Visual representation of variational auto encoders with 2D latent space.
This Tkinter-based graphical user interface (GUI) showcases an innovative fusion of graphical interactivity and machine learning. The left side of the interface features a dynamic graph that captures the x and y coordinates of the mouse cursor. These coordinates undergo scaling and conversion to a standardized range between -1 and 1.
The real magic happens as these processed coordinates serve as input for a Variational Autoencoder (VAE), pre-trained on the CelebA dataset using Google Colab. The VAE generates human faces based on the transformed cursor data. For visualization purposes, the latent dimension is set to 2, providing an engaging insight into the model's capabilities.
Feel free to explore higher latent dimensions, unlocking the potential for more detailed and diverse facial reconstructions. This project offers a unique intersection of user interaction, data processing, and machine learning, providing a captivating demonstration of technology in action.