Game Engine Based 2D Emotion Segmentation Generation Method

Abstract

This paper proposes a low-cost production and utilization technique for labeling emotion data in game engines, which can be used to support rapidly developing deep learning technologies. The proposed system extracts realistic images from game environments and automatically creates quantified two-dimensional (2D) emotion segmentation images linked to the extracted images. The segmentation data are learned through an image-to-image translation network. This 2D emotion segmentation mapping technique is trained using many training data, which allows stable learning. Industries that require spatial emotion interpretation can utilize the results of this study.