Data Format Description
=======================

The data for the different cars and crash types can be found in the corresponding folders
(e.g., 21_linear_unusual_hard) identifying the class of the cars.
The sub-folders are structured as follows:

  - anno/ contains the original annotation file (numpy)
  - lidar/ contains the point cloud (bin)
  - crashd_infos_kittiformat.pkl contains all annotations of the folder in KITTI format (pickle)

_________________________
The original annotation file contains a numpy array with dimensions (Number of cars, 8, 3) with
following description:

* `front_bottom_left`:	Bottom left point of the front rectangle as an (x, y ,z) triplet
* `front_bottom_right`:	Bottom right point of the front rectangle as an (x, y, z) triplet
* `front_top_left`: 	Top left point of the front rectangle as an (x, y, z) triplet
* `front_top_right`: 	Top right point of the front rectangle as an (x, y, z) triplet
* `rear_bottom_left`: 	Bottom left point of the rear rectangle as an (x, y, z) triplet
* `rear_bottom_right`: 	Bottom right point of the rear rectangle as an (x, y, z) triplet
* `rear_top_left`: 	Top left point of the rear rectangle as an (x, y, z) triplet
* `rear_top_right`: 	Top right point of the rear rectangle as an (x, y, z) triplet

_______________________
crashd_infos_kittiformat.pkl contains a list of dictionaries for each sample containing:

'frame': 	Sample id (str); The id is the same for the scenes with damaged and corresponding repaired vehicles.
	 	These can be found in the corresponding folders named (clean_ex***).
'type': 	Object type: 'car'.
'location': 	3D object location x,y,z in LiDAR coords. [m]
'dimensions': 	3D object dimensions: height, width, length [m]
'rotation_y': 	Rotation around Y-axis in LiDAR coords. [-Pi; Pi]
'occluded': 	Integer (0,1,2,3) indicating occlusion state;
		Included for completeness, all cars are always visible.
'truncated': 	Integer (0,1,2) indicating the level of truncation.
		Included for completeness, no truncation occurs.