PyTorch toolbox to work with spherical surfaces.

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

Notes

Debuging messages can be displayed by changing the log level using setup_logging(level='debug').

Examples

>>> import torch
>>> from surfify.nn import IcoRePaConv
>>> from surfify.utils import icosahedron, neighbors_rec
>>> ico2_vertices, ico2_triangles = icosahedron(order=2)
>>> neighbors = neighbors_rec(
        ico2_vertices, ico2_triangles, size=5, zoom=5)[:2]
>>> module = IcoRePaConv(
        in_feats=8, out_feats=8, neighs=neighbors)
>>> ico2_x = torch.zeros((10, 8, len(ico2_vertices)))
>>> ico2_x = module(ico2_x)
>>> ico2_x.shape

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

Init IcoRePaConv.

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]¶

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.