Helper Module for Deep Learning.

3D MRI Brain Generation with Generative Adversarial Networks (BGGAN) with Variational Auto Encoder (VAE).

class pynet.models.braingengan.BGCodeDiscriminator(out_channels=1, code_size=1000, n_units=4096)[source]¶

This is the code discriminator part of the BGGAN.

__init__(out_channels=1, code_size=1000, n_units=4096)[source]¶

Init class.

Parameters

out_channels: int, default 1

number of channels in the output tensor.

code_size: int, default 1000

the code sier.

n_units: int, default 4096

the number of hidden units.

debug(name, tensor)[source]¶

forward(x)[source]¶

class pynet.models.braingengan.BGDiscriminator(in_shape, in_channels=1, out_channels=1, start_filts=64, with_logit=True)[source]¶

This is the discriminator part of the BGGAN.

__init__(in_shape, in_channels=1, out_channels=1, start_filts=64, with_logit=True)[source]¶

Init class.

Parameters

in_shape: uplet

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

in_channels: int, default 1

number of channels in the input tensor.

out_channels: int, default 1

number of channels in the output tensor.

start_filts: int, default 64

number of convolutional filters for the first conv.

with_logit: bool, default True

apply the logit function to the result.

debug(name, tensor)[source]¶

forward(x)[source]¶

class pynet.models.braingengan.BGEncoder(in_shape, in_channels=1, start_filts=64, latent_dim=1000)[source]¶

This is the encoder part of the BGGAN.

__init__(in_shape, in_channels=1, start_filts=64, latent_dim=1000)[source]¶

Init class.

Parameters

in_shape: uplet

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

in_channels: int, default 1

number of channels in the input tensor.

start_filts: int, default 64

number of convolutional filters for the first conv.

latent_dim: int, default 1000

the latent variable sizes.

debug(name, tensor)[source]¶

forward(x)[source]¶

class pynet.models.braingengan.BGGenerator(in_shape, out_channels=1, start_filts=64, latent_dim=1000, mode='trilinear', with_code=False)[source]¶

This is the generator part of the BGGAN.

__init__(in_shape, out_channels=1, start_filts=64, latent_dim=1000, mode='trilinear', with_code=False)[source]¶

Init class.

Parameters

in_shape: uplet

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

out_channels: int, default 1

number of channels in the output tensor.

start_filts: int, default 64

number of convolutional filters for the first conv.

latent_dim: int, default 1000

the latent variable sizes.

mode: str, default ‘trilinear’

the interpolation mode.

with_code: bool, default False

change the architecture if code discriminator is used.

debug(name, tensor)[source]¶

forward(noise)[source]¶