# 🔍 DeepFake Image Detector

Can you tell what's real? Our CNN can.

A deep learning system that detects AI-generated (fake) face images with high confidence — featuring a full interactive GUI, batch detection, and live analytics.

---

✨ What It Does

Upload a face image → get an instant REAL or FAKE verdict with confidence score, probability breakdown, and visual analytics.

	Feature
🤖	CNN classifier trained on thousands of real and AI-generated face images
🖼️	Single & batch detection — analyze one image or hundreds at once
📊	Live analytics dashboard — confidence charts, distribution graphs
📋	Prediction history — full log with CSV export
⚡	Runs on Kaggle — no installation or setup needed

---

🚀 Quickstart — Run in 4 Steps on Kaggle

No installation. No configuration. Just click and run.

1. Go to kaggle.com and create a free account
        ↓
2. Add the dataset → search "deepfake-and-real-images" in Kaggle datasets
        ↓
3. Open notebooks/deepfake_detector.ipynb in a new Kaggle notebook
        ↓
4. Click "Run All" — training + evaluation + interactive app launches automatically ✅

---

🧠 Model Architecture

Input Image (128 × 128 × 3)
         │
         ▼
┌─────────────────────────────┐
│  Conv2D(32)  + BatchNorm    │  ← Block 1
│  MaxPool    + Dropout(0.25) │
└─────────────────────────────┘
         │
         ▼
┌─────────────────────────────┐
│  Conv2D(64)  + BatchNorm    │  ← Block 2
│  MaxPool    + Dropout(0.25) │
└─────────────────────────────┘
         │
         ▼
┌─────────────────────────────┐
│  Conv2D(128) + BatchNorm    │  ← Block 3
│  MaxPool    + Dropout(0.30) │
└─────────────────────────────┘
         │
         ▼
┌─────────────────────────────┐
│  Dense(128)  + BatchNorm    │  ← Classifier Head
│  Dropout(0.50)              │
│  Dense(1, sigmoid)          │
└─────────────────────────────┘
         │
         ▼
   🤖 FAKE (0)  or  👤 REAL (1)

Setting	Value
Optimizer	Adam (lr = 0.001)
Loss	Binary Cross-Entropy
Input Size	128 × 128 × 3
Early Stopping	patience = 5
LR Scheduler	ReduceLROnPlateau

---

🖥️ Interactive App

The app runs directly inside the Kaggle notebook with 3 tabs:

🔍 Detect Tab

Upload a single image or a batch — choose between two modes:

• Basic Mode → Label + confidence donut chart
• Diagnostic Mode → Label + confidence + full probability bar chart

📊 Dashboard Tab

Live statistics that update after every detection:

• Real vs Fake distribution pie chart
• Confidence score histogram across all predictions

📋 History Tab

• Scrollable log of every prediction made
• Timestamp, filename, result, confidence, raw score
• One-click CSV export

---

📁 Project Structure

deepfake-detector-system/
│
├── 📓 notebooks/
│   └── deepfake_detector.ipynb   ← ⭐ Start here
│
├── 🧠 src/
│   ├── data_loader.py            ← Dataset paths + generators
│   ├── model.py                  ← CNN architecture
│   ├── train.py                  ← Training loop + callbacks
│   ├── evaluate.py               ← Metrics + confusion matrix
│   └── predict.py                ← Single & batch inference
│
├── 🖥️ app/
│   └── app.py                    ← Interactive GUI
│
├── 📦 data/                      ← Dataset (not tracked by Git)
├── 💾 models/                    ← Saved .h5 files (not tracked)
├── 📤 outputs/                   ← Evaluation outputs
│
├── requirements.txt
└── README.md

---

💻 Run Locally

Click to expand local setup instructions

1. Clone the repo

git clone https://github.com/huda-usman/deepfake-detector-system
cd deepfake-detector-system

2. Install dependencies

pip install -r requirements.txt

3. Download the dataset

Get it from Kaggle and place it as:

data/
└── Dataset/
    ├── Train/   → Fake/  Real/
    ├── Validation/ → Fake/  Real/
    └── Test/    → Fake/  Real/

4. Train

python src/train.py

5. Evaluate

python src/evaluate.py

6. Predict a single image

python src/predict.py --image path/to/face.jpg

7. Launch the app

jupyter notebook notebooks/deepfake_detector.ipynb

---

📦 Requirements

Package	Version
Python	≥ 3.9
TensorFlow	≥ 2.10
OpenCV	≥ 4.6
scikit-learn	≥ 1.1
ipywidgets	≥ 8.0
pandas	≥ 1.5
matplotlib	≥ 3.6

pip install -r requirements.txt

---

🗂️ Dataset

deepfake-and-real-images on Kaggle

Split	Fake	Real
Train	✅	✅
Validation	✅	✅
Test	✅	✅

---

🙋‍♀️ Connect with Me

Developed by Huda Usman

⭐ If you found this project interesting, please give it a star! ⭐