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

Table 2

Performance comparisons with different feature combinations.

Accuracy
Features (%)		All	Y/N	Number	Other
Grid		70.56	86.27	52.21	61.85
Region		70.63	86.82	53.06	60.72
Grid + region		70.87	86.97	52.88	61.45

Y/N, number, and other are three different types of questions. Y/N includes yes or no questions and only answers to yes or no, for example: “Is it... ?”, “Does it... ?” Number questions include counting questions, answering only numbers, e.g., “how many...?” Other includes asking about color, type, etc. For example, “what color is....” and “what food is...”