Helper Module for Deep Learning.

NvNet: combination of Vnet and VAE (variation auto-encoder).

class pynet.models.nvnet.DecoderBlock(in_channels, out_channels, stride=1, kernel_size=3, padding=1, num_groups=8, activation='relu', normalization='group_normalization')[source]¶

Decoder block.

__init__(in_channels, out_channels, stride=1, kernel_size=3, padding=1, num_groups=8, activation='relu', normalization='group_normalization')[source]¶

Initialize self. See help(type(self)) for accurate signature.

class pynet.models.nvnet.DownSampling(in_channels, out_channels, stride=2, kernel_size=3, padding=1, dropout_rate=None, bias=False)[source]¶

A convolution and a padding.

__init__(in_channels, out_channels, stride=2, kernel_size=3, padding=1, dropout_rate=None, bias=False)[source]¶

Initialize self. See help(type(self)) for accurate signature.

forward(x)[source]¶

class pynet.models.nvnet.EncoderBlock(in_channels, out_channels, stride=1, kernel_size=3, padding=1, num_groups=8, activation='relu', normalization='group_normalization')[source]¶

Encoder block

__init__(in_channels, out_channels, stride=1, kernel_size=3, padding=1, num_groups=8, activation='relu', normalization='group_normalization')[source]¶

Initialize self. See help(type(self)) for accurate signature.

debug(name, tensor)[source]¶

forward(x)[source]¶

class pynet.models.nvnet.LinearUpSampling(in_channels, out_channels, scale_factor=2, mode='trilinear', align_corners=True)[source]¶

Interpolate to upsample.

__init__(in_channels, out_channels, scale_factor=2, mode='trilinear', align_corners=True)[source]¶

Initialize self. See help(type(self)) for accurate signature.

forward(x, skipx=None, cat=True)[source]¶

class pynet.models.nvnet.NvNet(input_shape, in_channels, num_classes, activation='relu', normalization='group_normalization', mode='trilinear', with_vae=True)[source]¶

NvNet: combination of Vnet and VAE (variation auto-encoder).

The variational auto-encoder branch reconstruct the input image jointly with segmentation in order to regularized the shared encoder.

Reference: https://arxiv.org/pdf/1810.11654.pdf. Code: https://github.com/athon2/BraTS2018_NvNet.

__init__(input_shape, in_channels, num_classes, activation='relu', normalization='group_normalization', mode='trilinear', with_vae=True)[source]¶

Init class.

Parameters

input_shape: uplet

the tensor data shape (X, Y, Z).

in_channels: int

number of channels in the input tensor.

num_classes: int

the number of features in the output segmentation map.

activation: str, default ‘relu’

the activation function.

normalization: str, default ‘group_normalization’

the normalization function.

mode: str, default ‘trilinear’

the interpolation mode.

with_vae: bool, default True

enable/disable vae penalty.

forward(x)[source]¶

class pynet.models.nvnet.OutputTransition(in_channels, out_channels)[source]¶

Decoder output layer: output the prediction of the segmentation.

__init__(in_channels, out_channels)[source]¶

Initialize self. See help(type(self)) for accurate signature.

forward(x)[source]¶

class pynet.models.nvnet.VAE(shapes, in_channels=256, out_channels=4, kernel_size=3, activation='relu', normalization='group_normalization', mode='trilinear')[source]¶

Variational Auto-Encoder: to group the features extracted by Encoder.

__init__(shapes, in_channels=256, out_channels=4, kernel_size=3, activation='relu', normalization='group_normalization', mode='trilinear')[source]¶

Initialize self. See help(type(self)) for accurate signature.

forward(x)[source]¶

class pynet.models.nvnet.VDResampling(in_channels=256, out_channels=256, dense_features=(10, 12, 8), stride=2, kernel_size=3, padding=1, activation='relu', normalization='group_normalization')[source]¶

Variational Auto-Encoder Resampling block.

__init__(in_channels=256, out_channels=256, dense_features=(10, 12, 8), stride=2, kernel_size=3, padding=1, activation='relu', normalization='group_normalization')[source]¶

Initialize self. See help(type(self)) for accurate signature.

forward(x)[source]¶

num_flat_features(x)[source]¶

class pynet.models.nvnet.VDecoderBlock(in_channels, out_channels, kernel_size=3, activation='relu', normalization='group_normalization', mode='trilinear')[source]¶

Variational Decoder block.

__init__(in_channels, out_channels, kernel_size=3, activation='relu', normalization='group_normalization', mode='trilinear')[source]¶

Initialize self. See help(type(self)) for accurate signature.

forward(x, shape=None)[source]¶

class pynet.models.nvnet.VDraw(in_channels=256, out_channels=128)[source]¶

Generate a Gaussian distribution with the given mean(128-d) and std(128-d).

__init__(in_channels=256, out_channels=128)[source]¶

Initialize self. See help(type(self)) for accurate signature.

forward(x)[source]¶

reparametrization(q)[source]¶