The Perception-Distortion Tradeoff Supplementary Material

In this supplemental, we first provide a proof for Theorem 1, the derivation of Example 1, and the derivation of the MMSE and MAP estimators which appear in Sections 3.1 and 3.2. We then briefly discuss the real-vs.-fake study setting and its relation to Bayesian hypothesis testing. In the following sections, we specify all training and architecture details for the WGAN experiment which was presented in Section 5, and also include details regarding the super-resolution algorithms comparison in Section 6. Finally, we include additional comparisons between super-resolution algorithms using two extra no-reference methods, and perform a comparison on RGB images as well.