LION.models.iterative_unrolled.ItNet

Classes

ItNet(operator_params[, model_parameters])

ItNet (Genzel, Macdonald and März, 2022): n_iters iterations of a U-Net followed by a data-consistency step, x - step * FDK(A x - y), starting from FDK.

---

class LION.models.iterative_unrolled.ItNet.ItNet(operator_params, model_parameters=None)

Bases: LIONmodel

ItNet (Genzel, Macdonald and März, 2022): n_iters iterations of a U-Net followed by a data-consistency step, x - step * FDK(A x - y), starting from FDK. CT only (it uses FDK). Takes the measurements.

model_parameters (default_parameters()):

• n_iters: the iterations (4);

• Unet_params: each iteration’s U-Net (UNet.default_parameters());

• step_size: each iteration’s step (one value, a list of n_iters, or None: 1);

• learned_step: learn the steps (True);

• step_positive: learn them as powers of 10, so positive (False).

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

static default_parameters()

4 iterations with the paper’s step sizes, learned (see the class).

static cite(cite_format='bib')

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

forward(sino)

The reconstructions of the sinograms sino (B, 1, angles, detectors).