Research map / Image generation / Fast diffusion sampling
Diffusion Sampling Path Tells More: An Efficient Plug-and-Play Strategy for Sample Filtering
Sixian Wang, Zhiwei Tang, Tsung-Hui Chang
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 |
|---|---|---|---|
| filter to top 10% by ASD | EDM2-S · ImageNet | PickScore: 20.38 -> 20.61 | Improved |
| filter to top 10% by ASD | EDM2-S · ImageNet | HPSv2: 26.13 -> 26.57 | Improved |
| cut off ASD accumulation at tau=10 instead of full trajectory (ablation) | CFG-Rejection · ImageNet | HPSv2: 26.13 -> 26.31 | Improved |
| extend ASD window from tau=10 to tau=20 (ablation) | CFG-Rejection · ImageNet | HPSv2: 26.31 -> 26.55 | Improved |
| filter by ASD (4 from 20) vs random selection | SD1.5 guidance 5 · GenEval | overall score: 0.4152 -> 0.4508 | Improved |
| filter by ASD (4 from 20) vs random selection | SD1.5 guidance 9 · GenEval | overall score: 0.4322 -> 0.4642 | Improved |
| use early cutoff tau=10 vs tau=20 (ablation) | CFG-Rejection SD1.5 · GenEval | overall score: 0.4594 -> 0.4445 | Hurt |
| filter by ASD (4 from 20) vs random selection | SD1.5 guidance 5 · DPG-Bench | overall score: 62.45 -> 64.48 | Improved |
| early cutoff tau=10 vs random (ablation) | CFG-Rejection SD1.5 · DPG-Bench | overall score: 62.45 -> 62.8 | No clear effect |
| increase candidate pool from 20 to 50 (ablation) | CFG-Rejection SD1.5 · DPG-Bench | overall score: 64.48 -> 65.15 | Improved |
| use CFG-Rejection vs Best-of-N under 5 s budget | CFG-Rejection · ImageNet | PickScore: 20.92 -> 21.04 | Improved |
| use CFG-Rejection vs Best-of-N under 15 s budget | CFG-Rejection · ImageNet | PickScore: 21.23 -> 21.12 | Hurt |
Closest papers by meaning
- Rectified Diffusion Guidance for Conditional Generation (CVPR 2025 · similarity 0.92)
- CCS: Controllable and Constrained Sampling with Diffusion Models via Initial Noise Perturbation (NeurIPS 2025 · similarity 0.91)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.90)
- CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models (ICLR 2025 · similarity 0.90)
- Your Diffusion Model is Secretly a Noise Classifier and Benefits from Contrastive Training (NeurIPS 2024 · similarity 0.90)
- Diffusion Rejection Sampling (ICML 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.
