Dense mapping

[Efesendil]

Is there a way to save the dense SLAM map output ? Or is it only possible to save sparse feature map with the current implementation ?

[zwdoescode]

Hi @Efesendil Could you please clarify on the dense SLAM map output?
Are you referring to 3D reconstruction of the environment?

[Efesendil]

Not exactly the reconstruction as a mesh, or using nvblox. Is there a way by default to build the dense point cloud map ?
For example, getting the dense point cloud scan of the key frame used, and then transforming these dense scans with the SLAM poses. Once we have the final optimized trajectory and if we can save the dense scans of the chosen key frames, we should be able to build the map, right ?

[hrabeti-nvidia]

SLAM systems rarely output very dense point clouds. Have you tried FoundationPose + nvblox?

[Efesendil]

I agree when using monocular or stereo cameras, since a dense point-cloud map is not very useful in those cases. However, RGB-D cameras can output accurate point clouds that can be used to build a dense point-cloud map of the environment. That’s why having access to the corresponding raw point clouds of the keyframes would be valuable.
I understand that this repository is mainly intended for navigation, but adding such a feature would broaden its potential use.