meddlr.metrics

Meddlr metrics extend the torchmetrics interface to support tracking and reporting metrics on a per-example, per-channel basis. This can be useful for more nuanced model monitoring, such as determining which set of examples is the model underperforming on or which categories have the least accuracy.

Because these metrics are based on the torchmetrics interface, they are compatible everywhere where torchmetrics metrics are used.

Base

meddlr.metrics.Metric	Interface for new metrics.
meddlr.metrics.MetricCollection	The class that manages multiple metrics.

Image Metrics

meddlr.metrics.MSE	Mean squared error with complex-valued support.
meddlr.metrics.NRMSE	Normalized root-mean-squared error with complex-valued support.
meddlr.metrics.PSNR	Peak signal-to-noise ratio with complex-valued support.
meddlr.metrics.RMSE	Root-mean-squared error with complex-valued support.
meddlr.metrics.SSIM	Structural similarity index measure with complex-valued support.

Segmentation Metrics

meddlr.metrics.ASSD	Average symmetric surface distance.
meddlr.metrics.CV	Coefficient of variation.
meddlr.metrics.DSC	Dice score coefficient.
meddlr.metrics.VOE	Volumetric overlap error.

Functional

All metrics in Meddlr are also available as functions in the meddlr.metrics.functional module.

meddlr.metrics.functional.mse	Computes mean square error.
meddlr.metrics.functional.nrmse	Computes normalized root mean squared error.
meddlr.metrics.functional.psnr	Computes peak signal-to-noise ratio.
meddlr.metrics.functional.rmse	Computes root mean square error.
meddlr.metrics.functional.ssim	Computes structural similarity index (SSIM).
meddlr.metrics.functional.assd	Computes average symmetric surface distance.
meddlr.metrics.functional.cv	Computes coefficient of variation.
meddlr.metrics.functional.dice	Computes dice score coefficient.
meddlr.metrics.functional.voe	Computes volumetric overlap error.