I don’t know this class! It seems I wrote a long explanation, and you answered it in a single sentence! I’ll test it, but it seems to meet my needs for accessing raw data… thank you!
Hi jmv
I have been following this thread, while I don’t really understand c#, i do understand the issues that I face, when I am trying to work on models (3d experiments) with millions of faces, or vertices. I often use mesh as a way to generate other ideas, to build meshes with say a million points for sample locations (closest vertices, rt tree), which is always absolutely faster than surface).
are you developing something that will be shared publically? I love to play with new tools.
And are you loading LLM’s / checkpoints locally? and somehow integrating this with meshes for segmentation?
So no, unfortunately, this won’t be an open-source project.
I wasn’t able to test Steve’s solution. For now, the simplest integration with Rhino I could find is to build an external application along with a bridge plugin that transmits an STL file. But I think we can do much better than that.
I highly encourage you to try out the CuPy library. I knew nothing about Python, let alone GPU programming, and I’ve come to realize that this library is simply amazing! It’s actually easier to write a GPU algorithm than an OpenGL rendering shader…
The performance is incredible, even with my GTX 1080 Ti. I’m running about fifteen different analyses on meshes with millions of faces—performing real-time curvature analyses, creating adjacency matrices, etc.—and all of it takes just a few milliseconds.
As for LLMs or checkpoints, I’m not actually using any true AI-related technology. I only drew that parallel because artificial intelligence has brought attention to libraries that either already existed or were poorly documented. Today, there are highly accessible libraries that significantly simplify development for GPU parallel processing.