Loss Functions for Neural Networks for Image Processing

Abstract

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.

This paper was published in IEEE Transactions on Computational Imaging on December 23, 2016

References in corpus (2)

• Improving neural networks by preventing co-adaptation of feature detectors
• Caffe: Convolutional Architecture for Fast Feature Embedding

Cited by in corpus (35)

• Deep Learning for Single Image Super-Resolution: A Brief Review
• SurfaceNet: Adversarial SVBRDF Estimation from a Single Image
• Deep-Learning Inversion of Seismic Data
• An Introduction to Image Synthesis with Generative Adversarial Nets
• TomoGAN: Low-Dose Synchrotron X-Ray Tomography with Generative Adversarial Networks
• Deep Shading: Convolutional Neural Networks for Screen-Space Shading
• MLFcGAN: Multi-level Feature Fusion based Conditional GAN for Underwater Image Color Correction
• Convolutional Autoencoders for Human Motion Infilling
• Super-resolution PET imaging using convolutional neural networks
• Closed Loop Predictive Control of Adaptive Optics Systems with Convolutional Neural Networks
• Unsupervised Monocular Depth Estimation with Left-Right Consistency
• High-Resolution Image Inpainting using Multi-Scale Neural Patch Synthesis
• Adaptive-CS-Net: FastMRI with Adaptive Intelligence
• Single Image Reflection Separation with Perceptual Losses
• End-to-End Learning-Based Ultrasound Reconstruction
• Trusted Artificial Intelligence: Towards Certification of Machine Learning Applications
• Model-based Behavioral Cloning with Future Image Similarity Learning
• Real-Time Cardiac Cine MRI with Residual Convolutional Recurrent Neural Network
• On the Sins of Image Synthesis Loss for Self-supervised Depth Estimation
• Anomaly Detection via Self-organizing Map
• Multi-Scale Progressive Fusion Learning for Depth Map Super-Resolution
• Attention-based Residual Speech Portrait Model for Speech to Face Generation
• ALANET: Adaptive Latent Attention Network forJoint Video Deblurring and Interpolation
• Blind Image Super-Resolution with Spatial Context Hallucination
• Deeply Aggregated Alternating Minimization for Image Restoration
• Efficient and Phase-aware Video Super-resolution for Cardiac MRI
• Exploiting Style and Attention in Real-World Super-Resolution
• Monocular Depth Estimation with Directional Consistency by Deep Networks
• ATAC-Net: Zoomed view works better for Anomaly Detection
• Learning to See via Epiretinal Implant Stimulation in silico with Model-Based Deep Reinforcement Learning
• An investigation of pre-upsampling generative modelling and Generative Adversarial Networks in audio super resolution
• Functional Neural Networks for Parametric Image Restoration Problems
• Reconstructing unseen modalities and pathology with an efficient Recurrent Inference Machine
• Generalized Categorisation of Digital Pathology Whole Image Slides using Unsupervised Learning
• Deep Residual Networks with a Fully Connected Recon-struction Layer for Single Image Super-Resolution