Research map / Probabilistic and scientific modeling / Variational and generative sampling
Poisson Flow 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 |
|---|---|---|---|
| Use the Poisson-flow backward ODE instead of VP-ODE | DDPM++ on CIFAR-10 · CIFAR-10 | FID: 2.97 -> 2.48 | Helped |
| Use the Poisson-flow backward ODE instead of VP-ODE | DDPM++ deep on CIFAR-10 · CIFAR-10 | FID: 2.86 -> 2.35 | Helped |
| Use the Poisson-flow backward ODE instead of VP-SDE | DDPM++ on CIFAR-10 · CIFAR-10 | NFE: 1000 -> 104 | Helped |
| Use the Poisson-flow backward ODE instead of VP-SDE | DDPM++ deep on CIFAR-10 · CIFAR-10 | NFE: 1000 -> 110 | Helped |
| Anchor the backward ODE with the added dimension z | PFGM on CIFAR-10 · CIFAR-10 | FID | Helped |
| Use the exponentially decaying z parameterization (ablation) | PFGM sampling · Image generation | sampling speed: 2× faster | Helped |
| Reduce Euler NFE from 100 to 10 | PFGM on CIFAR-10 · CIFAR-10 | FID | Hurt |
| Use PFGM rather than VP-ODE with Euler steps | CIFAR-10 sampling · CIFAR-10 | FID | Helped |
| Use PFGM rather than DDIM with Euler steps | CIFAR-10 sampling · CIFAR-10 | FID | Mixed |
| Use PFGM instead of VE/VP-ODE without a corrector | NCSNv2 · CIFAR-10 | FID: greater than 90 | Helped |
Closest papers by meaning
- Beckmann Transport Models: From Autonomous Flows to One-Step Maps (NeurIPS 2026 · similarity 0.86)
- Expanding Flow Maps (NeurIPS 2026 · similarity 0.84)
- One-Shot Generative Flows: Existence and Obstructions (NeurIPS 2026 · similarity 0.84)
- Spherical Flows for Sampling Categorical Data (NeurIPS 2026 · similarity 0.84)
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · similarity 0.84)
- Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization (ICLR 2024 · similarity 0.83)
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.