Instance segmentation of point cloud data

Hi, I am new to machine learning.

I am making a tool in web front end where user can go and manually add bounding box to the element inside a scene in 3D scanned Model.

In 3d scanned model, we can not know which small meshes are part of one mesh. So it is not possible to add physics model because we dont know bounding box of element in scene.

Like, from code when you traverse 3D model of room generated from scanner, it is not possible to get chair in room.

In recent works in some research paper, i saw people working on instance segmentation of point cloud data. I am new to machine learning so if i dont understand many thing especially what i will get in output.

Can anyone answer few questions to me on that?