Why can't i achieve the accuracy in the paper

UZ-SLAMLab / ORB_SLAM3

ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM

GNU General Public License v3.0

6.41k stars 2.52k forks source link

Why can't i achieve the accuracy in the paper #826

Open gxf193558 opened 10 months ago

gxf193558 commented 10 months ago

When I used the euroc data set to run monocular and stereocular, I used the evo tool to evaluate and found that the rmse was basically greater than the value in the paper, and the error in the paper was in the decimeter level. For example, for the mh01 sequence, the rmse result I got in the mono case was 0.043 while that in the paper was 0.016

shneka-swamy commented 9 months ago

Hey,

I am facing the same issue with TUM-VI dataset as well. Did you get any feasible solutions?

Mechazo11 commented 9 months ago

@gxf193558 and @shneka-swamy if I recall correctly, the authors mentioned using a "processed" ground-truth. You may want to procure that. In my experiments, using the evo toolbox, the best rmse I got was 0.051 m averaged over 10 runs. I did not use the "processed" ground truth and directly used the ones I got from EuRoC MAV without any modifications to them.