Decoupling Spatio-Temporal Adapter for Fine-Grained Badminton Action Localization
Researchers have developed a new method called the Decoupling Spatio-Temporal Adapter (DSTA) to improve the localization of fine-grained actions in professional badminton videos. This approach effectively models complex spatio-temporal dynamics by decomposing motion representation into temporal, vertical, and horizontal spatial variations. The DSTA method achieves state-of-the-art performance on a new benchmark dataset, Fine-Badminton, and the existing ShuttleSet benchmark, while maintaining efficiency in terms of computational and parameter costs. AI
IMPACT Enhances fine-grained action recognition in sports, potentially improving sports analytics and training tools.