piqa.mdsi#

Mean Deviation Similarity Index (MDSI)

This module implements the MDSI in PyTorch.

Original

https://www.mathworks.com/matlabcentral/fileexchange/59809-mdsi-ref-dist-combmethod

References

Mean Deviation Similarity Index: Efficient and Reliable Full-Reference Image Quality Evaluator (Nafchi et al., 2016)
https://arxiv.org/abs/1608.07433

Functions#

mdsi

Returns the MDSI between \(x\) and \(y\), without color space conversion and downsampling.

Classes#

MDSI

Measures the MDSI between an input and a target.

Descriptions#

piqa.mdsi.mdsi(x, y, kernel, combination='sum', value_range=1.0, c1=0.0021530180699730873, c2=0.0008458285274894271, c3=0.008458285274894272, alpha=0.6, beta=0.1, gamma=0.2, rho=1.0, q=0.25, o=0.25)#

Returns the MDSI between \(x\) and \(y\), without color space conversion and downsampling.

Parameters:

  • x (Tensor) – An input tensor, \((N, 3, H, W)\).

  • y (Tensor) – A target tensor, \((N, 3, H, W)\).

  • kernel (Tensor) – A gradient kernel, \((2, 1, K, K)\).

  • combination (str) – Specifies the scheme to combine the gradient and chromaticity similarities (GS, CS): 'sum' or 'prod'.

  • value_range (float) – The value range \(L\) of the inputs (usually 1 or 255).

Note

For the remaining arguments, refer to Nafchi et al. (2016).

Returns:

The MDSI vector, \((N,)\).

Return type:

Tensor

Example

>>> x = torch.rand(5, 1, 256, 256)
>>> y = torch.rand(5, 1, 256, 256)
>>> kernel = gradient_kernel(prewitt_kernel())
>>> l = mdsi(x, y, kernel)
>>> l.shape
torch.Size([5])
class piqa.mdsi.MDSI(downsample=True, kernel=None, reduction='mean', **kwargs)#

Measures the MDSI between an input and a target.

Before applying mdsi, the input and target are converted from RBG to LHM and downsampled to a 256-ish resolution.

Parameters:

  • downsample (bool) – Whether downsampling is enabled or not.

  • kernel (Tensor) – A gradient kernel, \((2, 1, K, K)\). If None, use the Prewitt kernel instead.

  • reduction (str) – Specifies the reduction to apply to the output: 'none', 'mean' or 'sum'.

  • kwargs – Keyword arguments passed to mdsi.

Example

>>> criterion = MDSI()
>>> x = torch.rand(5, 3, 256, 256, requires_grad=True)
>>> y = torch.rand(5, 3, 256, 256)
>>> l = criterion(x, y)
>>> l.shape
torch.Size([])
>>> l.backward()
forward(x, y)#
Parameters:

  • x (Tensor) – An input tensor, \((N, 3, H, W)\).

  • y (Tensor) – A target tensor, \((N, 3, H, W)\).

Returns:

The MDSI vector, \((N,)\) or \(()\) depending on reduction.

Return type:

Tensor