koide3
Hmmm, I'm not sure, but I guess you may need to specify CUDA compute capabilities in CMakeLists.txt as your GPU is a bit old one.
Hi @bpoebiapl , Thanks for reporting the problem. To make fast_gicp properly align your data, I had to downsample input data (with 0.4m voxel_grid) and limit the correspondence distance (```reg.setMaxCorrespondenceDistance(1.0)```)....
I usually use ```libeigen3-dev``` installed via apt. What is the version of your OS?
Yes. I've tested this package on only Ubuntu 18 and 20, and maybe it cannot be built on 16.
Hi @aiai84 , You need to build the library using cmake. I haven't tested this package on windows, though.
Yeah. There should be ```Ptr``` definitions. I'll add ```Ptr``` definitions.
@darrenjkt Can you provide example PCD files? I'm going to improve the stability of the registration algorithm and interested in the cause of your problem.
In short, fast_gicp is much accurate but a little bit slower than ndt_omp. Please take a look at our paper for details: https://easychair.org/publications/preprint/ftvV
I've not seen this problem before. Can you provide some data for reproducing the problem?
Do you mean running like this? With the following line, I couldn't reproduce the problem: ```rosrun fast_gicp gicp_align 1.pcd 1.pcd```