Here is a sample of fake images evaluated with HYPE. Images with higher HYPE are easier to mistake for real images. All of these images are generated with the state-of-the-art StyleGAN, trained on the FFHQ dataset and sampled with the truncation trick.
Abstract
Generative models often use human evaluations to determine and justify progress. Unfortunately, existing human evaluation methods are ad-hoc: there is currently no standardized, validated evaluation that:
(1) measures perceptual fidelity
(2) is reliable
(3) separates models into clear rank order,
(4) ensures high-quality measurement without intractable cost
In response, we construct Human eYe Perceptual Evaluation (HYPE), a human metric that is:
(1) grounded in psychophysics research in perception
(2) reliable across different sets of randomly sampled outputs from a model
(3) results in separable model performances
(4) efficient in cost and time. We introduce two methods.
The first, HYPE-Time, measures visual perception under adaptive time constraints to determine the minimum length of time (e.g., 250ms) that model output such as a generated face needs to be visible for people to distinguish it as real or fake.
The second, HYPE-Infinity, measures human error rate on fake and real images with no time constraints, maintaining stability and drastically reducing time and cost.
We test HYPE across four state-of-the-art generative adversarial networks (GANs) on unconditional image generation using two datasets, the popular CelebA and the newer higher-resolution FFHQ, and two sampling techniques of model outputs. By simulating HYPE's evaluation multiple times, we demonstrate consistent ranking of different models, identifying StyleGAN with truncation trick sampling (27.6% HYPE-Infinity deception rate, with roughly one quarter of images being misclassified by humans) as superior to StyleGAN without truncation (19.0%) on FFHQ.
Competition Leaderboard
HYPE tests generative models for how realistic their images look to the human eye. How realistic is your generative model? Our competition currently looks at generative models trained on CelebA or FFHQ.
FFHQ (1024x1024)
CelebA (64x64)
Submit your model on CodaLab Worksheets and add to the leaderboards! See our HYPE Submission Guide for instructions.
Sharon Zhou*, Mitchell L. Gordon*, Ranjay Krishna, Austin Narcomey, Li Fei-Fei, Michael S. Bernstein
Stanford University