e1250/safety_detection

Safety Detection

A fine-tuned YOLO model for detecting fire and smoke in images and video streams, built for real-time safety monitoring.

Model Details

• Architecture: YOLOv26 (fine-tuned)
• Framework: PyTorch
• Epochs: 52
• Experiment Tracking: ClearML

Classes

ID	Label
0	fire
1	smoke

Dataset

Fine-tuned on the Home Fire Dataset from Kaggle.

• Training Logs: ClearML Experiment

Evaluation Metrics

Metric	Value
mAP@50	0.930
mAP@50-95	0.626
Precision	0.913
Recall	0.891

Usage

from ultralytics import YOLO

model = YOLO("path/to/model.pt")
results = model("image.jpg")

Limitations

• Trained on home fire scenarios — performance may degrade in industrial or outdoor environments
• Detection confidence decreases at stricter IoU thresholds (mAP@50-95: 0.626)
• Not validated for production safety-critical systems without further testing