1. Introduction
Lossy image compression considers the tradeoff be-tween the number of bits used to store an input image and how close the reconstruction (that we obtain from the bits) is to that input image. As we use more bits, we will be able to get closer to the input. This idea is formalized in the fundamental rate-distortion tradeoff [34], where “rate” stands for bitrate, and “distortion” is formalized as a pair-wise metric between the input image and the reconstruction (e.g., the mean-squared error, MSE).