benchmark results are not accordance with the results presented in the dvo_slam paper

[somebodyus]

Dear author,

I successfully compile the dvo_slam package in ROS fuerte on Ubuntu 12.04.5. I use the default ROS openni calibration parameters.

I benchmark the dataset "rgbd_dataset_freiburg1_xyz" using the following command, roslaunch dvo_benchmark benchmark.launch keep_alive:=true Then I do evaluation using the following command, rosrun rgbd_benchmark_tools evaluate_ate.py --plot PLOT_ate --verbose assoc_opt_traj_final.txt groundtruth.txt

The results of ATE for fr1/xyz are as follows, compared_pose_pairs 790 pairs absolute_translational_error.rmse 0.016561 m absolute_translational_error.mean 0.014532 m absolute_translational_error.median 0.013264 m absolute_translational_error.std 0.007941 m absolute_translational_error.min 0.000828 m absolute_translational_error.max 0.049756 m

The results are not accordance with the results presented in your dvo_slam (IROS2013) paper. The RMSE value for fr1/xyz indicated in the Table.III of your paper is 0.011, but I got 0.016561.

Maybe there are some settings I don't have. Could you tell me how to do correct benchmark?

PS: Could you tell me how to find how many key frames that have been set during the SLAM?

Thank you very much.

[HeYijia]

I also faced the similar problem when I using the dvo package. Did anyone figure out? Many thanks in advance.

[amesh90]

same here but the error is much diferent compared_pose_pairs 791 pairs absolute_translational_error.rmse 0.029669 m absolute_translational_error.mean 0.026008 m absolute_translational_error.median 0.023651 m absolute_translational_error.std 0.014277 m absolute_translational_error.min 0.003234 m absolute_translational_error.max 0.082670 m

[nifengzhiyix2]

same question, and the error is different in some datasets. Can anyone help? Many thanks! such as: fr1/room: in paper is 0.053, the output is 0.192

[ZhouXiner]

Me too. In my test, the fr1/room is 0.2448