I am at the moment trying to segment pointclouds or rather meshes (so 3D-Segmentation) but nothing quite works. Here’s my specific problem:

My “task” right now is to segment a 3D model of a robot but I am looking for tools that are adaptable or even work with anything. I have tried PointNext with S3DIS Dataset as a proof of concept and was not surprised when it didn’t work. I mean it’s hard to find representations of rooms or walls or whatever in a typical 6-DoF-Robot. Then I started thinking: The Dataset would have to be pretty specific to segment my robot into its arms and joints and TCP or it would have to be just composed of primitives like cylinders or spheres or whatever. So my first question is:
Do you think this kind of task would work with a ML based method, especially if I wanted to segment ANY random object? Are you aware of a dataset or method that could help me?

Next I experimenting with a different aproach: Pointcloud Library.
I am getting okay-ish results with Region Growing but I don’t know nothing quite hits the Spot. Is this problem so absurd? Is there nothing we can do to segment a random robot or some random object?
Am I just blind or doing something wrong? Do you think PCL is the way to go and if so, what segmentation tool can you recommend?

I am very thankful for every little nudge in the right direction 🙂

I have tried PointNext with S3DIS and didn’t expect much but now am anxious to even try with a different dataset.

I have also tried PCL with region growing and was expecting as much as I got – a kind of okay result. Am now wondering where to focus my energy on.