I noticed in the Limitations Section of the DPVSlam paper monocular depth estimators are mentioned
DPV-SLAM, like most monocular systems, sometimes suffers from scale-drift in outdoor environments, though orthogonal work has addressed this problem using monocular depth networks [47].
Have you all looked into using Metric3D or Unidepth for DPVSlam? In particular Unidepth looks interesting because it also provides estimates for camera intrinsic. I've seen that Metric3D has been done for DROID slam but I'm curious to see how it would work for sparse points like DPVSlam
I noticed in the
LimitationsSection of the DPVSlam paper monocular depth estimators are mentionedHave you all looked into using Metric3D or Unidepth for DPVSlam? In particular Unidepth looks interesting because it also provides estimates for camera intrinsic. I've seen that Metric3D has been done for DROID slam but I'm curious to see how it would work for sparse points like DPVSlam