Ya using voxels was definitely a big drawback to my approach here but was more intended to just see what's possible.
I think a better approach would be something similar to the StructureNet paper I mention in the post and use graph based models to actually attempt to capture the topology. But they did it with super explicitly defined part trees as training data the hard part would be finding a way to do that in a unsupervised manner so you could actually make use of the massive amount of unlabelled 3D models available.
This would be really incredible! Could maybe be a way to quickly make video game assets for testing and such. Although to make the meshes performant in a game engine you'd need to decimate them quite a bit as the marching cubes I used here produces really dense meshes.
I hadn't seen that paper, looks really interesting! It definitely would improve the resolution but I think fundamentally it still runs into the issues I had with voxels where it mostly just learns low level correlations and not high level topology.
I think a better approach would be something similar to the StructureNet paper I mention in the post and use graph based models to actually attempt to capture the topology. But they did it with super explicitly defined part trees as training data the hard part would be finding a way to do that in a unsupervised manner so you could actually make use of the massive amount of unlabelled 3D models available.