MACHINE LEARNING AS AN ALTERNATIVE TO 3D PHOTOMODELING EMPLOYED IN ARCHITECTURAL SURVEY AND AUTOMATIC DESIGN MODELLING
Keywords: Instant Nerf, image-based survey, neural networks, artificial intelligence, volume rendering, real time rendering
Abstract. This paper presents some experiments on the use of an alternative technique, Nerf, based on artificial intelligence, which can be used in the survey of architectural structures or parts of them. The Nvidia video cards supported by the new RTX architectures are now able to manage an enormous amount of data both in the calculation of lighting and the rendering of digital shaders, and in the management of the number of polygons that can be displayed in real time in the scene. The support of artificial intelligence has further improved the three-dimensional digitization process in order to support the hardware power of algorithms that use neural networks to optimize and reduce calculation times, improving various aspects of the graphics and also of the digital representation, also influencing the representation on “image based” 3D acquisition methods, currently the most used low-cost systems in the photomodeling of objects or three-dimensional scenes. At the current state of the art, many commercial companies and research organizations, first Nvidia with its Instant Nerf, are betting on algorithms based on the Neural Radiance Field. The paper highlights the criticalities of this new system and shows the discrete results and the expandable research scenarios that can involve the impossible survey with photomodeling methodology related to works of art or architecture, due to the intrinsic nature of the materials that they compose them or because it is impossible or very difficult to photograph them.