Research map / Image generation / Fast diffusion sampling
AC-Sampler: Accelerate and Correct Diffusion Sampling with Metropolis-Hastings Algorithm
Minsang Park, Gyuwon Sim, Hyungho Na, Jiseok Kwak, Sumin Lee, Richard Lee Kim, Donghyeok Shin, Byeonghu Na, Yeongmin Kim, Il-chul Moon
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 AC-Sampler | EDM (Heun) · CIFAR-10 unconditional | FID / NFE: FID 2.01 -> 1.97; NFE 35 -> 26.19 | Improved |
| add AC-Sampler | DPM-v3 · CIFAR-10 unconditional | FID: 12.41 -> 9.88 | Improved |
| add AC-Sampler | DG · CIFAR-10 unconditional | FID: 1.93 -> 1.84 | Improved |
| add AC-Sampler at low NFE | EDM (Heun) · CIFAR-10 (NFE 17) | FID: 3.23 -> 2.38 | Improved |
| add MH accept-reject correction (ablation) | AC-Sampler · CIFAR-10 unconditional | FID | Improved |
| propose-until-accept instead of conventional MH keeping rejected samples (ablation) | AC-Sampler · CIFAR-10 (EDM) | FID: 3.22 -> 1.97 | Improved |
| add AC-Sampler to ScoreSDE KAR2 with 198 NFE | ScoreSDE (DLG) · CelebA-HQ 256 | FID: 8.14 -> 6.60 | Improved |
| extend MALA to joint (x, t) space instead of marginal (ablation) | AC-Sampler · CelebA-HQ 256 (KAR2, NFE 40) | FID: 45.74 -> 8.45 | Improved |
| add AC-Sampler | DiT (DDPM) · ImageNet 256x256 | FID / Precision / Recall: FID 2.35 -> 2.31; Precision 0.829 -> 0.817; Recall 0.576 -> 0.592 | Mixed |
| add AC-Sampler | SD-v1.5 (DDIM) · COCO text-to-image | FID: 24.34 -> 23.16 | Improved |
| add AC-Sampler | SD-v1.5 (DDIM) · GenEval | overall score: 0.4219 -> 0.4453 | Improved |
| lower target timestep tau from 13 to 3 (ablation) | AC-Sampler · CIFAR-10 | FID: 1.97 -> 43.96 | Hurt |
Closest papers by meaning
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.91)
- CCS: Controllable and Constrained Sampling with Diffusion Models via Initial Noise Perturbation (NeurIPS 2025 · similarity 0.91)
- TADA: Improved Diffusion Sampling with Training-free Augmented DynAmics (NeurIPS 2025 · similarity 0.90)
- Simple and Fast Distillation of Diffusion Models (NeurIPS 2024 · similarity 0.90)
- Informed Correctors for Discrete Diffusion Models (NeurIPS 2025 · similarity 0.90)
- Boosting Diffusion Models with Moving Average Sampling in Frequency Domain (CVPR 2024 · similarity 0.90)
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.
