Dataset Format and Difference between tracking and tracking-2023

[WesleyHsieh0806]

Hi, thanks for your great work. Some questions regarding the annotation format: In the main paper, you claimed that there are 10 columns

The ground truth and detections are stored in their own subfolder in comma-separate csv files with 10 columns. These values correspond in order to: frame ID, track ID, top left coordinate of the bounding box, top y coordinate, width, height, confidence score for the detection (always 1. for the ground truth) and the remaining values are set to -1 as they are not used in our dataset, but are needed to comply with the MOT20 requirements

However, there are only 9 columns in MOT20. I am curious that if the last column has any meaning?	MOT20 dataset format

Also, is there any difference between tracking and tracking-2023? Do they only differ in the challenge set?

mySoccerNetDownloader.downloadDataTask(task="tracking", split=["train", "test", "challenge"])
mySoccerNetDownloader.downloadDataTask(task="tracking-2023", split=["train", "test", "challenge"])

[SilvioGiancola]

Hi @WesleyHsieh0806 , The main difference between tracking (from 2022) and tracking-2023 are the set of sequences in the challenge set that are different from each other and the fat that we do not provide the GT bounding boxes anymore.

To clarify the splits, we provide the following:

• train set, with GT bbox and tracklet information to train on
• test set, with GT bbox and tracklet information to test offline
• challenge set from tracking (from 2022), with GT bbox only to evaluate online on https://eval.ai/web/challenges/challenge-page/1539/overview
• challenge set from tracking-2023, without any bbox to evaluate online on https://eval.ai/web/challenges/challenge-page/1951/overview

As per the MOT format, I will let @paulxinzhou answer that detail :)

Cheers,

[paulxinzhou]

Hi @WesleyHsieh0806, you are right. I think we have an extra -1 column at the end compared to MOT20.