Peaksignalnoiseratiowithblockedeffect

peaksignalnoiseratiowithblockedeffect

Metric PeakSignalNoiseRatioWithBlockedEffect from torchmetrics (torchmetrics.image.PeakSignalNoiseRatioWithBlockedEffect)

When to invoke this skill

The user has predictions + ground truth and asks to evaluate with PeakSignalNoiseRatioWithBlockedEffect, or mentions torchmetrics.image.PeakSignalNoiseRatioWithBlockedEffect directly, or wants the standard torchmetrics implementation.

Reference signature

from torchmetrics.image import PeakSignalNoiseRatioWithBlockedEffect

# PeakSignalNoiseRatioWithBlockedEffect(data_range: Union[float, tuple[float, float]], block_size: int = 8, **kwargs: Any) -> None

Library docstring

Computes `Peak Signal to Noise Ratio With Blocked Effect`_ (PSNRB).

.. math::
    \text{PSNRB}(I, J) = 10 * \log_{10} \left(\frac{\max(I)^2}{\text{MSE}(I, J)-\text{B}(I, J)}\right)

Where :math:`\text{MSE}` denotes the `mean-squared-error`_ function. This metric is a modified version of PSNR that
better supports evaluation of images with blocked artifacts, that oftens occur in compressed images.

.. attention::
    Metric only supports grayscale images. If you have RGB images, please convert them to grayscale first.

As input to ``forward`` and ``update`` the metric accepts the following input

- ``preds`` (:class:`~torch.Tensor`): Predictions from model of shape ``(N,1,H,W)``
- ``target`` (:class:`~torch.Tensor`): Ground truth values of shape ``(N,1,H,W)``

As output of `forward` and `compute` the metric returns the following output

- ``psnrb`` (:class:`~torch.Tensor`): float scalar tensor with aggregated PSNRB value

Args:
    data_range: the range of the data. If a tuple is provided then the range is calculated as the difference and
        input is clamped between the values.
    block_size: integer indication the block size
    kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.

Example:
    >>> from torch import rand
    >>> metric = PeakSignalNoiseRatioWithBlockedEffect(data_range=1.0)
    >>> preds = rand(2, 1, 10, 10)
    >>> target = rand(2, 1, 10, 10)
    >>> metric(preds, target)
    tensor(7.2893)

Quick recipe

import torchmetrics.image as _m
score = _m.PeakSignalNoiseRatioWithBlockedEffect(y_true, y_pred)

Don'ts

• Don't reimplement when the library version handles edge cases (NaN, ties, empty inputs) better than a hand-rolled formula.
• Always check the library version's argument order — sklearn is (y_true, y_pred) while torchmetrics is (preds, target).