mlsanityData to 14 Oct 2026

The Role of ImageNet Classes in Fr'echet Inception Distance

arXiv 2203.06026 ↗First posted 11 Mar 2022cs.CV cs.AI cs.LGcited by 213

Not on the map. Why?

Cited by 76 in Diffusion models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Fr'echet Inception Distance (FID) is the primary metric for ranking models in data-driven generative modeling. While remarkably successful, the metric is known to sometimes disagree with human judgement. We investigate a root cause of these discrepancies, and visualize what FID "looks at" in generated images. We show that the feature space that FID is (typically) computed in is so close to the ImageNet classifications that aligning the histograms of Top-N classifications between sets of generated and real images can reduce FID substantially -- without actually improving the quality of results. Thus, we conclude that FID is prone to intentional or accidental distortions. As a practical example of an accidental distortion, we discuss a case where an ImageNet pre-trained FastGAN achieves a FID comparable to StyleGAN2, while being worse in terms of human evaluation.

Key facts. Cited by 213 papers in the map (data through 14 Oct 2026). Builds on 56 of its 72 references that are in the map. First posted on arXiv 11 Mar 2022.

Builds on

56 of 72 references.

References · 56

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  3. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  4. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  5. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  6. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  7. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  8. 2019EfficientNet: Rethinking Model Scaling for Convolutional Neural Networkscited by 6,442
  9. 2018A Style-Based Generator Architecture for Generative Adversarial Networkscited by 6,013
  10. 2022Generative Adversarial Networkscited by 5,627
  11. 2015Rethinking the Inception Architecture for Computer Visioncited by 5,475
  12. 2021Zero-Shot Text-to-Image Generationcited by 4,288
38 more
  1. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,101
  2. 2013Auto-Encoding Variational Bayescited by 4,053
  3. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,614
  4. 2019Analyzing and Improving the Image Quality of StyleGANcited by 3,588
  5. 2020Taming Transformers for High-Resolution Image Synthesiscited by 3,253
  6. 2016Improved Techniques for Training GANscited by 2,970
  7. 2020Unsupervised Learning of Visual Features by Contrasting Cluster Assignmentscited by 2,857
  8. 2018Large Scale GAN Training for High Fidelity Natural Image Synthesiscited by 2,626
  9. 2016Density estimation using Real NVPcited by 2,042
  10. 2016Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Networkcited by 1,908
  11. 2018Glow: Generative Flow with Invertible 1x1 Convolutionscited by 1,545
  12. 2021Image Super-Resolution via Iterative Refinementcited by 1,511
  13. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,432
  14. 2019Generating Diverse High-Fidelity Images with VQ-VAE-2cited by 1,387
  15. 2015LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loopcited by 1,355
  16. 2019Semantic Image Synthesis with Spatially-Adaptive Normalizationcited by 1,299
  17. 2018Demystifying MMD GANscited by 1,293
  18. 2020Training Generative Adversarial Networks with Limited Datacited by 1,186
  19. 2021Alias-Free Generative Adversarial Networkscited by 1,184
  20. 2019StarGAN v2: Diverse Image Synthesis for Multiple Domainscited by 1,098
  21. 2019Improved Precision and Recall Metric for Assessing Generative Modelscited by 944
  22. 2016Conditional Image Generation with PixelCNN Decoderscited by 862
  23. 2016Pixel Recurrent Neural Networkscited by 816
  24. 2022StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasetscited by 492
  25. 2020Reliable Fidelity and Diversity Metrics for Generative Modelscited by 434
  26. 2018Assessing Generative Models via Precision and Recallcited by 406
  27. 2017Are GANs Created Equal? A Large-Scale Studycited by 267
  28. 2021Generating Images with Sparse Representationscited by 266
  29. 2020Swapping Autoencoder for Deep Image Manipulationcited by 215
  30. 2021Projected GANs Converge Fastercited by 208
  31. 2019U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translationcited by 186
  32. 2022Learning Fast Samplers for Diffusion Models by Differentiating Through Sample Qualitycited by 182
  33. 2019Blind Super-Resolution Kernel Estimation using an Internal-GANcited by 175
  34. 2019Effectively Unbiased FID and Inception Score and where to find themcited by 165
  35. 2021Towards Faster and Stabilized GAN Training for High-fidelity Few-shot Image Synthesiscited by 140
  36. 2021Pros and Cons of GAN Evaluation Measures: New Developmentscited by 123
  37. 2021Ensembling Off-the-shelf Models for GAN Trainingcited by 86
  38. 2018An empirical study on evaluation metrics of generative adversarial networkscited by 81

50 most cited of 56 All 56 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 21; 2023: 67; 2024: 53; 2025: 39; 2026: 33.

