search

robotology / superquadric-lib

GNU Lesser General Public License v2.1
16 stars 3 forks source link

Improve point cloud downsampling #3

Open giuliavezzani opened 5 years ago

giuliavezzani commented 5 years ago

Currently two methods of downsampling are available:

  1. Uniform in the indices of the points.
  2. Random sampling.

Unfortunately, 1. is not really uniform in the space due to how the points obtained from SFM are sorted. With 2. instead, we are able to obtain pretty spatial-uniform subsampling but paying the lack of reproducibility due to the randomness of the method.

In the future, it would be interesting to add a deterministic sub-sampling method guaranteeing uniformity in space.

giuliavezzani commented 5 years ago

The current downsampling is done randomly, but always using the same seed in order to allow reproducibility.