ICDDPM: Image-conditioned denoising diffusion probabilistic model for real-world complex point cloud single view reconstruction

Luda Zhao, Yihua Hu, Xing Yang, Zhenglei Dou, Qilong Wu

Abstract: Highlights•Propose a point cloud generation model ICDDPM leveraging 2-D images as condition.•Design latent shape vectors to integrate encoder and conditional diffusion model.•Redesign the classic diffusion model to cater point cloud single-view reconstruction.•Design the loss functions for the end-to-end training and interfering.•ICDDPM achieved state-of-the-art performance in the majority of public datasets.