Image Classification Project: Real vs Fake Image Detector

This project implements a Convolutional Neural Network (CNN) using TensorFlow and Keras to classify images as either "REAL" or "FAKE". The model uses multiple convolutional layers with data augmentation techniques to improve classification accuracy.

Project Structure

• src/main.py: Main Python script containing model architecture, training and testing functions
• archive/train/: Directory containing training images organized in class subdirectories
• trained_model.keras: Saved model file after training (stored using Git LFS)

Features

• Binary image classification (REAL/FAKE)
• Data augmentation (rotation, shifting, zoom, flip) to improve model generalization
• Early stopping to prevent overfitting
• Interactive CLI for training and testing
• Confidence score reporting for predictions

Requirements

tensorflow==2.19.0
numpy==2.1.3
keras==3.10.0
pillow==11.2.1
scipy==1.11.4

Installation

1. Clone this repository:

   git clone https://github.com/yourusername/image-classification.git
   cd image-classification

2. Install the required dependencies:

   pip install -r requirements.txt

3. Pull the large files (if using the pre-trained model):

   git lfs pull

Usage

Run the main script:

python src/main.py

Training Mode

Select "train" when prompted to train the model:

• Images should be in the archive/train directory with appropriate class subdirectories
• Default hyperparameters: batch_size=32, epochs=10
• The model architecture uses three convolutional layers, followed by dense layers
• Training progress and validation accuracy will be displayed
• The trained model will be saved as trained_model.keras

Testing Mode

Select "test" when prompted to test the model:

• You will be asked to provide the path to an image for classification
• The model will process the image and provide:
  • Classification result (REAL or FAKE)
  • Confidence percentage
  • Raw prediction value

Git LFS Usage

This repository uses Git Large File Storage (Git LFS) for managing large files such as:

• The trained model file (trained_model.keras)
• Any large dataset files

To work with this repository, ensure you have Git LFS installed:

git lfs install

Model Architecture

The CNN architecture consists of:

• 3 convolutional layers with ReLU activation and max pooling
• Flatten layer to convert 2D feature maps to 1D features
• 2 dense layers with the final layer using sigmoid activation for binary classification
• Adam optimizer with binary crossentropy loss