Loss Functions for Neural Networks for Image Processing
This work addresses the under-explored impact of loss functions in neural networks for image processing, offering incremental improvements for researchers and practitioners in computer vision.
The paper tackled the problem of image restoration by exploring alternative loss functions beyond the default L2, showing that perceptually-motivated losses significantly improve output quality for human observers, with concrete improvements reported.
Neural networks are becoming central in several areas of computer vision and image processing and different architectures have been proposed to solve specific problems. The impact of the loss layer of neural networks, however, has not received much attention in the context of image processing: the default and virtually only choice is L2. In this paper, we bring attention to alternative choices for image restoration. In particular, we show the importance of perceptually-motivated losses when the resulting image is to be evaluated by a human observer. We compare the performance of several losses, and propose a novel, differentiable error function. We show that the quality of the results improves significantly with better loss functions, even when the network architecture is left unchanged.