The RMS measured by the rpg dataset is different from the result in the paper

[ZDJeffrey]

Hello Dr. Zhou. I used 538b576 to test the RMS of the Relative Pose Error of the rpg data set, but the results are very different from the results in the paper, and the results of the upenn data set are similar to the results in the paper.
I use rostopic to record pose_pub and groundtruth, then use evo_traj to convert rosbag to tum format, and finally use evaluate_rpe.py to calculate the error.
I would like to ask if additional processing is required for the rpg dataset, or is it a problem with the testing method I am using. At the same time, I would like to ask how you measure Relative Pose Error and Absolute Trajectory Error.

dataset	R in the paper(deg/s)	R(deg/s)	t in the paper(cm/s)	t(cm/s)
rpg_bin	1.2	10.1	3.1	19.3
rpg_box	3.4	22.2	7.2	27.5
rpg_desk	3.1	12.7	4.5	20.6
rpg_monitor	1.7	14.0	3.2	15.8
upenn_flying1	1.0	1.1	6.5	6.8
upenn_flying3	1.2	1.1	7.1	6.5

[ZDJeffrey]

I have found the T_maker_left to transform points and get the RMS of the Relative Pose Error of the rpg data set similar to the results in the paper. However, the RMS of the Absolute Trajectory Error tested after using T_maker_left is still different from the results in the paper , especially when using rpg monitor.

dataset	Paper(cm)	Test(cm)
rpg bin	2.8	3.7
rpg box	5.8	6.1
rpg desk	3.2	4.5
rpg monitor	3.3	8.0
upenn flying1	13.9	14.0
upenn flying3	11.1	9.5