FBNet: Feedback Network for Point Cloud Completion

The rapid development of point cloud learning has driven completion into a new era. However, the information flows most existing methods are solely feedforward, and high-level is rarely reused to improve low-level feature learning. To this end, we propose novel Feedback Network (FBNet) for completion, in which present features efficiently refined by rerouting subsequent fine-grained ones. Firstly, partial inputs fed Hierarchical Graph-based (HGNet) generate coarse shapes. Then, cascade several Feedback-Aware Completion (FBAC) Blocks unfold them across time recurrently. connections between two adjacent steps exploit shape generations. main challenge building feedback dimension mismatching features. address this, elaborately designed Cross Transformer exploits efficient from via cross attention strategy then refines with enhanced Quantitative qualitative experiments on datasets demonstrate superiority proposed FBNet compared state-of-the-art task. source code model available at https://github.com/hikvision-research/3DVision/ .