Pith. sign in

REVIEW 7 cited by

Rethinking FID: Towards a Better Evaluation Metric for Image Generation

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2401.09603 v2 pith:UEFV3TYN submitted 2023-11-30 cs.CV

classification cs.CV
keywords distancegeneratedimageimagesmetricmodelssampletext-to-image
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

As with many machine learning problems, the progress of image generation methods hinges on good evaluation metrics. One of the most popular is the Frechet Inception Distance (FID). FID estimates the distance between a distribution of Inception-v3 features of real images, and those of images generated by the algorithm. We highlight important drawbacks of FID: Inception's poor representation of the rich and varied content generated by modern text-to-image models, incorrect normality assumptions, and poor sample complexity. We call for a reevaluation of FID's use as the primary quality metric for generated images. We empirically demonstrate that FID contradicts human raters, it does not reflect gradual improvement of iterative text-to-image models, it does not capture distortion levels, and that it produces inconsistent results when varying the sample size. We also propose an alternative new metric, CMMD, based on richer CLIP embeddings and the maximum mean discrepancy distance with the Gaussian RBF kernel. It is an unbiased estimator that does not make any assumptions on the probability distribution of the embeddings and is sample efficient. Through extensive experiments and analysis, we demonstrate that FID-based evaluations of text-to-image models may be unreliable, and that CMMD offers a more robust and reliable assessment of image quality.

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Drop-In Perceptual Optimization for 3D Gaussian Splatting

    cs.CV 2026-03 accept novelty 6.5 of 10

    WD-R, a lightly regularized Wasserstein Distortion loss, is preferred by humans 2.3× over the original 3DGS loss and 1.5× over Perceptual-GS while matching or reducing splat count and generalizing to other 3DGS framew...

  2. MorphUNet: Alpha-Controlled Biometric Transport for Diffusion-Based Face Morphing Attacks

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Parent-separated dual cross-attention inside a diffusion U-Net yields stronger, more balanced two-identity face morphs than StableMorph, MIPGAN-II, and MorDIFF on FEI and FRLL.

  3. ROMS-IMLE: A Minimalist Approach to Competitive Single-Step Generative Modelling

    cs.LG 2026-07 conditional novelty 6.0 of 10

    A single-step IMLE generator with per-stage supervision and a robust loss reports FID 2.56 on ImageNet-256 by filtering ~5% of samples at test time.

  4. LeDiFlow: Learned Distribution-guided Flow Matching to Accelerate Image Generation

    cs.CV 2025-05 conditional novelty 6.0 of 10

    A learned distribution prior, predicted by a VAE-style decoder, lets flow-matching image models generate in fewer ODE steps than with a Gaussian prior.

  5. Sealing The Backdoor: Unlearning Adversarial Text Triggers In Diffusion Models Using Knowledge Distillation

    cs.CV 2025-08 unverdicted novelty 5.0 of 10

    SKD-CAG erases adversarial text triggers from diffusion models by distilling the model's own clean outputs through cross-attention guidance, claiming 100% and 93% removal for pixel and style backdoors.

  6. Evaluation of Human Visual Privacy Protection: A Three-Dimensional Framework and Benchmark Dataset

    cs.CV 2025-07 conditional novelty 5.0 of 10

    The HR-VISPR dataset plus a privacy, utility, and practicality framework ranks 11 visual anonymization methods, with privacy scores based on classifier accuracy loss rather than measured human perception.

  7. Why Settle for Mid: A Probabilistic Viewpoint to Spatial Relationship Alignment in Text-to-image Models

    cs.CV 2025-06 conditional novelty 5.0 of 10

    A probabilistic overlap measure for object positions yields a human-aligned spatial relationship metric and a training-free generation guidance method for text-to-image models.

Pith tools