Data Format

We fixed the sensor update rate as 1Hz to optimize sensor fusion efficiency. The radar requires three seconds to complete a full 360-degree rotation, however, our data provided updates one-third of the full Cartesian image per second.

The filename of each file corresponds to the acquisition timestamp.

W-band radar, which commonly referred as imaging radar operates at a frequency of 4Hz with polar coordinate images.

The filename of each file corresponds to the acquisition timestamp.

The data structure for the LiDAR contains x, y, z, intensity. It operates at a frequency of 1Hz. (synchronized with X-band radar)

The filename of each binary file corresponds to the acquisition timestamp.

The stereo camera operates at a frequency of  10Hz.

The filename of each binary file corresponds to the acquisition timestamp.

Each calibration file represent the extrinsic calibration parameter of the each sensors from the base frame (GNSS).

The calibration data is provided in SE(3) format with rotation matrix and translation vector.

The intrinsic parameters of each camera also provided.

GNSS file contains [timstamp, latitude, longitude, altitude, roll, pitch, yaw] 

The ground truth of detected objects with its annotation labels in X-band radar, W-band radar, and stereo camera images are provided with JSON file format.

For the annotation label, stereo camera images are all synchronized with X-band radar

For the annotation label of stereo camera, images from the left camera is denoted as "Cam0", and right camera is denoted as "Cam1".

JSON file contains the [timestamp, id, category_id, xmin, ymin, width, height].

id represents the tracked id of detected object.

category_id is fixed with 1, which corresponding to the objects that should be avoided during navigation (e.g. vessels or buoys).

[xmin, ymin width, height] represents the 2D bounding box location of detected object in each image.

For the W-band radar, the bounding box location is represented in 1024x1024 Cartesian coordinate image (not polar coordinate image).