A dual-prior driven Gaussian splatting framework for high-fidelity reconstruction of museum artifacts

Standard 3D Gaussian Splatting (3DGS) relies on multi-view images, creating a barrier for museum artifacts archived only as colored point clouds without corresponding registered images. We propose a 3DGS paradigm that operates directly on colored point clouds, enabling high-fidelity reconstruction independent of external image sets. Our approach is guided by dual priors. First, a feature-aware sampling algorithm constructs a precise geometric prior with absolute scale from the input point cloud, serving as the geometric foundation. Second, an ideal visual prior provides supervision by generating synthetic views from the point cloud, which are then refined using an enhancement chain that includes anti-aliasing and super-resolution. Experiments on public and self-built datasets validate our method’s superior rendering quality, achieving a significant PSNR improvement of 3.25 dB over the baseline method on our self-built dataset. The proposed framework thus offers an innovative and robust pipeline for effectively leveraging vast archives of existing 3D scan data.