Abstract: In this paper, we propose a Transformer encoder-decoder architecture, called PoinTr, which reformulates point cloud completion as a set-to-set translation problem and employs a ...