Research map / Image generation / Fast diffusion sampling
Preconditioned Score-Based Generative Models
Hengyuan Ma, Xiatian Zhu, Jianfeng Feng, Li Zhang
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 |
|---|---|---|---|
| replace vanilla sampler with PDS preconditioned sampling | NCSN++ · CIFAR-10 T=50 | FID: 306.91 -> 4.90 | Improved |
| replace vanilla sampler with PDS | NCSN++ · CIFAR-10 T=1000 | FID: 2.23 -> 1.99 | Improved |
| replace vanilla sampler with PDS | NCSN++ · CelebA 64x64 T=100 | FID: 433.98 -> 3.49 | Improved |
| replace vanilla sampler with PDS | NCSN++ · CelebA 64x64 T=1000 | FID: 3.33 -> 3.25 | No clear effect |
| replace vanilla sampler with PDS | NCSN++ · ImageNet 64x64 T=1000 | FID: 18.93 -> 15.63 | Improved |
| replace vanilla sampler with PDS | NCSN++ · COCO T=100 | FID: 464.48 -> 19.57 | Improved |
| add PDS on top of NCSN++ at 50 iterations versus PNDM (cosine) | NCSN++ w/ PDS · CIFAR-10 T=50 | FID: 4.90 vs 3.68 | Hurt |
| use PDS instead of subspace diffusion (SDGM) | NCSN++ · CIFAR-10 T=100 | FID: 64.05 -> 2.90 | Improved |
| perturb alpha by -10% (ablation) | PDS · CIFAR-10 T=50 | FID: 4.90 -> 7.34 | Hurt |
| perturb alpha by +/-10% (ablation) | PDS · CIFAR-10 T=1000 | FID: 1.99 -> 2.00 and 2.01 | No clear effect |
| add PDS per-iteration overhead | NCSN++ · FFHQ 1024x1024 | time per iteration: 0.96 -> 1.03 seconds | Hurt |
| fewer iterations enabled by PDS on FFHQ | NCSN++ w/ PDS · FFHQ 1024x1024 | time for 8 samples: 1920 -> 68 seconds | Improved |
Closest papers by meaning
- Accelerating Convergence of Score-Based Diffusion Models, Provably (ICML 2024 · similarity 0.90)
- Evaluating the design space of diffusion-based generative models (NeurIPS 2024 · similarity 0.90)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.90)
- Preconditioned Langevin Dynamics with Score-based Generative Models for Infinite-Dimensional Linear Bayesian Inverse Problems (NeurIPS 2025 · similarity 0.89)
- One-Step Diffusion Distillation through Score Implicit Matching (NeurIPS 2024 · similarity 0.89)
- Simple and Fast Distillation of Diffusion 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.
