direct.nn.lpd package

Submodules

direct.nn.lpd.config module

class direct.nn.lpd.config.LPDNetConfig(model_name: str = '???', engine_name: str | None = None, num_iter: int = 25, num_primal: int = 5, num_dual: int = 5, primal_model_architecture: str = 'MWCNN', dual_model_architecture: str = 'DIDN', primal_mwcnn_hidden_channels: int = 16, primal_mwcnn_num_scales: int = 4, primal_mwcnn_bias: bool = True, primal_mwcnn_batchnorm: bool = False, primal_unet_num_filters: int = 8, primal_unet_num_pool_layers: int = 4, primal_unet_dropout_probability: float = 0.0, dual_conv_hidden_channels: int = 16, dual_conv_n_convs: int = 4, dual_conv_batchnorm: bool = False, dual_didn_hidden_channels: int = 64, dual_didn_num_dubs: int = 6, dual_didn_num_convs_recon: int = 9, dual_unet_num_filters: int = 8, dual_unet_num_pool_layers: int = 4, dual_unet_dropout_probability: float = 0.0)

Bases: ModelConfig

dual_conv_batchnorm: bool = False

dual_conv_hidden_channels: int = 16

dual_conv_n_convs: int = 4

dual_didn_hidden_channels: int = 64

dual_didn_num_convs_recon: int = 9

dual_didn_num_dubs: int = 6

dual_model_architecture: str = 'DIDN'

dual_unet_dropout_probability: float = 0.0

dual_unet_num_filters: int = 8

dual_unet_num_pool_layers: int = 4

num_dual: int = 5

num_iter: int = 25

num_primal: int = 5

primal_model_architecture: str = 'MWCNN'

primal_mwcnn_batchnorm: bool = False

primal_mwcnn_bias: bool = True

primal_mwcnn_hidden_channels: int = 16

primal_mwcnn_num_scales: int = 4

primal_unet_dropout_probability: float = 0.0

primal_unet_num_filters: int = 8

primal_unet_num_pool_layers: int = 4

direct.nn.lpd.lpd module

class direct.nn.lpd.lpd.DualNet(num_dual, **kwargs)

Bases: Module

Dual Network for Learned Primal Dual Network.

static compute_model_per_coil(model, data)

Computes model per coil.

Parameters:

model: nn.Module

Model to compute.

data: torch.Tensor

Multi-coil input.

Returns:

output: torch.Tensor

Multi-coil output.

Return type:

Tensor

forward(h, forward_f, g)

Define the computation performed at every call.

Should be overridden by all subclasses. :rtype: Tensor

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class direct.nn.lpd.lpd.LPDNet(forward_operator, backward_operator, num_iter, num_primal, num_dual, primal_model_architecture='MWCNN', dual_model_architecture='DIDN', **kwargs)

Bases: Module

Learned Primal Dual network implementation inspired by [1].

References

[1]

Adler, Jonas, and Ozan Öktem. “Learned Primal-Dual Reconstruction.” IEEE Transactions on Medical Imaging, vol. 37, no. 6, June 2018, pp. 1322–32. arXiv.org, https://doi.org/10.1109/TMI.2018.2799231.

forward(masked_kspace, sensitivity_map, sampling_mask)

Computes forward pass of LPDNet.

Parameters:

masked_kspace: torch.Tensor

Masked k-space of shape (N, coil, height, width, complex=2).

sensitivity_map: torch.Tensor

Sensitivity map of shape (N, coil, height, width, complex=2).

sampling_mask: torch.Tensor

Sampling mask of shape (N, 1, height, width, 1).

Returns:

output: torch.Tensor

Output image of shape (N, height, width, complex=2).

Return type:

Tensor

class direct.nn.lpd.lpd.PrimalNet(num_primal, **kwargs)

Bases: Module

Primal Network for Learned Primal Dual Network.

forward(f, backward_h)

Define the computation performed at every call.

Should be overridden by all subclasses. :rtype: Tensor

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

direct.nn.lpd.lpd_engine module

class direct.nn.lpd.lpd_engine.LPDNetEngine(cfg, model, device, forward_operator=None, backward_operator=None, mixed_precision=False, **models)

Bases: MRIModelEngine

LPDNet Engine.

forward_function(data)

This method performs the model’s forward method given data which contains all tensor inputs.

Must be implemented by child classes.

Return type:

Tuple[Tensor, None]

Module contents