Physics-Guided Spectral-Geometric Restoration and Closed-Loop Verification on Simulated Cultural Heritage Benchmarks
Year: 2026
Type: dataset
Abstract: SG-RestorAgent simulated cultural heritage benchmark This dataset accompanies the manuscript Physics-Guided Spectral-Geometric Restoration and Closed-Loop Verification on Simulated Cultural Heritage Benchmarks. The archive contains 1,200 simulated, co-registered RGB, hyperspectral and three-dimensional micro-relief patches from 120 source-artwork clusters. Cluster-disjoint training, validation and test splits contain 720, 180 and 300 patches, respectively. These are simulated benchmarks, not newly acquired heritage scans. Contents benchmark/: multimodal patch tensors, split assignments, fitted model checkpoints, patch catalogues, validation and test evaluations, verification trajectories and pigment spectra. derived/: evaluation tables, cluster-aware statistics, calibration and sensitivity results, figure inputs and verification ledger. README.md: dataset description and reproduction instructions. Extract the associated software archive as code/ beside data/. Run python code/scripts/run_evaluation.py, followed by python code/scripts/make_figures.py. The accompanying software README maps manuscript tables and figures to their source files. Comparisons use shared-stem parametric surrogate implementations. Files are restricted; a read-only peer-review link is supplied in the manuscript. Dataset licence: CC BY 4.0. Cite this record and the associated article when reusing the data. Download and reconstruction The complete data ZIP is stored as 16 lossless parts. Download all A0136-data.zip.part01 through part16, manifest.json and merge_data.py into one directory. Run python merge_data.py to verify each part and reconstruct A0136-data.zip, then extract the ZIP. No dataset values or source files were changed by splitting.
Author: Zhang, Xiaoliang
Institution: Jiaozuo Normal College
Language: English
Subfield: Archeology (arts and humanities)
Field: Arts and Humanities
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7235189233
