Thalos Fire Safety Detection – v1.0 (Roboflow → HuggingFace export)
This repository contains the Fire & Smoke Safety Detection model used in the Thalos Safety Intelligence pipeline.
It identifies fire-related hazards such as:
• open flame
• ignition sources
• sparks / ember activity
• smoke
• early-stage fire risks in job sites, workshops, and industrial settings
This model was originally trained in Roboflow (Object Detection Large / YOLOv8-derived) and exported as a PyTorch .pt file.
It is now hosted on HuggingFace for independent, cost-controlled inference inside the Thalos YOLO service.
⸻
🔥 Usage (Python)
from ultralytics import YOLO
model = YOLO("thalostech2025/thalos-fire-safety-v1/fire_weights.pt")
results = model("example.jpg")
results.show()
⸻
🌐 Load From URL (used by Thalos YOLO Service)
import torch
import requests
from io import BytesIO
response = requests.get(HF_URL)
model = torch.load(BytesIO(response.content), map_location="cpu")
⸻
🏷 Labels / Classes
The following classes are included:
open_flame
smoke
fire_risk
spark
ignition_source
(Labels may vary slightly depending on the v1 Roboflow dataset.)
⸻
📄 License
This model is distributed under the AGPL-3.0 license, consistent with Roboflow export requirements.
⸻
⚠️ Important Notes
• This repository only includes the weights, not the training data.
• Thalos uses this model for server-side inference in the YOLO service.
• A distilled / optimized v2.0 model may be released for faster inference and improved recall.
• Fire detection can be sensitive to lighting; multi-lighting augmentation is planned for v2.
⸻
✨ Maintainer
Thalos Tech (2025)
Safety Intelligence & Hazard Detection Platform
⸻
