You may download the dataset from: https://github.com/haleqiu/tartanair-shibuya
- Eigen (Tested on v3.3)
- OpenCV (Tested on Opencv 3.4)
- Pangolin (v0.5)
Run the ./build.sh
in root directory to build third party libraries, AirDOS and construct a binary representation of ORB vocabulary file.
Currently, the project only supports Stereo
mode. The stereo_human
executable in Examples/Stereo
./Examples/Stereo/stereo_human \
./Vocabulary/ORBvoc.txt \
./Examples/Stereo/config/tartanair.yaml \
/.../TartanAir_shibuya/RoadCrossing07 \
./Evaluation/data/trajectory_output.txt
For evaluation, please check https://github.com/castacks/tartanair_tools.git.
Note: the TartanAir Tools expect to read the trajectory format with 7 columns, representing the translation and quaternion.
The trajectory exported by the program contains 8 columns, where first column is the timestamp of pose.
To evaluate the exported trajectory using TartanAir Tools, you need to remove the first column of exported trajectory file.
We also provided an evaluation tool based on evo
package, for more detail, please check this README.md.
AirDOS: Dynamic SLAM benefits from Articulated Objects
@inproceedings{qiu2022airdos,
author={Qiu, Yuheng and Wang, Chen and Wang, Wenshan and Henein, Mina and Scherer, Sebastian},
booktitle={IEEE International Conference on Robotics and Automation (ICRA)},
title={AirDOS: Dynamic SLAM benefits from Articulated Objects},
year={2022},
pages={8047-8053},
doi={10.1109/ICRA46639.2022.9811667}
}