Which variables are in "additional_info"? about ab3dmot

[xinshuoweng]

Yes. The additional_info is not used by the main algorithm.

In details,
column 0 is frame number
column 1 is the object type
columns 2, 3, 4, 5 are 2D box
column 6 is confidence
columns 7-13 are 3D box
column 14 is the orientation.

[abhigoku10]

@autonomobil @xinshuoweng thanks for sharing the information of this , i wanted to know the details of the columns of 7-13 3D box values since the object detectors you have mentioned will it provided directly the values in the kitti data format , if not is there any code base which does this conversion since my object detector is different

[xinshuoweng]

Yes. The 3D bounding box in columns 7-13 is exactly the same as the KITTI format.

In brief,
columns 7-9 are the dimensions (height width length in meters)
columns 10-12 are the location (x, y, z in meters in camera coordinate)
columns 13 is rotation_y (rotation ry around Y-axis in camera coordinate [-pi, pi])

You can check out the details on the KITTI 3d object detection benchmark. There is a readme file when you download the object development toolkit (devkit_object.zip)

[vikyzeng]

@xinshuoweng I am really confused with the data format. In the main.py , you apply det[3] as theta. Which one is correct?

[xinshuoweng]

Can you please let me know which line you are talking about? I cannot find det[3]

[vikyzeng]

In main.py line230-> KalmanBoxTracker.update.

new_theta = bbox3D[3]
if new_theta >= np.pi: new_theta -= np.pi * 2    # make the theta still in the range
if new_theta < -np.pi: new_theta += np.pi * 2
bbox3D[3] = new_theta

where bbox3D is det

[xinshuoweng]

This is explained inline 340, inside the function of “update(self, dets_all)”, each det is [x, y, z, theta, l, w, h]. So det[3] is theta. What are you confused about?

Basically, the order of the value in det is swapped in line 348, I guess that is what you are confused about. As long as you feed the data based on the format I have written in the readme, you are good to go.