Download
MPL Dataset Toolbox
Since the data sequences are provided in individual rosbag files given their large size, we provide an additional, easy-to-use toolbox to guide this downloading process. The most used functions are listed below:
- Event Visualizer: transform raw event stream into accumulated event frames.
- Data Validator: validate the downloaded rosbag or the merged rosbag for its completeness.
- Bag Merger: merge multiple, single-topic rosbags chronologically into one complete rosbag.
- [NEW 2022-05] k4a Projector: reproject depth readings into other sensor frames.
Data Format
We provide all data sequences in two different formats (rosbag files and text files) with identical content. All rosbag files are compressed in bz2 format, and can be easily decompressed by using the bag merger function in MPL dataset toolbox.
- /prophesee/left/events (prophesee_event_msgs/EventArray): Events from the left event camera.
- /prophesee/right/events (prophesee_event_msgs/EventArray): Events from the right event camera.
- /camera/left/image_mono (sensor_msgs/Image): Grayscale images from the left regular camera.
- /camera/right/image_mono (sensor_msgs/Image): Grayscale images from the right regular camera.
- /kinect/depth_image (sensor_msgs/Image): Depth images from the RGB-D sensor.
- /ouster/point_cloud (sensor_msgs/PointCloud2): Point clouds from the LiDAR.
- /imu/data (sensor_msgs/Imu): Full 9-axis readings from the IMU.
-
/gt/pose (geometry_msgs/PoseStamped): Ground truth signals.
- left_event.hdf5 (DSEC format): Events from the left event camera stored in hdf5 format.
- right_event.hdf5 (DSEC format): Events from the right event camera stored in hdf5 format.
- left_camera.zip: Grayscale images with their exposure time from the left regular camera.
- right_camera.zip: Grayscale images with their exposure time from the right regular camera.
- kinect_depth.zip: Depth images from the RGB-D sensor.
- lidar.zip: Point clouds from the LiDAR.
- imu.txt: Full 9-axis readings from the IMU.
- gt.txt (TUM format): Alternative ground truth signals for quantitative evaluation.
Disclaimer on the Depth Stream
- The unit of the depth intensity is in millimeters from the origin of the camera.
- Dropped frames are to be expected, owing to the singularity of the depth calculation in the k4a driver. We have deliberately left a blank on the depth stream to maintain the overall correctness.
Small-scale Data Sequences
board-slow
corner-slow
robot-normal
robot-fast
desk-normal
desk-fast
sofa-normal
sofa-fast
mountain-normal
mountain-fast
hdr-normal
hdr-fast
Large-scale Data Sequences
corridors-dolly
corridors-walk
units-dolly
units-scooter
school-dolly
