imresize_matlab

deepinv.physics.functional.imresize_matlab(x, scale=None, sizes=None, kernel='cubic', sigma=2, padding_type='reflect', antialiasing=True)

MATLAB imresize reimplementation.

A standalone PyTorch implementation for fast and efficient bicubic resampling. The resulting values are the same to MATLAB function imresize(‘bicubic’) with reflect padding.

Code reproduced with modifications from sanghyun-son/bicubic_pytorch

Parameters:

• x (torch.Tensor) – input tensor of shape (H,W), (C,H,W) or (B,C,H,W)

• scale (float) – imresize scale factor. > 1 = upsample.

• sizes (tuple) – optional output image size following MATLAB convention.

• kernel (str) – downsampling kernel, choose between ‘cubic’ (for MATLAB bicubic) or ‘gaussian’.

• sigma (float) – Gaussian kernel size. Ignored if kernel is not gaussian.

• padding_type (str) – reflect padding only.

• antialiasing (bool) – whether to perform antialiasing.

Returns:

torch.Tensor: output resized image.

Example:

>>> import torch
>>> from deepinv.physics.functional import imresize_matlab
>>> x = torch.randn(1, 1, 8, 8)
>>> y = imresize_matlab(x, scale=2)
>>> y.shape
torch.Size([1, 1, 16, 16])

Return type:

Tensor