model

Submodules

model

class segmentation_rt.dl.model.model.Model(expr_dir, structures, seed=None, batch_size=None, epoch_count=1, niter=150, niter_decay=50, beta1=0.5, lr=0.0002, ngf=64, n_blocks=9, input_nc=1, use_dropout=False, norm='batch', max_grad_norm=500.0, monitor_grad_norm=True, save_epoch_freq=2, print_freq=60, display_epoch_freq=1, testing=False, resume=False)

Bases: object

Initialize a chosen model with parameters.

This class allows the training of the networks and afterward for testing.Only resnet available, more coming soon.

Parameters

  • expr_dir (str) – output folder.

  • seed (int) – manual seed.

  • batch_size (int) – input batch size.

  • epoch_count (int) – the starting epoch count.

  • niter (int) – # of iter at starting learning rate.

  • niter_decay (int) – # of iter to linearly decay learning rate to zero.

  • beta1 (float) – momentum term of adam.

  • lr (float) – initial learning rate for adam.

  • ngf (int) – # of gen filters in first conv layer.

  • n_blocks (int) – # of residual blocks in the global generator network.

  • input_nc (int) – the number of channels in input images.

  • ngf – the number of filters in the last conv layer.

  • n_blocks – the number of ResNet blocks.

  • use_dropout (bool) – if use dropout layers.

  • norm (str) – normalization.

  • max_grad_norm (float) – max grad norm to which it will be clipped (if exceeded).

  • monitor_grad_norm (bool) – flag set to monitor grad norms.

  • save_epoch_freq (int) – frequency of saving checkpoints at the end of epochs.

  • print_freq (int) – frequency of showing training results on console.

  • testing (bool) – if test phase.

eval()

Sets the module in evaluation mode.

load(checkpoint_path, optimizer=False)

Loads an object saved with torch.save from a file.

Parameters

  • checkpoint_path (str) – path to the checkpoint.

  • optimizer (bool) – with optimizer.

save(checkpoint_name)

Save the model and optimizer.

Parameters

checkpoint_name (str) – name of the checkpoint.

save_options()

Save model options.

test(dataset, export_path=None, checkpoint=None, save=False)

Model prediction for a SingleDataset.

Parameters

  • dataset (DataLoader) – training dataset.

  • export_path (str) – export path.

Returns

MAE of the dataset.

Return type

float

train(train_dataset, test_dataset=None)

Train the model with a dataset.

Parameters

  • train_dataset (DataLoader) – training dataset.

  • test_dataset (DataLoader) – test dataset. If given output statistic during training.

train_instance(ct, segmentation)

Training instance (batch).

Parameters

  • ct (Tensor) – input tensor.

  • segmentation (Tensor) – ground truth tensor.

Returns

losses and visualization data.

update_learning_rate()

Update learning rate.

visualize_training(visuals, affine, epoch, index)

Save training image for visualization.

Parameters

  • affine –

  • visuals (dict) – images.

  • epoch (int) – epoch

  • index (float) – index

networks

class segmentation_rt.dl.model.networks.ResnetBlock(dim, norm_layer, use_dropout, use_bias)

Bases: torch.nn.modules.module.Module

Initialize the Resnet block.

A resnet block is a conv block with skip connections We construct a conv block with build_conv_block function, and implement skip connections in <forward> function. Original Resnet paper: https://arxiv.org/pdf/1512.03385.pdf.

param int dim

the number of channels in the conv layer.

Parameters

  • norm_layer – normalization layer.

  • use_dropout (bool) – if use dropout layers.

  • use_bias (bool) – if use bias.

forward(x)

Standard forward.

training: bool
class segmentation_rt.dl.model.networks.ResnetGenerator(input_nc, output_nc, ngf=64, n_blocks=9, norm_layer=functools.partial(<class 'torch.nn.modules.batchnorm.BatchNorm3d'>, affine=True), use_dropout=False)

Bases: torch.nn.modules.module.Module

Resnet-based generator that consists of Resnet blocks between a few downscaling/up-sampling operations.

Parameters

  • input_nc (int) – the number of channels in input images.

  • output_nc (int) – the number of channels in output images.

  • ngf (int) – the number of filters in the last conv layer.

  • n_blocks (int) – the number of ResNet blocks.

  • norm_layer – normalization layer.

  • use_dropout (bool) – if use dropout layers.

forward(x)

Standard forward.

training: bool
segmentation_rt.dl.model.networks.build_conv_block(dim, norm_layer, use_dropout, use_bias)

Construct a convolutional block.

Parameters

  • dim (int) – the number of channels in the conv layer.

  • norm_layer (nn.Module) – normalization layer.

  • use_dropout (bool) – if use dropout layers.

  • use_bias (bool) – if the conv layer uses bias or not

Returns

Returns a conv block (with a conv layer, a normalization layer, and a non-linearity layer (ReLU))

Return type

nn.Sequential

segmentation_rt.dl.model.networks.define_generator(input_nc, output_nc, ngf, n_blocks, device, use_dropout=False)

Create a generator.

Parameters

  • input_nc (int) – the number of channels in input images.

  • output_nc (i) – the number of channels in output images.

  • ngf (int) – the number of filters in the last conv layer.

  • n_blocks (int) – the number of ResNet blocks.

  • use_dropout (bool) – if use dropout layers.

  • device – device.

Returns

a generator.

Return type

torch.nn.Module

segmentation_rt.dl.model.networks.dice_loss(output, target)

Dice Loss between two tensors.

Parameters

  • output (torch.Tensor) – input tensor.

  • target (torch.Tensor) – ground truth tensor.

Returns

dice loss.

segmentation_rt.dl.model.networks.dice_score(output, target)

Dice Score between two tensors.

param output

input tensor.

type output

torch.Tensor

param target

ground truth tensor.

type target

torch.Tensor

return

dice score.

segmentation_rt.dl.model.networks.print_network(net, out_f=None)

Prints the number of learnable parameters.

Parameters

  • net – the network.

  • out_f – file

Returns

number of learnable parameters.

Return type

int

segmentation_rt.dl.model.networks.weights_init(m)

Initialize network weights.

Parameters

m (torch.nn.Module) – Module