SIM2REAL

One of the key components in the AI training is the training data. In particular, for the utilization of computer vision, appropriate images and corresponding label data are essential. However, there are several limitations to obtaining real-world image data. The quantity of data may be insufficient due to the infrequency of occurrences of real phenomena, or it may not exist at all. Even with easily accessible cases, personal information de-identification processes for data ethics or complex data labeling can consume significant human resources. As a solution to these challenges, SIM2REAL proposes the acquisition of image and label data in a virtual simulation environment. This approach can achieve high quality, even distribution, evasion of personal information issues, and automated labeling. This is highly effective for anyone who want to develop and test a computer vision AI model and it can save a large amount of resources for AI training and testing.