RTAB-Map and COLMAP

[huan_c]

Hi everyone,

I,m using RTAB-Map and I'm trying to export the poses and use them in COLMAP, but the origin is different, in adittion, the poses from COLMAP are 1-N image referenced. Is there any way to play with this poses exportation?

Thanks in advance.

[matlabbe]

Hi,

In rtabmap-databaseViewer, to extract images with ID:

File-> Extract Images... , select Id

to extract poses in camera frame (which is often used in SFM frameworks), x->right, y->down, z->forward:

File -> Export Poses, select RGBD SLAM with ID to know the correspondences with the images extracted, select current Map Graph view, select Camera frame.

cheers,
Mathieu

[FooTheBar]

Is there also a way to pass the optimized poses from COLMAP into RTAB-Map and then generate a map/Database with RTAB? (like using an odometry with infine weight or nor running the pose-graph optimization at all?)

[matlabbe]

If IDs are matching between the new optimized poses and nodes in the database, the "simpliest" way would be to update opt_ids and opt_poses fields in the database directly. A python script with sqlite3 could be a way to implement this (this script can be used to compress and uncompress the opt_poses blob).

If the modified database is used in localization mode, it should keep the modified poses.

cheers,
Mathieu

[PLAN8]

Hi, I'm also very keen to figure out this pipeline in order to try using RTAB for Gaussian Splatting software.

I haven't figured out how to import any of the exported data into COLMAP, any chance you could explain how you go about getting the camera poses and sparse cloud into Colmap GUI please?

Thank you!