1)
Both pointnet2_ssg_sem and pointnet2_msg_sem have a typo in the definition and usage of the fully connected layer.
As you can see from the attached links the variable is called lyaer instead of layer.
2)
I'm working on customizing this implementation for running with another PC based dataset and I think that it would be helpful to make explicit in the models files the number responsibible for the channels and classes.
For example here for the labels variable (my dataset has 5 labels so I guessed that this was the labels variable) and I also guess that here and here the features channels number should change in case someone did not have normals (e.g. from RGB+normals(x,y,z) to only RGB).
If I'm wrong about the last part then, this would be a proof that some more explanation/comments are due on how to change correctly the channels and labels numbers.
In any case, thanks for the great work :)
Cheers
Dario
Hi, I would like to point out two things:
1) Both
pointnet2_ssg_semandpointnet2_msg_semhave a typo in the definition and usage of the fully connected layer. As you can see from the attached links the variable is called lyaer instead of layer.https://github.com/erikwijmans/Pointnet2_PyTorch/blob/acda965224f35854bc331cd5fe140393216b0a71/pointnet2/models/pointnet2_msg_sem.py#L68
https://github.com/erikwijmans/Pointnet2_PyTorch/blob/acda965224f35854bc331cd5fe140393216b0a71/pointnet2/models/pointnet2_ssg_sem.py#L57
fortunatelly here is called in the same way https://github.com/erikwijmans/Pointnet2_PyTorch/blob/acda965224f35854bc331cd5fe140393216b0a71/pointnet2/models/pointnet2_ssg_sem.py#L90
2) I'm working on customizing this implementation for running with another PC based dataset and I think that it would be helpful to make explicit in the models files the number responsibible for the channels and classes. For example here for the labels variable (my dataset has 5 labels so I guessed that this was the labels variable) and I also guess that here and here the features channels number should change in case someone did not have normals (e.g. from RGB+normals(x,y,z) to only RGB).
If I'm wrong about the last part then, this would be a proof that some more explanation/comments are due on how to change correctly the channels and labels numbers.
In any case, thanks for the great work :) Cheers Dario