Galaxy photometry underlies virtually all extragalactic science, yet many existing tools are limited to single-band fitting with hard to modify workflows. AstroPhot is a flexible tool for jointly fitting many sources in an image simultaneously across multiple bands/epochs, leveraging automatic differentiation and GPU acceleration (PyTorch/JAX) to make previously intractable analyses easy to set up. AstroPhot provides a large zoo of parametric models for stars, galaxies, and other sources, combinable to represent arbitrarily complex scenes. Key features:
- GPU acceleration — roughly an order of magnitude speedup; hardware is auto-detected
- Multi-band fitting — simultaneously models multiple images, epochs, and wavelength bands
- SED extraction — broadband spectral energy distributions via forward modelling, with no need for aperture corrections or PSF matching
- Crowded fields — handles large models with many overlapping objects
- Simultaneous PSF fitting — the PSF may be included as part of a model and optimized alongside other components
- Uncertainty analysis — built-in support for MCMC and Levenberg-Marquardt optimisers, with simultaneous PSF parameter optimisation