pcaGAN: Improving Posterior-Sampling cGANs via Principal Component Regularization

In ill-posed imaging inverse problems, there can exist many hypotheses that fit both the observed measurements and prior knowledge of the true image.