Abstract. In the domain of Cultural Heritage Restoration, the demand for high-fidelity 3D models with accurate textures and geometry is critical for documentation, analysis, and conservation. While photogrammetry remains the standard in restoration field due to its ability to produce reliable, high-resolution textures, it faces limitations when dealing with reflective, transparent, or textureless surfaces. Recent advancements in in neural rendering techniques, particularly Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting, offer promising alternatives by enabling photorealistic scene reconstruction from unstructured image sets. This study presents a comparative evaluation of photogrammetry, NeRFs, and Gaussian Splatting, assessing their performance and suitability for restoration-oriented applications, with particular attention to scenarios where the limitations of photogrammetry hinder comprehensive documentation. The evaluation is conducted on a set of heterogeneous case studies, ranging from sculptures and design objects to architectural heritage buildings.
Carmeliti et al. (Thu,) studied this question.