Research map / Image generation / Fast diffusion sampling
DPM-Solver-v3: Improved Diffusion ODE Solver with Empirical Model Statistics
Kaiwen Zheng, Cheng Lu, Jianfei Chen, Jun Zhu
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 |
|---|---|---|---|
| Use DPM-Solver-v3 instead of the strongest listed baseline, UniPC | CIFAR10 with ScoreSDE · CIFAR10 (ScoreSDE), 5 NFE | FID: 23.71 -> 12.76 | Improved |
| Use DPM-Solver-v3 instead of the strongest listed baseline, UniPC | CIFAR10 with ScoreSDE · CIFAR10 (ScoreSDE), 10 NFE | FID: 3.93 -> 3.40 | Improved |
| Use DPM-Solver-v3 instead of the strongest listed baseline, UniPC | CIFAR10 with ScoreSDE · CIFAR10 (ScoreSDE), 20 NFE | FID: 2.73 -> 2.71 | Improved |
| Use DPM-Solver-v3 instead of the strongest listed baseline, DPM-Solver++ | CIFAR10 with EDM · CIFAR10 (EDM), 5 NFE | FID: 24.54 -> 12.21 | Improved |
| Use DPM-Solver-v3 instead of the strongest listed baseline, UniPC | CIFAR10 with EDM · CIFAR10 (EDM), 10 NFE | FID: 2.85 -> 2.51 | Improved |
| Use DPM-Solver-v3 instead of the strongest listed baseline, UniPC | CIFAR10 with EDM · CIFAR10 (EDM), 20 NFE | FID: 2.01 -> 2.02 | No clear effect |
| Add empirical model statistics to choose the ODE formulation and approximated parameterization (ablation) | DPM-Solver-v3 · CIFAR10 | FID | Improved |
| Use a pseudo-order solver at very small NFE rather than estimating derivatives from all previous values (ablation) | DPM-Solver-v3 · Few-step sampling | sample quality | Improved |
| Use a half-corrector only in the time region t≤0.5 instead of a full-corrector (ablation) | DPM-Solver-v3 with large guidance scale · Stable Diffusion, guidance scale 7.5 | MSE | Improved |
| Add a corrector at every step (ablation) | DPM-Solver-v3 with large guidance scale · Stable Diffusion, guidance scale 7.5 | sample quality | Hurt |
| Use DPM-Solver-v3 instead of DPM-Solver++ | Stable Diffusion at guidance scale 7.5 · Stable Diffusion, guidance scale 7.5 | MSE | Improved |
| Use DPM-Solver-v3 instead of UniPC | Stable Diffusion at guidance scale 7.5 · Stable Diffusion, NFE<10 | MSE | Improved |
Later papers that compared with this method
2 later accepted papers report DPM-Solver-v3 as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Diffusion Sampling Correction via Approximately 10 Parameters (ICML 2025 · better on 1 of 1 shared result)
- EVODiff: Entropy-aware Variance Optimized Diffusion Inference (NeurIPS 2025 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- UniPC — UniPC: A Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models (NeurIPS 2023) · better on 5 of 6 shared results
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
- Accelerating Diffusion Sampling with Optimized Time Steps (CVPR 2024 · similarity 0.93)
- PFDiff: Training-Free Acceleration of Diffusion Models Combining Past and Future Scores (ICLR 2025 · similarity 0.93)
- Learning to Discretize Denoising Diffusion ODEs (ICLR 2025 · similarity 0.93)
- A Unified Sampling Framework for Solver Searching of Diffusion Probabilistic Models (ICLR 2024 · similarity 0.92)
- Enhanced Diffusion Sampling via Extrapolation with Multiple ODE Solutions (ICLR 2025 · similarity 0.92)
- On Exact Inversion of DPM-Solvers (CVPR 2024 · similarity 0.91)
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.
