Rugby sevens tracking dataset

Tracking rugby player is a challenging task. It is a sport of physical contact where player occlusions are very frequent on camera during rucks, tackles and scrums. The players can also adopt a wide range of body postures from sprinting to laying on the ground in a foetal position.

Our dataset is composed of three 40 seconds extracts of matches from the World Rugby 2021 Dubai tournament. Each video clip is annotated with tracking data. Video are encoded at the 1920x1080 resolution and sampled 50 frames per seconds. The camera pan, tilt and zoom to follow the game action.

License and Citation

Videos are the courtesy of World Rugby. This dataset can only be used for non-commercial research purposes.

To get the R7 tracking dataset, please download the license formular, fulfill, sign and send it by eMail to : r7-tracking-dataset@cea.fr Once we received your fulfiled request, download instructions will be sent to you back.

Click here to Download the formular

If you use this dataset for a scientific publication, please cite the following article:

A. Maglo, A. Orcesi and Q.C. Pham, "Efficient tracking of team sport players with few game-specific annotations," 2022 8th International Workshop on Computer Vision in Sports (CVsports 2022)

https://arxiv.org/abs/2204.04049

@INPROCEEDINGS{rugby7_tracking_dataset_2022,
	author={Maglo, Adrien and Orcesi, Astrid and Pham, Quoc Cuong},
	booktitle={2022 8th International Workshop on Computer Vision in Sports (CVsports 2022)},
	title={Efficient tracking of team sport players with few game-specific annotations},
	year={2022},
	url={https://arxiv.org/abs/2204.04049}}

Ground truth and tracklet format

Ground truth annotations are stored in the MOT challenge format in the test folder. The folder structure follows the convention of the MOT challenge.

In the ground truth annotation files, one line corresponds to a tracking bounding box. Column values are separated by commas:

• Frame id
• Track id
• X
• Y
• Width
• Height

The last three column values are unused.

We also release the tracklets we generated for our paper in the tracklets folder in the same format described above. The aim is to match each tracklet to a player identity or discard it.

Videos

Argentina France sample

The track ids format is the following:

• 1XXX: French players
• 2XXX: Argentinian players
• 3XXX: referees
• 4XXX: public (not all persons visible in frames are annotated)

France Chile sample

The track ids format is the following:

• 1XXX : French players
• 2XXX : Chilean players
• 3XXX : referees
• 4XXX : public (not all persons visible in frames are annotated)

France Kenya sample

The track ids format is the following:

• 1XXX: French players
• 2XXX: Kenyan players
• 3XXX: referees
• 4XXX: public (not all persons visible in frames are annotated)

Contact

If you have any question about this dataset, you can contact us by email at: r7-tracking-dataset@cea.fr