Module atomic 

Atomic, domain-agnostic metrics for comparing scalar intensity images.

Inputs use reference and candidate terminology. Signed quantities use the residual candidate - reference; valid_mask, when supplied, includes pixels whose value is true. These functions are for evaluation, not optimization objectives.

Enums

IntensityMetricError

Errors returned by atomic intensity metrics.

Functions

amplitude_nrmse

Compare square-root intensities, normalized by the reference amplitude L2 norm.

bias

Return the mean signed residual, candidate - reference.

correlation

Return the Pearson correlation coefficient of valid pixels.

fitted_gain

Fit the least-squares scalar gain in candidate ≈ gain × reference.

mae

Return the mean absolute error.

mean_poisson_deviance

Return mean Poisson deviance, flooring candidate intensities at epsilon.

mse

Return the mean squared error.

nrmse

Return ||candidate - reference||_2 / ||reference||_2.

poisson_deviance

Return summed Poisson deviance, flooring candidate intensities at epsilon.

psnr

Return peak signal-to-noise ratio in dB for an explicit intensity range.

relative_l1

Return sum(abs(candidate - reference)) / sum(abs(reference)).

rmse

Return the root mean squared error.

ssim

Return canonical single-scale SSIM using an 11×11 Gaussian window (σ=1.5).