English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

3D point cloud is an important 3D representation for capturing real world 3Dobjects. However, real-scanned 3D point clouds are often incomplete, and it isimportant to recover complete point clouds for downstream applications. Mostexisting point cloud completion methods use Chamfer Distance (CD) loss fortraining. The CD loss estimates correspondences between two point clouds bysearching nearest neighbors, which does not capture the overall point densitydistribution on the generated shape, and therefore likely leads to non-uniformpoint cloud generation. To tackle this problem, we propose a novel PointDiffusion-Refinement (PDR) paradigm for point cloud completion. PDR consists ofa Conditional Generation Network (CGNet) and a ReFinement Network (RFNet). TheCGNet uses a conditional generative model called the denoising diffusionprobabilistic model (DDPM) to generate a coarse completion conditioned on thepartial observation. DDPM establishes a one-to-one pointwise mapping betweenthe generated point cloud and the uniform ground truth, and then optimizes themean squared error loss to realize uniform generation. The RFNet refines thecoarse output of the CGNet and further improves quality of the completed pointcloud. Furthermore, we develop a novel dual-path architecture for bothnetworks. The architecture can (1) effectively and efficiently extractmulti-level features from partially observed point clouds to guide completion,and (2) accurately manipulate spatial locations of 3D points to obtain smoothsurfaces and sharp details. Extensive experimental results on various benchmarkdatasets show that our PDR paradigm outperforms previous state-of-the-artmethods for point cloud completion. Remarkably, with the help of the RFNet, wecan accelerate the iterative generation process of the DDPM by up to 50 timeswithout much performance drop.

A Conditional Point Diffusion-Refinement Paradigm for 3D Point Cloud Completion