# Edge Forest Fire Challenge 2025 — Data

来源：https://www.kaggle.com/competitions/edge-forest-fire-challenge-2025/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

The Forest Fire Dataset for Binary Classification is a comprehensive, multi-source image collection designed specifically to train and evaluate machine learning models for wildfire and forest fire detection. It provides a clean, organized, and labeled dataset ideal for binary classification tasks, helping participants build reliable and efficient models for real-world deployment.

The dataset is structured into two primary classes:

Fire: Images containing visible fire, flames, or smoke indicating active or potential wildfire.

NoFire: Images representing forested or natural landscapes without any visible fire or smoke.

Each image varies in size, resolution, and lighting conditions to ensure diversity and realism, reflecting the wide range of scenarios that occur in natural environments. The dataset supports experimentation with both traditional computer vision and deep learning-based approaches.

🗂 Dataset Structure

The dataset is divided into three subsets to support effective model training and evaluation:

Train (70%) – Used for training the model and learning key visual patterns.

Validation (10%) – Used for fine-tuning hyperparameters and avoiding overfitting.

Test (20%) – Used to evaluate final model performance and generalization capability.

Each subset contains separate folders for the fire and nofire categories:
dataset/
│
├── train/
│   ├── fire/
│   └── nofire/
│
├── val/
│   ├── fire/
│   └── nofire/
│
└── test/
    ├── fire/
    └── nofire/
Data Collection and Preparation

This dataset was carefully curated by aggregating and cleaning images from multiple reliable sources to ensure quality and balance. The preparation process involved three major phases:

Aggregation:
Multiple open-access forest fire datasets, academic repositories, and publicly available web sources were combined to create a unified dataset.

Manual Classification:
A large collection of previously unlabeled images was manually reviewed and categorized into fire and nofire classes. This manual step greatly expanded the dataset’s diversity and ensured labeling accuracy.

Final Split:
The cleaned dataset was randomly divided into training, validation, and test sets (70/10/20 ratio) to maintain class balance and reproducibility.

Usage Guidelines for Participants

Participants are encouraged to:

Download and explore the dataset freely.

Preprocess, shuffle, or augment the data as they find appropriate for their model’s performance.

Publish modified versions (if any) with proper documentation and credit to the original dataset.

Clearly describe all data preprocessing and augmentation steps in their submission notebook under the “Data Preparation” section.

All images in this dataset are sourced from publicly available and academic datasets. The combined collection has been cleaned and labeled solely for educational and research purposes as part of the EdgeFire Challenge 2025.