Performance Analysis and Applicability Study of 3D Modeling Using 3D Gaussian Splatting in The 3D Model Production Process

Recently, the creation of new digital content and the implementation of new rendering processes and learning systems based on radiance fields are rapidly spreading. Among these, 3D Gaussian splatting has emerged as an attractive attention technique in computer graphics and content production. This method enables the rapid reconstruction of 3D models by training neural networks on images captured from various viewpoints. The application of this technology in digital content production has been the focus of numerous studies. In this study, the process of producing 3D models on representative platforms such as Luma AI, Polycam, and Volinga, which utilize 3D Gaussian splatting, was investigated. The 3D models were subsequently categorized based on size for each platform and the resulting models were produced and analyzed. The models were visualized using the Unreal engine, evaluated for efficiency, and the model quality and applicability were verified by five producers with an average of 14 years of experience. As a result, it was concluded that 3D Gaussian splatting is significantly more efficient in terms of the production time and cost than existing 3D model production methods and that it is useful for producing background elements, implying its high applicability. This research shows that 3D Gaussian splatting can complement the tools used in traditional 3D model production fields where games, movies, and dramas are created, confirming the potential for its commercialization.