Wildfire Detection: YOLOv11-Nano & YOLOv26-Nano
This repository contains the PyTorch weights (.pt) for YOLOv11-Nano and YOLOv26-Nano models, which are optimized for real-time edge-based wildfire and smoke detection. These models were trained and benchmarked to balance strict hardware constraints (limited memory, low computational power, and energy budgets) with high detection accuracy.
Model Details
- Architecture: YOLOv11-Nano and YOLOv26-Nano
- Task: Object Detection (Bounding Box)
- Classes:
Smoke,Fire - Target Deployment: Autonomous edge devices like Unmanned Aerial Vehicles (UAVs) and embedded systems (e.g., Raspberry Pi).
Performance Metrics
The models were rigorously evaluated on the Smoke-Fire Dataset. Below are the baseline performance benchmarks measured on an NVIDIA T4 GPU:
| Model | Params (M) | mAP@50 | mAP@50-95 | Latency (ms) |
|---|---|---|---|---|
| YOLOv11-Nano | 2.58 | 0.760 | 0.443 | 2.4 |
| YOLOv26-Nano | 2.37 | 0.755 | 0.450 | 2.4 |
How to Use
You can easily download and run these models using the huggingface_hub and ultralytics libraries.
Install Dependencies
pip install ultralytics huggingface_hub
- Python Inference Code
from huggingface_hub import hf_hub_download
from ultralytics import YOLO
# Choose the model you want to use:
# Option 1: "yolo11n.pt"
# Option 2: "yolo26n.pt"
model_name = "yolo11n.pt"
# Download the weights directly from Hugging Face
model_path = hf_hub_download(repo_id="elvinguseinov/wildfire-detection-yolo11-yolo26-nano", filename=model_name)
# Load the model
model = YOLO(model_path)
# Run inference on an image or video
results = model.predict("path/to/test_image.jpg")
# Display results
results[0].show()