HeriNeRF‐JP: Occlusion‐Aware Prior‐Guided Neural Radiance Fields for Sparse‐View Reconstruction of Japanese Cultural Heritage

ABSTRACT

Sparse‐view heritage reconstruction can produce plausible but unsupported geometry. HeriNeRF‐JP utilizes photometric weighting and occlusion‐aware photometric weighting along with heritage‐prior modulation. The 25‐view summaries of a Japanese benchmark are reported with 28.24‐dB PSNR and 3.94‐cm Chamfer distance, with descriptive ablations showing component sensitivity. This is an expert vignette study with 65 professionals which cannot be reconstructed analytically and is therefore only retained as archived descriptive evidence. Uncertainty maps are not calibrated probabilities. The framework can be used for further survey prioritization, but implementation data are not available, compromising the ability to reproduce the survey results and inferred geometry needs to be verified by independent surveys.