LION.models.learned_regularizer.AR

Classes

AR(operator_params[, model_parameters])

Adversarial regulariser (Lunz, Öktem and Schönlieb, 2018): a CNN mapping an image to a number, low for clean images and high for bad reconstructions, trained by ARSolver and used by LearnedRegularizerReconstructor.

---

class LION.models.learned_regularizer.AR.AR(operator_params, model_parameters=None)

Bases: LIONmodel

Adversarial regulariser (Lunz, Öktem and Schönlieb, 2018): a CNN mapping an image to a number, low for clean images and high for bad reconstructions, trained by ARSolver and used by LearnedRegularizerReconstructor. Six 5x5 convolutions (four strided) and two fully connected layers, as in the paper. Takes images of the operator’s domain shape (its size sets the fully connected layer).

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

forward(image)

The regulariser’s value for each image of image (B, 1, H, W): (B, 1).

static default_parameters()

None of its own: the architecture is the paper’s.

static cite(cite_format='bib')

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