인공지능 기반 스타일 전이 엔진을 활용한 360VR 영상콘텐츠 제작연구

Abstract

(Background and Purpose) This study was intended to create the contents providing unique visual experiences, which reinterpreted the space from the master artists' viewpoint by considering the characteristics of 'artificial intelligence' and 'VR'. Through that, the purpose of this study was to identify the considerations and limitations in the process of creation of 360VR video contents using the AI-based style transfer engines so as to suggest the applicability of the result of the style transfer engines and the course of development. (Method) In this study, content production stage was planned based on theoretical consideration of 360VR image and AI-based style transfer engines. The production process of 360VR video contents, using the AI-based high resolution style transfer engine developed in a way applicable to 360VR contents production, was classified into 3 stages, i.e., content planning, resource production, and video editing, and described, respectively. The considerations for each stage of 360VR video content production, which used the style transfer engine results, were analyzed based on the characteristics of 360VR video contents and AI style transfer engine. (Results) The results of analyses of the considerations during the production process showed that the 360-degree image on the VR was felt closer to the 360-degree viewed on the computer screen, and placed greater emphasis on ultra-high resolution, and furthermore, various results were rendered according to the transference and input style of 2D images in connection with the characteristics of the AI style transfer engine. At the content planning stage, sufficient time should be given for the appreciation and conversion of the 360VR video scenes. In the resource production stage, the resource of engine results should be edited with a focus on the portion where images are converged up and down and the portion where the line is created on both ends of engine results without being connected according to the input style, and it should be considered that the results are felt closer on the part of the experiencing person. (Conclusions) This study presents a detailed description of the production process of 360VR video contents using AI-based high resolution style transfer engine, thus providing guidance to the beginners working on similar contents. In addition, it identifies the limitations involved in the analyses of engine characteristics based on the styles selected arbitrarily for engine tests or direct correction of connections in the engine results as pinpointed by the producer, presenting the style transfers engine developers and art work producers, who utilize the style transfer results, with the usability of the engines applying new media and the course of development, as well as the researchers in related fields, thus increasing the value of this study.