- ## 1. Overview
  
  The MSSPE-Dataset is a large-scale dataset for determining stellar atmospheric parameters. It integrates LAMOST spectra, SDSS photometric images/magnitudes, and WISE images/magnitudes.  
- Total Samples: 240,436
- Subsets:
	- 01: Low-SNR (5 <= SNR_g <= 50, 96,275 samples)
	- 02: High-SNR (SNR_g > 50, 144,161 samples)
- Data Split: Training, Validation, and Test sets at a 7:1:2 ratio.
- ## 2. Directory Structure
  
  The dataset is organized by SNR level, then by split (train/val/test), and finally by modality folders:  
  
  ``
  MSSPE-Dataset/
  ├── 01/ (Low-SNR)
  │   ├── train/
  │   │   ├── labels/             # Ground truth (CSV files)
  │   │   ├── lamost_spec/        # 1D spectra (NPY files)
  │   │   ├── sdss_photo/         # Photometric images (NPY files)
  │   │   ├── sdss_mag/           # SDSS magnitudes (NPY files)
  │   │   ├── wise_photo/         # WISE images (NPY files)
  │   │   └── wise_mag/           # WISE magnitudes (NPY files)
  │   ├── val/ ... (same structure as train)
  │   └── test/ ... (same structure as train)
  └── 02/ (High-SNR)
      ├── train/ ...
      ├── val/ ...
      └── test/ ...``  
- ## 3. File Formats & Content
- labels/: Contains .csv files with stellar parameters (Teff, log g, [Fe/H]).
- lamost_spec/: .npy files containing 1D spectral flux vectors.
- sdss_photo/ & wise_photo/: .npy files containing multi-band image cutouts.
- sdss_mag/ & wise_mag/: .npy files containing multi-band magnitude values.
- ## 4. Usage Note
  
  This dataset incorporates real-world observational incompleteness. While LAMOST spectra are provided for all sources, the SDSS and WISE photometric modalities may be incomplete for certain samples, allowing for the evaluation of model robustness under missing data scenarios.