This repo is a vanilla implementation of 3D farthest-point-sampling (FPS) algorithm in paper:

"Eldar, Yuval, Michael Lindenbaum, Moshe Porat, and Yehoshua Y. Zeevi. "The farthest point strategy for progressive image sampling." IEEE Transactions on Image Processing 6, no. 9 (1997): 1305-1315."

The most important equation is eq. 2.6.

Two demos are avaible in this repo:

1. main_sample.py: demonstrates how the points are sampled in FPS and provides a function to visulise sampling process step by step.
2. main_group.py: A simple point grouping method which groups points using a fix radius sphere over the FPS sampled points.

conda install numpy
conda install -c open3d-admin open3d=0.7

To simply run the fps sampling demo:

python main_sample.py

To simply run the fps grouping demo:

python main_group.py

Other parameters can be set:

• --n_samples: num of samples.
• --data: choose an example data to load, available options are "bunny", "circle", "eclipse", or you can set it to a path points to your ply file.
• --manually_step: (only in main_sample) step the sampling process manully by pressing "N/n" key.
• --group_radius: (only in main_group) set the grouping radius.

Example:

python main_sample.py --data="circle" --n_samples=50 --manually_step=True
python main_group.py --data="circle" --n_samples=50 --group_radius=0.06