HYPE: A Benchmark for Human eYe Perceptual Evaluation of Generative Models

Authors: Sharon Zhou, Mitchell Gordon, Ranjay Krishna, Austin Narcomey, Li F. Fei-Fei, Michael Bernstein

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test HYPE across six state-of-the-art generative adversarial networks and two sampling techniques on conditional and unconditional image generation using four datasets: Celeb A, FFHQ, CIFAR-10, and Image Net. We find that HYPE can track the relative improvements between models, and we confirm via bootstrap sampling that these measurements are consistent and replicable.
Researcher Affiliation Academia Sharon Zhou , Mitchell L. Gordon , Ranjay Krishna , Austin Narcomey , Li Fei-Fei , Michael S. Bernstein Stanford University {sharonz, mgord, ranjaykrishna, aon2, feifeili, msb}@cs.stanford.edu
Pseudocode No The paper describes methods in text and uses diagrams, but does not include explicit pseudocode or algorithm blocks.
Open Source Code No We deploy HYPE at https://hype.stanford.edu, where researchers can upload a model and retrieve a HYPE score. This describes a deployed service, not necessarily open-source code for the methodology.
Open Datasets Yes Datasets. We evaluate on four datasets. (1) Celeb A-64 [37]... (2) FFHQ-1024 [26]... (3) CIFAR-10 [31]... (4) Image Net-5... from the Image Net dataset [13]
Dataset Splits No The paper describes the datasets used to train the GANs and the sampling strategy for the human evaluation (50 real and 50 fake images), but does not specify explicit train/validation/test splits for these datasets for model training or for the evaluation itself.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU types) used to run the experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers needed to replicate the experiments.
Experiment Setup Yes Image exposures are in the range [100ms, 1000ms], derived from the perception literature [17]. All blocks begin at 500ms and last for 150 images (50% generated, 50% real)... Exposure times are raised at 10ms increments and reduced at 30ms decrements, following the 3-up/1-down adaptive staircase approach...