This repository contains some examples of how to do training on video sequences using tensorflow. In contrast to a lot of methods that train for video classification or other applications, the feature extractor can also be trained during training in the style of an RNN. This repo contains the whole pipeline from loading videos into tfrecords, to masking the loss and optimizing the Dynamic RNN by passing the sequence lengths.