error run pointnet.py

[rancheng]

The model inconsistency happens in multiple places, until now I still can not have a smooth run on the example of pointnet.py

I installed minkowski==0.4.0 in my system with CUDA 10.1

https://github.com/StanfordVL/MinkowskiEngine/blob/a1fab7a14564d77245be57a9cc789e69f14c482a/examples/pointnet.py#L84-L88

here the error shows that the class ME.MinkowskiInstanceNorm doesn't have dimension argument.

But when navigate into file MinkowskiEngine/MinkowskiInstanceNorm.py

https://github.com/StanfordVL/MinkowskiEngine/blob/a1fab7a14564d77245be57a9cc789e69f14c482a/MinkowskiEngine/MinkowskiNormalization.py#L310-L318

There's still dimension here. I don't know why you delete this argument or will it break other models if we restore those arguments.

Another problem is that in here:

https://github.com/StanfordVL/MinkowskiEngine/blob/a1fab7a14564d77245be57a9cc789e69f14c482a/examples/pointnet.py#L82

Do you mean:

self.broadcast = ME.MinkowskiBroadcastMultiplication()

and if so, your tensor should be reshaped into 3x3 rather than left as 1x9

I have modified the code and I can open a pull request if it helps.

[chrischoy]

Thanks for submitting the issue.
I found a few problems with the code and it now works.
Please pull from the latest source for the update.

Also, please let me know if you have further questions.