AI vs Real Image Analysis

This project implements a deep learning model to distinguish between AI-generated artwork and real artwork using Convolutional Neural Networks (CNN). The analysis provides insights into the visual characteristics that differentiate AI-generated art from human-created art.

Project Overview

The project uses a CNN model trained on a dataset of AI-generated artwork and real artwork to classify images into these two categories. The model employs various data augmentation techniques and early stopping to prevent overfitting.

Features

• Image classification between AI-generated and real artwork
• Data augmentation for improved model generalization
• CNN architecture with multiple convolutional layers
• Early stopping mechanism to prevent overfitting
• Visualization of training and validation metrics
• Support for various image dimensions

Requirements

• Python 3.x
• TensorFlow
• OpenCV
• NumPy
• Pandas
• Matplotlib
• Pillow

Installation

1. Clone the repository:

git clone https://github.com/yourusername/AI-vs-Real-Image-analysiss.git
cd AI-vs-Real-Image-analysiss

2. Install the required packages:

pip install -r requirements.txt

Project Structure

AI-vs-Real-Image-analysiss/
├── images/                 # Directory containing sample images
├── AIvsRealAArt.ipynb     # Jupyter notebook with analysis
├── aivsrealimageanalysis.py  # Main Python script
└── README.md              # Project documentation

Usage

1. Prepare your dataset:

   • Place AI-generated artwork in the AiArtData directory
   • Place real artwork in the RealArt directory

2. Run the analysis:

python aivsrealimageanalysis.py

Model Architecture

The CNN model consists of:

• Three convolutional layers with ReLU activation
• MaxPooling layers after each convolutional layer
• Flatten layer
• Dense layer with 512 units and ReLU activation
• Dropout layer (0.5)
• Output layer with softmax activation

Training Process

The model is trained with the following specifications:

• Image size: 255x255 pixels
• Batch size: 64 for training, 16 for validation
• Optimizer: Adam with learning rate 0.001
• Early stopping with patience of 20 epochs
• Data augmentation including rotation, width/height shifts, shear, zoom, and horizontal flips

Results

The model provides:

• Training and validation accuracy metrics
• Training and validation loss curves
• Classification performance on test data

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Dataset contributors
• TensorFlow team for the deep learning framework
• OpenCV team for image processing capabilities