Awesome

Event Camera Rotation Dataset (ECRot)

The Event Camera Rotation Dataset (ECRot) is a VGA-resolution event camera dataset designed for the development of event-based rotational motion related algorithms. ECRot contains ten real-world sequences recorded with a DVXplorer (640 x 480 pixels) and six synthetic sequences generated with a DAVIS240C camera model (240 x 180 pixels) and a DVS128 model.

We provide all sequences in ROSbag files that contain events using dvs_msgs/EventArray message types, and IMU data using sensor_msgs/Imu message types respectively. The camera calibration files are provided in yaml files.

For real data, rotational motion is generated in two modes: Motorized mount (7 sequences) and Hand-held (3 sequences). The motorized mount rotates the event camera approximately around its vertical axis, while the rotation center slightly deviates from the camera's optical center. In the hand-held sequences, there are also some minor unavoidable translations. To show the advantages of event cameras over standard cameras, these sequences contain challenging illumination conditions (e.g., looking directly at the sun, reflections in the river and windows) as well as dynamic objects (e.g., moving pedestrians, bicycles, cars, leaves and water on the river).

The six synthetic sequences cover indoor, outdoor, daylight, night, human-made and natural scenarios. The size of the input panoramas varies from 2K (playroom), 4K (bicycle), 6K (city and street), to 7K (town and bay) resolution. They were generated using the ESIM simulator. The input panoramas were downloaded from the Internet.

Citation

This data is associated to the paper CMax-SLAM: Event-based Rotational-Motion Bundle Adjustment and SLAM System using Contrast Maximization, IEEE T-RO 2024, by Shuang Guo and Guillermo Gallego.

If you use it in your research, please cite it as follows:

@Article{Guo24tro,
  author        = {Shuang Guo and Guillermo Gallego},
  title         = {{CMax}-{SLAM}: Event-based Rotational-Motion Bundle Adjustment and {SLAM} System using Contrast Maximization},
  journal       = {{IEEE} Transactions on Robotics},
  year          = 2024,
  volume        = {},
  number        = {},
  pages         = {1--20}
}

Sequences

Synthetic data

Sequence	Camera model	Duration [s]	File size
Playroom	DVS128	2.5	47.8 MB
Bicycle	DAVIS240C	5.0	264.3 MB
City	DAVIS240C	5.0	1.0 GB
Street	DAVIS240C	5.0	699.1 MB
Town	DAVIS240C	5.0	743.6 MB
Bay	DAVIS240C	5.0	827.7 MB

Real-world data (with a DVXplorer, VGA resolution)

Sequence	Setup	Duration [s]	Rotation range [deg]	File size
Brandenburg Gate	Motorized mount	8.0	360	1.40 GB
Charlottenburg Palace	Motorized mount	8.4	360	1.18 GB
Victory Column	Motorized mount	10.0	90	62.2 MB
TU Berlin Main Building	Motorized mount	8.5	360	1.41 GB
TU Berlin MAR Building	Motorized mount	4.0	$\approx$ 90	421 MB
Square Center	Motorized mount	8.8	360	1.37 GB
Square Side	Motorized mount	8.0	360	1.46 GB
Crossroad	Hand-held	10.2	360	792 MB
River	Hand-held	5.5	random	862 MB
Bridge	Hand-held	7.5	random	1.08 GB