First, we captured or collected (e.g., from the Adobe FiveK dataset) a set of images of the same scene under different exposure conditions. For the normally exposed images, we further refined them ...