Training an image captioning model with visual attention. An example of combining 2 modalities into a single model.