Deepfake Detection Using MesoNet Architecture

This project focuses on detecting deepfakes in images using the MesoNet network architecture, which is specifically designed for identifying facial manipulations. The model leverages key deep learning frameworks including TensorFlow, Keras, PyTorch, and OpenCV for efficient processing and analysis.

Requirements

Python 3.5
Numpy 1.14.2
Keras 2.1.5

Frameworks and Tools:

TensorFlow, Keras: For model development and training.
PyTorch: Supporting model optimization and performance tuning.
OpenCV: Used for image processing and manipulation tasks.

Dataset Overview:

Training Set:

Forged Images: 5,111
Real Images: 7,250

Validation Set:

Forged Images: 2,889
Real Images: 4,259

Download link for the dataset

Balance:

Both the training and validation sets maintain a balanced distribution between forged and real images. This balance ensures that the model is effectively trained to recognize deepfake patterns and generalize across different data scenarios.

Pretrained models

You can find the pretrained weight in the weights folder. The _DF extension correspond to a model trained to classify deepfake-generated images and the _F2F to Face2Face-generated images.

How to Run

Clone the Repository

git clone https://github.com/Deepfake-Detection-using-Mesonet.git
cd Deepfake-Detection-using-Mesonet

Download and Prepare the Dataset

/data
  /train
    /real
    /forged
  /validation
    /real
    /forged

Train the Model

python train.py --epochs 50 --batch_size 32

Validate the Model

python validate.py --model mesonet_model.h5 --data_dir ./data/validation

Run Inference on New Data

python predict.py --image_path ./path_to_image.jpg --model mesonet_model.h5

Purpose:

The primary goal of this project is to develop a robust detection framework that identifies deepfake images with high accuracy. By training the model on a diverse dataset, the system is validated for real-world deployment, providing a critical tool for cybersecurity and digital forensics.

Key Features:

MesoNet Architecture: A compact Convolutional Neural Network (CNN) architecture designed for detecting facial forgeries with precision.
Cybersecurity Application: The system enhances the integrity of digital media by identifying manipulated content, safeguarding against threats like misinformation, identity theft, and fraud.

Authors

Maimuna Khatoon - Github

Charvi Jaiswal - Github

Sheryl Sokhi - Github

Tulika Tripathi - Github

References

Afchar, D., Nozick, V., Yamagishi, J., & Echizen, I. (2018, September). MesoNet: a Compact Facial Video Forgery Detection Network. In IEEE Workshop on Information Forensics and Security, WIFS 2018.

maimuna01 / Deepfake-Detection-using-Mesonet