The sequential decoding paradigm attains better detection performance compared to previous works that predict all attributes of the 3D objects in parallel, thanks to its competence in exploring intrinsic dependencies among object attributes.

Point2Seq: Detecting 3D Objects as Sequences, CVPR22