Hugging Face's logo

Datasets:
matjs
/
ignis

Tasks:
Image Segmentation
Languages:
English
Size:
n<1K
Tags:
geografy
wildfire
nature
preservation
License:
Dataset card Files
xet
Community
1
ignis / README.md
matjs's picture
matjs
feat: cover image
5e14f15
preview code
|
raw
history blame contribute delete
2.94 kB
metadata
license: cc-by-4.0
task_categories:
  - image-segmentation
language:
  - en
tags:
  - geografy
  - wildfire
  - nature
  - preservation
pretty_name: IGNIS - Intelligent Geospatial Network for Incendiary Surveillance
size_categories:
  - n<1K

IGNIS - Intelligent Geospatial Network for Incendiary Surveillance

A dataset for image segmentation of wildfires in satellite/aerial imagery. The dataset contains paired images and labels, where each label highlights wildfire-affected regions.

IGNIS Dataset Sample Image

Dataset Summary

This dataset was created to support research in wildfire detection, monitoring, and environmental risk assessment. It can be used for training and evaluating segmentation models.

  • Task: Image Segmentation
  • Domain: Remote sensing / Environmental monitoring
  • License: CC BY 4.0

Supported Tasks

  • Image Segmentation – Identify wildfire regions pixel-by-pixel.

  • Potential Applications:

    • Early wildfire detection
    • Environmental monitoring
    • Risk modeling and prevention systems

Dataset Structure

Data Splits

The dataset is divided into:

  • train
  • validation
  • test

Data Fields

  • image (Image) – RGB image
  • label (Label) – TXT file containing coordinates for the polygons following YOLOv11 format

Example:

{
  "image": "train/images/image_001.jpg",
  "label": "train/labels/image_001.txt"
}

Dataset Creation

Motivation

Wildfires are an increasing threat worldwide. This dataset was built to help researchers and engineers develop segmentation models that can detect wildfire-affected areas in aerial/satellite imagery.

This dataset is originally a personal project, but anyone with expert knowledge in meteorological, geographic, geophisical and related areas might feel free to reach out and help expand the dataset and increase its quality.

Source Data

  • Collection Process: Images were sourced from open satellite/aerial datasets.
  • Annotation Process: Masks were generated using a mix between manual labelling and automatic polygon generation thanks to Roboflow's tools.

Annotations

  • Annotation Guidelines: Each class is labeled as:

    • 0 → Burned Ground (burnt)
    • 1 → Smoke Cloud (smoke_cloud)
    • 2 → Smoke Column (smoke_column)
    • 3 → Wildfire (wildfire)

Licensing Information

  • Dataset License: CC BY 4.0

Citation

If you use this dataset, please cite:

@dataset{ignis,
  title = {Intelligent Geospatial Network for Incendiary Surveillance},
  author = {Matheus J. G. Silva},
  year = {2025},
  url = {https://huggingface.co/datasets/matjs/ignis}
}

Acknowledgements