mlsanityData to 14 Oct 2026

Perception-Distortion Trade-off in the SR Space Spanned by Flow Models

arXiv 2209.08564 ↗First posted 18 Sep 2022cs.CV cs.LG eess.IVcited by 2

Not on the map. Why?

Abstract (arXiv)

Flow-based generative super-resolution (SR) models learn to produce a diverse set of feasible SR solutions, called the SR space. Diversity of SR solutions increases with the temperature (τ) of latent variables, which introduces random variations of texture among sample solutions, resulting in visual artifacts and low fidelity. In this paper, we present a simple but effective image ensembling/fusion approach to obtain a single SR image eliminating random artifacts and improving fidelity without significantly compromising perceptual quality. We achieve this by benefiting from a diverse set of feasible photo-realistic solutions in the SR space spanned by flow models. We propose different image ensembling and fusion strategies which offer multiple paths to move sample solutions in the SR space to more desired destinations in the perception-distortion plane in a controllable manner depending on the fidelity vs. perceptual quality requirements of the task at hand. Experimental results demonstrate that our image ensembling/fusion strategy achieves more promising perception-distortion trade-off compared to sample SR images produced by flow models and adversarially trained models in terms of both quantitative metrics and visual quality.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 18 of its 22 references that are in the map. First posted on arXiv 18 Sep 2022.

Builds on

18 of 22 references.

References · 18

  1. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  2. 2022Generative Adversarial Networkscited by 5,627
  3. 2016Perceptual Losses for Real-Time Style Transfer and Super-Resolutioncited by 2,694
  4. 2015Variational Inference with Normalizing Flowscited by 2,170
  5. 2016Density estimation using Real NVPcited by 2,042
  6. 2016Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Networkcited by 1,908
  7. 2017Enhanced Deep Residual Networks for Single Image Super-Resolutioncited by 1,552
  8. 2018Glow: Generative Flow with Invertible 1x1 Convolutionscited by 1,545
  9. 2018Image Super-Resolution Using Very Deep Residual Channel Attention Networkscited by 1,264
  10. 2018ESRGAN: Enhanced Super-Resolution Generative Adversarial Networkscited by 1,254
  11. 2014Image Super-Resolution Using Deep Convolutional Networkscited by 1,224
  12. 2014NICE: Non-linear Independent Components Estimationcited by 1,058
6 more
  1. 2015Accurate Image Super-Resolution Using Very Deep Convolutional Networkscited by 903
  2. 2017The Perception-Distortion Tradeoffcited by 587
  3. 2020SRFlow: Learning the Super-Resolution Space with Normalizing Flowcited by 209
  4. 2019RankSRGAN: Generative Adversarial Networks with Ranker for Image Super-Resolutioncited by 145
  5. 2020Unsupervised Real-world Image Super Resolution via Domain-distance Aware Trainingcited by 83
  6. 2020ESRGAN+ : Further Improving Enhanced Super-Resolution Generative Adversarial Networkcited by 27

Built on by

Cited by year

Citing papers per year: 2024: 1; 2025: 1.

2 papers.

  1. 2025Leveraging Vision-Language Models to Select Trustworthy Super-Resolution Samples Generated by Diffusion Modelscited by 2
  2. 2024Trustworthy SR: Resolving Ambiguity in Image Super-resolution via Diffusion Models and Human Feedbackcited by 1