Research map / Image generation / Fast diffusion sampling
Hierarchical Schedule Optimization for Fast and Robust Diffusion Model Sampling
Aihua Zhu, Rui Su, Qinglin Zhao, Li Feng, Meng Shen, Shibo He
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 NFE-adaptive spacing penalty (SPF) to the fitness (ablation) | HSO (DDIM) · LAION-Aesthetics / DDIM | FID: 165.48 -> 19.76 | Improved |
| add bi-level global search to the local-only baseline (ablation) | DM-NonUni baseline · LAION-Aesthetics 6.5+, NFE=4, DDIM | FID: 71.57 -> 29.22 | Improved |
| add bi-level global search to the local-only baseline (ablation) | DM-NonUni baseline · LAION-Aesthetics 6.5+, NFE=4, UniPC | FID: 18.07 -> 11.44 | Improved |
| replace UniPC-specific objective with MEP (no bi-level search) (ablation) | DM-NonUni baseline · LAION-Aesthetics 6.5+, NFE=4, DDIM | FID: 71.57 -> 37.27 | Improved |
| replace UniPC-specific objective with MEP (no bi-level search) (ablation) | DM-NonUni baseline · LAION-Aesthetics 6.5+, NFE=4, UniPC | FID: 18.07 -> 26.33 | Hurt |
| add MEP objective on top of bi-level search (full HSO) (ablation) | HSO · LAION-Aesthetics 6.5+, NFE=4 | FID: DDIM 29.22 -> 24.80; UniPC 11.44 -> 15.70 | Mixed |
| use HSO schedule instead of the default PixArt-alpha schedule | PixArt-alpha 512px · MS-COCO 30k val, UniPC, NFE=5 | FID: 37.65 -> 18.05 | Improved |
| use HSO instead of DM-NonUni schedule | Stable Diffusion v2.1 with DDIM · LAION-Aesthetics 6.5+, NFE=4 | FID: 68.92 -> 24.77 | Improved |
| use HSO instead of DM-NonUni schedule | Stable Diffusion v2.1 with UniPC · LAION-Aesthetics 6.5+, NFE=5 | FID: 13.91 -> 11.94 | Improved |
| use a fixed rule-based schedule independent of NFE | schedule optimization · Stable Diffusion v2.1 | optimal initialization parameter rho | Hurt |
Closest papers by meaning
- Align Your Steps: Optimizing Sampling Schedules in Diffusion Models (ICML 2024 · similarity 0.94)
- Adaptive Non-Uniform Timestep Sampling for Accelerating Diffusion Model Training (CVPR 2025 · similarity 0.91)
- Score-Optimal Diffusion Schedules (NeurIPS 2024 · similarity 0.90)
- A Unified Sampling Framework for Solver Searching of Diffusion Probabilistic Models (ICLR 2024 · similarity 0.90)
- SADA: Stability-guided Adaptive Diffusion Acceleration (ICML 2025 · similarity 0.90)
- Inference-Time Alignment of Diffusion Models with Direct Noise Optimization (ICML 2025 · 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.
