Helper Module for Deep Learning.

Module that provides spherical layers.

class pynet.models.spherical.layers.DiNeIcoConvLayer(in_feats, out_feats, neigh_indices, n_ring=1)[source]¶

The convolutional layer on icosahedron discretized sphere using n-ring filter (based on the Direct Neighbor (DiNe) formulation).

__init__(in_feats, out_feats, neigh_indices, n_ring=1)[source]¶

Init.

Parameters

in_feats: int

input features/channels.

out_feats: int

output features/channels.

neigh_indices: array (N, k)

conv layer’s filters’ neighborhood indices, where N is the ico number of vertices and k the considered nodes neighbors.

forward(x)[source]¶

class pynet.models.spherical.layers.IcoGenericUpConvLayer(in_feats, out_feats, up_neigh_indices, down_indices)[source]¶

The transposed convolution layer on icosahedron discretized sphere using n-ring filter (slow).

__init__(in_feats, out_feats, up_neigh_indices, down_indices)[source]¶

Init.

Parameters

in_feats: int

input features/channels.

out_feats: int

output features/channels.

up_neigh_indices: array

upsampling neighborhood indices at sampling i + 1.

down_indices: array

downsampling indices at sampling i

forward(x)[source]¶

class pynet.models.spherical.layers.IcoPoolLayer(down_neigh_indices, down_indices, pooling_type='mean')[source]¶

The pooling layer on icosahedron discretized sphere using 1-ring filter.

__init__(down_neigh_indices, down_indices, pooling_type='mean')[source]¶

Init.

Parameters

down_neigh_indices: array

downsampling neighborhood indices at sampling i + 1.

down_indices: array

downsampling indices at sampling i.

pooling_type: str, default ‘mean’

the pooling type: ‘mean’ or ‘max’.

forward(x)[source]¶

class pynet.models.spherical.layers.IcoUpConvLayer(in_feats, out_feats, up_neigh_indices, down_indices)[source]¶

The transposed convolution layer on icosahedron discretized sphere using 1-ring filter.

__init__(in_feats, out_feats, up_neigh_indices, down_indices)[source]¶

Init.

Parameters

in_feats: int

input features/channels.

out_feats: int

output features/channels.

up_neigh_indices: array

upsampling neighborhood indices at sampling i + 1.

down_indices: array

downsampling indices at sampling i

forward(x)[source]¶

class pynet.models.spherical.layers.IcoUpSampleFixIndexLayer(in_feats, out_feats, up_neigh_indices)[source]¶

The upsampling layer on icosahedron discretized sphere using fixed indices 0 (padding new vertices with 0).

__init__(in_feats, out_feats, up_neigh_indices)[source]¶

Init.

Parameters

in_feats: int

input features/channels.

out_feats: int

output features/channels.

up_neigh_indices: array

upsampling neighborhood indices.

forward(x)[source]¶

class pynet.models.spherical.layers.IcoUpSampleLayer(in_feats, out_feats, up_neigh_indices)[source]¶

The upsampling layer on icosahedron discretized sphere using interpolation.

__init__(in_feats, out_feats, up_neigh_indices)[source]¶

Init.

Parameters

in_feats: int

input features/channels.

out_feats: int

output features/channels.

up_neigh_indices: array

upsampling neighborhood indices.

forward(x)[source]¶

class pynet.models.spherical.layers.IcoUpSampleMaxIndexLayer(in_feats, out_feats, up_neigh_indices, down_indices)[source]¶

The upsampling layer on icosahedron discretized sphere using max indices.

__init__(in_feats, out_feats, up_neigh_indices, down_indices)[source]¶

Init.

Parameters

in_feats: int

input features/channels.

out_feats: int

output features/channels.

up_neigh_indices: array

upsampling neighborhood indices at sampling i + 1.

down_indices: array

downsampling indices at sampling i.

forward(x, max_pool_indices)[source]¶

class pynet.models.spherical.layers.RePaIcoConvLayer(in_feats, out_feats, neighs)[source]¶

Define the convolutional layer on icosahedron discretized sphere using rectagular filter in tangent plane.

__init__(in_feats, out_feats, neighs)[source]¶

Init.

Parameters

in_feats: int

input features/channels.

out_feats: int

output features/channels.

neighs: 2-uplet

neigh_indices: array (N, k, 3) - the neighbors indices. neigh_weights: array (N, k, 3) - the neighbors distances.

forward(x)[source]¶