APE percentage computation in publication

[JanQuenzel]

Hey there, I would like to compare your results from Tab. 3 ( https://arxiv.org/pdf/2205.11784.pdf ) against my algorithm. Unfortunately, I am missing some information on how the percentage for mean APE was computed. Lets say we have ground truth position P_i and odometry estimate Q_i. After SE(3) alignment with T* = argmin_T sum_i^N ||P_i - T * Q_i || ( e.g. using the evo package with --align option ) one gets the translation error e_i=||P_i-T*Q_i|| between individual scan positions. Mean APE would now be mAPE=1/N sum_i e_i, but this is now in meters. How did you normalize this error then? Dividing by the total trajectory length computed from the ground truth poses? Or similar to the KITTI RPE evaluation by 100m?

Furthermore, what was the reasoning for not including all sequences to Tab. 3 ?

Thanks in advance!

[femust]

see here in our localization_analyzer package

[whw-create]

see here in our localization_analyzer package hello, i want to know what's the coordinate of the gt file based on? IMU or Lidar or base_link？