label_instruction.txt

Some illustrations of BLVD.
Label illustration: In the label files, we correspond each frame (*.jpg) with its 3D point cloud (*.txt). There are 13 values in each row of the label files (e.g., "000001.txt") which are saved in the "Label file". The meaning of these values are:

1:IDs of objects;

2:Driving scenario category of ego vehicle 
{
(1)straght driving in highway with low density of participants;
(2)overtaking in highway with low density of participants;
(3)straght driving in highway with high density of participants;
(4)straght driving in urban road with low density of participants;
(5)overtaking in urban road with low density of participants;
(6)straght driving in urban road with high density of participants, and
(7)intersection.
}

3:Behaviour categories of ego vehicle 
{
(1)straight accelerating;
(2)straight decelerating;
(3)turning right;
(4)turning left;
(5)uniformly straight driving;
(6)changing line to left;
(7)changing line to right, and 
(8)stopping.
};

4:Object categories: pedestrian(1), vehicle(2), and rider(3);

5:Interactive behaviour of other object surrounding ego vehicle (specified in the original paper);
The values from the 6-th one to the 13-th one are denoted in the 3D point cloud.

6-11:The location information of 3D bounding box, containing the 3D box and direction;

12:The height value of the top surface of 3D bounding box;

13:The height value of the bottom surface of 3D bounding box;

Illustration of training and testing list: 
In BLVD, we denoted the training list and testing list, in which the first value in each row of the list (e.g., 9/M_H_m23_7_103) can be ignored. In other words, the field of "M_H_m23_7_103" is useful.