch.randn(256, 1, 64, 64)
>>> img = img.to(device='cpu', dtype=torch.float64)
>>> had = walsh2_torch(img)
>>> print(had.device,'+',had.dtype)
cpu + torch.float64