this is interesting. how does the semantic segmentation work? do you generate 3d models and then separate them? or are the separated from the initial generation?

I think since your target is blender it is better as a blender add on, so I can generate directly into my scene. then you can publish it on blender marketplaces.