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

Figure 8

Visualization example of the attention mechanism in JGRCAN. The 1^st line of the figure is the input image and question; the 2^nd line is the important information that the region feature pays attention to; the 3^rd line is the important information that the grid feature pays attention to; and the important information is circled in red.

(a)

(b)

(c)

(d)