CG Utilization for Creation of Regression Model Training Data in Deep Learning

摘要

Nowadays, deep learning is utilized in a wide range of fields and has achieved great results. However, in order to use deep learning, it is necessary to prepare a large amount of training data, and this becomes a big obstacle in some fields. In this study, we show the case to estimate quantities from pictures by using the regression model of deep learning, which is trained by CG images generated automatically. And, to improve the estimation accuracy of this case, we propose the following methods: mixing pictures into the CG images which is training data; converting the color of pictures for estimation according to the CG images. And, we evaluate these methods and show their effectiveness.