Team history

The first version of SoccerNet was released in 2018 by Silvio Giancola at the CVsports workshop at CVPR.  At the beginning, the dataset contained 500 broadcast games on which 3 classes of events where annotated for the action spotting task (goals, cards and substitutions). Quickly, Anthony Cioppa and Adrien Deliège joined the SoccerNet team with a first collaboration on an Action Spotting technique published at CVPR.  

With their supervisors Marc Van Droogenbroeck and Bernard Ghanem, they funded a vast annotation campaign in collaboration with Prof. Thomas B. Moeslund (Aalborg University) to extend the set of events from 3 classes to 17 classes,  covering all possible actions in the games,. Furthermore, they annotated camera cuts and replay information to propose the novel task of replay grounding. They were joined by Meisam Jamshidi Seikavandi ,Jacob V. Dueholm and Kamal Nasrollahi to work on these tasks and propose the first baselines. This dataset, called at that time SoccerNet-v2, was published at the CVsports workshop at CVPR 2021 and won a best paper award. At this same conference, they organised two challenges related to action spotting and replay grounding at the ActivityNet workshop.

Following the success of the challenges, Marc Van Droogenbroeck with the DeepSport project funded a novel vast annotation campaign to annotate player bounding boxes,  camera correspondances and field lines on the SoccerNet broadcast games. Baidu Research, with Zhiyu Cheng, Le Kang and Xin Zhou, joined the crew for the new player tracking task. EVS Broadcast Equipment, with Floriane Magera in the lead, took care of defining the camera calibration task and proposed a first baseline method. Vladimir Somers from UCLouvain, EPFL and SportRadar described the re-identification task and proposed a first benchmark.  This allowed the team to propose 4 challenges including: Player Tracking, Camera Calibration, Player Re-Identification and Action Spotting at CVPR 2022.

In 2023, Hassan Mkhallati joined the crew as lead researcher on the dense video captioning task. Vladimir Somers also proposes a new jersey number recognition task. Thanks to footovision, the SoccerNet team is able to propose a new challenge on ball action spotting, with fine-grained dense temporal annotations. Finally Floriane Magera proposes an improvement of the camera calibration challenge by providing more data and more annotations. All these tasks are featured as guest challenges of the 2023 CVSports workshop at CVPR.

In 2024, two new members, Victor and Abolfazl, joined the team of the game state reconstruction task, and NASK Science, represented by Karolina, Mateusz, Zuzanna, and Michał, proposed a new ball action spotting task.

They Trust US

We would like to thank our funding organisms

And we would also like to thank our sponsors

Finally, we would also like to thank our collaborators