The Frechét Inception Distance
The Fréchet Inception Distance (FID) was proposed by Martin Heusel et al., in their paper, GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium, as an alternative to the Inception score. One of the drawbacks of the Inception score is that it is a discriminative model; in other words, the Inception network is not robust to adversarial examples and can output high confidence on such adversarial examples. In their paper, the proposed solution to this problem is to compare the statistics of real samples with generated samples.
Similar to the Inception score, FID uses a pretrained Inception network. The FID is a distance measure between the means and covariance matrices of the activations ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access