Is L2 a Good Loss Function for Neural Networks for Image Processing
Hang Zhao, Orazio Gallo, Iuri Frosio, Jan Kautz · arXiv (Cornell University) · 2015
Neural networks are becoming central in several areas of computer vision and image processing. Different architectures have been proposed to solve specific problems. The impact of the loss layer of neural networks, however, has not received much attention by the research community: the default and most common choice is L2. This can be particularly limiting in the context of image processing, since L2 correlates poorly with perceived image quality. In this paper we bring attention to alternative choices. We study the performance of several losses, including perceptually-motivated losses, and propose a novel, differentiable error function. We show that the quality of the results improves significantly with better loss functions, even for the same network architecture.