kaira.metrics.image.PeakSignalNoiseRatio

Inheritance diagram of PeakSignalNoiseRatio

Inheritance diagram for PeakSignalNoiseRatio

class kaira.metrics.image.PeakSignalNoiseRatio(data_range: float = 1.0, reduction: str | None = None, *args: Any, **kwargs: Any)[source]

Bases: BaseMetric

Peak Signal-to-Noise Ratio (PSNR) Module.

PSNR measures the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the quality of its representation. Higher values indicate better quality [Hore and Ziou, 2010]. While PSNR doesn’t perfectly correlate with human perception, it is widely used for its simplicity and clear physical meaning [Wang and Bovik, 2009].

Methods

__init__

Initialize the PeakSignalNoiseRatio module.

compute_with_stats

Compute PSNR with mean and standard deviation.

forward

Calculate PSNR between predicted and target images.
__init__(data_range: float = 1.0, reduction: str | None = None, *args: Any, **kwargs: Any) → None[source]

Initialize the PeakSignalNoiseRatio module.

Parameters:

  • data_range (float) – The range of the input data (typically 1.0 or 255)

  • reduction (Optional[str]) – Reduction method. The underlying torchmetrics implementation requires reduction=None, so this parameter controls post-processing reduction.

  • *args – Variable length argument list passed to the base class and torchmetrics.

  • **kwargs – Arbitrary keyword arguments passed to the base class and torchmetrics.

forward(x: Tensor, y: Tensor, *args: Any, **kwargs: Any) → Tensor[source]

Calculate PSNR between predicted and target images.

Parameters:

  • x (Tensor) – Predicted images

  • y (Tensor) – Target images

  • *args – Variable length argument list (currently unused).

  • **kwargs – Arbitrary keyword arguments (currently unused).

Returns:

PSNR values for each sample or reduced according to reduction parameter

Return type:

Tensor

compute_with_stats(x: Tensor, y: Tensor, *args: Any, **kwargs: Any) → Tuple[Tensor, Tensor][source]

Compute PSNR with mean and standard deviation.

Parameters:

  • x (Tensor) – Predicted images

  • y (Tensor) – Target images

  • *args – Variable length argument list (currently unused).

  • **kwargs – Arbitrary keyword arguments (currently unused).

Returns:

Mean and standard deviation of PSNR values

Return type:

Tuple[Tensor, Tensor]