RTAB-Map's visual odometry drifts more than Tango's visual odometry. This may be caused by bad depth/RGB image synchronization and/or bad depth registration. Can you share the resulting database with orbbec?
For the scale problem, it is related to camera calibration (in particular fx anf fy parameters). You may recalibrate the RGB camera to compare de K matrix. You may also evaluate if the depth values make sense too.
For the orbbec database, there are some registration errors, the depth image is not exactly matching the color image (look at the red chair):
For the realsense database, adding more local (between consecutive nodes) loop closures would not add really more accuracy than the odometry provide. Here is a comparison without and with loop closures respectively:
Having a real loop closure between the beginning and the end would help more to correct the odometry drift. For this kind of path (no real loop closure), I would try to tune OdomF2M parameters:
$ rtabmap --params | grep OdomF2M
Param: OdomF2M/BundleAdjustment = "1" [Local bundle adjustment: 0=disabled, 1=g2o, 2=cvsba.]
Param: OdomF2M/BundleAdjustmentMaxFrames = "10" [Maximum frames used for bundle
Param: OdomF2M/MaxSize = "2000" [[Visual] Local map size: If > 0 (example 5000), the odometry will maintain a local map of X maximum words.]
As odometry F2M does local bunble adjustment between local frames, it should give better optimized transforms locally. If you still want to add loop closures on consecutive nodes in real-time, set RGBD/ProximityDetectionByTime to true.
$ rtabmap --params | grep ProximityByTime
Param: RGBD/ProximityByTime = "false" [Detection over all locations in STM.]
There is indeed no ROS service to call "Detect more loop closures", though it would be possible to add the interface, I opened an issue about that.
Finally, RTAB-Map on Google Tango uses the visual-inertial odometry approach from Google, which gives better odometry than RTAB-Map's odometry approach, explaining why you can get better results on Tango than with the cameras alone.