LION.models.post_processing.FBPConvNet

Classes

ConvBlock(channels[, relu_type, relu_last, ...])	Convolutions through the channels listed, each with batch normalisation and relu_type (the last one's activation optional).
Down()	Downscaling with maxpool
FBPConvNet(operator_params[, model_parameters])	FBPConvNet (Jin et al., 2017): a U-Net that removes the artefacts of an FBP reconstruction, learning the residual.
Up(channels[, stride, relu_type])	Upscaling with transpose conv

---

class LION.models.post_processing.FBPConvNet.ConvBlock(channels, relu_type='ReLU', relu_last=True, kernel_size=3)

Bases: Module

Convolutions through the channels listed, each with batch normalisation and relu_type (the last one’s activation optional).

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

forward(x)

The block’s output for x.

class LION.models.post_processing.FBPConvNet.Down

Bases: Module

Downscaling with maxpool

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

forward(x)

x halved in size (max pooling).

class LION.models.post_processing.FBPConvNet.FBPConvNet(operator_params, model_parameters=None)

Bases: LIONmodel

FBPConvNet (Jin et al., 2017): a U-Net that removes the artefacts of an FBP reconstruction, learning the residual. This class reconstructs with FDK itself, so it takes the measurements (CT only); FBPConvNetImage takes the reconstruction.

model_parameters (default_parameters()):

• down_1_channels to down_4_channels, latent_channels, up_1_channels to up_4_channels, last_block: each block’s channels, a list (its input, then each convolution’s output);

• activation: “ReLU”, “LeakyReLU” or “None”.

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

static default_parameters()

The paper’s U-Net: 64 to 1024 channels over five scales (see the class).

static cite(cite_format='bib')

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

forward(x)

The reconstructions of the sinograms x: FDK plus the network’s residual.

class LION.models.post_processing.FBPConvNet.Up(channels, stride=2, relu_type='ReLU')

Bases: Module

Upscaling with transpose conv

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

forward(x)

x doubled in size (transposed convolution).