Method for quantitative evaluation of the realism of CG images using deep learning
Masa-aki Sato, Masataka Imura · 2017
With the development of three-dimensional computer graphics (3DCG) technology, expressing various objects and phenomena graphically became possible. However, there is no method for quantitatively evaluating the realism of the generated CG images. In recent years, deep learning has been widely used for its image discrimination performance beyond that of human beings. We propose a deep learning-based method that enables quantitative evaluation of the realism of CG images with has high image discrimination ability. The results of implementation using convolution neural networks (CNNs) are presented.