3D MOT code

[chumpblocckami]

In the official website is referencing multiple times the 3D MOT, which essentially runs the evaluation using x,y,z coordinates instead of bounding boxes.

After inspecting the repo, i found no evidence of the 3D MOT evaluation script, nor anything related to x,y,z coordinates in MOT challenge.

Can you please reference me to the point where x,y,z coordinates are used in a way that i can evaluate my world coordinates tracker?

Thanks a lot!

Happy coding

[michael-hoss]

This would indeed be nice.

The docs recommend to convert own data to the MOTChallenge format. When I dig deeper, I find:

TrackEval/docs/MOTChallenge-Official/Readme.md

Line 76 in 12c8791

Similarly, the bounding boxes are ignored for the 3D challenge.

But then, the code here only supports the 2D Box:

TrackEval/scripts/run_mot_challenge.py

Line 84 in 12c8791

dataset_list = [trackeval.datasets.MotChallenge2DBox(dataset_config)]

[michael-hoss]

It also looks like #71 would enable 3D evaluation.

But I also just read that at least the official KITTI 3D Object Detection benchmark still requires you to submit 2D bounding boxes in the image plane