Dynamic Fusion With Intra- and Inter-Modality Attention Flow for Visual Question Answering | Synapse