Figure 1. We show how the proposed concept IRP differs from
the traditional image metric visual quality. Given a sequence of
images captured under varying exposures in dynamic scenes (first
row), both IQA metric NIQE [23] and human observers favorite
either (a) or (b) as relatively good shots, but they do not neces-
sarily lead to better restored results (second row). As a compari-
son, a perceptually poor image (c) leads to the best restored image
quality. The result indicates the necessity of developing the IRP
metric which predicts if the restored image quality will be good
even before it is processed. All images are processed by the same
restoration model MPR [40] to ensure fairness.