Structural-Adaptive Contrastive Chamfer Distance for Robust Point Cloud Completion

Point cloud completion is vital in 3D computer vision, yet real-world data often presents incomplete representations due to occlusions and sparsity. While Chamfer Distance (CD) is a prevalent loss function, it often suffers from gradient degradation, poor global structure sensitivity, and inadequate geometric distribution modeling, leading to artifacts and suboptimal reconstructions. To address these issues, we propose the novel Structural-Adaptive Contrastive Chamfer Distance (SA-CD). Our SA-CD integrates a structural decomposition of CD into local fitting and global coverage terms, adaptively balancing their importance based on training stage and geometric complexity. Concurrently, a point-level contrastive learning module enhances feature discriminability by maximizing similarity with ground truth neighbors and minimizing with challenging negative samples. This framework guides accurate geometric reconstruction and robust feature learning. Experiments on diverse datasets show SA-CD consistently outperforms state-of-the-art methods. Ablation studies validate component contributions, and human evaluations confirm superior perceptual quality. Supplementary Material File (sa_cd.pdf) - Download - 2.39 MB Information & Authors Information Version history Copyright This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License

Authors Metrics & Citations Metrics Article Usage 109views 69downloads Citations Download citation Boyuan Xie, Yixuan Cao. Structural-Adaptive Contrastive Chamfer Distance for Robust Point Cloud Completion. Authorea. 02 April 2026. DOI: https://doi.org/10.22541/au.177516444.46231426/v1 DOI: https://doi.org/10.22541/au.177516444.46231426/v1 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu.