Train AI model with .blend file examples of source and target then automate conversion?

[jmjava]

I have have the need to modify large numbers of meshes representing lakes, sand traps, etc in a golf course. The "finishing" process involves taking a generated basic mesh and applying some mesh transformations to add new loops and bevels etc. I would have a training set of original .blend files along with the manually modified .blend files

Can the blenderProc be adapted to this usage? If not, are there any other community tools that could?

[cornerfarmer]

No this is not possible with blenderproc. Are you sure you need to train a model on that mapping? If the transformations you apply are deterministic, you can also just program them using the blender/blenderproc api.

[jmjava]

That is an interesting suggestion. The idea was to automate a series of steps that are generally applied semi randomly. I was hoping i could get the model to recognize the allowed changes and produce something that would output a new mesh when given existing mesh. I guess i could insert some randomness into api calls just that is what i was hoping to avoid. Bard had suggested something like this: https://g.co/bard/share/b0e263a0f601. hoping its not an AI hallucination does that look like its in the ballpark

[cornerfarmer]

So with blenderproc, the only way to do this is via api calls that include randomness. As far as I know, there is now magic tool which can learn complex mesh transformations from a few labels. Of course you can try to train a network on it, but I doubt you will get precise/high resolution results. But this of course depends on your specific use case.

[jmjava]

Right, not sure if this will work. I just noticed the changes to the mesh to add the variation I want are always on the outer edge and its adding loops etc, hoping the model could learn that. Its not changing the shape of the mesh..only adding adjusting some of the outer vertices.