Browse Teach Curate

Frechet Inception Distance for Image Generation

Frechet Inception Distance (FID) is the standard scalar metric for judging the quality and diversity of generated images. It passes real and generated images through a fixed Inception network, models each set of feature activations as a multivariate Gaussian, and computes the Frechet (Wasserstein-2) distance between the two Gaussians; lower FID means generated images are statistically closer to real ones. It underpins benchmark comparisons across GANs and diffusion models (e.g. on COCO and CUB).

jeremy Video

Frechet Inception Distance in Image Generation Evaluation using Pre-trained CNNs

Frechet Inception Distance (FID) is a statistical metric used in generative modeling to evaluate image quality by computing the distance between feature vector distributions derived from synthetic an…