I\'m trying to train a model using video data in TensorFlow. The current data pipeline uses a data generator, which loads the video using the module moviepy and then does some v