🎨 GANs in TensorFlow (Google Colab)

This project implements a basic Generative Adversarial Network (GAN) using TensorFlow, inspired by a DigitalOcean tutorial. The GAN is trained on the MNIST dataset to generate realistic images of handwritten digits. Designed to run smoothly on Google Colab with GPU acceleration.

---

🚀 Features

• 🗂 Dataset: MNIST (28×28 grayscale images)
• ⚙ Generator:
  • Input: 100-dimensional random noise vector
  • Output: 784-dimensional vector reshaped to 28×28 image
• 🛡 Discriminator:
  • Input: 784-dimensional image
  • Output: Probability of image being real (0–1)
• 🎯 Training:
  • Binary cross-entropy loss
  • Optimizer: Adam (lr = 0.0001)
  • Alternating Generator/Discriminator updates
• 🖼 Visualization:
  • Saves 4×4 image grid after each epoch

---

📋 Prerequisites

Google Colab comes pre-installed with most libraries. If needed, install manually:

!pip install tensorflow numpy matplotlib
🧪 Setup Instructions (Google Colab)
🔗 Open Google Colab: https://colab.research.google.com

➕ New Notebook: File → New Notebook

📦 Install Dependencies (if required):

python
Copy
Edit
!pip install tensorflow numpy matplotlib
📤 Upload Code:

Paste the contents of main.py into a cell, or

Use sidebar → "Files" tab → Upload main.py

💾 Optional: Mount Google Drive (for saving outputs):

python
Copy
Edit
from google.colab import drive
drive.mount('/content/drive')
🔄 Clone Your GitHub Repo (optional):

bash
Copy
Edit
!git config --global user.name "your-username"
!git config --global user.email "[email protected]"
!git clone https://github.com/your-username/your-repo-name.git
💡 Usage
Train the GAN:

Default: 50 epochs

Losses printed every epoch

View Outputs:

Images saved as image_at_epoch_000X.png

Found in "Files" tab or in Google Drive (if mounted)

💾 Save & Push to GitHub:
bash
Copy
Edit
%cd /content/your-repo-name
!cp /content/image_at_epoch_*.png .
!git add .
!git commit -m "Add GAN code and generated images from Colab"
!git push origin main
🔐 You may need a GitHub personal access token.

📁 Project Structure
bash
Copy
Edit
gan_project/
├── main.py                # Main GAN training script
├── image_at_epoch_*.png  # Generated images
└── requirements.txt       # (Optional) Python dependencies
🧠 How It Works
🔄 Data Preprocessing
Normalize MNIST images to [-1, 1]

Flatten 28×28 images → 784D vectors

🧬 Generator Architecture
text
Copy
Edit
Input: 100D noise
→ Dense(256) → LeakyReLU
→ Dense(512) → LeakyReLU
→ Dense(784, activation='tanh')
🛡 Discriminator Architecture
text
Copy
Edit
Input: 784D image
→ Dense(512) → LeakyReLU
→ Dense(256) → LeakyReLU
→ Dense(1, activation='sigmoid')
🎓 Training Strategy
Discriminator learns real vs fake

Generator improves to fool Discriminator

Alternate training with binary cross-entropy loss

⚙️ Colab Tips
Enable GPU: Runtime → Change runtime type → Select "GPU"

Stability: If training is unstable, try tweaking:

Learning rate

Batch size

Architecture layers

Storage: Save images to Google Drive before Colab resets

📦 Requirements
txt
Copy
Edit
tensorflow>=2.0
numpy
matplotlib
🤝 Contributing
Fork the repo

Create a new branch:

bash
Copy
Edit
git checkout -b feature/your-feature
Commit & Push:

bash
Copy
Edit
git add .
git commit -m "Add your feature"
git push origin feature/your-feature
Open a Pull Request

📄 License
This project is licensed under the MIT License.
See the LICENSE file for details.

🙏 Acknowledgments
Based on a DigitalOcean tutorial

Built using TensorFlow, NumPy, and Matplotlib

Powered by Google Colab

🌐 Live Demo (Optional)
You can run it in Colab by uploading main.py.