Autoencoding Video Frames

This report details the implementation of an autoencoder trained with a learned similarity metric - one that is capable of modelling a complex dis- tribution of natural images - training it on frames from selected films, and using it to reconstruct video sequences by passing each frame through the autoencoder and re-sequencing the output frames in-order. This is primarily an artistic exploration of the representational capacity of the current state of the art in generative models and is a novel application of autoencoders. This model is trained on, and used to reconstruct the films Blade Runner and A Scanner Darkly, producing new artworks in their own right. Experiments passing other videos through these models is carried out, demonstrating the potential of this method to become a new technique in the production of experimental image and video.