Can NURBS 3D models be fully digitized?

Rhino
, , , ,
1

what i mean is “Can neural networks be used to generate 3D NURBS models?”

In 3D models using subdivision surfaces, information is generally represented through point clouds or mesh surfaces.

So, can NURBS 3D models be fully digitized? I understand they usually involve control points, knot vectors, weights, and orders, but this format is only applicable to curves. Is there a universal data format for NURBS that not only works for curves but also for surfaces, and even complex surfaces?

If such a format exists, using it entirely with Ai neural networks could bring new developments to NURBS models.

2

The IGES file Specification should be what you are looking for
https://wiki.eclipse.org/IGES_file_Specification#Rational_B-Spline_Curve_.28Type_126.29

1 Like
3

NURBS surfaces also are described by control points, knot vectors, weights, and orders/degrees.

IGES is one of several data exchange format for NURBS.

2 Likes
4

Where are you gonna get the training data?

2 Likes
6

After parametric modeling with Grasshopper, traversing a large number of parameters can yield a vast amount of data

7

That’s not useful. See: how the LLM’s are going to collapse from too much AI-generated slop contaminating their training data.

8

This is good, but essentially it is an indirect form of data. I hope to be able to directly define the NURBS 3D model

9

This problem would not exist if the training data was artificial.

10

I have no idea what you’re going for here, but I don’t think one model with a bunch of parameters changed is going to do much to teach the AI very much. It could perhaps be trained to tell you something about the model.

I would posit that sufficient training data for a text-to-NURBS model…model does not exist, never mind that all the quality data is proprietary. They can’t even really do the math involved.

1 Like
11

Hi Community,
Yes, if you are interested, here is one of our recent work.
https://www.researchgate.net/publication/385921092_NeuroNURBS_Learning_Efficient_Surface_Representations_for_3D_Solids