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

Figure 4

Specific process of region feature extraction.