Latent space explorer

I've trained an autoencoder on the MNIST-dataset.
Move your mouse over the second image to explore the 2D latent space.
The autoencoder has never seen the labels, and has purely organized the images based on how similar they look.
But I've plot each datapoint with the label as color, so you can see the distribution.

You can also manually input a coordinate:
latentX: latentY:
Notice that it was meant to be between 0 and 100,
but you can go outside this range for some weird images.