Image–catalog complementarity in quasar–galaxy ambiguous regimes of photometric source classification

This dataset supports the paper “Image–catalog complementarity in quasar–galaxy ambiguous regimes of photometric source classification”. The archive contains the processed benchmark data used in the experiments. The benchmark consists of 12,000 spectroscopically labeled sources for star, galaxy, and quasar classification, with 4,000 sources in each class. The sample is constructed as a balanced medium-hard benchmark and is split into fixed train, validation, and test subsets with a 70:15:15 ratio. The archived files include processed catalog features, WISE/NEOWISE variability features, fixed train/validation/test split files, feature-configuration metadata, an image-cutout index, and the optical cutouts used in the image-based and image–catalog fusion experiments. The original spectroscopic labels are based on publicly available SDSS spectroscopic catalogs and quasar catalogs. The catalog measurements and optical image cutouts are based on publicly available DESI Legacy Imaging Surveys data products. Gaia astrometric and photometric measurements and WISE/NEOWISE infrared measurements are publicly available from their respective survey archives. These original public survey products are not redistributed in full in this archive. File contents include: - catalog_features.parquet: processed catalog-feature table.- variability_features.parquet: processed WISE/NEOWISE variability-feature table.- multimodal_medium_hard_12000.parquet: final 12,000-source multimodal benchmark table.- multimodal_medium_only_6000.parquet: medium-difficulty subset.- multimodal_hard_only_6000.parquet: hard-difficulty subset.- medium_hard_12000_with_split_seed42.parquet: full benchmark table with fixed split labels.- medium_hard_12000_train_seed42.parquet: training subset.- medium_hard_12000_val_seed42.parquet: validation subset.- medium_hard_12000_test_seed42.parquet: test subset.- split_config_medium_hard_12000_seed42.json: configuration file for the fixed data split.- feature_config.json: catalog-feature configuration file.- cutout_index_medium_hard_checked_12000.parquet: image-cutout index and quality-check information.- cutouts_medium_hard.zip: optical image cutouts used in the experiments.