The dataset was introduced by Silvio Giancola et al. at the CVPR 2018 Workshop on Computer Vision in Sports. It was designed to solve the problem of —temporally localizing sparse events like goals or cards within long video broadcasts.
: The paper proposes using recent developments in action recognition and detection to provide baselines, reaching a mean Average Precision (mAP) of 67.8% for classifying 1-minute temporal segments. SoccerStar-v1-pc_UQ.7z
Since the original v1 release, the dataset has expanded significantly into newer versions: The dataset was introduced by Silvio Giancola et al
A Scalable Dataset for Action Spotting in Soccer Videos - arXiv SoccerStar-v1-pc_UQ.7z