The kitti dataformat and tum dataformat of same result,but have a different APE

[goergehu]

Thank you for your tools. I run some code to have KITTI dataformat result and TUM dataformat result at same time. And I use your tools to evaluate these two results, but they have different results.

like this for kitti:

Loaded 4524 poses from: 00.txt
Loaded 4524 poses from: global_kitti.txt
--------------------------------------------------------------------------------
Aligning using Umeyama's method...
Rotation of alignment:
[[ 0.99685285  0.06100972 -0.05061831]
 [-0.06320537  0.99707494 -0.0429725 ]
 [ 0.0478485   0.0460366   0.99779314]]
Translation of alignment:
[-1.03516941  0.14592958 13.2319137 ]
Scale correction: 1.0
--------------------------------------------------------------------------------
Compared 4524 absolute pose pairs.
Calculating APE for translation part pose relation...
--------------------------------------------------------------------------------
APE w.r.t. translation part (m)
(with SE(3) Umeyama alignment)

       max  3.729264
      mean  2.107056
    median  2.112665
       min  0.372393
      rmse  2.220705
       sse  22310.241984
       std  0.701316

and like this for TUM:

Loaded 4541 stamps and poses from: ../00.txt
Loaded 4524 stamps and poses from: global.txt
Synchronizing trajectories...
Found 4524 of max. 4524 possible matching timestamps between...
    ../00.txt
and:    global.txt
..with max. time diff.: 0.01 (s) and time offset: 0.0 (s).
--------------------------------------------------------------------------------
Aligning using Umeyama's method...
Rotation of alignment:
[[ 0.9979565  -0.04333859 -0.04695297]
 [ 0.04618998  0.99704083  0.06144969]
 [ 0.04415089 -0.06349287  0.99700519]]
Translation of alignment:
[13.32838423  1.40974435 -0.6254514 ]
Scale correction: 1.0
--------------------------------------------------------------------------------
Compared 4524 absolute pose pairs.
Calculating APE for translation part pose relation...
--------------------------------------------------------------------------------
APE w.r.t. translation part (m)
(with SE(3) Umeyama alignment)

       max  3.049987
      mean  1.181944
    median 1.042915
       min  0.201317
      rmse  1.342138
       sse  8149.241122
       std  0.635881

I have transform their coordinate before evaluate, the results confused me a lot, Could you help me?

[MichaelGrupp]

Did you check if they are really the same?

I have transform their coordinate before evaluate

What does this mean?

[goergehu]

Thank you for your reply. I run my code used KITTI-00, and I got results of poses. I save poses as TUM and KITTI format, but the number of poses is less than KITTI groudTruth because of Initialization. Then I delete some begining groudTruth poses so as to have the same number pose for KITTI dataformat results. Finally, I evaluate the APE of these two poses, they have different results, as described before.

By the way, the TUM poses is evaluated in LiDAR Frame, but the KITTI poses is evaluated in CAMERA Frame(why I say transform corrodinate).

[MichaelGrupp]

As said, you would need to check if your data is the same in both formats. If that's not the case, differences are expected.

TUM poses is evaluated in LiDAR Frame, but the KITTI poses is evaluated in CAMERA Frame

This most likely means that they are not the same, as the sensors are probably not in the same location on the vehicle.