EvSLAM Challenge

Folder Structure

The data used in the EvSLAM challenge can be downloaded via this link, and the folder structure is as follows:


  EvSLAM/
  |-- rosbag/
  |   `-- seq001.bag
  |   `-- seq002.bag
  |   `-- seq003.bag
  |   `-- seq004.bag
  |   `-- seq005.bag
  |   `-- seq006.bag
  |   `-- seq007.bag
  |   `-- test.bag
  |-- calibration/
  │   `-- calib_results_cam_drone.txt
  │   `-- calib_results_imu_drone.txt
  |   `-- calib_results_cam_others.txt
  |   `-- calib_results_imu_others.txt
  │   `-- calib_drone.bag
  |   `-- calib_others.bag
  |-- timestamps/
  │   `-- seq001_ref_time.txt
  │   `-- seq002_ref_time.txt
  │   `-- seq003_ref_time.txt
  │   `-- seq004_ref_time.txt
  │   `-- seq005_ref_time.txt
  │   `-- seq006_ref_time.txt
  │   `-- seq007_ref_time.txt
  |   `-- test_ref_time.txt
  |-- gt/
  │   `-- test_gt.txt

The rosbag folder contains the data bags for each sequence, the calibration folder contains the calibration files for different vehicles, the timestamps folder contains the timestamps associated with the trajectories we expect participants to provide, and the gt folder stores the ground truth data for the test sequences.

Data Format

Our dataset includes data from two event cameras, an Intel RealSense D435, and a motion capture system.

'/dvxplorer_left/events':
(dvs_msgs/EventArray): Events from the left event camera.
'/dvxplorer_right/events':
(dvs_msgs/EventArray): Events from the right event camera.
'/dvxplorer_left/imu':
(sensor_msgs/Imu): Acceleration and angular velocity from IMU.
'/camera/infra1/image_rect_raw/compressed':
(sensor_msgs/CompressedImage): Left-view grayscale images captured by the RealSense camera.
'/camera/infra2/image_rect_raw/compressed':
(sensor_msgs/CompressedImage): Right-view grayscale images captured by the RealSense camera.
'/camera/color/image_raw/compressed':
(sensor_msgs/CompressedImage): Color images captured by the RealSense camera.
'/camera/depth/image_rect_raw/compressed':
(sensor_msgs/CompressedImage): Depth images captured by the RealSense camera.

Calibrations

The calibration format for calib_results_cam_XX.txt is as follows:

Common settings:
Camera model: pinhole + radtan distortion,
Resolution: 640×480
'/dvxplorer_left/events' (cam0):
Extrinsics: T_cam_imu (IMU → cam0)
Intrinsics: [fx, fy, cx, cy]
Distortion: [k1, k2, p1, p2]
Overlaps: cam1–cam4
Time offset: timeshift_cam_imu
'/dvxplorer_right/events' (cam1):
Extrinsics: T_cam_imu (IMU → cam1)
Intrinsics & distortion: same as common
Overlaps: cam0, cam2–cam4
Relative extrinsics: T_cn_cnm1 (cam1 ← cam0)
Time offset: timeshift_cam_imu
'/camera/color/image_raw/compressed' (cam2):
Extrinsics: T_cam_imu (IMU → cam2)
Intrinsics & distortion: same as common
Overlaps: cam0–cam1, cam3–cam4
Relative extrinsics: T_cn_cnm1 (cam2 ← cam1)
Time offset: timeshift_cam_imu
'/camera/infra1/image_rect_raw/compressed' (cam3):
Extrinsics: T_cam_imu (IMU → cam3)
Intrinsics & distortion: same as common
Overlaps: cam0–cam2, cam4
Relative extrinsics: T_cn_cnm1 (cam3 ← cam2)
Time offset: timeshift_cam_imu
'/camera/infra2/image_rect_raw/compressed' (cam4):
Extrinsics: T_cam_imu (IMU → cam4)
Intrinsics & distortion: same as common
Overlaps: cam0–cam3
Relative extrinsics: T_cn_cnm1 (cam4 ← cam3)
Time offset: timeshift_cam_imu

The calibration format for calib_results_imu_XX.txt is as follows:

'/dvxplorer_left/imu':
Extrinsic calibration: T_i_b (IMU → body frame),
Accelerometer noise density: accelerometer_noise_density,
Accelerometer random walk: accelerometer_random_walk,
Gyroscope noise density: gyroscope_noise_density,
Gyroscope random walk: gyroscope_random_walk,
Model: model,
Time offset: time_offset,
Update rate: update_rate Hz.

Sensor Suite

Event cameras: x -> right, y -> down, z -> forward
Frame-based Camera: x -> right, y -> down, z -> forward
IMU: x -> right, y -> down, z -> forward

The specific sensor models and their parameters are shown in the table below:

Sensor	Descriptison
Event Cameras	2 x DVXplorer 640x480, 10cm baseline
Frame-based Camera	Intel RealSense D435 2 x 640x480 Grayscale, 5cm baseline, 30Hz, 1 x 640x480 RGB, 30Hz
IMU	Bosch BMI-160 800Hz
Motion Capture Device	NOKOV Mars26H 120Hz (for drone) and 60Hz (for others)