From the DSEC dataset, download the following files:
- interlaken_00_b_events_left.zip
- interlaken_00_b_events_right.zip
- interlaken_00_b_calibration.zip for camera and hand-eye calibration.
Camera poses obtained using LiDAR-IMU odometry are available here in ROSBag format. Thanks to Mathias Gehrig for the data.
Extract the zip files. Convert left and right events from h5 format to ROSBag. Clone and install our h52bag converter. Then, convert using:
./events_h52bag interlaken_00_b_events_left/events.h5 interlaken_00_b_events_left /dvs/left/events 480 640
./events_h52bag interlaken_00_b_events_right/events.h5 interlaken_00_b_events_right /dvs/right/events 480 640
Since each h5 file is big (>500M events), this will generate multiple ROSBag files containing events from both cameras. In this example, we'll use the files interlaken_00_b_events_left_1.bag and interlaken_00_b_events_right_1.bag, which comprises a subset of the whole interlaken_00_b sequence.
Set the correct path of the input events, poses and the unzipped calibration files by editing the configuration file mapper_emvs_stereo/cfg/DSEC/interlaken_00_b_2/dsec.conf.
Finally, run mapper_emvs_stereo:
roscd mapper_emvs_stereo
cd cfg/DSEC/interlaken_00_b_2
rosrun mapper_emvs_stereo run_emvs --flagfile=dsec.conf
This will process the subsequence interlaken_00_b_1 and generate a sequence of time-stamped output files (depth maps and confidence maps) in the current folder.
 |
 |
| Depth map | Confidence map |
Download the rpg_monitor_edited.bag file from the ESVO 2020, which contains stereo events and camera poses.
Set the correct path of the filtered bag file in the --bag_filename parameter in the file mapper_emvs_stereo/cfg/rpg_eccv18/monitor_edited/alg1/rpg_monitor_edited_fixedts.conf. Make sure that the topic names are correct. Finally, run mapper_emvs_stereo
roscd mapper_emvs_stereo
cd cfg/rpg_eccv18/monitor_edited/alg1/
rosrun mapper_emvs_stereo run_emvs --flagfile=rpg_monitor_edited_fixedts.conf
The output files will be saved in the current directory.
The raw depth points are stored in 014.000000depth_points_fused_2.txtfile in the format [row column depth].
The color-coded inverse depth map is saved as 014.000000inv_depth_colored_dilated_fused_2.png.
The suffix _2 denotes the fusion function used (Harmonic mean -HM- in this case).
 |
 |
| Depth map | Confidence map |
From the TUM-VIE dataset, download the following files:
- mocap-desk2-events_left.h5
- mocap-desk2-events_right.h5
- mocap-desk2-vi_gt_data.tar.gz for camera poses.
- camera-calibrationA.json for camera and hand-eye calibration.
Convert left and right events from h5 format to ROSBag. Clone and install our h52bag converter. Then, convert using:
./events_h52bag mocap-desk2-events_left.h5 mocap-desk2-events_left /dvs/left/events 720 1280 700000000
./events_h52bag mocap-desk2-events_right.h5 mocap-desk2-events_right /dvs/left/events 720 1280 700000000
This should generate 2 bag files for the events, namely mocap-desk2-events_left_0.bag and mocap-desk2-events_right_0.bag. This was tested with 32GB RAM. If you run out of memory, use a lower number for events_per_bag instead 700000000. This will split the output into multiple ROSBag files.
Extract the contents of mocap-desk2-vi_gt_data.tar.gz into a folder mocap-desk2-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-desk2-vi_gt_data
This should generate pose.bag as output inside the mocap-desk2-vi_gt_data folder.
Set the correct path of the input events, poses and the calibration file by editing the configuration file mapper_emvs_stereo/cfg/tumvie/desk2_full/tum-vie.conf.
Finally, run mapper_emvs_stereo:
roscd mapper_emvs_stereo
cd cfg/rpg_eccv18/tumvie/desk2_full
rosrun mapper_emvs_stereo run_emvs --flagfile=tum-vie.conf
This will process the whole desk2 sequence and generate a sequence of time-stamped output files (depth maps and confidence maps) in the current folder.
 |
 |
| Depth map | Confidence map |