search

Huangying-Zhan / DF-VO

Depth and Flow for Visual Odometry
https://df-vo.readthedocs.io
MIT License
555 stars 117 forks source link

strange result with mono_sc depth model #15

Open Kewenjing1020 opened 3 years ago

Kewenjing1020 commented 3 years ago

Hi, I got the normal result with the kitti_odom/stereo model, but when I did the evaluation with command python run.py -c options/kitti/kitti_mono_sc_0.yml

Here's the result I got:

Evaluate result in result/mono_sc/0? [y/n]y
Sequence: 9
Translational error (%):  286.9784259392447
Rotational error (deg/100m):  0.31097519037206917
ATE (m):  864.6199638028094
RPE (m):  4.013671579698525
RPE (deg):  0.03655727300821237
Sequence: 10
Translational error (%):  336.7985781227918
Rotational error (deg/100m):  0.3488643421558913
ATE (m):  830.726195238146
RPE (m):  2.9854127555332712
RPE (deg):  0.04222684748013362

The result is quite strange, do you know why?

Huangying-Zhan commented 3 years ago

Hi @Kewenjing1020 , since mono_sc depth model is not aligned with real-world scale, you have to use 7DoF alignment (to align the result to the real-world scale) in the evaluation.