Fine-Badminton: A Fine-Grained Badminton Dataset for Temporal Action Localization

Fine-Badminton is a fine-grained badminton dataset for temporal action localization. It contains a representative subset of 10 professionally recorded badminton singles matches, 980,703 frames (approximately 10.72 hours), and 9,817 frame-level temporal annotations spanning 29 badminton stroke categories. The dataset is designed to support research in temporal action localization, fine-grained action recognition, and sports video understanding. The video data were collected from publicly available badminton broadcasts on YouTube. Copyright of the original videos remains with the respective content owners. Fine-Badminton is distributed solely for non-commercial academic research purposes.