Research map / Image generation / Fast diffusion sampling
OMS-DPM: Optimizing the Model Schedule for Diffusion Probabilistic Models
Enshu Liu, Xuefei Ning, Zinan Lin, Huazhong Yang, Yu Wang
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 |
|---|---|---|---|
| Adding searched mixed model schedules (OMS-DPM) instead of a single model | DPM-Solver on CIFAR-10 · CIFAR-10 FID across budgets | FID: e.g., 3.25 vs 3.56 at 7.0s budget; 6.08 vs 8.73 at 0.7s | Improved |
| Adding mixed model schedules (OMS-DPM) instead of a single model | DDIM across four datasets · CIFAR-10, CelebA, ImageNet-64, LSUN-Church | FID: gains range from about 0.01 to several points across budgets | Improved |
| Searched schedule vs. random schedule sampling | DPM-Solver across datasets · all four datasets | FID: e.g., 3.25 vs 9.33 on CIFAR-10 at 7.0s | Improved |
| Predictor-based search vs. best schedule in predictor training set | DPM-Solver and DDIM across datasets · all four datasets | FID: e.g., 3.25 vs 3.33 on CIFAR-10 at 7.0s; 13.94 vs 32.57 on LSUN-Church at 4s | Improved |
| Mixing Stable Diffusion checkpoints (4 models) with OMS-DPM search | Stable Diffusion DPM-Solver · MS-COCO 256x256 FID | FID: 10.57 vs 11.81 at 24 NFE; 11.34 at 12 NFE beats single model at 24 NFE (11.81) | Improved |
| Larger model zoo size (6 vs 2) in search (ablation) | DPM-Solver on CIFAR-10 · CIFAR-10 at 0.7s budget | FID: 6.08 (N=6) vs 6.05 (N=2) | No clear effect |
| Reducing predictor training data from 3662 to 915 schedules (ablation) | DPM-Solver predictor on CIFAR-10 · CIFAR-10 at 1.4s budget | FID: 3.97 (915) vs 3.48 (3662) | Hurt |
Later papers that compared with this method
2 later accepted papers report OMS-DPM as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Towards More Accurate Diffusion Model Acceleration with A Timestep Tuner (CVPR 2024 · better on 1 of 1 shared result)
- Mixture of Efficient Diffusion Experts Through Automatic Interval and Sub-Network Selection (ECCV 2024 · better on 0 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Align Your Steps: Optimizing Sampling Schedules in Diffusion Models (ICML 2024 · similarity 0.89)
- Learning to Discretize Denoising Diffusion ODEs (ICLR 2025 · similarity 0.89)
- On Inference Stability for Diffusion Models (AAAI 2024 · similarity 0.88)
- Denoising Diffusion Step-aware Models (ICLR 2024 · similarity 0.88)
- Early-Bird Diffusion: Investigating and Leveraging Timestep-Aware Early-Bird Tickets in Diffusion Models for Efficient Training (CVPR 2025 · similarity 0.88)
- Accelerating Diffusion Sampling with Optimized Time Steps (CVPR 2024 · 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.
