Getting very poor estimation when reproducing the work on fr3_walking_rpy sequence

[GUESSVPN]

Hi, it seems that something was wrong in my reproducing your excellant work. I got almost the same estimation to your paper in any other four sequences except for fr3_walking_rpy one. The numerical RPE values of this sequence came even larger than that of original ORB-SLAM2. I am confused, could you tell me what is the cause of this ?

compared_pose_pairs 9695 pairs
translational_error.rmse 0.570246 m
translational_error.mean 0.453428 m
translational_error.median 0.362260 m
translational_error.std 0.345808 m
translational_error.min 0.000000 m
translational_error.max 1.396033 m
rotational_error.rmse 11.530595 deg
rotational_error.mean 9.154053 deg
rotational_error.median 7.271315 deg
rotational_error.std 7.011271 deg
rotational_error.min 0.000000 deg
rotational_error.max 28.179311 deg

[zoeyuchao]

Do you mean fr3_walking_rpy doesn't work well (even larger than ORB-SLAM2), but others work well? Maybe you can check if you use the right sequence and right pos results? It seems really strange.

[GUESSVPN]

I downloaded the squence from TUM again and I still got worse RPE result. As you can see in the picture I pinned below, the ATE value is nearly the same with your paper, but the RPE..., that's so weird 😭.

[luyitry]

Hi, I used Evo tools and fr3_walking_xyz sequences to reproduce the data but didn't get the same results in the paper. Could you tell me how to reproduce the work in fr3_walking_xyz sequences? Thanks a lot!

[Wei0703]

@zoeyuchao I have the same question, and result like the following picture, could you tell me what causes this?

@GUESSVPN I have the same issue with you, did you solve it?

Thanks for your reading.
Best regards!