now I got depth map and corresponding poses , how can i use Kimera-Semantics to reconstruct, only could transform the input into cloud points?