Visual Experience-Based Question Answering with Complex Multimodal Environments

Table 5

Performance analysis of 3D localization depending on different input information.

Input information	Output	Mean absolute error		Accuracy (%)
Depth Image + 2D Bbox	Position	0.344	0.522	17.89
	Size	0.178
Depth Image + 2D Bbox + Agent Pose	Position	0.109	0.249	62.52
	Size	0.140
Depth Image + 2D Bbox + Agent Pose + Object Class	Position	0.089	0.216	78.93
	Size	0.127

Bold values represent the best results.