search

castacks / tartanvo

TartanVO: A Generalizable Learning-based VO
BSD 3-Clause "New" or "Revised" License
235 stars 55 forks source link

Kitti evaluation #3

Open thohemp opened 2 years ago

thohemp commented 2 years ago

Hello,

awesome work!

When I run the test script for the kitti sequence 10 with your provided weights I receive relative errors that differ from those in the TartanVO paper.

==> ATE: 11.8605, KITTI-R/t: 0.0280, 0.0911

In the paper the reported errors are 6.89 (t_r) and 2.73 (r_r).

I used the evaluation script from here: https://github.com/Huangying-Zhan/kitti-odom-eval and received:

Sequence: 10 Translational error (%): 8.008632651428766 Rotational error (deg/100m): 3.2037836856229243 ATE (m): 58.1216975619417 RPE (m): 0.02597469391189542 RPE (deg): 0.09051218838992246

Would be glad about a hint how you evaluated the results.

ghost commented 1 year ago

Have you found this answer?