Computer Graphics for Machine Learning in Computer Vision

Abstract

There has to be much training data to apply machine learning methods for various computer vision tasks. In some of the tasks it is difficult to make ground truth labels for training images. Some researchers have tried to make training images with ground truth labels by using computer graphics. And in reinforcement learning where the agent can learn by interacting with the environment, autonomous robot with various sensors has to work in real environment. But it is difficult to make the environment. On the other hand, in the virtual environment using computer graphics, it is easy to control the environment and to get various information in the environment. Therefore the virtual environment using computer graphics is easy to use for reinforcement learning. This paper describes usage of computer graphics for making training dataset and making reinforcement learning environment.