When do I need to perform post-prediction trajectory alignment to get ATE or RPE?

MichaelGrupp / evo

Python package for the evaluation of odometry and SLAM

https://michaelgrupp.github.io/evo/

GNU General Public License v3.0

3.33k stars 745 forks source link

When do I need to perform post-prediction trajectory alignment to get ATE or RPE? #630

Closed RandyChen233 closed 5 months ago

RandyChen233 commented 5 months ago

Hi all,

Is is true that I don't need to perform any trajectory alignment if I get the trajectory prediction from an end-to-end deep learning based visual odometry network? For example, I deployed the Deep VO model and got the following results for sequence 04 in KITTI:

DeepVO_04

The origin is already aligned. Should I just calculate the ATE simply by doing the root-mean-squared differences? I'm a little confused because the author claims the following:

MichaelGrupp commented 5 months ago

This question seems to be not about evo, but about a specific paper or application. Looks like you should contact the author of the paper for clarification instead, I can't help you with this.