According to the detail in https://ftc-2024.github.io/, the format of the training data's bounding boxes follows the MOT format which is:
<frame>, <id>, <bb_left>, <bb_top>, <bb_width>, <bb_height>, <conf>, <x>, <y>, <z>
where bb_left, bb_top are x_min and y_min of the bounding box respectively.
But after visualizing the ground truth, it seems to me that the format of the provided data in [https://drive.google.com/drive/folders/1Xn3IY46t7DkG2bntbO6sL3oElRPeQA1d] is given in the form of
<frame>, <id>, <bb_top>, <bb_left>, <bb_height>, <bb_width>, <conf>, <x>, <y>, <z>
Can you please clarify this?
Posted by: xmba15 @ Dec. 22, 2023, 7 a.m.I am also a participant and had a similar problem, and I can confirm that the order of <top>,<left> is reversed from original MOT challenge.
I realized this because I got very bad results in the platform after submitting even though my bbs in video seemed very good.
Seeing the example code shared by the organizers (specifically the python code to translate the code for submission), I noticed that the order was reversed from the original MOTA competition as you pointed out.
Posted by: maburto @ Jan. 10, 2024, 2:12 a.m.