Layers module

Layer module.

class MARBLE.layers.Diffusion(tau0=0.0)

Diffusion with learned t.

initialise.

forward(x, L, Lc=None, method='spectral')

Forward.

class MARBLE.layers.AnisoConv(**kwargs)

Anisotropic Convolution

Initialize.

forward(x, kernels)

Forward.

message_and_aggregate(K_t, x)

Message passing step. If K_t is a txs matrix (s sources, t targets), do matrix multiplication K_t@x, broadcasting over column features. If K_t is a t*dimxs*dim matrix, in case of manifold computations, then first reshape, assuming that the columns of x are ordered as [dx1/du, x1/dv, …, dx2/du, dx2/dv, …].

class MARBLE.layers.InnerProductFeatures(C, D)

Compute scaled inner-products between channel vectors.

Input: (V x C*D) vector of (V x n_i) list of vectors with sum_in_i = C*D Output: (VxC) dot products

Initialize internal Module state, shared by both nn.Module and ScriptModule.

reset_parameters()

Reset parameters.

forward(x)

Forward.