Erik Wijmans

I’ll take a look. Which version of visdom are you using?

Hmmm. Those are the exact same versions I am using with no trouble. It is also very odd that initializing the visdom client has anything influence on the `_ext` module....

I am not sure what the "raw response of a submap" is.

You can remove the FC layer: https://github.com/erikwijmans/Pointnet2_PyTorch/blob/master/pointnet2/models/pointnet2_ssg_cls.py#L79 and then just output the features https://github.com/erikwijmans/Pointnet2_PyTorch/blob/master/pointnet2/models/pointnet2_ssg_cls.py#L112

If I recall correctly, I was able to reproduce the results on ScanNet by setting the hyper-parameters to the same as in https://github.com/charlesq34/pointnet2/blob/master/scannet/train.py **and** adding the class balanced cross-entropy loss...

Segfaults can be very challenging to debug and I haven't seen this this one happen. One thing: ``` gdb python (gdb) r -m pointnet2.train.train_cls (gdb) bt (gdb) f 0 ```...

Refer to here for PointNet results on s3DIS: https://github.com/charlesq34/pointnet/tree/master/sem_seg

2 things that could be causing this discrepancy: 1. You need to change to use 10k points instead of the default to match the results in the paper. 2. Batch...

I have not tested building this on Windows and I don't have access to a windows machine to debug this on, so I can't be of much help. If you...

What version of pytorch are you using? master requires `pytorch>=1.0`. If you are on `pytorch>=1.0`, then there seems to be an issue with your pytorch install and I'd suggest re-installing.