Requires numpy==1.18.0
Set BASE_DIR to directory the dataset should be generated
Run python ncap.py
Directories 0-11 are the infrastructure sensor units and car is the vehicle under Test. Each directory contains camera images, LiDAR point clouds and information for each frame. The information is saved in the following format.
actor_type min_x min_y max_x max_y dist
Where actor_type is one of
[car, motorcycle, bycicle, pedestrian]
min_x, min_y, max_x, max_y is the bounding Box of the vehicle
and dist is the distance of the camera to the actor.
The directory WorldData contains information about the world for each frame.
dist car_speed vru_speed
Where dist is the distance between the VRU and car in meters and car_speed and vru_speed are the speeds of the actors in m/s.
Eval is not part of the simulation. It contains the predicted bounding boxes for each frame.
cam_index actor_type min_x min_y max_x max_y confidence
Where cam_index is the camera in the scene for which the prediction was made. It is one of
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, car]
actor_type, min_x, min_y, max_x, max_y are the same as in Cam Data and confidence is the models' confidence score for that prediction.