GPUPhot: Kernel-Based Algorithms for Point-Source Detection and Photometry with a Spatially Variable PSF
M. R. Alarcon, S. Lemes-Perera, M. Serra-Ricart, J. Licandro
Production-proven, openly released GPU-native pipeline with a neat exact decomposition of the variable-PSF matched filter, but validation is self-referential and lacks head-to-head accuracy benchmarks.
Abstract
Modern imaging surveys generate data at rates that exceed the throughput of photometry pipelines built on iterative, per-source fitting, a construct that does not map onto parallel hardware. We present a set of algorithms for point-source detection and photometry reformulated so that the entire reduction runs natively on a Graphics Processing Unit (GPU), implemented in GPUPhot. Their design rests on a single principle: every stage is cast as a convolution with a fixed kernel or an element-wise operation on full-frame maps. The algorithms are designed to operate on wide-field images with a spatially variable Point Spread Function (PSF), where the field-dependent PSF is described through a small set of fixed, flux-conserving principal-component eigenmodes with position-dependent coefficients, so that the spatially varying matched filter decomposes exactly into a few genuine convolutions while retaining its maximum-likelihood foundation. We show that, even in fields with strong background gradients and markedly inhomogeneous PSFs, the framework achieves sub-pixel astrometric accuracy and few-millimagnitude photometric precision for bright sources. The pipeline is already in production at the telescopes of the Teide Observatory, where it has processed 3.7 million images since 2023, amounting to 1.2 PB of data, at effective end-to-end rates below 15 s per 150-Mpx image, from ingestion of the raw frame to a source catalog with astrometric and photometric solutions. Because the same primitives underlie every stage, the individual algorithms are self-contained and applicable well beyond photometry. They form the basis of a complete, openly developed software package, released publicly with this paper.
AI Impact Assessments
(1 model)Core Contribution
GPUPhot is a GPU-native pipeline for point-source detection and photometry on wide-field images with a spatially variable PSF. Its organizing principle is that every stage is cast either as a convolution with a fixed kernel or as an element-wise operation on full-frame maps. The central technical idea is the PCA-eigenmode PSF model, P(x;µ) = P0 + Σ c_k(µ) e_k(x). With this model the position-dependent matched filter decomposes exactly into K+1 genuine FFT convolutions weighted by coefficient maps. The effective-area normalization is closed-form, and the zero-sum eigenmodes preserve flux. The paper also gives GPU-friendly formulations of cosmic-ray and RTN cleaning, tile-percentile background estimation, empirical clustered aperture corrections, and an adaptive aperture radius calibrated as a power law in detection significance.
Individual components are largely known. These include Zackay & Ofek matched filtering, PCA PSF variation (Jarvis & Jain), L.A.Cosmic-style rejection, and Astrometry.net. The novelty lies in the unified, consistently kernel-based design and in the demonstration that it works at production scale: 3.7M images and 1.2 PB since 2023, with under 15 s per 150-Mpx frame.
Methodological Rigor
Strengths:
Weaknesses:
Potential Impact
For time-domain and survey astronomy, throughput is a real bottleneck, with LSST, WFST, ATLAS, and similar projects producing large data rates. A public, production-proven GPU pipeline that handles spatially variable PSFs without per-source iteration is useful. The kernel-based decomposition is a reusable design pattern. Adoption will depend on whether others can use the pipeline on their instruments, and on independent benchmarks, which are currently lacking. The impact is likely concentrated in small-to-medium survey teams, NEO and transient programs, and GPU-focused astronomy-software developers, rather than in major collaborations with entrenched pipelines such as the LSST stack. It does not push the accuracy frontier: aperture photometry with an empirical correction is less accurate than full PSF-fitting photometry in crowded fields, and that tradeoff is not quantified.
Timeliness & Relevance
The paper is highly timely: Rubin's operations, GPU availability, and growing real-time alert needs make GPU-native reduction a live topic. Existing GPU efforts are stage-wise ports or single-stage accelerations (image subtraction), so a full end-to-end design with an open release fills a niche.
Strengths
Limitations
Overall
This is a solid, useful tool/system paper whose significance rests mainly on its deployment evidence and its reusable design pattern. It is not a conceptual breakthrough, and its evidence is weaker than the claims. I expect moderate citations from the astronomical-software and survey-pipeline community, with some uptake as a package.
Generated Nov 4, 2026
Comparison History (0)
Production-proven, openly released GPU-native pipeline with a neat exact decomposition of the variable-PSF matched filter, but validation is self-referential and lacks head-to-head accuracy benchmarks.