---
title: Deepfake Detector
emoji: 🎭
colorFrom: blue
colorTo: red
sdk: streamlit
sdk_version: 1.57.0
app_file: app.py
pinned: true
license: mit
---

# 🎭 Deepfake Detection System

> CNN-based deepfake detector trained on FaceForensics++ achieving **90.4% accuracy** and **0.96 AUC**.

Upload a video → get a Real/Fake verdict with confidence score and frame-by-frame analysis.

---

## 📊 Results

| Metric | Score |
|--------|-------|
| Accuracy | **90.4%** |
| AUC | **0.96** |
| Fake Precision | 93% |
| Real Recall | 94% |

---

## 🏗️ How It Works
Video → Extract Frames → RetinaFace Face Detection → EfficientNetB0 → Real/Fake

---

## 🗂️ Dataset
- **FaceForensics++** — 1,000 original YouTube videos
- 6 manipulation types: Deepfakes, Face2Face, FaceSwap, NeuralTextures, FaceShifter, DeepFakeDetection

---

## 🧠 Model
- **Base:** EfficientNetB0 pretrained on ImageNet
- **Training:** Two-phase transfer learning + fine-tuning
- **Face Detection:** RetinaFace with padding

---

## 🔗 Links
- [GitHub Repo](https://github.com/yyouretoast/deepfake-detection)
- [Model Weights](https://huggingface.co/yyouretoast/deepfake-detection)

---

## 👤 Author
**Yassin Mohamed**
[![LinkedIn](https://img.shields.io/badge/LinkedIn-0077B5?style=flat&logo=linkedin&logoColor=white)](https://www.linkedin.com/in/yassinyasser/)
[![Email](https://img.shields.io/badge/Email-D14836?style=flat&logo=gmail&logoColor=white)](mailto:yyasso2005@gmail.com)