A Noising-Denoising Framework for Point Cloud Upsampling via Normalizing Flows

Xin Hu, Xin Wei, Jian Sun

Abstract: Highlights•We study a challenging but realistic task that is how to flexibly upsample the sparse point cloud in arbitrary ratios, even without the given supervised high resolution point cloud.•We propose a noising-denoising framework for 3D point cloud upsampling, trained in supervised and self-supervised settings. Once our network is trained, it can be applied for upsampling in different ratios.•Extensive experiments demonstrate that our proposed approach achieves competitive upsampling results on public datasets.