DeepShield is a comprehensive AI-powered system designed to detect and defend against deepfake multimedia and fake news. This ecosystem leverages cutting-edge technologies to tackle the challenges posed by manipulated visual, audio, video, and textual content.

• Deepfake Image Detection: Detect inconsistencies and manipulations in images.
• Deepfake Voice Detection: Identify tampered audio using advanced RNN and transformer models.
• Deepfake Video Detection: Analyze frames and audio together to identify deepfake videos.
• Fake News Detection: Evaluate credibility using NLP techniques like sentiment analysis and entity recognition.
• Adversarial Attack and Defense: Test model robustness with evasion attacks and build defenses using adversarial training.

deepfake-detection/
├── data/
│   ├── images/
│   ├── audio/
│   ├── videos/
│   └── fake_news/
├── models/
│   ├── image_detection/
│   ├── voice_detection/
│   ├── video_detection/
│   ├── fake_news_detection/
│   └── adversarial_defense/
├── notebooks/
│   ├── image_detection.ipynb
│   ├── voice_detection.ipynb
│   ├── video_detection.ipynb
│   ├── adversarial_attack_defense.ipynb
│   └── fake_news_detection.ipynb
├── requirements.txt
└── README.md

1. Clone the repository:

   git clone https://github.com/HarshaBeth/deepshield-detection.git

2. Navigate to the project directory:

   cd deepshield-detection

3. Install the required dependencies:

   pip install -r requirements.txt

Explore interactive notebooks in the notebooks/ folder for experimentation:

• image_detection.ipynb
• voice_detection.ipynb
• video_detection.ipynb
• fake_news_detection.ipynb
• adversarial_attack_defense.ipynb

• Images Dataset - 140k real and fake faces
• ASVspoof 2019 dataset
• Fake News Detection Dataset "WELFake"
• Deep Fake Detection (DFD) "Videos"

The models are stored in this Google Drive because of their large size. You can download them from Models.

• Adversarial Robustness Toolbox (ART) for adversarial testing and defense.