You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am running on a docker image, using ORBSLAM3 ROS Front-End and COVINS Server Back-End (not COVINS-G). The computer is MSI GP68HX, i9-13950HX rtx4060 16G.
I found that on MH_01 sequence of EuRoc, placerec_be.cpp can run Optimization::OptimizeRelativePose(). However, the number of matching point pairs, numCorrespondences, is 6, which is much smaller than the number of outlier-"free" optimization 12. Therefore, loop detection (ComputeSE3) is not successful.
On MH_03 or MH_04 sequences of EuRoc, placerec_be.cpp does not run Optimization::OptimizeRelativePose(). Therefore, loop detection is not successful.
On MH_02 or MH_05 sequences of EuRoc, due to the reset of the ORBSLAM3 Front-End, the server eventually got an error "ProcessKeyframeMessages:144: Received full msgs for existing KF KF 0|0
/bin/bash: line 2: 1520 Segmentation fault (core dumped) rosrun covins_backend covins_backend_node". I set comm.start_sending_after_kf: 50 and orb.imu_stamp_max_diff: 2.0.
Therefore, I cannot run MH_01, MH_02 and MH_03 in multiple terminals on one device in the form of multi-session. The trajectory accuracy on EuRoc, single agent mode is much lower than ORBSLAM3. I think the reason is that the loop detection is not successful. But I am using the default configuration, can you give me some help?
When running multi-agent mode, such as running the MH_01 sequence and MH_03 sequence, the place recognition is successful.
Anyway, thank you very much
The text was updated successfully, but these errors were encountered:
I am running on a docker image, using ORBSLAM3 ROS Front-End and COVINS Server Back-End (not COVINS-G). The computer is MSI GP68HX, i9-13950HX rtx4060 16G.
I found that on MH_01 sequence of EuRoc, placerec_be.cpp can run Optimization::OptimizeRelativePose(). However, the number of matching point pairs, numCorrespondences, is 6, which is much smaller than the number of outlier-"free" optimization 12. Therefore, loop detection (ComputeSE3) is not successful.
On MH_03 or MH_04 sequences of EuRoc, placerec_be.cpp does not run Optimization::OptimizeRelativePose(). Therefore, loop detection is not successful.
On MH_02 or MH_05 sequences of EuRoc, due to the reset of the ORBSLAM3 Front-End, the server eventually got an error "ProcessKeyframeMessages:144: Received full msgs for existing KF KF 0|0
/bin/bash: line 2: 1520 Segmentation fault (core dumped) rosrun covins_backend covins_backend_node". I set comm.start_sending_after_kf: 50 and orb.imu_stamp_max_diff: 2.0.
Therefore, I cannot run MH_01, MH_02 and MH_03 in multiple terminals on one device in the form of multi-session. The trajectory accuracy on EuRoc, single agent mode is much lower than ORBSLAM3. I think the reason is that the loop detection is not successful. But I am using the default configuration, can you give me some help?
When running multi-agent mode, such as running the MH_01 sequence and MH_03 sequence, the place recognition is successful.
Anyway, thank you very much
The text was updated successfully, but these errors were encountered: