This dataset provides restored stellar spectra and re-derived physical parameters for defective observations within the LAMOST DR10 survey. Due to instrumental and observational limitations, many conventional spectra contain missing or corrupted contiguous regions that compromise downstream analysis. 

To address this, we applied PRISM (Progressive Restoration and Inpainting for Spectral Measurement), a deep-learning model designed to restore defective spectra while preserving their scientific integrity. Utilizing a two-stage architecture and a progressive context-propagating strategy, PRISM successfully reconstructs large gaps and local details. The downstream analysis of these restored spectra yields physical parameters whose accuracy closely matches pristine observations.

File Structure

label.csv: Contains observation metadata and re-derived physical parameters.
    basename: The unique filename for the spectrum.
    obsid, lmjd, mjd, planid, spid, fiberid: Standard LAMOST observation metadata.
    pred_logg: Re-derived surface gravity (log g).
    pred_teff: Re-derived effective temperature (Teff).
    pred_feh: Re-derived metallicity ([Fe/H]).

restored_spec.zip: A compressed archive of the spectral files. Each spectrum is saved as a .csv file matching the basename and contains three columns:
    wavelength: Wavelength array.
    original_flux: The original LAMOST DR10 flux (containing defects/gaps).
    restored_flux: The continuous flux repaired by the PRISM model.

SDS-main.zip: Paper source code.