🎨 VAE-Based Image Interpolator

This project implements a Variational Autoencoder (VAE) based image interpolator. It generates new "in-between" images by interpolating between two different image classes using a weighted combination of their latent space distributions.

---

🚀 Features

✅ Variational Autoencoder (VAE) for learning latent representations
✅ Latent Space Interpolation for smooth class transitions
✅ Customizable Weighting to control the class influence
✅ Scalable for different datasets and resolutions

---

📦 Installation

Ensure you have installed Python and the necessary dependencies.

---

🔧 How to Use

1️⃣ Train the VAE Model

Train the model on your dataset:

python model.ipynb --dataset path/to/dataset

2️⃣ Perform Interpolation

Generate interpolated images between two classes:

python interpolate.py --class1 label1 --class2 label2 --alpha 0.5

• alpha: Controls the weighting between the two classes (0.0 = full class1, 1.0 = full class2).

---

🎯 Example

Suppose you have a dataset of handwritten digits (MNIST). You can interpolate between 0 and 1 as follows:

python interpolate.py --class1 0 --class2 1 --alpha 0.5

This will generate an image that blends digit 0 and digit 1.

---

📂 Folder Structure

├── train.py               # Training script  
├── interpolate.py         # Interpolation script  
├── models/               # Model checkpoints  
├── datasets/             # Dataset storage  
├── requirements.txt      # Dependencies  
└── README.md             # Project documentation  

---

💡 Future Improvements

✨ Implement different distance metrics for interpolation
✨ Extend to Conditional VAE (cVAE) for better class control
✨ Test on high-resolution datasets

---

📜 License

This project is licensed under the MIT License.