Exploring_VAE_Latent_Space
Explores and visualizes a VAE's latent space with you mouse and keyboard.
Creates latent dimensions from mouse position, decodes them using a VAE trained on MNIST using Louis Tiao's code, and visualizes them using Pygame.
All credit to Louis Tiao for the code to train the VAE and code for nll and KLDivergenceLayer (http://louistiao.me/posts/implementing-variational-autoencoders-in-keras-beyond-the-quickstart-tutorial/). Edit the code to save the encoder and decoder in seperate h5/hdf5 files, then load them into the Models folder. To change the number of dimensions, edit latent_dim in Louis Tiao's original code.
Move cursor position on screen to change x and y values, and, if enabled, use left and right arrow keys to change z value in or up and down to change the t value.