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

Table 3

Experiment results compared with other state-of-the-art models on the test-dev and test-std sets.

Accuracy		Test-dev				Test-std
Methods (%)		All	Y/N	Number	Other	All
Bottom-up [3]		65.32	81.82	44.21	56.05	65.67
BAN [16]		69.52	85.31	50.93	60.26	69.84
BAN + counter [16]		70.04	85.42	54.04	60.52	70.35
MuRel [29]		68.76	84.77	49.84	57.85	68.41
DFAF [30]		70.22	86.09	53.32	60.49	70.34
MCAN [2]		70.63	86.82	53.26	60.72	70.90
JGRCAN		70.87	86.97	53.15	61.45	71.18