Paper: CaddieSet: A Golf Swing Dataset with Human Joint Features and Ball Information
Authors: Seunghyeon Jung (Dongguk University), Seoyoung Hong (Kimcaddie Inc), Jiwoo Jeong (Dongguk University), Seungwon Jeong (Dongguk University), Jaerim Choi (Kimcaddie Inc), Hoki Kim (Chung-Ang University), Woojin Lee (Dongguk University)
Venue: 11th International Workshop on Computer Vision in Sports (CVsports) at CVPR 2025
Original article from Money Today
Kimcaddie, a screen golf and golf lesson booking platform, announced on the 14th that it presented a paper on its self-developed golf swing analysis technology at CVSports, a workshop of CVPR 2025, the world’s most prestigious computer vision conference.
The paper is titled “CaddieSet: A Golf Swing Dataset with Human Joint Features and Ball Information.”
The research was a collaborative effort involving Jaerim Choi and Seoyoung Hong from the Kimcaddie AI team, researchers Seunghyeon Jung, Jiwoo Jeong, and Seungwon Jeong from Prof. Woojin Lee’s lab at the School of Computer Science at Dongguk University, and Prof. Hoki Kim from Chung-Ang University.
A Kimcaddie spokesperson stated, “This is the first quantitative approach based on actual user data in the field of golf motion recognition, and it has been recognized for both its technical capability and reliability.”
The paper divides golf swings into 8 phases and precisely extracts joint position data at each phase to compute a total of 15 motion metrics. By pairing ball trajectory information (direction, spin axis, speed, etc.) from the same swing, the study established a foundation for quantitatively analyzing the relationship between swing mechanics and ball movement.
The AI model trained on this dataset showed significantly improved ball trajectory prediction accuracy compared to conventional video-based analysis. It also features the ability to provide explainable analysis of a golfer’s problematic movements. The researchers explained that this lays the technical groundwork for AI to suggest specific corrective directions.
Prof. Woojin Lee of Dongguk University said, “This paper goes beyond a simple technical achievement—it represents a meaningful step toward broadening the reach of AI technology across the golf industry,” and added, “We will focus even more on practical technical research and field applications so that more golfers can improve their skills in a rational and systematic way.”