What is it about?
The ORDB dataset contains top-view doubles videos with rotated bounding boxes that capture racket orientation. The dataset contains 8,704 frames and 34,816 CVAT-labeled instances (x, y, w, h, θ, i) and is released under a CC BY 4.0 license. YOLOv11-OBB achieves an mAP of 0.80 at 0.5 and can be transferred to back-view videos.
Featured Image
Photo by Glen Carrie on Unsplash
Why is it important?
ORDB introduces the following: - Rotated bounding boxes for racket orientation - Carefully hand-validated top-view doubles annotations - A public CC BY 4.0 release - Strong baselines (YOLOv11-OBB mAP@0.5=0.80), which prove the benefits of rotation-aware detection
Perspectives
Use ORDB to go beyond detection. Fuse racket orientation with player pose and shuttle trajectories to recognize strokes and model tactics. Use self/active learning to reduce labeling, improve cross-view/domain adaptation, and enable real-time coaching on edge devices.
Norio Tagawa
Read the Original
This page is a summary of: Oriented Racket Database (ORDB): An Annotated Dataset for Rotated Racket Detection in Badminton, October 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3728423.3759409.
You can read the full text:
Contributors
The following have contributed to this page
