Open bingogome opened 2 months ago
bingogome
commented
2 months ago I realized I did not install the G2O dependency. However, after installing G2O, I got some very jumpy results. Both translation and rotation became very unstable. Here is a recording. https://drive.google.com/file/d/1Z8zhv7aNKZkijUcokgZz_GstBfVU1Px4/view?usp=drive_link
matlabbe
commented
2 months ago It should perform roughly the same on Windows and Ubuntu. If you search OMP_WAIT_POLICY on this page, you will some issues depending on the OS config where there could be high artificial CPU usage. Other thing is how the dependencies are compiled between the OS. In your case, you seem to be using mostly system dependencies on Ubuntu, so it should be okay.
For your first videos, fast motion like that and pointing the camera to white wall are more likely to cause red screens, see description here https://github.com/introlab/rtabmap/wiki/Kinect-mapping#lost-odometry-red-screens
For the jumps on the last post, it looks like related to IMU (the extrinsic between imu and camera may be wrong). Is it the same camera on both Windows and Ubuntu? Which kind of camera is it? You can try without IMU by detting IMU filtering to None here:
bingogome
commented
2 months ago Thank you for getting back to me!
For the jumps on the last post, it looks like related to IMU (the extrinsic between imu and camera may be wrong). Is it the same camera on both Windows and Ubuntu? Which kind of camera is it? You can try without IMU by detting IMU filtering to None here:
In all my tests, I use the same camera connected to the same PC with a dual boot. It is a RealSense 435i. I was also expecting the same behavior since the only difference I can think of is the dependency. Later, I also checked the settings options. I found one difference is the feature extraction. On Windows, I have GFTT+BRIEF, but it was grayed out on Ubuntu. On Ubuntu, it was GFTT+ORL.
Thank you for your suggestion to disable IMU. I will report back after the tests.
Edit Sep 09: Disabling IMU filter did the job, thank you so much! That being said, you also mentioned the calibration results might be wrong. Would it still be the case? Should I calibrate the camera? Even that's the case, I am still not sure what was the reason that it worked fine on Windows (no jumps in the last video) but jumped on Ubuntu?
Edit Sep 12: It looks like there is something wrong in the firmware of RealSense 2. The newest version 5.16.0.1 is the cause of the jumpy odometry. Older version 5.15.1.0 fixed it. However, this still does not explain the difference on Windows (no jumps even with 5.16.0.1 firmware) and Ubuntu.
Hi,
Thank you for your amazing work!
I noticed some performance difference on Windows and Ubuntu while I move the camera. It seems on Windows the perform is better. Is this somewhat expected? On Windows, I used the released installer and on Ubuntu built from the source. I wonder if there is a way to look at the dependency, setting difference etc?
Also, the Odometry Viewer on Ubuntu has some disruption. It doesn't behave the same as in the main app.
Some screen recordings:
My CMake: