This model is a fine-tuned YOLO11m object detector trained to identify key entities in football (soccer) match images.
It was trained on the martinjolif/football-player-detection dataset and is suitable for sports analytics tasks such as player tracking or match analysis for example.

Detected Classes

player
goalkeeper
referee
ball

Training Details

Base model: Ultralytics YOLO11m
Task: Object Detection
Dataset: martinjolif/football-player-detection
License: AGPL-3.0

Evaluation Results

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

Class	Images	Instances	Precision	Recall	F1-score	mAP50	mAP50-95
ball	23	24	0.8912	0.6833	0.7735	0.6799	0.3380
goalkeeper	19	19	0.9473	0.9463	0.9468	0.9413	0.8024
player	25	500	0.9576	0.9940	0.9755	0.9937	0.8737
referee	25	56	0.9059	0.9643	0.9342	0.9888	0.7741

Prediction Examples:

Notes

Excellent performance on player detection, with very high recall and mAP.
Ball detection is more challenging due to small object size and motion blur.

Intended Use

Football match analytics
Player, referee, and ball tracking
Annotation assistance for sports datasets
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

wget https://huggingface.co/martinjolif/yolo-football-player-detection/blob/main/yolo-football-player-detection.pt

bash

from ultralytics import YOLO

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

Citation

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