This repository contains a Convolutional Autoencoder trained on the MovingMNIST dataset but by sequences of 3 frames, where the aim is to predict the fourth one.
You will find the model, and all the relative information about it in the Notebook. It reports also the training history as well as a discussion on the possible alternatives (without considering fine-tuned SOTA models).