PQDT: Pseudo-Query Dual Transformer for Robust Point Cloud Restoration
The Pseudo-Query Dual Transformer (PQDT) framework introduces a unified point-only backbone to restore high-quality 3D geometry from degraded point clouds like those with noise or incompleteness. It uses a two-stage encoding process where auxiliary pseudoqueries act as noiseresistant anchors, and a Dynamic Query Selection (DQS) mechanism refines these queries. This adaptive approach balances input data fidelity with learned shape priors to achieve state-of-the-art results in general 3D restoration.