The paper "Event Detection in Table Tennis Videos using 2D Keypoints" authored by Rainer Lienhart, Daniel Kienzle, Shin'ichi Satoh, and Anastasiia Bilinska has been accepted at the 9th International ACM Workshop on Multimedia Content Analysis in Sports (ACM MMSports 2026). The workshop will take place on November 10, 2026, in Rio de Janeiro (Brazil) in conjunction with ACM Multimedia (ACM MM) 2026.  

Previous methods for ball trajectory analysis in table tennis usually assume that key events, such as ball-racket contact, have already been precisely identified in advance. The authors present EventNet, a novel, frame-accurate method for the automatic detection of such events in uncut video recordings. Instead of full visual video frames, the method uses a compact input representation of 2D keypoints (ball and table corners) to detect key events. Thanks to this compact representation, an adapted transformer architecture, and targeted data augmentation for camera viewpoints, EventNet generalizes to unseen game scenes.