fadinammour / galaxy2galaxy Goto Github PK

Generate a dataset with CFHT inputs and noiseless HST targets. The noiseless targets allow to perform more accurate measurements on the shape errors. Real galaxy images always contain noise this is why the galaxies of this dataset are parametric.

The chosen profile for the simulated galaxies is the Sersic one.

To do:

• Add ParametricCFHT2HST problem.
• Generate a subsample of the dataset.
• Check the generated subsample.
• Generate the full dataset.

Goal: Add a CFHT2HST problem that generates a dataset which inputs are CFHT like images and targets are HST images.

• create a COSMOS PSF that is the smallest isotropic PSF to reconstruct HST images. This PSF will be the one used to generate the HST target images from the COSMOS catalog.

• create a CFHT like PSF sampler. The PSF samples should mimic the ones of CFIS.

• simulate CFIS like images with realistic PSFs (from the task above) and realistic noise. For the latter, the r band filter of CFHT will be used and the chosen sky setting is Dark sky, Moon 0%. The information about the telescope can be found on the CFHT site and on the LSST GitHub page.

Information on Galsim flux units can be found here.

• declare a CFHT2HST problem.

• add the HST PSF.

• add CFHT PSFs.

• add noise to CFHT images.

• generate, check and validate the dataset.

We aim to generate a radio dataset using galaxy2galaxy. The main goal is to test an ADMM + U-Net implementation for deconvolving and denoising radio galaxies. The algorithm has already been tested for optical galaxies. Radio galaxy images mainly differ from optical galaxy images because of the presence of noise in Fourier space.

The following tasks need to be achieved -

• Create a separate branch for the radio dataset
  - Created on my local machine
• Rename the problem
• Generate a 1-shard dataset
• Remove the unused fields from cosmos.py and galsim_utils.py
• Adapt the MeerKAT PSF
• Adapt the galaxy images
• Preprocessing function for noise addition