This paper presents a comprehensive study of deep correlation features on image style classification. Inspired by that correlation between feature maps can effectively describe image texture, we design various correlations and transform them into style vectors, and investigate classification performance brought by different variants. In addition to intra-layer correlation, inter-layer correlation is proposed as well, and its effectiveness is verified. After showing the effectiveness of deep correlation features, we further propose a learning framework to automatically learn correlations between feature maps. Through extensive experiments on image style classification and artist classification, we demonstrate that the proposed learnt deep correlation features outperform several variants of CNN features by a large margin, and achieve the state-of-the-art performance.

We participated in Taiwan's National Intercollegiate Athletic Games (NIAG) in early May 2021, and formed a Sport Technology Team of more than 30 scholars, students, and engineers to provide novel systems and solutions that make the athletic games rich in sport technologies. Some of the features could be the first time shown to the Internet audience for large-scale athletic games. The technologies involved include table tennis ball trajectory and bounce distribution, badminton shuttlecock tracking and trajectory, augmented-reality enriched video streaming on social networks, real-time three-dimensional broadcasting with free view-angle, in-stadium video stream pushing by a private fifth-generation (5G) network with multiaccess edge computing, AI-based sport data analytics during live streaming, etc. All the technologies and applications are integrated in a novel technology-enhanced broadcasting system (TEBS) that is dedicated to sport events. This article introduces the respective technologies that we have developed, deployed, and demonstrated in the 2021 NIAG. We stress the layered architecture design and integration of the TEBS, as well as experimental results from real games in the smart stadium and swimming pool. We also discuss the technical challenges and our approaches to tackle them, as well as lessons learned.

To enrich the viewing experience of baseball games and provide some clues for enhancing pitcher's performance, we propose a Kalman filter-based approach to track ball trajectory from single-view pitching sequences. Without setting extraordinary equipments in stadiums or other sensing instruments, this approach robustly extracts ball trajectory for pitching sequences captured from TV channels or downloaded from the Internet. To validate the detected ball trajectories, we investigate the characteristics of ball trajectories on the basis of a baseball physical model. The effectiveness of ball trajectory extraction and ball position detection are presented