JS Autoencoder demo

Doing cool stuff with an autoencoder trained on the Fashion MNIST dataset.

Demo

autoencoder.mp4

Running it

1. Save weights for use in the web app -> python generate_js_weights.py

2. Save all the images needed -> python save_images_txt.py

3. Start flask server python app.py

Sampling from the net's latent space

New images can be generated by sampling the autoencoder's latent space.

Generate

Interpolating between images

Interpolation between images can be performed by linearly interpolating between the latent vectors of those images and then feeding that into the decoder part of the autoencoder.

Interpolate