||Point cloud data is a very important source for reconstructing the 3D geometry of underlying objects. Recent advances in 3D scanner technology opened the door to easy acquisition and access of point cloud data for the masses, where not only the geometry recovery, but also the semantic reconstruction, are critical for serving applications in various domains.
We are interesting in semantic reconstruction from point cloud. We argue that the hidden semantics are often more resistant to acquisition errors, which once recovered can be beneficial to the reconstruction of 3D geometry. More specifically, we are interested in semantic reconstruction of growing plants from 4D point cloud (analyze and detect the growth events), indoor scene from live RGBD scanning (interactively extract and recognize indoor scene objects during the scanning). The semantic reconstruction will create effective representation and enable high-level abstraction and manipulation of underlying objects.