Content
summary Summary

A new AI system can now watch soccer games, identify key plays, and even provide commentary similar to that of a human announcer.

Ad

The technology comes from researchers at Shanghai Jiao Tong University and Alibaba, who built it using the largest collection of soccer footage ever assembled.

Their dataset, called "SoccerReplay-1988," includes nearly 2,000 complete soccer matches from Europe's top leagues and the Champions League, spanning from 2014 to 2024. That adds up to more than 3,300 hours of game footage, with an average of 76 commentary clips per match.

Teaching AI to see soccer through the eyes of a commentator

The researchers used this massive dataset to create MatchVision, which they say is the first all-in-one system for analyzing soccer matches. What makes it special is its ability to handle multiple tasks at once - from tracking game events to generating natural-sounding commentary that considers what's happening on the field.

Ad
Ad

The system can recognize 24 different types of game events, including goals, fouls, and tactical moves like possession changes. When it comes to analyzing fouls, MatchVision examines footage from multiple camera angles to determine both the type of foul and how serious it was.

A table shows AI-identified key moments in matches and AI-generated commentary based on that.
The table shows how MatchVision identifies key moments in matches and generates commentary for each scene. | Image: Rao, Wu, et al.

In tests, MatchVision achieved up to 84% accuracy in identifying game events. The team says it outperformed existing systems not only in identifying plays, but also in generating commentary and calling fouls. They plan to make both the dataset and the model available on GitHub.

One interesting finding shows how AI and human commentators focus on different aspects of the game. While the AI tends to zero in on technical details and tactical moves, human announcers are more likely to capture the emotional flow of the game and provide broader context.

Comparative presentation of soccer scenes with AI and human commentaries on a yellow card, corner kick, goal and goalkeeping save.
Side-by-side examples comparing how human commentators (GT) and AI (ours) describe three key match moments - a controversial yellow card, a corner kick sequence, and a goal-scoring play.| Image: Rao, Wu, et al.

The researchers believe their work represents a significant step forward in automated sports analysis. Looking ahead, they suggest the system could help create automatic game highlights or assist referees - building on existing AI technologies already used for things like offside calls.

Ad
Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Summary
  • Researchers from Shanghai Jiao Tong University and Alibaba have created "SoccerReplay-1988", the largest dataset for AI-based soccer analysis, containing nearly 2,000 complete matches and over 3,300 hours of video footage from major European leagues and the Champions League.
  • Based on this dataset, the researchers developed the MatchVision" AI system, which is capable of recognizing 24 different match events, automatically commenting on them and assessing the severity of fouls by evaluating video footage from different camera angles.
  • MatchVision achieves a recognition accuracy of up to 84%, surpassing previous systems. The dataset and model will be available soon on GitHub.
Sources
Max is managing editor at THE DECODER. As a trained philosopher, he deals with consciousness, AI, and the question of whether machines can really think or just pretend to.
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.