Closed aleenajohnsony closed 3 years ago
Note that the first max pooling use torch.max(f, dim=2, keepdim=True)[0]
, the keyword argument keepdim
is set True
, it means that we keep the dimension, so the output dimension of the first max pooling is (B, 256, 1). In the second max pooling, I didn't set the keepdim
to True
, because I just want every point cloud to be a latent code, i.e. a vector. So, in a mini-batch data, every point cloud is represented a vector. The final dimension of a batch of point cloud is (B, 1024).
This is from encoder. Why is the dimension 2? I couldnot find a valid dimension. def forward(self, x): n = x.size()[2]