Game State Reconstruction

Game State Reconstruction

The SoccerNet Game State Recognition task is a novel high level computer vision task that is specific to sports analytics. It aims at recognizing the state of a sport game, i.e., identifying and localizing all sports individuals (players, referees, ..) on the field based on a raw input videos.

Our task.

SoccerNet GS (Game State) takes raw broadcast videos as input and outputs the position of all individuals on the court, their role (player, goalkeeper, referee, ...), team and their jersey number. The output of the game state task can therefore be used to display a minimap or radar view of the game, as illustrated in the image.

All of our classes.

The individuals to identify and localize are among the following classes: {player, goalkeeper, referee, other}. Additonally, the jersey number must be recognized for each player and goalkeeper. Finally, each player and goalkeeper must be associated to a team: {team left, team right}.

Our data.

The public annotated data consists of 57 train, 59 validation and 50 test broadcast videos clips of 30 seconds each from the main camera available at 1080p. The challenge set for 2024 is composed of X separate clips of 30 seconds.

Our Metric.

For our benchmark and challenge, we consider GS-HOTA as the main metric. GS-HOTA is a novel evaluation metric designed specifically for our new task and derived from the HOTA metric for multi-object tracking. GS-HOTA measures the ability to localize sports individuals on the court coordinates space, and identify them by their role (referee, player, goalkeeper, ...), team and jersey number.

For more details, check out our development kit on github