Visual Experience-Based Question Answering with Complex Multimodal Environments

Table 2

Performance analysis of visual experience-based question answering depending on different VQAS configurations.

Configurations	Accuracy (%)
	Question types						Total
	Existence	Counting	Attribute	Relationship	Include	AgentHas
VQAS	91.96	79.60	68.53	61.16	56.24	63.21	72.37
VQAS with GT 2D objects	99.74	91.18	78.91	73.94	72.93	78.30	83.37
VQAS with GT scene graph	99.74	100.0	99.91	93.23	99.70	100.0	98.62
VQAS with GT query	92.22	79.60	68.53	64.18	56.24	63.21	72.95