import torch import torch.nn as nn import torch.nn.functional as F class PointNet(nn.Module): def __init__(self, hidden_dim=32): super(PointNet, self).__init__() self.conv_in = nn.Conv3d(1, 16, 3, padding=1, padding_mode='replicate') # out: 32 ->m.p. 16 self.conv_0 = nn.Conv3d(16, 32, 3, padding=1, padding_mode='replicate') # out: 16 self.conv_0_1 = nn.Conv3d(32, hidden_dim, 3, padding=1, padding_mode='replicate') # out: 16 ->m.p. 8 self.conv_1 = nn.Conv3d(hidden_dim, hidden_dim, 3, padding=1, padding_mode='replicate') # out: 8 self.conv_1_1 = nn.Conv3d(hidden_dim, hidden_dim, 3, padding=1, padding_mode='replicate') # out: 8 self.actvn = nn.ReLU() self.maxpool = nn.MaxPool3d(2) self.conv_in_bn = nn.BatchNorm3d(16) self.conv0_1_bn = nn.BatchNorm3d(hidden_dim) self.conv1_1_bn = nn.BatchNorm3d(hidden_dim) def forward(self, x): x = x.unsqueeze(1) net = self.actvn(self.conv_in(x)) net = self.conv_in_bn(net) net = self.maxpool(net) # out 16 net = self.actvn(self.conv_0(net)) net = self.actvn(self.conv_0_1(net)) net = self.conv0_1_bn(net) net = self.maxpool(net) # out 8 net = self.actvn(self.conv_1(net)) net = self.actvn(self.conv_1_1(net)) fea = self.conv1_1_bn(net) return fea