search

HViktorTsoi / PV-LIO

A probabilistic voxelmap-based LiDAR-Inertial Odometry.
GNU General Public License v2.0
351 stars 49 forks source link

Any suggestion on memory usage for large-scale dataset? #7

Open wlt027 opened 1 year ago

wlt027 commented 1 year ago

Hi,thanks for nice work! I have test my own data and get good result.But for large-scale data or long-term data,I found that this process use too much memory,so could you please give some suggestions? ![Uploading Selection_115.png…]()

HViktorTsoi commented 1 year ago

Hello, excessive memory consumption is caused by the large size of the VoxelMap. The structure of the VoxelMap is actually an std::unordered_map. You can try keeping track of the occupancy status of each map cell and remove those cells that have been unoccupied for a long time(or too far away), thus maintaining a smaller local map. We are also planning to add this feature.

HViktorTsoi commented 1 year ago

Besides, if dense point cloud publishing is enabled, the RVIZ process may also consume a lot of memory.