This website presents supplementary materials accompanying the paper:

A Light-Weight Neural Network for Fast and Interactive Edge Detection in 3D Point Clouds
Chems-Eddine Himeur, Thibault Lejemble, Thomas Pellegrini, Mathias Paulin, Loic Barthe, Nicolas Mellado
CNRS, IRIT, Université de Toulouse, France.
ACM Transactions on Graphics. Volume 41. Issue 1. February 2022. Article No.: 10. pp 1–21. https://doi.org/10.1145/3481804

Paper thumbnail(click to download preprint)

The dataset webpages present interactive visualizations of our quantitative experiments and comparisons. Additional plots, side-by-side comparisons high-resolution versions of the paper figures are available further on this page. Code source, building system and data of this website can be found at: github.com/STORM-IRIT/pcednet-supp.

Abstract: In recent years, Convolutional Neural Networks (CNN) have proven to be efficient analysis tools for processing point clouds, e.g., for reconstruction, segmentation, and classification. In this article, we focus on the classification of edges in point clouds, where both edges and their surrounding are described. We propose a new parameterization adding to each point a set of differential information on its surrounding shape reconstructed at different scales. These parameters, stored in a Scale-Space Matrix (SSM), provide a well-suited information from which an adequate neural network can learn the description of edges and use it to efficiently detect them in acquired point clouds. After successfully applying a multi-scale CNN on SSMs for the efficient classification of edges and their neighborhood, we propose a new lightweight neural network architecture outperforming the CNN in learning time, processing time, and classification capabilities. Our architecture is compact, requires small learning sets, is very fast to train, and classifies millions of points in seconds.

@article{10.1145/3481804,
  author     = {Himeur, Chems-Eddine and Lejemble, Thibault and Pellegrini, Thomas and Paulin, Mathias and Barthe, Loic and Mellado, Nicolas},
  title      = {PCEDNet: A Lightweight Neural Network for Fast and Interactive Edge Detection in 3D Point Clouds},
  year       = {2021},
  issue_date = {February 2022},
  publisher  = {Association for Computing Machinery},
  address    = {New York, NY, USA},
  volume     = {41}, number = {1},
  issn       = {0730-0301},
  url        = {https://doi.org/10.1145/3481804}, doi = {10.1145/3481804},
  journal    = {ACM Trans. Graph.},
  month      = {nov}, articleno = {10}, numpages = {21},
  keywords   = {low resource computing, energy efficiency, datasets, edge detection, neural networks, Point clouds processing}
}
                  

Naming conventions

Classifications

  • PCED(D): PCED (trained on Default)
  • PCED(D): PCED-2C (trained on Default)
  • PCED(E): PCED-2C (trained on EC)
  • PCED(A): PCED-2C (trained on ABC)
  • PCED(D)-ABL-REDUC4: PCED (trained on Default) - 4 scales
  • PCED(D)-ABL-REDUC8: PCED (trained on Default) - 8 scales
  • PCED(D)-ABL-REDUC32: PCED (trained on Default) - 32 scales
  • PCED(D)-ABL-REDUC64: PCED (trained on Default) - 64 scales
  • PCED(D)-ABL-REDUC128: PCED (trained on Default) - 128 scales
  • PCED(S): PCED-2C (trained on Shrec)
  • PCED(D) Ext: PCED (trained on Default - P = Sharp+Smooth)
  • FC(D): FC (trained on Default)
  • FC(D): FC-2C (trained on Default)
  • FC(A): FC-2C (trained on ABC)
  • FC(S): FC-2C (trained on Shrec)
  • FC(D) S+S: FC (trained on Default - P = Sharp+Smooth)
  • CNN(D): CNN (trained on Default)
  • CNN(D): CNN-2C (trained on Default)
  • CNN(A): CNN-2C (trained on ABC)
  • CNN(S): CNN-2C (trained on Shrec)
  • CNN(D) S+S: CNN (trained on Default - P = Sharp+Smooth)
  • CA: Covariance Analysis
  • FEE: Feature Edge Extraction
  • PCPNet(D): PCPNet (trained on Default)
  • PCPNet(A): PCPNet (trained on ABC)
  • ECNet: Edge-aware Point set Consolidation Network (Pre-trained)
  • GT: Ground Truth

Please refer to the paper for technical details and references.

Experiments

  • ABC: ABC Dataset
  • ABC_NOISE_0.04: ABC Dataset (noisy version)
  • SHREC: Feature Curve Extraction on Triangle Meshes (SHREC'19) - Sharp edges
  • DEFAULT: Default Dataset: simple geometrical objects with noise

Additional plots

High resolution figures

Full-scale interactive image comparison

Usage: for each figure, select one method on each side using the left and right dropdown menus. Move the vertical slider horizontally to compare the two images.

Figure 09 bottom

Left image:

Right image:

Figure 09 middle

Left image:

Right image:

Figure 09 top

Left image:

Right image:

Figure 14 bottom

Left image:

Right image:

Figure 14 top

Left image:

Right image:

Figure 15

Left image:

Right image:

Figure 16

Left image:

Right image:

Figure 17

Left image:

Right image:

Figure 17 Close up

Left image:

Right image:

Figure 18

Left image:

Right image:

Figure 18 Close up

Left image:

Right image:

Figure 19

Left image:

Right image:

Figure 19 Close up

Left image:

Right image:

Figure 20

Left image:

Right image:

Figure 20 Close up 1

Left image:

Right image:

Figure 20 Close up 2

Left image:

Right image:

Figure 21 bottom

Left image:

Right image:

Figure 21 top

Left image:

Right image:

Figure 23

Left image:

Right image:

Figure 23 Close up

Left image:

Right image:

Figure 25

Left image:

Right image:

Figure 26

Left image:

Right image:

Figure 26 Close up

Left image:

Right image:

Figure 27

Left image:

Right image:

Figure 27 Close up

Left image:

Right image:

Figure 27 Side view

Left image:

Right image: