Research map / Image generation / Fast diffusion sampling
HYPE: A Benchmark for Human eYe Perceptual Evaluation of Generative Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add the truncation trick to StyleGAN sampling (ablation) | StyleGAN trained on FFHQ · FFHQ | HYPE∞: 19.0% -> 27.6% | Helped |
| Add the truncation trick to StyleGAN sampling (ablation) | StyleGAN trained on FFHQ · FFHQ | perceptual threshold: 240.7 -> 363.2 | Helped |
| Add the truncation trick to BigGAN sampling (ablation) | BigGAN · ImageNet: Lemon | HYPE∞: 20.2% -> 18.4% | Hurt |
| Add the truncation trick to BigGAN sampling (ablation) | BigGAN · ImageNet: Samoyed | HYPE∞: 19.7% -> 19.9% | No clear effect |
| Add the truncation trick to BigGAN sampling (ablation) | BigGAN · ImageNet: Library | HYPE∞: 22.9% -> 17.4% | Hurt |
| Add the truncation trick to BigGAN sampling (ablation) | BigGAN · ImageNet: French Horn | HYPE∞: 6.9% -> 7.3% | No clear effect |
| Add the truncation trick to BigGAN sampling (ablation) | BigGAN · ImageNet: Baseball Player | HYPE∞: 2.2% -> 1.9% | No clear effect |
| Use StyleGAN instead of earlier GANs | Unconditional image generation · CIFAR-10 | HYPE∞: 14.8% -> 23.3% | Helped |
| Train StyleGAN for longer | StyleGAN · Training progress | HYPE∞: 29.5% -> 45.9% -> 50.3% | Helped |
| Use HYPE∞ instead of the timed staircase evaluation (ablation) | Human evaluation of generated images · Face generation | evaluation time and cost: 6x reduction; 10 minutes instead of 60 minutes | Helped |
| Use HYPE∞ rather than FID to rank models by human realism | Generative models on CelebA and FFHQ · CelebA and FFHQ | rank correlation with human judgment: ρ=-0.029, p=0.96 | Mixed |
| Apply the truncation trick to StyleGAN trained on CelebA (ablation) | StyleGAN · CelebA | HYPE∞ | Helped |
Closest papers by meaning
- PerceptionRubrics: Calibrating Multimodal Evaluation to Human Perception (ICML 2026 · similarity 0.89)
- Evaluating Multiview Object Consistency in Humans and Image Models (NeurIPS 2024 · similarity 0.88)
- LL-Bench: Rethinking Low-Level Vision Evaluation in the Era of Large-Scale Generative Models (NeurIPS 2026 · similarity 0.88)
- Intriguing Properties of Generative Classifiers (ICLR 2024 · similarity 0.87)
- DEEM: Diffusion models serve as the eyes of large language models for image perception (ICLR 2025 · similarity 0.87)
- Multidimensional Observer Model and Perceptual Dimensions of Human Image Quality Assessment (NeurIPS 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.