YOLOv26m Finetuned for SportsMOT (Football)
This repository contains a YOLOv26m model fine-tuned on the SportsMOT dataset, specifically optimized for football player detection and tracking in broadcast videos.
Performance Metrics
The model was evaluated on the validation set with the following results:
| Metric | Value |
|---|---|
| mAP50 | 0.9705 |
| mAP50-95 | 0.8377 |
| Precision | 0.9395 |
| Recall | 0.9416 |
| Inference Speed | 6.7ms (TensorRT/FP16) |
Detailed Validation Metrics
The following metrics were captured at the best epoch:
- Box Loss (Val): 0.617
- Class Loss (Val): 0.397
- F1-Score: 0.94 at 0.528 confidence
Training Results)
Validation Predictions
Usage
You can use this model directly with the ultralytics library:
from ultralytics import YOLO
# Load the model directly from Hugging Face
model = YOLO("HLouy/yolov26m-sportsmot-football")
# Run tracking on a football match video
results = model.track(source="path/to/video.mp4", conf=0.3, iou=0.5, save=True)