Pipe-Run Extraction and Reconstruction from Point Clouds Supplementary Material

This document contains additional comparisons of our approach and commercial software [1] on large real-world datasets. 1 Additional ExperimentsSection 2.3 of the paper introduced commercial software [1] as a state-of-the-art technique of pipe-run reconstruction. We purchased a license and used theirautomatical pipeline to reconstruct pipe-runs from the same dataset in Figure11 of the paper. The comparison is shown in Figure 1 of this supplementarydocument. Our results significantly outperform that of [1] due to the followingreasons: we introduce global similarities which increase the robustness of primi-tive detection under input scans with high-level noise and data incompleteness;our joint detection is more robust and faithful to the input data. References1. ClearEdge3D: Edgewise plant. http://www.clearedge3d.com/ (Dec 2012),http://www.clearedge3d.com/Products.aspx?show=EdgeWisePlant 2 Rongqi Qiu, Qian-Yi Zhou and Ulrich Neumann Fig. 1. Comparisons of pipe-run reconstruction by different methods: (a) input pointcloud; (b) commercial software [1]; (c) our approach.