mlsanityData to 14 Oct 2026

Making Reconstruction FID Predictive of Diffusion Generation FID

arXiv 2603.05630 ↗First posted 5 Mar 2026cs.CV cs.LGcited by 13

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

It is well known that the reconstruction FID (rFID) of a VAE is poorly correlated with the generation FID (gFID) of a latent diffusion model. We propose interpolated FID (iFID), a simple variant of rFID that exhibits a strong correlation with gFID. Specifically, for each dataset element, we retrieve its nearest neighbor in latent space, interpolate between their latent representations, decode the interpolated latent, and compute the FID between the decoded samples and the original dataset. We provide an intuitive explanation for why iFID correlates well with gFID, and why reconstruction metrics can be negatively correlated with gFID, by connecting iFID to recent results on diffusion generalization and hallucination. Theoretically, we show that iFID evaluates decoded interpolations aligned with the ridge set around which diffusion samples concentrate, thereby measuring a quantity closely related to diffusion sample quality. Empirically, iFID is the first metric shown to strongly correlate with diffusion gFID across diverse VAEs, achieving Pearson and Spearman correlations of approximately 0.85. The project page is available at https://tongdaxu.github.io/pages/ifid.html.

Key facts. Cited by 13 papers in the map (data through 14 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 58 of its 66 references that are in the map. First posted on arXiv 5 Mar 2026.

Builds on

58 of 66 references.

Closest ancestors · 37

Same direction, newest first

  1. 2026Unified Latents (UL): How to train your latents19 Feb
  2. 2026Diffusion Model's Generalization Can Be Characterized by Inductive Biases toward a Data-Dependent Ridge Manifold5 Feb
  3. 2026Laminating Representation Autoencoders for Efficient Diffusion4 Feb
  4. 2025Generalization of Diffusion Models Arises with a Balanced Representation Space24 Dec
  5. 2025One Layer Is Enough: Adapting Pretrained Visual Encoders for Image Generation8 Dec
  6. 2025Delving into Latent Spectral Biasing of Video VAEs for Superior Diffusability5 Dec
  7. 2025Latent Diffusion Model without Variational Autoencoder17 Oct
  8. 2025Diffusion Transformers with Representation Autoencoders13 Oct
  9. 2025Selective Underfitting in Diffusion Models1 Oct
  10. 2025AlignTok: Aligning Visual Foundation Encoders to Tokenizers for Diffusion Models29 Sep
  11. 2025On the Edge of Memorization in Diffusion Models25 Aug
  12. 2025Latent Diffusion Models with Masked AutoEncoders14 Jul
25 more
  1. 2025Why Diffusion Models Don't Memorize: The Role of Implicit Dynamical Regularization in Training23 May
  2. 2025REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers14 Apr
  3. 2025Improving the Diffusability of Autoencoders20 Feb
  4. 2025Masked Autoencoders Are Effective Tokenizers for Diffusion Models5 Feb
  5. 2025Memorization and Regularization in Generative Diffusion Models27 Jan
  6. 2025Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models2 Jan
  7. 2024Understanding and Mitigating Memorization in Generative Models via Sharpness of Probability Landscapes5 Dec
  8. 2024Towards a Mechanistic Explanation of Diffusion Model Generalization28 Nov
  9. 2024Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think9 Oct
  10. 2024Going beyond Compositions, DDPMs Can Produce Zero-Shot Interpolations29 May
  11. 2024Improved Distribution Matching Distillation for Fast Image Synthesis23 May
  12. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesis5 Mar
  13. 2024Critical windows: non-asymptotic theory for feature emergence in diffusion models3 Mar
  14. 2024SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers16 Jan
  15. 2023The Journey, Not the Destination: How Data Guides Diffusion Models11 Dec
  16. 2023Closed-Form Diffusion Models19 Oct
  17. 2023Compositional Abilities Emerge Multiplicatively: Exploring Diffusion Models on a Synthetic Task13 Oct
  18. 2023Generalization in diffusion models arises from geometry-adaptive harmonic representations4 Oct
  19. 2023UDPM: Upsampling Diffusion Probabilistic Models25 May
  20. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  21. 2021Diffusion Priors In Variational Autoencoders29 Jun
  22. 2021Score-based Generative Modeling in Latent Space10 Jun
  23. 2021Maximum Likelihood Training of Score-Based Diffusion Models22 Jan
  24. 2020Score-Based Generative Modeling through Stochastic Differential Equations26 Nov
  25. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 21

