3D Supernova Remnant

What happens if you take a 2D image of a supernova remnant and deproject it in 3D using a Gaussian process prior?

→

This experiment uses the diffuse X-ray image of SN1006 from Westerkamp et al. 2024 .

We actually learn three Gaussian processes: a 1D radial profile for each spectral band, a 3D radial distortion field that shapes the bubble, and a 3D brightness scaling that captures surface variation. The prior draws look kind of cool:

Philipp Frank and I hacked this together as a demo using NIFTy, but additional constraints like velocity information could make this scientifically interesting!