Pretrained registry #736
Replies: 2 comments 3 replies
-
Hey @RobotJente ! The error is because we updated the OmegaConf version, but the models still are using the old omegaconf version. Check out these issues which detail how to solve it and let me know if it helps: #669 #642 |
Beta Was this translation helpful? Give feedback.
-
Thank you for your quick reply! I tried my best to solve it by following the method from the issues you linked. The converting of the weights worked I think. However, I got the following error when running the line
My current package versions: The way I installed torch-points3d was through pypi, if that helps, but I also have the repo cloned and in my PythonPath. I tried installing torch-points3d through poetry/requirements.txt but was having problems because the torch cuda version and my local cuda version did not match. I recently saw the last Pypi release was a year ago, do you think it's a good idea to try to get the development version working using poetry? I'm a bit lost in all the packages and package versions, sorry! I hope this gives some information about my problem, and thank you for all your help so far! Edit: I also tried it with the hydra-core and omegaconf versions from the requirements.txt, and got the same error |
Beta Was this translation helpful? Give feedback.
-
Hi! I was not sure whether to ask this here or open an issue, so I hope this is okay. I was having trouble loading the pretrained models. I think a related issue is #729 because I ran into this while exploring some of the notebooks. I was running a few of the notebooks and the cell importing/downloading the pretrained model was complaining about this:
AttributeError: Can't get attribute 'UntypedNode' on <module 'omegaconf.nodes' from '/home/jente/.virtualenvs/ml-env/lib/python3.8/site-packages/omegaconf/nodes.py'>
The code I was running is
model = PretainedRegistry.from_pretrained("minkowski-registration-3dmatch").cuda()
as found in demo_registration_3dm.ipynb
Does somebody know where I could start looking to solve this? The stack trace suggests it might be a pickle error, but it could also be that I'm missing something obvious and it's a mistake on my part. This error happens for every pretrained model I tested, I didn't test all but I did test models from different sources with different archtitectures
Beta Was this translation helpful? Give feedback.
All reactions