dfcosmic: A Python package for cosmic ray removal

TL;DR

dfcosmic is a Python package that efficiently removes cosmic ray artifacts from astronomical images using a GPU-accelerated port of the L.A.Cosmic algorithm, significantly improving processing speed.

Contribution

It introduces a Python implementation of L.A.Cosmic with GPU support, enabling faster cosmic ray removal for large-scale astronomical imaging.

Findings

Achieves at least 20% speed increase over previous Python versions

Supports efficient processing on both CPUs and GPUs

Designed for high-throughput astronomical imaging pipelines

Abstract

Astronomical images often show sharp features that are caused by cosmic ray (CR) hits, hot pixels, or non-Gaussian noise. L.A.Cosmic (van Dokkum 2001) is a widely used edge detection algorithm that identifies and replaces such features. Here we describe dfcosmic, a direct python port of L.A.Cosmic utilizing PyTorch and C++ to enable efficient performance on both CPUs and GPUs. The code was developed for the MOTHRA array, which is projected to produce more than 1000 large format CMOS images every 15 minutes. Compared to previous python implementations, dfcosmic achieves a speed gain of at least 20%.