213 papers.

Sort:

By area:

  1. 2023Scaling up GANs for Text-to-Image Synthesiscited by 492
  2. 2022Paint by Example: Exemplar-based Image Editing with Diffusion Modelscited by 489
  3. 2023Consistency Trajectory Models: Learning Probability Flow ODE Trajectory of Diffusioncited by 459
  4. 2023Improved Techniques for Training Consistency Modelscited by 439
  5. 2022Rodin: A Generative Model for Sculpting 3D Digital Avatars Using Diffusioncited by 309
  6. 2023Preserve Your Own Correlation: A Noise Prior for Video Diffusion Modelscited by 285
  7. 2023Pixel-Aware Stable Diffusion for Realistic Image Super-resolution and Personalized Stylizationcited by 254
  8. 2023StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesiscited by 212
  9. 2023Glaze: Protecting Artists from Style Mimicry by Text-to-Image Modelscited by 205
  10. 2023Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion modelscited by 205
  11. 2022Pretraining is All You Need for Image-to-Image Translationcited by 160
  12. 2023Concept Sliders: LoRA Adaptors for Precise Control in Diffusion Modelscited by 153
38 more
  1. 2023TextMesh: Generation of Realistic 3D Meshes From Text Promptscited by 135
  2. 2024Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformerscited by 118
  3. 2024InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Priorcited by 117
  4. 2022Generating Long Videos of Dynamic Scenescited by 113
  5. 2024Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesiscited by 105
  6. 2024Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenizationcited by 100
  7. 2024WAVES: Benchmarking the Robustness of Image Watermarkscited by 97
  8. 2023Improving Statistical Fidelity for Neural Image Compression with Implicit Local Likelihood Modelscited by 94
  9. 2023WoVoGen: World Volume-aware Diffusion for Controllable Multi-camera Driving Scene Generationcited by 93
  10. 2022RoentGen: Vision-Language Foundation Model for Chest X-ray Generationcited by 89
  11. 2023Differentially Private Diffusion Models Generate Useful Synthetic Imagescited by 86
  12. 2023Domain-Agnostic Tuning-Encoder for Fast Personalization of Text-To-Image Modelscited by 85
  13. 2025What matters for Representation Alignment: Global Information or Spatial Structure?cited by 84
  14. 2022A Morphology Focused Diffusion Probabilistic Model for Synthesis of Histopathology Imagescited by 74
  15. 2024EchoScene: Indoor Scene Generation via Information Echo over Scene Graph Diffusioncited by 67
  16. 2024On the Content Bias in Fr'echet Video Distancecited by 52
  17. 2022Efficient Diffusion Models for Vision: A Surveycited by 51
  18. 2022DeepPrivacy2: Towards Realistic Full-Body Anonymizationcited by 51
  19. 2024Image Watermarks are Removable Using Controllable Regeneration from Clean Noisecited by 50
  20. 2024Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Stepcited by 47
  21. 2023Perceptual Artifacts Localization for Image Synthesis Taskscited by 43
  22. 2023360-Degree Panorama Generation from Few Unregistered NFoV Imagescited by 42
  23. 2023CycleNet: Rethinking Cycle Consistency in Text-Guided Diffusion for Image Manipulationcited by 41
  24. 2025AlphaFlow: Understanding and Improving MeanFlow Modelscited by 41
  25. 2023Continuous-Time Functional Diffusion Processescited by 39
  26. 2022StudioGAN: A Taxonomy and Benchmark of GANs for Image Synthesiscited by 38
  27. 2024Rethinking Score Distillation as a Bridge Between Image Distributionscited by 38
  28. 2022Semi-Parametric Neural Image Synthesiscited by 37
  29. 2023Diffusion Models and Semi-Supervised Learners Benefit Mutually with Few Labelscited by 37
  30. 2024Finding NeMo: Localizing Neurons Responsible For Memorization in Diffusion Modelscited by 37
  31. 2026Representation Fr'echet Loss for Visual Generationcited by 37
  32. 2022How Much is Enough? A Study on Diffusion Times in Score-based Generative Modelscited by 35
  33. 2025CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generationcited by 35
  34. 2022LidarCLIP or: How I Learned to Talk to Point Cloudscited by 33
  35. 2023Disentangled Clothed Avatar Generation from Text Descriptionscited by 33
  36. 2024Facing the Elephant in the Room: Visual Prompt Tuning or Full Finetuning?cited by 31
  37. 2024On the Challenges and Opportunities in Generative AIcited by 31
  38. 2023GeoDream: Disentangling 2D and Geometric Priors for High-Fidelity and Consistent 3D Generationcited by 30

50 most cited of 213 All 213 (JSON)