# DeepFake-450K Dataset

## Overview
The **DeepFake-450K** dataset is a large-scale image collection developed to support deepfake image detection research. It consists of **451,440 facial images**, evenly divided between real and synthetically generated content. This dataset was curated to train, validate, and benchmark machine learning models, particularly deep learning classifiers for multimedia forensics.

## Contents
The dataset includes:

- `real/`: Real face images from diverse sources (Celeb-DF, DeeperForensics, Kaggle)
- `fake/`: Deepfake face images generated using various GAN and facial manipulation methods (e.g., FaceSwap, Neural Reenactment)

## Structure
```
DeepFake-450K/
│
├── real/
│   ├── celebdf/
│   ├── deeperforensics/
│   └── kaggle_real/
│
├── fake/
│   ├── faceforensics/
│   ├── wilddeepfake/
│   └── kaggle_fake/
│
└── metadata.csv
```

## Variables in metadata.csv
- **filename**: Name of the image file
- **label**: Real or Fake
- **source**: Dataset origin
- **resolution**: Original size before preprocessing
- **manipulation**: If fake, method used to generate it (if known)
- **compression**: Compression applied (high, medium, low)

## Licensing & Usage
This dataset is intended for **research and educational use only**. Please ensure proper citation if used in publications or derivative works.

## Contact
For queries, please contact: [sami.alanazi@cranfield.ac.uk]