Skip to content

This project leverages the power of deep learning to differentiate between real and fake face images. An Inception ResNet V1 model was trained on a dataset of 140,000 images, equally divided between real and fake faces, to achieve high accuracy in identifying fraudulent attempts at digital impersonation.

Notifications You must be signed in to change notification settings

mahe115/Deepfake-image-classification

Repository files navigation

Real vs. Fake Face Detection with Inception ResNet V1

This project leverages the power of deep learning to differentiate between real and fake face images. An Inception ResNet V1 model was trained on a dataset of 140,000 images, equally divided between real and fake faces, to achieve high accuracy in identifying fraudulent attempts at digital impersonation.

Project Overview

The increasing sophistication of digital impersonation techniques poses significant security risks. This project aims to bolster the reliability of facial recognition systems by developing a model capable of distinguishing real faces from fakes with high precision.

Prediction Results

Prediction Result

Dataset

The model was trained on a dataset consisting of 140,000 images, split evenly between real and fake faces. This dataset was specifically chosen for its balance and variety, providing a comprehensive foundation for training.

Model Details

  • Architecture: Inception ResNet V1
  • Training Platform: Kaggle Notebook
  • Training Data: 70,000 real images and 70,000 fake images
  • Evaluation Metrics: Accuracy, Precision, Recall, F1 Score, Confusion Matrix, ROC-AUC

Training Results

Loss Function Graph

Loss Function Over Epochs

Accuracy Graph

Accuracy Over Epochs

Installation

Before running this project, ensure Python 3.x is installed. Then, install the required dependencies:

Usage

To replicate the training process or to train the model with new data, follow the instructions provided in the Kaggle notebook:

Evaluation and Results

The model's performance is thoroughly evaluated using multiple metrics. The results highlight the model's capability to accurately classify real and fake faces. Detailed results and analysis are available in the training notebook.

Ethical Considerations

We are committed to ethical AI practices. This project addresses potential biases in facial recognition technology and ensures diverse representation within the training dataset to mitigate discriminatory outcomes.

Future Work

Future enhancements may include:

  • Data augmentation to increase model robustness.
  • Implementing ensemble learning for improved accuracy.
  • Continuous updates to the model to counter new impersonation techniques.

Acknowledgments

  • Special thanks to Kaggle for hosting both the dataset and the platform for training.
  • Appreciation to all contributors to the Inception ResNet V1 architecture and related technologies.

About

This project leverages the power of deep learning to differentiate between real and fake face images. An Inception ResNet V1 model was trained on a dataset of 140,000 images, equally divided between real and fake faces, to achieve high accuracy in identifying fraudulent attempts at digital impersonation.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published