mlsanityData to 14 Oct 2026

Evaluating generative networks using Gaussian mixtures of image features

arXiv 2110.05240 ↗First posted 8 Oct 2021cs.CV cs.LGcited by 11

Not on the map. Why?

Abstract (arXiv)

We develop a measure for evaluating the performance of generative networks given two sets of images. A popular performance measure currently used to do this is the Fr'echet Inception Distance (FID). FID assumes that images featurized using the penultimate layer of Inception-v3 follow a Gaussian distribution, an assumption which cannot be violated if we wish to use FID as a metric. However, we show that Inception-v3 features of the ImageNet dataset are not Gaussian; in particular, every single marginal is not Gaussian. To remedy this problem, we model the featurized images using Gaussian mixture models (GMMs) and compute the 2-Wasserstein distance restricted to GMMs. We define a performance measure, which we call WaM, on two sets of images by using Inception-v3 (or another classifier) to featurize the images, estimate two GMMs, and use the restricted 2-Wasserstein distance to compare the GMMs. We experimentally show the advantages of WaM over FID, including how FID is more sensitive than WaM to imperceptible image perturbations. By modelling the non-Gaussian features obtained from Inception-v3 as GMMs and using a GMM metric, we can more accurately evaluate generative network performance.

Key facts. Cited by 11 papers in the map (data through 14 Oct 2026). Builds on 23 of its 46 references that are in the map. First posted on arXiv 8 Oct 2021.

Builds on

23 of 46 references.

References · 23

  1. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  2. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  3. 2014Explaining and Harnessing Adversarial Examplescited by 7,364
  4. 2018A Style-Based Generator Architecture for Generative Adversarial Networkscited by 6,013
  5. 2015Rethinking the Inception Architecture for Computer Visioncited by 5,475
  6. 2016Image-to-Image Translation with Conditional Adversarial Networkscited by 4,527
  7. 2013Auto-Encoding Variational Bayescited by 4,053
  8. 2019Analyzing and Improving the Image Quality of StyleGANcited by 3,588
  9. 2017Progressive Growing of GANs for Improved Quality, Stability, and Variationcited by 2,988
  10. 2018Large Scale GAN Training for High Fidelity Natural Image Synthesiscited by 2,626
  11. 2016Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Networkcited by 1,908
  12. 2018Demystifying MMD GANscited by 1,293
11 more
  1. 2018ESRGAN: Enhanced Super-Resolution Generative Adversarial Networkscited by 1,254
  2. 2016Generative Visual Manipulation on the Natural Image Manifoldcited by 348
  3. 2017Compressed Sensing using Generative Modelscited by 339
  4. 2018Pros and Cons of GAN Evaluation Measurescited by 224
  5. 2020PIPAL: a Large-Scale Image Quality Assessment Dataset for Perceptual Image Restorationcited by 178
  6. 2019Effectively Unbiased FID and Inception Score and where to find themcited by 165
  7. 2019A Wasserstein-type distance in the space of Gaussian Mixture Modelscited by 75
  8. 2017Optimal transport for Gaussian mixture modelscited by 45
  9. 2018An Improved Evaluation Framework for Generative Adversarial Networkscited by 26
  10. 2021Conditional Frechet Inception Distancecited by 16
  11. 2016A Distance for HMMs based on Aggregated Wasserstein Metric and State Registrationcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 2; 2024: 4; 2025: 4; 2026: 1.

11 papers.

  1. 2024Geometry Fidelity for Spherical Imagescited by 9
  2. 2025A Pragmatic Note on Evaluating Generative Models with Fr'echet Inception Distance for Retinal Image Synthesiscited by 9
  3. 2025Slicing the Gaussian Mixture Wasserstein Distancecited by 7
  4. 2023Gromov-Wasserstein-like Distances in the Gaussian Mixture Models Spacecited by 4
  5. 2024Unifying and extending Precision Recall metrics for assessing generative modelscited by 1
  6. 2024Offline Evaluation of Set-Based Text-to-Image Generationcited by 1
  7. 2023Using Skew to Assess the Quality of GAN-generated Image Featurescited by 0
  8. 2024Distribution-aware Noisy-label Crack Segmentationcited by 0
  9. 2025Differentiable Expectation-Maximisation and Applications to Gaussian Mixture Model Optimal Transportcited by 0
  10. 2025A New Perspective on Precision and Recall for Generative Modelscited by 0
  11. 2026What FID Hides: Detecting, Ranking, and Diagnosing Deviations in Generative Evaluationcited by 0