CP3: Unifying Point Cloud Completion by Pretrain-Prompt-Predict Paradigm

Point cloud completion aims to predict complete shape from its partial observation. Current approaches mainly consist of generation and refinement stages in a coarse-to-fine style. However, the generation stage often lacks robustness to tackle different incomplete variations, while the refinement stage blindly recovers point clouds without the semantic awareness. To tackle these challenges, we unify point cloud Completion by a generic PretrainPrompt-Predict paradigm, namely CP3. Inspired by prompting approaches from NLP, we creatively reinterpret point cloud generation and refinement as the prompting and predicting stages, respectively. Then, we introduce a concise self-supervised pretraining stage before prompting. It can effectively increase robustness of point cloud generation, by an Incompletion-Of-Incompletion (IOI) pretext task. Moreover, we develop a novel Semantic Conditional Refinement (SCR) network at the predicting stage. It can discriminatively modulate multi-scale refinement with the guidance of semantics. Finally, extensive experiments demonstrate that our CP3 outperforms the state-of-the-artmethods with a large margin. code will be available at https://github.com/ MingyeXu/cp3.

Automated summary

Point cloud completion aims to predict complete shape from partial observations. A generic PretrainPrompt-Predict paradigm, CP3, is proposed to tackle the challenges in current approaches. CP3 creatively interprets point cloud generation and refinement stages as prompting and predicting stages, respectively, and introduces a self-supervised pretraining stage and a Semantic Conditional Refinement (SCR) network to increase robustness and improve refinement with semantics. Experimental results show that CP3 outperforms state-of-the-art methods significantly. Code is available at https://github.com/MingyeXu/cp3.