StatueDiff: Regionally Enhanced Cross-Scale Diffusion for Single-View 3D Reconstruction of Classical Statues

Abstract. Promoted by techniques such as deep learning, automated 3D reconstruction of cultural artefacts and historical sites has made significant progress in many application fields in recent years, such as the creative industry. In museums, digital technologies are driving the transition from static display forms to more immersive and dynamic exhibition experiences. However, most artefacts in museum environments are still displayed in traditional showcases, with only a limited number of institutions adopting digital twins and presenting 3D models through their online platforms. This situation is particularly evident in the case of artefacts such as statues, with an emphasis on passive viewing and limited interaction with the audience. The development of permanent exhibitions that move beyond static visual presentation thus remains a largely unexplored frontier in interactive technology in museums. To address these challenges, this paper proposes a diffusion-based framework for Single-View 3D Reconstruction, using sculptural heritage objects as a case study. This deep learning approach takes a colour photograph as input and aims to generate a photorealistic 3D reconstruction with detail suitable for digital exhibitions in museums. Our preliminary results show that this approach has the potential to create a novel integration of art and technology, contributing to new possibilities for audience participation and immersive experiences within cultural heritage exhibitions.