JGRCAN: A Visual Question Answering Co-Attention Network via Joint Grid-Region Features

Figure 2

Grid feature extraction method.