ProxIQA: A Proxy Approach to Perceptual Optimization of Learned Image Compression

  title={ProxIQA: A Proxy Approach to Perceptual Optimization of Learned Image Compression},
  author={Li-Heng Chen and C. Bampis and Z. Li and A. Norkin and A. Bovik},
  journal={IEEE Transactions on Image Processing},
The use of $\ell _{p}$ (p = 1,2) norms has largely dominated the measurement of loss in neural networks due to their simplicity and analytical properties. However, when used to assess the loss of visual information, these simple norms are not very consistent with human perception. Here, we describe a different “proximal” approach to optimize image analysis networks against quantitative perceptual models. Specifically, we construct a proxy network, broadly termed ProxIQA, which mimics the… Expand
UGC-VQA: Benchmarking Blind Video Quality Assessment for User Generated Content
Nonlinear Transform Coding
Predicting Eye Fixations Under Distortion Using Bayesian Observers
RAPIQUE: Rapid and Accurate Video Quality Prediction of User Generated Content
Regression or Classification? New Methods to Evaluate No-Reference Picture and Video Quality Models
Adaptive Debanding Filter
Learning to Distort Images Using Generative Adversarial Networks


Variational image compression with a scale hyperprior
Learning to generate images with perceptual similarity metrics
End-to-end Optimized Image Compression
The Unreasonable Effectiveness of Deep Features as a Perceptual Metric
Improved Lossy Image Compression with Priming and Spatially Adaptive Bit Rates for Recurrent Networks
Generative Adversarial Networks for Extreme Learned Image Compression
Full Resolution Image Compression with Recurrent Neural Networks
EnhanceNet: Single Image Super-Resolution Through Automated Texture Synthesis
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
  • C. Ledig, Lucas Theis, +6 authors W. Shi
  • Computer Science, Mathematics
  • 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  • 2017
Perceptual Losses for Real-Time Style Transfer and Super-Resolution