Single-image 3D reconstruction of painted potteries using AI diffusion and feedforward models
Year: 2025
Type: article
Abstract: The study of painted potteries is crucial for archaeological research. While traditional documentation methods like manual line drawings and 2D photography remain relevant, 3D technologies offer significant advantages. In this paper, we introduce a novel single-view image to 3D documentation approach for painted potteries. Experiments on various datasets show that integrating diffusion and feedforward models yields robust 3D reconstruction performance. In this hybrid pipeline, the diffusion model first generates novel view synthesis from the input image, then the reconstruction model uses those views to create a relatively accurate 3D mesh. In our tests the InstantMesh method excels at recovering precise object shape. These findings demonstrate that AI-driven 3D generation from a single-view image is an effective solution for documenting painted potteries, especially when artifacts are physically inaccessible. It not only enhances the richness and dimensionality of digital documentation but also broadens accessibility for conservation, analysis, and cultural dissemination.
Source: npj Heritage Science
Authors: Yuehan Wang, Haozhe Cui, Jiaying Du, Haichuan Di, Le Yan, Fang Xu, Kaiqi Cui, Yiduo Liu, Dongdong Tu
Institution: ShanghaiTech University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Funder: ShanghaiTech University
OpenAlex ID: W4415600642