Fire and non-fire forest dataset

Published: 15 July 2025| Version 2 | DOI: 10.17632/5vb2kds993.2
Contributors:
Al Rafi Aurnob, Sharia Arfin Tanim, Tapu Biswas, Firoz Mridha, Mejdl Safran, Sultan Alfarhood, Dunren Che

Description

# Dataset Description: Forest Fire Image Dataset (Generated with Stable Diffusion 2.1) This dataset was created using the Stable Diffusion 2.1 model to support research in automated forest fire detection. It contains synthetically generated images across two distinct classes: * Fire * Non-Fire #Dataset Composition Total Classes: 2 (Fire, Non-Fire) Images per Class: 1000 Total Images: 2,000 Image Resolutions: 768 × 768 pixels Image Format: All images are in PNG format #Class Details Fire Class: Includes a wide variety of forest fire scenarios, primarily focused on surface-level forest fires. Images depict fires from multiple angles and viewpoints, such as: * Ground-level perspectives * Aerial (sky-to-surface) views * Close-ups and panoramic scenes This diversity ensures robust representation of real-world fire conditions. Non-Fire Class: Comprises natural forest and landscape scenes without any signs of fire, providing essential contrast for binary classification or detection tasks. #Purpose This dataset is intended for use in: * Forest fire detection using computer vision and AI * Training and evaluating deep learning models * Exploring the effectiveness of synthetic data in hazard monitoring tasks

Files

Institutions

  • American International University Bangladesh
  • King Saud University

Categories

Image Segmentation, Image Classification, Forest, Fire, Computer Generated Imagery

Licence