PICASSO

Python Inpainter for Cosmological and AStrophysical SOurces

This package provides a suite of inpainting methodologies aimed at reconstructing holes on images (128x128 pixels) extracted from a HEALPIX map.

Three inpainting techniques are included in PICASSO and can be divided into two main groups:

1. diffusive-based methods (Nearest-Neighbours)
2. learning-based methods that rely on training DCNNs to fill the missing pixels with the predictions learned from a training data-set (Deep-Prior and Generative Adversarial Networks, GAN ).

For further details see Puglisi et al. (2020) < >.

Requirements

• tensorflow
• keras
• astropy
• reproject
• mpi4py (for parallel inpainting)
• argparse
• neuralgym with pip install git+https://github.com/JiahuiYu/neuralgym

Install

git clone https://github.com/giuspugl/picasso
cd picasso
python setup.py install

Usage

Scripts are provided to the user in order to perform:

• projection from full sky HEALPIX maps to flat thumbnails images image_stacker
• inpainting on GPUs inpaint_gpu
• parallel inpainting on multiple processes (with mpi4py ) inpaint_mpi
• projection from flat images to HEALPIX inpaint_gpu

Pretrained models

GAN training weights

Download the model directories (rename checkpoint.txt to checkpoint because google drive automatically add ext after download)

Support

If you encounter any difficulty in installing and using the code or you think you found a bug, please open an issue.