Challenges and Capabilities of 3DGS in Detecting Tiny Details for Cultural Heritage

Abstract. 3D reality-based techniques are well-known in Cultural Heritage documentation and exploitation. Recently, Neural radiance Field (NeRF) and 3D Gaussian Splatting (3DGS) have been arose as a possible new technique for 3D modelling reality, if not even substitutes of photogrammetry. NeRF is a neural field used to build a three-dimensional depiction of a scene using photographs. Gaussian splatting provides for an efficient and precise representation of geometry and appearance characteristics by modelling things and surfaces as a set of Gaussians. This paper presents a comparison of the models derived from two different techniques (Structure from Motion – SfM, 3DGS) for the acquisition and detection of cracks and tiny details in the bathroom of the Royal Palace of Caserta. The idea is to analyse the resultant data to understand if 3DGS is mature enough to be utilised for the accurate reconstruction of this type of objects, to provide data useful for structural analysis. The Postshot AI tool by Jawset was tested using the different parameters for the training of the AI and using the same set of data. The results are promising in terms of speed and graphical results but not so satisfactory as for real data accuracy needed for deepen analysis.