Source code for torch_geometric.nn.conv.x_conv

from __future__ import division

from math import ceil

import torch
from torch.nn import Sequential as S, Linear as L, BatchNorm1d as BN
from torch.nn import ELU, Conv1d
from torch_geometric.nn import Reshape

from ..inits import reset

try:
    from torch_cluster import knn_graph
except ImportError:
    knn_graph = None


[docs]class XConv(torch.nn.Module): r"""The convolutional operator on :math:`\mathcal{X}`-transformed points from the `"PointCNN: Convolution On X-Transformed Points" <https://arxiv.org/abs/1801.07791>`_ paper .. math:: \mathbf{x}^{\prime}_i = \mathrm{Conv}\left(\mathbf{K}, \gamma_{\mathbf{\Theta}}(\mathbf{P}_i - \mathbf{p}_i) \times \left( h_\mathbf{\Theta}(\mathbf{P}_i - \mathbf{p}_i) \, \Vert \, \mathbf{x}_i \right) \right), where :math:`\mathbf{K}` and :math:`\mathbf{P}_i` denote the trainable filter and neighboring point positions of :math:`\mathbf{x}_i`, respectively. :math:`\gamma_{\mathbf{\Theta}}` and :math:`h_{\mathbf{\Theta}}` describe neural networks, *i.e.* MLPs, where :math:`h_{\mathbf{\Theta}}` individually lifts each point into a higher-dimensional space, and :math:`\gamma_{\mathbf{\Theta}}` computes the :math:`\mathcal{X}`- transformation matrix based on *all* points in a neighborhood. Args: in_channels (int): Size of each input sample. out_channels (int): Size of each output sample. dim (int): Point cloud dimensionality. kernel_size (int): Size of the convolving kernel, *i.e.* number of neighbors including self-loops. hidden_channels (int, optional): Output size of :math:`h_{\mathbf{\Theta}}`, *i.e.* dimensionality of lifted points. If set to :obj:`None`, will be automatically set to :obj:`in_channels / 4`. (default: :obj:`None`) dilation (int, optional): The factor by which the neighborhood is extended, from which :obj:`kernel_size` neighbors are then uniformly sampled. Can be interpreted as the dilation rate of classical convolutional operators. (default: :obj:`1`) bias (bool, optional): If set to :obj:`False`, the layer will not learn an additive bias. (default: :obj:`True`) **kwargs (optional): Additional arguments of :class:`torch_cluster.knn_graph`. """ def __init__(self, in_channels, out_channels, dim, kernel_size, hidden_channels=None, dilation=1, bias=True, **kwargs): super(XConv, self).__init__() if knn_graph is None: raise ImportError('`XConv` requires `torch-cluster`.') self.in_channels = in_channels if hidden_channels is None: hidden_channels = in_channels // 4 assert hidden_channels > 0 self.hidden_channels = hidden_channels self.out_channels = out_channels self.dim = dim self.kernel_size = kernel_size self.dilation = dilation self.kwargs = kwargs C_in, C_delta, C_out = in_channels, hidden_channels, out_channels D, K = dim, kernel_size self.mlp1 = S( L(dim, C_delta), ELU(), BN(C_delta), L(C_delta, C_delta), ELU(), BN(C_delta), Reshape(-1, K, C_delta), ) self.mlp2 = S( L(D * K, K**2), ELU(), BN(K**2), Reshape(-1, K, K), Conv1d(K, K**2, K, groups=K), ELU(), BN(K**2), Reshape(-1, K, K), Conv1d(K, K**2, K, groups=K), BN(K**2), Reshape(-1, K, K), ) C_in = C_in + C_delta depth_multiplier = int(ceil(C_out / C_in)) self.conv = S( Conv1d(C_in, C_in * depth_multiplier, K, groups=C_in), Reshape(-1, C_in * depth_multiplier), L(C_in * depth_multiplier, C_out, bias=bias), ) self.reset_parameters()
[docs] def reset_parameters(self): reset(self.mlp1) reset(self.mlp2) reset(self.conv)
[docs] def forward(self, x, pos, batch=None): """""" pos = pos.unsqueeze(-1) if pos.dim() == 1 else pos (N, D), K = pos.size(), self.kernel_size row, col = knn_graph(pos, K * self.dilation, batch, loop=True, flow='target_to_source', **self.kwargs) if self.dilation > 1: dil = self.dilation index = torch.randint(K * dil, (N, K), dtype=torch.long, device=row.device) arange = torch.arange(N, dtype=torch.long, device=row.device) arange = arange * (K * dil) index = (index + arange.view(-1, 1)).view(-1) row, col = row[index], col[index] pos = pos[col] - pos[row] x_star = self.mlp1(pos.view(N * K, D)) if x is not None: x = x.unsqueeze(-1) if x.dim() == 1 else x x = x[col].view(N, K, self.in_channels) x_star = torch.cat([x_star, x], dim=-1) x_star = x_star.transpose(1, 2).contiguous() x_star = x_star.view(N, self.in_channels + self.hidden_channels, K, 1) transform_matrix = self.mlp2(pos.view(N, K * D)) transform_matrix = transform_matrix.view(N, 1, K, K) x_transformed = torch.matmul(transform_matrix, x_star) x_transformed = x_transformed.view(N, -1, K) out = self.conv(x_transformed) return out
def __repr__(self): return '{}({}, {})'.format(self.__class__.__name__, self.in_channels, self.out_channels)