LION.losses.SURE

Classes

SURE(noise_std[, epsilon])

Stein's unbiased risk estimate as a self-supervised loss for denoisers of Gaussian noise of standard deviation noise_std (Metzler et al., 2018), its divergence estimated by Monte Carlo with a perturbation epsilon (default_epsilon if None).

---

class LION.losses.SURE.SURE(noise_std, epsilon=None)

Bases: Module

Stein’s unbiased risk estimate as a self-supervised loss for denoisers of Gaussian noise of standard deviation noise_std (Metzler et al., 2018), its divergence estimated by Monte Carlo with a perturbation epsilon (default_epsilon if None). Not usable yet: it raises, as CT noise isn’t Gaussian (issue #144).

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(noisy, model)

The loss of the denoiser model on the noisy images.

static default_epsilon(y)

The default perturbation: the maximum of y / 1000.

static cite(cite_format='bib')

Where SURE comes from: its citation, BibTeX (“bib”) or “MLA”.