kshitijd/mmu-norm-legacy-north
Legacy Survey DR9 North image cutouts — L1 (release v1) This L1 repository contains 2,191,927 objects matched across the release, in 726 shards (about 700 GB). Each object has 152×152-pixel cutouts at 0.262″ per pixel in three bands. Band names: DR9 North imaging uses BASS g/r and MzLS z. This repository uses bass-g, bass-r, and mzls-z; the original des-g/des-r/des-z tokens are kept in native_band_tokens. Schema image struct: band (3), flux (3×152×152… See the full description on the dataset page: https://huggingface.co/datasets/kshitijd/mmu-norm-legacy-north.
Legacy Survey DR9 North image cutouts — L1 (release v1)
This L1 repository contains 2,191,927 objects matched across the release, in 726 shards (about 700 GB). Each object has 152×152-pixel cutouts at 0.262″ per pixel in three bands.
Band names: DR9 North imaging uses BASS g/r and MzLS z. This repository uses bass-g, bass-r, and mzls-z; the original des-g/des-r/des-z tokens are kept in native_band_tokens.
Schema
image struct: band (3), flux (3×152×152, nanomaggies/pixel — verified against catalog fluxes, ratio 0.92–0.96), psf_fwhm, scale, native_band_tokens. Catalog columns per row: flux_g/r/z, fiberflux_*, psfdepth_*, ebv, z_spec, ra, dec, object_id.
Conversion: flux_nJy = pixel × 3630.7805.
Data notes
- Pixel inverse variance and pixel masks are unavailable in this dataset.
psfdepthgives a catalog point-source depth and does not provide a pixel noise map. - The images are sky-subtracted coadds, and Lanczos resampling introduces correlated noise.
- The cutouts do not include WCS information;
raanddecgive their centers. - Fluxes are observed values.
ebvsupports optional extinction-corrected views.
The images already have a fixed shape. Their L2 preparation uses the training-split scaling statistics in mmu-norm-index → l2_stats/v1/ssl_legacysurvey_north.json.
Credit: DESI Legacy Imaging Surveys DR9 (Dey et al. 2019); BASS; MzLS. Sourced via Multimodal Universe SSL subset.