HaarPSI#

class deepinv.loss.metric.HaarPSI(C=5.0, alpha=4.9, preprocess_with_subsampling=True, **kwargs)[source]#

Bases: Metric

HaarPSI metric with tuned parameters.

The metric was proposed by Reisenhofer et al. and the parameters are taken from Karner et al.. The metric computes similarities in the Haar wavelet domain and it is shown to closely match human evaluation. See original papers for more details. The metric range is \([0,1]\). The higher the metric, the better.

Code is adapted from this implementation by Sören Dittmer, Clemens Karner and Anna Breger, adapted from David Neumann, adapted from Rafael Reisenhofer.

Note

Images must be scaled to \([0,1]\). You can use norm_inputs = clip or min_max to achieve this.

The parameters should be set as follows depending on the image domain:

  • Natural images: \(C=30,\alpha=4.2\).

  • Medical images: \(C=5,\alpha=4.9\).

Note

By default, no reduction is performed in the batch dimension.

Example:

>>> import torch
>>> from deepinv.loss.metric import HaarPSI
>>> m = HaarPSI(norm_inputs="clip")
>>> x_net = x = torch.ones(3, 1, 8, 8) # B,C,H,W
>>> m(x_net, x)
tensor([1.0000, 1.0000, 1.0000])
Parameters:

  • C (float) – metric parameter \(C\in[5, 100]\).

  • alpha (float) – metric paramter \(\alpha\in[2, 8]\).

  • preprocess_with_subsampling (bool) – Determines if subsampling is performed.

  • complex_abs (bool) – perform complex magnitude before passing data to metric function. If True, the data must either be of complex dtype or have size 2 in the channel dimension (usually the second dimension after batch).

  • reduction (str) – a method to reduce metric score over individual batch scores. mean: takes the mean, sum takes the sum, none or None no reduction will be applied (default).

  • norm_inputs (str) – normalize images before passing to metric. l2 normalizes by \(\ell_2\) spatial norm, min_max normalizes by min and max of each input.

  • center_crop (int, tuple[int], None) – If not None (default), center crop the tensor(s) before computing the metrics. If an int is provided, the cropping is applied equally on all spatial dimensions (by default, all dimensions except the first two). If tuple of int, cropping is performed over the last len(center_crop) dimensions. If positive values are provided, a standard center crop is applied. If negative (or zero) values are passed, cropping will be done by removing center_crop pixels from the borders (useful when tensors vary in size across the dataset).

On this page

This Page