README.md

---
title: Deepfake Image Classification
colorFrom: blue
colorTo: pink
sdk: keras
tags:
- deepfake
- image-classification
- computer-vision
- efficientnet
license: mit
app_file: Deepfake_Image_Classification.ipynb
model-index:
- name: deepfake_image_classification_model
  results: []
---



# Deepfake Image Classification

This repository contains a TensorFlow/Keras model trained to classify **REAL vs FAKE (deepfake) images**.  
The model is based on **EfficientNetB4** with data augmentation and fine-tuning.  

## 📂 Files
- `deepfake_image_classification_model.keras` – trained Keras model.  
- `Deepfake_Image_Classification.ipynb` – full training notebook (Google Colab).  

## 🧠 Model Details
- Backbone: EfficientNetB4 (pretrained on ImageNet).  
- Input size: `224 x 224 x 3` RGB images.  
- Output: Binary classification (REAL = 0, FAKE = 1).  
- Loss: `binary_crossentropy`.  
- Optimizer: `Adam (lr=0.001)`.  

## 📊 Dataset
Trained on [Deepfake Faces dataset](https://www.kaggle.com/datasets/dagnelies/deepfake-faces).  
Balanced subset: 16,000 REAL + 16,000 FAKE images.  

## 🚀 Usage
Install dependencies:
```bash
pip install tensorflow
Load and use the model:

python
Copy code
import tensorflow as tf

# Load model
model = tf.keras.models.load_model("deepfake_image_classification_model.keras")

# Preprocess image
img = tf.keras.utils.load_img("example.jpg", target_size=(224, 224))
x = tf.keras.utils.img_to_array(img)
x = tf.expand_dims(x, axis=0)  # add batch dimension

# Predict
pred = model.predict(x)[0][0]
label = "FAKE" if pred > 0.5 else "REAL"
print(f"Prediction: {label} (score: {pred:.4f})")
📈 Performance
Validation Accuracy: ~92%

Test Accuracy: ~91%

📓 Training
For full details of the training pipeline (dataset loading, augmentation, callbacks, etc.), check the notebook:
Deepfake_Image_Classification.ipynb