All data sequences are accessible at https://vcu-rvi-dataset.github.io/2020/08/14/Dataset-Download/
If you run into issues like "blocked by Google Drive now (You need permission)" from the previous webpage, please use this link to access the datasets instead!
The ground truths of the device trajectories are accessible in the folder tools.
Implementations of VIO methods to use the benchmark with examples are shown below.
A preview for the data sequences in the benchmark is shown below
- VINS-Fusion: https://github.com/rising-turtle/VINS-Fusion
- VINS-RGBD: https://github.com/rising-turtle/VINS-RGBD
- VINS-Mono: https://github.com/rising-turtle/VINS-Mono
- Open_VINS: https://github.com/rising-turtle/open_vins
- DUI-VIO: https://github.com/rising-turtle/DUI_VIO
Topics:
- depth topic: /cam0/depth
- color topic: /cam0/color
- imu topic: /imu
Ground truths can be downloaded in the tools folder as well as files with helper functions.
In the folder tools, process_result.py (or process_result_old.py for old scipy version) can be used to process the VINS' trajectory output, by calling
python process_result.py [trajectory_result].csv [data_sequence_name] [trajectory_result].csv is the output of VINS-Mono/VINS-RGBD/DUI-VIO, some examples can be found in the folder test_results
[data_sequence_name] are the data sequence names defined in Table III and IV in the paper
e.g. python process_result.py lab_motion2.csv lab-motion2
This ouputs a new trajectory file in the TUM format, with suffix [trajectory_result]_tum.csv.
Next, use evo to evaluate the results compared to the ground truth
e.g. evo_ape tum lab_motion2_gt.csv lab_motion2_tum.csv -a -p
Or, we can use the following command to align to the ground truth with the first N frames:
e.g. evo_ape tum lab_motion2_gt.csv lab_motion2_tum.csv -a --n_to_align 150 -p
