Table of Contents
Score, the computer vision network running on Subnet 44, has launched Detect Football Event, a new mining task that asks participants to identify football actions in low-resolution match footage from youth league games.
The competition moves Score back into the sport where its subnet began, but with a more focused event-detection challenge. Miners must detect match events such as passes, shots, saves, and the ball going out of play, then submit predictions that are scored against the actual sequence of play.
Score said the task opened with a best score of 53.1% against an 85% target. The public Score console later showed Detect Football Event at a 46.6 private score against the same 85.0 target on a live leaderboard that updates as evaluations run.
Detect Football Event runs through Score's challenge console, where miners submit event predictions and receive a private score. Evaluations run hourly, with the current leader, target score, and next evaluation window displayed publicly.
Detect Football Event announcement video | Score on X
The 85% target gives the task a clear benchmark. Score can compare miner outputs against that threshold and route emissions toward submissions that improve the challenge.
The 85% target also makes sense when you consider that football video is a harder computer vision setting than many standard object-detection benchmarks because the model has to understand both where objects are and what is happening over time. A shot, pass, save, or stoppage depends on player movement, ball trajectory, timing, and context rather than a single object in a frame.
At the same time, Score's new task adds another constraint, because the source footage is not polished broadcast video. The clips come from youth football matches, with a single wide camera, small players, a low-resolution ball, and limited visual detail. That makes the challenge closer to the kind of messy footage businesses, clubs, and venues actually produce than to highly curated sports datasets.
