Fréchet inception distance

Fréchet inception distance (FID) compares image distributions

The FID metric evaluates the quality of generated images by comparing their statistical distribution to that of real images. It uses mean and covariance statistics of multiple images to assess the similarity between the generated and real image sets.

Example

If a generative model produces images with a mean and covariance similar to those of a reference set of real images, the FID score will be low, indicating high quality.

Remember this

Understanding FID helps in improving generative models by providing a quantitative measure of image quality.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

IS (Inception Score) measures: diversity and quality of generated images

The Inception Score (IS) measures diversity and quality of generated images

Effect size

Cohen's D benchmarks: 0.2 = small, 0.5 = medium, 0.8 = large effect

cosine similarity works better than Euclidean distance in high dimensions

Cosine similarity measures orientation, not magnitude, making it more robust to irrelevant dimensions in high-dimensional spaces

to normalize features: when features have different scales and you use distance-based methods

Why do some things need to be adjusted to compare fairly?

Short-time Fourier transform

STFT divides a signal into shorter segments for analysis

Euclidean geometry

Euclidean distance measures absolute position in space

Swipe through 100 ML concepts daily

Open Pocket Polymath