Research map / Image generation / Fast diffusion sampling
One step further with Monte-Carlo sampler to guide diffusion better
Minsi Ren, Wenhao Deng, Ruiqi Feng, Tailin Wu
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 |
|---|---|---|---|
| replace DSG guidance gradients with ABMS (additional backward Monte-Carlo step) | DSG · ImageNet 256 inpainting | FID: 23.63 -> 19.25 | Improved |
| replace DSG with ABMS | DSG · ImageNet 256 super resolution | FID: 34.28 -> 33.06 | Improved |
| replace DSG with ABMS | DSG · ImageNet 256 Gaussian deblurring | FID: 45.27 -> 41.65 | Improved |
| replace DSG with ABMS | DSG · ImageNet 256 super resolution | PSNR: 23.74 -> 23.80 | No clear effect |
| replace DSG with ABMS at guidance scale 0.1 | DSG · Stylized handwriting, category guidance | style score: 0.534 -> 0.878 | Improved |
| replace DSG with ABMS at guidance scale 0.5 | DSG · Stylized handwriting, category guidance | style score: 0.166 -> 0.756 | Improved |
| replace DSG with ABMS at guidance scale 0.01 | DSG · Stylized handwriting, category guidance | content score: 0.927 -> 0.981 | Improved |
| replace DSG with ABMS (M=3) | DSG · QM9 mu | MAE: 0.7811 -> 0.7274 | Improved |
| replace DSG with ABMS (M=3) | DSG · QM9 Cv | MAE: 0.8447 -> 0.8348 | Improved |
| increase Monte-Carlo samples M from 1 to 3 (ablation) | ABMS · ImageNet inverse problems | Distance vs FID | Improved |
| use ABMS with M=3 instead of DPS/DSG | DSG · Image inverse problem sampling speed | iterations per second: 4.8 iter/s -> 1.7 iter/s | Hurt |
| increase M from 1 to 5 (ablation) | ABMS · Image inverse problem sampling speed | iterations per second: 2.5 iter/s -> 1.3 iter/s | Hurt |
Closest papers by meaning
- On the Trajectory Regularity of ODE-based Diffusion Sampling (ICML 2024 · similarity 0.90)
- Multilevel and Sequential Monte Carlo for Training-Free Diffusion Guidance (NeurIPS 2026 · similarity 0.90)
- NoiseCtrl: A Sampling-Algorithm-Agnostic Conditional Generation Method for Diffusion Models (CVPR 2025 · similarity 0.90)
- Decoupling Training-Free Guided Diffusion by ADMM (CVPR 2025 · similarity 0.89)
- CoDi: Conditional Diffusion Distillation for Higher-Fidelity and Faster Image Generation (CVPR 2024 · similarity 0.89)
- CCS: Controllable and Constrained Sampling with Diffusion Models via Initial Noise Perturbation (NeurIPS 2025 · similarity 0.89)
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.
