Research map / Image generation / Fast diffusion sampling
Backward SDE–Based Diffusion for Physics-Constrained Generation
Zihao 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 |
|---|---|---|---|
| use hard terminal constraint instead of soft terminal penalty (ablation) | BSDE solver · 128x128 CT reconstruction | PSNR: 33.10 -> 35.36 | Improved |
| use hard terminal constraint instead of soft terminal penalty (ablation) | BSDE solver · 128x128 CT reconstruction | Relative measurement residual: 0.00203 -> 0.00128 | Improved |
| add BSDE correction network with terminal constraint (ablation) | BSDE solver · 128x128 CT reconstruction | PSNR: 17.38 -> 35.36 | Improved |
| add BSDE correction network with terminal constraint (ablation) | BSDE solver · 128x128 CT reconstruction | SSIM: 0.4188 -> 0.9460 | Improved |
| add BSDE correction network with terminal constraint (ablation) | BSDE solver · 128x128 CT reconstruction | Relative measurement residual: 0.2008 -> 0.00128 | Improved |
| set terminal timepoint tau 0.15 instead of 0.05 (ablation) | BSDE solver · 128x128 CT reconstruction | PSNR: 35.52 -> 35.21 | No clear effect |
| raise terminal timepoint tau from 0.15 to 0.35 (ablation) | BSDE solver · 128x128 CT reconstruction | PSNR: 35.21 -> 31.34 | Hurt |
| raise terminal timepoint tau from 0.35 to 0.45 (ablation) | BSDE solver · 128x128 CT reconstruction | PSNR: 31.34 -> 23.48 | Hurt |
| use BSDE terminal conditioning instead of TVDIP | SVCT reconstruction · SVCT | MAE: 4.56 -> 3.57 | Improved |
| use BSDE terminal conditioning instead of FBP | SVCT reconstruction · SVCT | SSIM: 0.366 -> 0.868 | Improved |
| use BSDE instead of DDS (CT-domain prior) on PSNR | SVCT reconstruction · SVCT | PSNR: 31.10 -> 30.99 | Hurt |
| use BSDE instead of SDE-diffusion perturbation editing for Cepheid | star lightcurve generation · Star lightcurves (Cepheid) | Average cosine similarity: 0.4338 -> 0.7095 | Improved |
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.
- SiTCoM — SITCOM: Step-wise Triple-Consistent Diffusion Sampling For Inverse Problems (ICML 2025) · better on 2 of 2 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
- Learning Image Priors Through Patch-Based Diffusion Models for Solving Inverse Problems (NeurIPS 2024 · similarity 0.90)
- Stability and Generalizability in SDE Diffusion Models with Measure-Preserving Dynamics (NeurIPS 2024 · similarity 0.88)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.88)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.87)
- Guided Diffusion Sampling on Function Spaces with Applications to PDEs (NeurIPS 2025 · similarity 0.87)
- InverseBench: Benchmarking Plug-and-Play Diffusion Priors for Inverse Problems in Physical Sciences (ICLR 2025 · 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.
