The Autoencoder encodes (128 x 128 x 3) images into a latent image of dimensions (16 x 16 x 8), hereby compressing the images. The Decoder model then decodes the latent space images with reversed weights to regenerate the supplied image
Here are the input images to the Autoencoder model, followed by its corresponding regenerated output image by the Decoder model:
Here are their corresponding latent space visualization (these are technically not images. They are (16 x 16 x 8) matrices, transposed to produce a visual image)
