A Study on Acceleration of Rendering Using Super-Resolution Processing

Abstract

In recent years, video contents such as computer graphics have become higher resolution. However, high-resolution rendering takes a long time. Therefore, in this research, we thought that by using the fact that super-resolution processing does not take much time, and by performing super-resolution processing such as TecoGAN on images rendered at low resolution, we could accelerate rendering while maintaining image quality similar to that of images rendered at high resolution. As a result of my experiments, my approach can significantly reduce the time required for acquiring the result high-resolution image by using the super-resolution process. On the other hand, there was a large difference in image quality depending on the scene, and there were scenes where the degradation could be seen briefly, so it was difficult to conclude that images after super-resolution processing could be used instead of high-resolution images. To objectively validate this difference, we analyzed some CG scenes. As these results, it was found that TecoGAN can generate super-resolution images better when an input image has more high-frequency components and a larger color variance.