CoolFace
Modelpublic

martinjolif/yolo-football-pitch-detection

sourceHugging Faceagpl-3.0updated 9mo agoView on Hugging Face
0likes
Model Card

version: 0.1.0

Football Player Detection (YOLOv8x)

This model is a fine-tuned YOLOv8x pose estimator trained to identify keypoints in football (soccer) match images. It was trained on the [martinjolif/football-pitch-detection](https://huggingface.co/datasets/martinjolif/football-pitch-detection) dataset and is suitable for sports analytics tasks such as identifying pitch keypoints.

Detected Classes

  • —32 keypoints

<img src="figure-keypoints.png" alt="figure1" width="800">

Training Details

  • —Base model: Ultralytics YOLOv8x
  • —Task: Pose Estimation
  • —Dataset: martinjolif/football-pitch-detection
  • —License: AGPL-3.0

Evaluation Results

Results on the test split using standard object detection metrics (Precision, Recall, F1, mAP50, mAP50-95):

(To do)

Prediction Examples:

<img src="figure-pred.jpg" alt="figure1" width="1400">

Notes

Intended Use

  • —Football match analytics
  • —Homography calibration
  • —Research in sports computer vision

Limitations

  • —Performance may drop on non-broadcast footage, unusual camera angles, or low-resolution images.
  • —Trained specifically for football (soccer); not intended for other sports without further fine-tuning.

Usage

bash
wget https://huggingface.co/martinjolif/yolo-football-pitch-detection/blob/main/yolo-football-pitch-detection.pt
bash
from ultralytics import YOLO

model = YOLO("yolo-football-pitch-detection.pt")

Citation

If you use this model or the dataset, please cite the dataset authors and Ultralytics YOLO accordingly.