We present a fresh perspective on shot noise corrupted images and noise
removal. By viewing image formation as the sequential accumulation of photons
on a detector grid, we show that a network trained to predict where the next
photon could arrive is in fact solving the minimum mean squ
Poisson flow generative models (PPFM) for low-dose CT image denoising outperforms diffusion models with a significantly reduced number of function evaluations (NFE=1).