LION.models.iterative_unrolled.LG

Classes

LG(operator_params[, model_parameters])	Learned Gradient (Adler and Öktem, 2017): n_iters updates of the image, each a CNN of the image, a memory, the data-fit gradient Aᵀ(y - A x) and a TV gradient, with learned step sizes.
LGblock(channels)	A learned-gradient block: 3x3 convolutions through the channels listed, with ReLUs between them.

---

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

Bases: LIONmodel

Learned Gradient (Adler and Öktem, 2017): n_iters updates of the image, each a CNN of the image, a memory, the data-fit gradient Aᵀ(y - A x) and a TV gradient, with learned step sizes. Starts from the initial reconstruction. Takes the measurements.

model_parameters (default_parameters()):

• channels: each block’s channels, a list: its input (the image, the memory, the two gradients) to its output (the memory, then the update); [8, 32, 32, 6];

• n_iters: the iterations (5).

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

static default_parameters()

5 iterations of blocks of 32 channels, a memory of 5 channels (see the class).

static cite(cite_format='bib')

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

gradientTVnorm2D(f)

The gradient of the (smoothed) isotropic total variation of the images f (B, C, H, W).

forward(x)

The reconstructions of the measurements x (B, 1, …).

class LION.models.iterative_unrolled.LG.LGblock(channels)

Bases: Module

A learned-gradient block: 3x3 convolutions through the channels listed, with ReLUs between them.

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

forward(x)

The block’s output for x.