Helper Module for Deep Learning.

Module that provides the spherical UNet architecture.

class pynet.models.spherical.unet.DownBlock(conv_layer, in_ch, out_ch, conv_neigh_indices, down_neigh_indices, down_indices, pool_mode='mean', first=False)[source]¶

Downsampling block in spherical UNet: mean pooling => (conv => BN => ReLU) * 2

__init__(conv_layer, in_ch, out_ch, conv_neigh_indices, down_neigh_indices, down_indices, pool_mode='mean', first=False)[source]¶

Init.

Parameters

conv_layer: nn.Module

the convolutional layer on icosahedron discretized sphere.

in_ch: int

input features/channels.

out_ch: int

output features/channels.

conv_neigh_indices: array

conv layer’s filters’ neighborhood indices at sampling i.

down_neigh_indices: array

conv layer’s filters’ neighborhood indices at sampling i + 1.

down_indices: array

downsampling indices at sampling i.

pool_mode: str, default ‘mean’

the pooling mode: ‘mean’ or ‘max’.

first: bool, default False

if set skip the pooling block.

forward(x)[source]¶

class pynet.models.spherical.unet.Ico(order, vertices, triangles, neighbor_indices, down_indices, up_indices, conv_neighbor_indices)¶

property conv_neighbor_indices¶

Alias for field number 6

property down_indices¶

Alias for field number 4

property neighbor_indices¶

Alias for field number 3

property order¶

Alias for field number 0

property triangles¶

Alias for field number 2

property up_indices¶

Alias for field number 5

property vertices¶

Alias for field number 1

class pynet.models.spherical.unet.SphericalUNet(in_order, in_channels, out_channels, depth=5, start_filts=32, conv_mode='1ring', up_mode='interp', cachedir=None)[source]¶

Define the Spherical UNet structure.

https://github.com/zhaofenqiang/Spherical_U-Net Spherical U-Net on Cortical Surfaces: Methods and Applications

__init__(in_order, in_channels, out_channels, depth=5, start_filts=32, conv_mode='1ring', up_mode='interp', cachedir=None)[source]¶

Initialize the Spherical UNet.

Parameters

in_order: int

the input icosahedron order.

in_channels: int

input features/channels.

out_channels: int

output features/channels.

depth: int, default 5

number of layers in the UNet.

start_filts: int, default 32

number of convolutional filters for the first conv.

conv_mode: str, default ‘1ring’

the size of the spherical convolution filter: ‘1ring’ or ‘2ring’. Can also use rectangular grid projection: ‘repa’.

up_mode: str, default ‘interp’

type of upsampling: ‘transpose’ for transpose convolution (1 ring), ‘interp’ for nearest neighbor linear interpolation, ‘maxpad’ for max pooling shifted zero padding, and ‘zeropad’ for classical zero padding.

cachedir: str, default None

set tthis folder tu use smart caching speedup.

forward(x)[source]¶

class pynet.models.spherical.unet.UpBlock(conv_layer, in_ch, out_ch, conv_neigh_indices, neigh_indices, up_neigh_indices, down_indices, up_mode)[source]¶

Define the upsamping block in spherical UNet: upconv => (conv => BN => ReLU) * 2

__init__(conv_layer, in_ch, out_ch, conv_neigh_indices, neigh_indices, up_neigh_indices, down_indices, up_mode)[source]¶

Init.

Parameters

conv_layer: nn.Module

the convolutional layer on icosahedron discretized sphere.

in_ch: int

input features/channels.

out_ch: int

output features/channels.

conv_neigh_indices: tensor, int

conv layer’s filters’ neighborhood indices at sampling i.

neigh_indices: tensor, int

neighborhood indices at sampling i.

up_neigh_indices: array

upsampling neighborhood indices at sampling i + 1.

down_indices: array

downsampling indices at sampling i.

up_mode: str, default ‘interp’

type of upsampling: ‘transpose’ for transpose convolution, ‘interp’ for nearest neighbor linear interpolation, ‘maxpad’ for max pooling shifted zero padding, and ‘zeropad’ for classical zero padding.

forward(x1, x2, max_pool_indices)[source]¶