kimren227 / csgstumpnet Goto Github PK

Hi there!

CSGStumpNet seems interesting! I see you currently save your model checkpoint in this repo. Would you be interested in sharing the model in the Hugging Face Hub?

The Hub offers free hosting of over 20K models, and it would make your work more accessible and visible to the rest of the ML community. Some of the benefits of sharing your models would be:

• versioning
• commit history and diffs
• repos provide useful metadata about their tasks, languages, metrics, etc
• we could add a widget for users to try the model directly in the browser

Creating the repos and adding new models should be a relatively straightforward process if you've used Git before. This is a step-by-step guide explaining the process in case you're interested. Please let us know if you would be interested and if you have any questions.

Happy to hear your thoughts,
Omar and the Hugging Face team

Hi, daxuan
I noticed that in model.py, the complement layer was not written after connection decoder.
Is it a trick to make the end-to-end method perform better under ShapeNet dataset or just an ignorance?
I'm asking this for the reason that my own dataset for mechanical parts have many through-hole structures, which is not topologically isomorphic to those in ShapeNet, and I'm trying to figure out why the training process always ignored the hole structure.
Till now, I've excluded the reason of sampling strategy and loss function.
Do you think it will help if the complement layer is added?

Hi, thanks for your excellent work! But I'm confused that why you use infinitely long cylinders and cones when computing the sdf. Why there is no "height" parameter associated with these two types of primitives? I think they need to be bounded to form the final shape. Could you please explain this for me? Thank you in advance!

Hi, thank you for this interesting work and congrats on publishing at ICCV.

I have a question about the implementation of the primitive loss term. In the paper, it is

But in the implementation it is:

Which seems like it is this expression instead:

I suppose this is equivalent to the first equation when all testing points are outside the primitive, but once there are testing points inside this loss term seems to pull the surface so that no testing points are too far inside (since it pushes the most negative value toward 0). Is this intentional?

Might the correct implementation be as follows?

primitive_loss = torch.mean((primitive_sdf**2).min(dim=1)[0]) * self.scale

Hi daxuan,
Why is the value avg_test_loss_recon divided twice by test_iter when printed to console as "loss_recon"? It occurs both in train.py and eval.py.
Best regards

Hi,

This is an interesting work and thanks for sharing your code!

I just want to know when will you share the pre-trained weights? I'd like to try it in my project!

Thank you!

I didn't find the code to export parametric shapes to openSCAD, could you please point it out or provide it?
Thank you so much!

Hi, I am interested in your work.

However, I find the first complement layer code is missing in your code model.py.

Will you plan to release the code of that part?

Looking forward to your reply, thank you!

Best.