mfinzi / lieconv Goto Github PK

Thank you for your great work.
I have a 3D voxel CNN model. Because of a lack of data, I have to do data argumentation many times. And also the dataset is very sparse, so it is not efficient for memory. I want to transfer my model into a se(3) model. However, I don't know how to prepare my dataset.
In addition, how is your model difference from e3nn and se(3)cnn?

Many thanks.

Hallo, Good afternoon.
Thank you for helping me regarding the issues before.
May I know how to save the model from your training and how to run the test program?
I am very sorry for asking a dumb question since this is my first time try to code.

Hello,
I am very interested in your code and I have it a try.
but this error happend when I am trying to run your code.

Traceback (most recent call last):
File "examples/train_springs.py", line 9, in
from lie_conv.datasets import SpringDynamics
File "/home/em308/LieConv/lie_conv/datasets.py", line 58, in
class ModelNet40(Dataset,metaclass=Named):
TypeError: metaclass conflict: the metaclass of a derived class must be a (non-strict) subclass of the metaclasses of all its bases

Since I am very new in programming, would you mind to tell me what can I do to solve this?
thank you so much for your help.

On line 329 in lieConv.py the group object is not instantiated, causing problems with LieResNet's forward method. The obvious two-character fix seemed to get it working.

Hello,

Thank you for the very nice paper. I was wondering how is the memory and computational complexity of lift approach to building equivariant networks compared with the non-lift approach? (those used in tensor field network and se3 transformer). And what is the memory and computational bottleneck in your opinion? Thank you!

With torchvision version '0.8.1+cu101' and torch version '1.7.0+cu101' I get the error

ImportError: cannot import name 'makedir_exist_ok' from 'torchvision.datasets.utils'

Indeed a search in torchvision github return no results for makedir_exist_ok.

I manually changed makedir_exist_ok to os.makedirs (for Python 3.7) in lie_conv/datasets.py and it worked, but posting here for reference