  1. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  2. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  3. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,403
  4. 2025Qwen-Image Technical Reportcited by 1,044
  5. 2024Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generationcited by 457
  6. 2022Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Modelscited by 424
  7. 2022BEiT v2: Masked Image Modeling with Vector-Quantized Visual Tokenizerscited by 303
  8. 2025X-Omni: Reinforcement Learning Makes Discrete Autoregressive Image Generative Models Great Againcited by 112
  9. 2024An analytic theory of creativity in convolutional diffusion modelscited by 104
  10. 2024Understanding Hallucinations in Diffusion Models through Mode Interpolationcited by 91
  11. 2025EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modelingcited by 89
  12. 2025GigaTok: Scaling Visual Tokenizers to 3 Billion Parameters for Autoregressive Image Generationcited by 57
9 more
  1. 2024SoftVQ-VAE: Efficient 1-Dimensional Continuous Tokenizercited by 56
  2. 2024Spherical Linear Interpolation and Text-Anchoring for Zero-shot Composed Image Retrievalcited by 44
  3. 2025Towards Scalable Pre-training of Visual Tokenizers for Generationcited by 28
  4. 2025The Prism Hypothesis: Harmonizing Semantic and Pixel Representations via Unified Autoencodingcited by 22
  5. 2024Image Understanding Makes for A Good Tokenizer for Image Generationcited by 15
  6. 2025From Navigation to Refinement: Revealing the Two-Stage Nature of Flow-based Diffusion Models through Oracle Velocitycited by 8
  7. 2025Distribution Matching Variational AutoEncodercited by 6
  8. 2025Enforcing Latent Euclidean Geometry in Single-Cell VAEs for Manifold Interpolationcited by 5
  9. 2025Laplacian Score Sharpening for Mitigating Hallucination in Diffusion Modelscited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 13.

13 papers.

Sort:

By area:

  1. 2026An Empirical Study of Training Pixel-Space Text-to-Image Diffusion Modelscited by 6
  2. 2026V-RAE: Rethinking Video Latent Spaces for Generationcited by 3
  3. 2026Aligning Latent Geometry for Spherical Flow Matching in Image Generationcited by 2
  4. 2026Efficient Image Synthesis with Sphere Latent Encodercited by 1
  5. 2026PoDAR: Power-Disentangled Audio Representation for Generative Modelingcited by 0
  6. 2026Smoothing Dark Areas in Molecular Latent Diffusioncited by 0
  7. 2026Stable Autoregressive Speech Generation with Low-Frame-Rate High-Dimensional Continuous Tokenscited by 0
  8. 2026AffineTok: Semantic Affine Consistency for Diffusion-Friendly Visual Tokenizercited by 0
  9. 2026CPR: Combining global composing, local performing and full-sequence refining in piano rendering with continuous autoregressive modellingcited by 0
  10. 2026A Latent Distribution Perspective on Evaluating and Improving Latent Generative Modelscited by 0
  11. 2026What Makes a Good Medical Image Tokenizer? Rethinking Reconstruction and Generation in Medical Image Tokenizationcited by 0
  12. 2026Scaling and Distilling Text Embeddings for Better Diffusibilitycited by 0
1 more
  1. 2026Image Tokenizers Should Be Trained Between Imagescited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023DiffNAS: Bootstrapping Diffusion Models by Prompting for Better Architectures
  2. 2026Inverting the Generation Process of Denoising Diffusion Implicit Models: Empirical Evaluation and a Novel Method
  3. 2023Adapt and Diffuse: Sample-adaptive Reconstruction via Latent Diffusion Models
  4. 2025Unifying and extending Diffusion Models through PDEs for solving Inverse Problems
  5. 2025Data augmentation using diffusion models to enhance inverse Ising inference
  6. 2024On Statistical Rates and Provably Efficient Criteria of Latent Diffusion Transformers (DiTs)
  7. 2025Disentangling Total-Variance and Signal-to-Noise-Ratio Improves Diffusion Models
  8. 2026Rethinking the Diffusion Model from a Langevin Perspective