SAINetset/SAINetset_v8.0

SAINetset - Wildfire Smoke Detection Dataset

Dataset of real-world images captured by SAI (Sistema de Alerta de Incendios / Fire Alert System) surveillance nodes for wildfire smoke detection in Cordoba, Argentina.

Current version: v8.0 (January 2026)

About SAI

The SAI (Fire Alert System) is an open-source early wildfire detection platform developed by AlterMundi, a civil association in Argentina. The system uses distributed camera nodes with YOLO-based AI (powered by Ultralytics) to detect smoke in real-time and alert local firefighters via Telegram.

In its first month of operation (September 2025), SAI successfully detected 6 real wildfires (including one at 32.8 km distance) with zero missed detections within the monitored area. Since then, the system has continued detecting fires regularly.

Dataset Description

This dataset contains only images from SAI nodes. It is released separately from third-party datasets for licensing clarity.

What's Included

Split	Images	Positives	Negatives	Pos:Neg Ratio
Train	4,462	200	4,262	1:21
Val	1,134	45	1,089	1:24
Total	5,596	245	5,351	1:22

Content Types

• True Positives (VP): Real smoke/fire detections from operating nodes
• False Positives (FP): Hard negatives (urban lights, lens flare, atmospheric conditions)
• Synthetic: AI-generated smoke added to real backgrounds (photorealistic)
• Donated: Community-contributed images from the region

Classes

ID	Class
0	smoke

Label Format

Standard YOLO format (.txt files):

<class_id> <x_center> <y_center> <width> <height>

All coordinates are normalized (0-1). Empty .txt files indicate negative samples (background).

Full Training Dataset

To reproduce SAINet training, combine this dataset with:

1. D-Fire Dataset - GitHub

   • ~21,000 images of fire and smoke
   • License: Check repository

2. Pyro-SDIS Dataset - Hugging Face

   • ~33,000 images from French fire services
   • License: Check repository

Usage

With Ultralytics YOLO

from ultralytics import YOLO

# Train with this dataset
model = YOLO("yolo12m.pt")
model.train(data="path/to/sainetset/data.yaml", epochs=100)

# Or validate
model.val(data="path/to/sainetset/data.yaml")

With Hugging Face Datasets

from datasets import load_dataset

# Load the dataset
ds = load_dataset("SAINetset/SAINetset_v8.0")

# Access samples
sample = ds["train"][0]
print(sample)

Direct Download

# Clone with Git LFS
git lfs install
git clone https://huggingface.co/datasets/SAINetset/SAINetset_v8.0

Structure

sainetset/
├── README.md
├── data.yaml
└── data/
    ├── train/
    │   ├── images/    # 4,462 images
    │   └── labels/    # 4,462 label files
    └── val/
        ├── images/    # 1,134 images
        └── labels/    # 1,134 label files

Related Resources

• SAINet Model: GitHub - sai-sainet-latest
• SAI Project: Website

Citation

If you use this dataset, please cite:

@misc{sainetset2026,
  title={SAINetset: Wildfire Smoke Detection Dataset from SAI Nodes},
  author={AlterMundi},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/SAINetset/SAINetset_v8.0}
}

License

This dataset is released under CC BY 4.0.

You are free to:

• Share - copy and redistribute the material
• Adapt - remix, transform, and build upon the material

Under the following terms:

• Attribution - You must give appropriate credit to AlterMundi

Contact

• Organization: AlterMundi
• Project: SAI - Sistema de Alerta de Incendios