This project demonstrates the use of a Variational AutoEncoder (VAE) to learn a latent space representation of simple synthetic data: black-and-white images of circles with varying radius, x, and y coordinates. The goal is to map these parameters to a 3-dimensional latent space, where each dimension ideally corresponds to one of the circle's properties.
The dataset consists of black-and-white images of circles, each of shape 32x32, generated using opencv's cv2.circle function. Each image contains a single circle, randomly positioned and sized within the frame. The parameters for the circles are:
Radius: Randomly selected within a defined range.X-coordinate: Random horizontal position of the circle's center.Y-coordinate: Random vertical position of the circle's center.
Below is a sample visualization of the generated data:
The encoder network compresses the flattened 32x32 input image into a 3-dimensional latent representation. It consists entirely of fully connected layers, producing two outputs:
- Mean vector (μ) of the latent distribution.
- Log variance vector (logσ2) of the latent distribution.
The decoder reconstructs the original image from the 3-dimensional latent representation. It also consists entirely of fully connected layers, ensuring the output matches the flattened dimensions of the original image (32x32).
The latent space is a 3-dimensional representation of the circle data. Ideally, each dimension captures a specific property:
The radius?
The Y-axis?
The X-axis?
- Add Beta-VAE
Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.
- Fork the repository
- Create your feature branch (
git checkout -b feature/AmazingFeature) - Commit your changes (
git commit -m 'Add some AmazingFeature') - Push to the branch (
git push origin feature/AmazingFeature) - Open a Pull Request
Made with ❤️ by [Fer14]