VigiLens: Deepfake Videos Detection Using Deep Learning

VigiLens is a deepfake video detection model created using advanced deep learning techniques. We're currently working on developing a user-friendly interface for the system.

📋 Table of Contents

1. 🤖 Introduction
2. ⚙️ Tech Stack
3. 🔋 Features
4. 🗂️ Directory Structure
5. 👩🏾‍💻 Contributors
6. ❗ Limitations & Future Scope

🤖 Introduction

VigiLens is a deepfake video detection model created using advanced deep learning techniques like RestNext and LSTM to predict whether the given video is Real or Fake(AI Synthesized). We implemented the model using a pre-trained ResNext CNN model to extract frame-level features and LSTM for temporal sequence processing to spot changes between the t and t-1 frame. This approach overcomes challenges faced by previous deepfake detection models, such as struggles with higher-resolution videos, data oversampling issues, and a lack of robustness. We're currently working on developing a scalable user-friendly interface for the system.

⚙️ Tech Stack

• Python
• Libraries: Tensorflow, OpenCV, Pytorch, numpy, matplotlib, Face Recognition, pandas
• Django

🔋 Features

👉 Dataset Diversity: Comprehensive datasets with varied facial expressions, lighting, and scenarios.

👉 Hybrid Model Architecture: Integration of CNN and RNN for feature extraction and temporal analysis.

👉 Optimized Training: Adam optimizer, cross-entropy loss, and model fine-tuning.

👉 Real-time Prediction: Quick and accurate classification of videos as real or deepfake.

👉 Confidence Metrics: Providing confidence levels for classification results.

🗂️ Directory Structure

The directory Structure is given below:

Deepfake_detection_using_deep_learning
    |
    |--- Web Application
    |--- Deepfake Detection Model 
    |--- Documentaion

1. Web Application
   • This directory will hold Web Application where a user can upload the video and submit it to the model for prediction. The trained model will perform the prediction and the result will be displayed on the screen.
2. Deepfake Detection Model
   • It contains procedure of creating and training a deepfake detection model using our approach.
3. Documentation
   • It has related documentation done for the project.

👩🏾‍💻 Contributors

• Sanskruti B.
• Trithi Amin
• Anusha Goyal

❗ Limitations & Future Scope

👉 Upscaling to Browser Plugin/Web Application: This project can be scaled up from a web-based platform to a browser plugin for automatic deepfake detection. Integration into large applications like WhatsApp and Facebook can provide convenient pre-detection capabilities for users, enhancing accessibility.

👉 Expanding Detection Capabilities: Although the current algorithm focuses on face deepfakes, there's room for improvement to detect full-body deepfakes. This enhancement would significantly increase the system's effectiveness and coverage, addressing broader deepfake scenarios.

👉 Audio Detection Limitation: Currently, the system is only capable of detecting videos without audio. Future enhancements may include audio analysis for more comprehensive deepfake detection.

🚀 Follow to see what we're building!

Have Suggestions? Want to colab? Shoot an email

Show some ❤️ by starring this repo! ⭐