Research map / Image generation / Fast diffusion sampling
Accelerating Score-Based Generative Models with Preconditioned Diffusion Sampling
Hengyuan Ma, Li Zhang, Xiatian Zhu, Jianfeng Feng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 preconditioned diffusion sampling to NCSN++ at T=100 | NCSN++ · CIFAR-10 | FID: 29.39 -> 3.26 | Improved |
| add preconditioned diffusion sampling to NCSN++ at T=200 | NCSN++ · CIFAR-10 | FID: 4.35 -> 2.61 | Improved |
| add PDS to NCSNv2 on LSUN church 96x96 | NCSNv2 · LSUN church 96x96 | Clean-FID: 217.9 -> 65.7 | Improved |
| add PDS to NCSNv2 on LSUN tower 128x128 | NCSNv2 · LSUN tower 128x128 | Clean-FID: 67.2 -> 43.8 | Improved |
| add PDS to NCSN++ on LSUN bedroom 256x256 | NCSN++ · LSUN bedroom 256x256 | Clean-FID: 393.7 -> 16.9 | Improved |
| add PDS to NCSN++ on LSUN church 256x256 | NCSN++ · LSUN church 256x256 | Clean-FID: 393.3 -> 15.0 | Improved |
| add PDS to NCSN++ on FFHQ 1024x1024 | NCSN++ · FFHQ 1024x1024 | Clean-FID: 463.2 -> 61.2 | Improved |
| add PDS to NCSN++ for sampling FFHQ | NCSN++ · FFHQ wall-clock, batch of 8 | time (s): 2030 -> 71 | Improved |
| add PDS to NCSN++ for sampling LSUN | NCSN++ · LSUN wall-clock, batch of 8 | time (s): 1173 -> 90 | Improved |
| add solenoidal term S with scale omega from 1 to 1000 (ablation) | PDS on NCSN++ · FFHQ 1024x1024 (66 iterations) | sample quality (visual) | No clear effect |
| add shift-based skew-symmetric solenoidal operators S1 to S6 (omega=1000) (ablation) | PDS on NCSN++ · FFHQ 1024x1024 | sampling quality and acceleration | No clear effect |
| increase lambda too high or lower r too much in the frequency filter (ablation) | PDS on NCSN++ · LSUN bedroom 256x256 (T=166) | sample quality (visual) | Hurt |
Closest papers by meaning
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.92)
- Accelerating Convergence of Score-Based Diffusion Models, Provably (ICML 2024 · similarity 0.91)
- Preconditioned Langevin Dynamics with Score-based Generative Models for Infinite-Dimensional Linear Bayesian Inverse Problems (NeurIPS 2025 · similarity 0.90)
- Simple and Fast Distillation of Diffusion Models (NeurIPS 2024 · similarity 0.90)
- One-Step Diffusion Distillation through Score Implicit Matching (NeurIPS 2024 · similarity 0.89)
- Evaluating the design space of diffusion-based generative models (NeurIPS 2024 · similarity 0.89)
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.
