Badminton Player + Shuttlecock Detector (YOLOv11s, fine-tuned)

A YOLOv11s object detector for badminton broadcast footage with two classes: player and shuttlecock. Used to track the two singles players (and, secondarily, the shuttle) in a hybrid CV + VLM tactical-analysis pipeline.

Task: object detection (task=detect)
Classes: player (0), shuttlecock (1)
Base checkpoint: yolo11s.pt (Ultralytics)
Framework: Ultralytics 8.4.47

Intended use

Per-frame detection of players (and shuttle) in single-camera broadcast video. Player boxes feed player tracking and court-quadrant assignment.

Note on the shuttle: this model's shuttlecock class is low-recall (small, fast object). For shuttle tracking we use a dedicated higher-resolution model (badminton-shuttlecock-yolov11) at imgsz=1280; this detector is primarily a player detector.

How to use

from ultralytics import YOLO
from huggingface_hub import hf_hub_download

w = hf_hub_download("<your-username>/badminton-players-yolov11", "game_object_yolo11s.pt")
model = YOLO(w)

res = model.predict("frame.jpg", imgsz=960, conf=0.25)[0]
for b in res.boxes:
    cls = model.names[int(b.cls)]   # 'player' | 'shuttlecock'
    xyxy = b.xyxy[0].tolist()

Training


Base model	`yolo11s.pt`
Epochs	150
Image size	960
Batch	4
Dataset	Roboflow Universe — Badminton Players Detection (`hongy20`)
Dataset link	https://universe.roboflow.com/hongy20/badminton-players-detection-gwgb1

Evaluation (validation split, from the training checkpoint)

Metrics are combined over both classes on the dataset's validation split:

Metric	Value
Precision (box)	0.946
Recall (box)	0.724
mAP@50 (box)	0.781
mAP@50-95 (box)	0.568

The combined mAP is pulled down by the hard shuttlecock class; player detection is the reliable output of this model. Per-class breakdown was not stored in the checkpoint.

Limitations

Singles broadcast viewpoint; not validated for doubles or other camera angles.
Shuttle detection is unreliable here — use the dedicated shuttle model.
Small validation split (see source dataset); treat metrics as in-domain.

License

Inherits AGPL-3.0 from Ultralytics YOLO.

Citation

@software{jocher2023yolo,
  author  = {Jocher, Glenn and Qiu, Jing and Chaurasia, Ayush},
  title   = {Ultralytics YOLO},
  url     = {https://github.com/ultralytics/ultralytics},
  version = {11.0.0}, year = {2024}
}
@misc{roboflow_badminton_players,
  title        = {Badminton Players Detection Dataset},
  author       = {hongy20},
  howpublished = {\url{https://universe.roboflow.com/hongy20/badminton-players-detection-gwgb1}},
  journal      = {Roboflow Universe}, publisher = {Roboflow}, year = {20XX}
}

Downloads last month: -