search

koide3 / small_gicp

Efficient and parallel algorithms for point cloud registration [C++, Python]
MIT License
370 stars 47 forks source link

Performance issues with smallgicp, while the old fast-gicp version does very well with the same data. PCD is too small warning. #89

Open VTrencsenyi opened 1 week ago

VTrencsenyi commented 1 week ago

The fast-gicp version of GICP works pretty well with my python pipeline and data. However, with small-gicp I keep getting the warning "target/source pcd is too small" and the resulting accuracy is very poor. Do you have any intuition where the difference may be coming from? Note, this is the case regardless of downsampling.

koide3 commented 1 week ago

The warning on the point cloud size is shown when the number of points is less or equals to 10, that indicates something strange is happening on your input point clouds. Can you provide minimum example code and data to reproduce the problem? https://github.com/koide3/small_gicp/blob/e669301de3d90c490ff169a9abd1be8c31816b3f/include/small_gicp/registration/registration.hpp#L34