3

I want to know what are the 14 values for each object in the kitti training labels. When I label the objects in matlab, i get 4 values for each object viz (x,y,width,height).

What's the difference between the two?

pmdav
  • 301
  • 4
  • 14
  • 1
    The first result if you Google "kitti training labels" is a GitHub issue with linked documentation which lists all of the attributes with descriptions: [here](https://github.com/NVIDIA/DIGITS/blob/v4.0.0-rc.3/digits/extensions/data/objectDetection/README.md). – Wolfie Jul 19 '18 at 10:28
  • You might want to accept the answer. – mrk Nov 11 '19 at 14:07

1 Answers1

14

frame Position of the sample within the sequence

track id Tracking ID of the object within the sequence

type Object type: ’Car’, ’Pedestrian’, ’Cyclist’, ’Tram’, ’Person_sitting’, ’Misc’ or ’DontCare’

truncated Integer (0,1,2) indicating the level of truncation.

occluded Integer (0,1,2,3) indicating occlusion state.

alpha Observation angle of object, ranging [-Pi; Pi]

bbox 2D (0-based) bounding box of the object: Left, top, right, bottom image coordinates

dimensions 3D object dimensions: height, width, length [m]

location 3D object location x,y,z in camera coords. [m]

rotation_y Rotation around Y-axis in camera coords. [-Pi; Pi]

In your case the four Labels you get are a transformed Version of the bbox 2D Labels of the KITTI groundtruth

mrk
  • 8,059
  • 3
  • 56
  • 78