Run roscore in a terminal and keep it running.
roscore
In a separate terminal, launch the camera tracking node
roslaunch dvs_tracking live_tracker_[dataset].launch
In a separate terminal, launch the mapping node
roslaunch mapper_emvs_stereo live_mapper_[dataset].launch
Finally, in another terminal, play the stereo events and GT camera pose for bootstrapping. Note that ES-PTAM is non-deterministic, i.e., results may vary each time you run it on the same rosbag file. This is due to stochastic operations involved in the tracking, and also, the parallelism of the system. The performance differs according to the condition of your PC. Try playing the bag at a slower rate (especially for higher camera resolution) to improve performance as well as reproducibility.
rosbag play [ROSBag file].bag -r [rate] --clock
Below, we provide details for running on different example datasets.
Download the rpg_monitor_edited.bag file from the ESVO project page, which contains stereo events and camera poses.
Run the tracking and mapping nodes using the following launch files.
- Tracking: live_tracker_eccv18_edited.launch
- Mapping: live_mapper_eccv18_edited.launch
Play the ROSBag file
rosbag play rpg_monitor_edited.bag -r 0.4 --clock
You can download pre-packaged ROSBag files from here which contains stereo event data.
Alternatively, you can generate stereo event ROSBag files from the original DSEC h5 data using our h52bag converter like:
events_h52bag zurich_city_04_a_events_left/events.h5 zurich_city_04_a_events_left /dvs/left/events 480 640
events_h52bag zurich_city_04_a_events_right/events.h5 zurich_city_04_a_events_right /dvs/right/events 480 640
Then, download camera poses (obtained via LiDAR-IMU odometry) from here. Thanks to Mathias Gehrig for the data.
Run the tracking and mapping nodes using the following launch files.
- Tracking: live_tracker_dsec.launch
- Mapping: live_mapper_dsec.launch
Play the ROSBag file.
Make sure that the ROS topics are according to what is expected in the launch files. If using the pre-package bag files above, they need to be remapped during playing.
You also need to align the playing of the events and pose ROSBags, hence the offset needs to be set using the -s flag.
rosbag play dsec_zurich_city_04_a.bag zurich_city_04-odometry/pose.bag /davis/left/events:=/dvs/left/events /davis/right/events:=/dvs/right/events -r 0.02 -s 132 -u 34 --clock
For other sequences of zurich_city_04, the offset -s and duration -u can be computed using the start_end_time.yaml files provided in our output trajectories and GT pose timestamps.
From the TUM-VIE dataset, download the following files:
- mocap-desk-events_left.h5
- mocap-desk-events_right.h5
- mocap-desk-vi_gt_data.tar.gz for camera poses.
Convert left and right events from h5 format to ROSBag using our h52bag converter:
./events_h52bag mocap-desk-events_left.h5 mocap-desk-events_left /dvs/left/events 720 1280
./events_h52bag mocap-desk-events_right.h5 mocap-desk-events_right /dvs/left/events 720 1280
Extract the contents of mocap-desk-vi_gt_data.tar.gz into a folder mocap-desk-vi_gt_data. Then, convert poses from mocap_data.txt to ROSBag using this script:
python mapper_emvs_stereo/scripts/mocap_txt2bag.py --path_prefix mocap-desk-vi_gt_data
This should generate pose.bag as output inside the mocap-desk-vi_gt_data folder.
Run the tracking and mapping nodes using the following launch files.
- Tracking: live_tracker_tumvie_calibA.launch
- Mapping: live_mapper_tumvie_calibA.launch
Play the ROSBag file, leaving out the first few seconds containing a lot of flashing lights and no camera motion:
rosbag play mocap-desk-events_left.bag mocap-desk-events_right.bag mocap-desk-vi_gt_data/pose.bag -r 0.01 -s 5 --clock