blendset.yaml

# Copyright (c) 2012-2026 by the GalSim developers team on GitHub
# https://github.com/GalSim-developers
#
# This file is part of GalSim: The modular galaxy image simulation toolkit.
# https://github.com/GalSim-developers/GalSim
#
# GalSim is free software: redistribution and use in source and binary forms,
# with or without modification, are permitted provided that the following
# conditions are met:
#
# 1. Redistributions of source code must retain the above copyright notice, this
#    list of conditions, and the disclaimer given in the accompanying LICENSE
#    file.
# 2. Redistributions in binary form must reproduce the above copyright notice,
#    this list of conditions, and the disclaimer given in the documentation
#    and/or other materials provided with the distribution.


# A simulation of sets of blended galaxies written to a MEDS file.
# As in blend.yaml, there are two output files, one with the blended images and one with
# the neighbors removed.  i.e. what you might get from a perfect deblender.
# This time each galaxy in the set is given a postage stamp where it is at the center.
# The same noise realization is used for each full set of blended objects, so this corresponds
# to multiple cutouts from a larger image where each detection is at the center, as you would
# get for a normal MEDS file from real data.

modules:
    # Use the extra features from the galsim.des module
    - galsim.des

    # A module in this directory that defines some custom types related to blending
    - blend

    # Another module in this directory that defines the ExcludedRandom type
    - excluded_random

psf:
    # Use a PSFEx PSF as measured from DES SV data.
    type: DES_PSFEx

    # The image_pos doesn't matter much for the MEDS output, but it is used to get the
    # right PSF from the PSFEx file.
    image_pos :
        type : XY
        x : { type : Random, min: 1, max: 2048 }
        y : { type : Random, min: 1, max: 4096 }


gal:
    # Choose random galaxies from the COSMOS catalog.
    type: COSMOSGalaxy
    index: { type: Random }

    # Either of these will work here, but the real galaxies are much slower on such a small
    # simulation, since they require a lot of setup time to read in each of the RealGalaxy
    # image files.  On a typically large simulation, this time gets amortized over a lot
    # of objects, so the difference isn't as huge.  So to keep the running time reasonable,
    # we use parameteric here.
    gal_type: parametric
    #gal_type: real

    # Randomly spin the original profile
    rotate: { type: Random }

    # This works out about right for the sky level we chose below.  The galaxies mostly seem
    # to have pretty decent S/N based on visual inspection of image.
    scale_flux: 1000

    # Just a constant shear for now.
    shear:
        type: G1G2
        g1: 0.03
        g2: -0.02

stamp:
    # In blend.py (loaded by the modules field above), we define a custom stamp type
    # called BlendSet.
    # The first stamp is similar to what we did in blend.yaml with the stamp type Blend.
    # It has one galaxy centered (more or less -- cf image.offset) on the stamp, and then
    # some number of neighbors also appearing on the stamp placed randomly within some range
    # of allowed separations.
    # However, then BlendSet repeats the image n_neighbors more times centering on each of the
    # neighbors in turn.
    type: BlendSet

    # Up to 5 neighbors.  Each neighbor has a prob of p=0.3 of being included.
    n_neighbors:
        type: RandomBinomial
        N: 5
        p: 0.3

    # The neighbors cannot be too close (assume they are rejected as unusable) or too far
    # (not really a blend).
    min_sep: 2   # arcsec -- clearly overlapping.  Too hard to deblend.
    max_sep: 10  # arcsec -- this would be off the edge of the stamp.  Not really blended.

    # Offset is just random within central pixel.
    offset:
        type: XY
        x: { type: Random, min: -0.5, max: 0.5 }
        y: { type: Random, min: -0.5, max: 0.5 }


image:
    pixel_scale : 0.26

    random_seed: 8675309

    # Poisson noise
    noise:
        type: Poisson
        sky_level: 800
        whiten: True  # Only relevant for gal_type = real, but it doesn't hurt to leave this in.


input:
    cosmos_catalog :
        # Use the same COSMOS catalog choices we made in meds.yaml
        template : "meds.yaml:input.cosmos_catalog"

        # Also impose a minimum size, which we didn't there, to avoid very small galaxies.
        # This is just a cosmetic choice for this example to make the galaxies easier to see.
        # You probably wouldn't want to do this for something that was attempting to be realistic.
        min_hlr : 0.5  # arcsec

    des_psfex :
        dir : 'des_data'
        file_name :
            type : NumberedFile
            root : 'DECam_00154912_'
            digits : 2
            ext : '_psfcat.psf'
            num :
                # Pick a random chip, but not either of our bad chips.
                type : ExcludedRandom  # This is a custom type defined in exclude_random.py
                min: 1
                max : 62
                exclude: [ 61, 31 ]


# We have two separate runs, one writing to a MEDS file and the other writing a Tiled image.
# These have different output fields, defined in the two sections below.  We also put the
# parts of the iamge field that differ there as well.

---

# The first output file is a MEDS file, as would be made from DES data.

image:
    size: 64


output:
    type: MEDS
    nobjects: 100
    nstamps_per_object: 1

    nfiles: 1

    dir: output
    file_name:
        type: FormattedStr
        format: "blendset-v001-%03d.fits"
        items:
        - { type: Sequence, nitems: '@output.nfiles' }

    # Need a different specification of the deblend output to get it to write a MEDS file.
    deblend_meds:
        file_name:
            template: :output.file_name   # template with no filename uses current config.
            format: "blendset-deblend-v001-%03d.fits"

---

# The second output file is a tiled image, just to make it easier to view the stamps in ds9.

image:
    type: Tiled
    stamp_size : 64
    nx_tiles : 10
    ny_tiles : 10

output:
    type: Fits

    nfiles: 1

    dir: output
    file_name:
        type: FormattedStr
        format: "blendset-tiled-v001-%03d.fits"
        items:
        - { type: Sequence, nitems: '@output.nfiles' }

    # This is the same output specification we used in blend.yaml.
    deblend:
        file_name:
            type: FormattedStr
            format: "blendset-tiled-deblend-v001-%03d.fits"
            items:
            - { type: Sequence, nitems: '@output.nfiles' }