Research map / Image generation / Fast diffusion sampling
DM4CT: Benchmarking Diffusion Models for Computed Tomography Reconstruction
Jiayang Shi, Daniel Pelt, Joost Batenburg
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 Poisson-noise measurements instead of noise-free ones | DDS · Medical CT (config i vs ii) | PSNR (dB): 31.43 -> 20.12 | Hurt |
| use Poisson-noise measurements instead of noise-free ones | DDS · Industrial CT (config i vs ii) | PSNR (dB): 22.87 -> 18.23 | Hurt |
| replace Poisson noise with matched-level Gaussian noise (ablation) | DDS · Noise-model comparison | reconstruction quality | Improved |
| increase DC step size eta moderately (ablation) | DPS · CT reconstruction | PSNR and data fit | Improved |
| increase DC step size eta too far (ablation) | DPS · CT reconstruction | PSNR | Hurt |
| add explicit data-consistency optimization steps to latent diffusion (ReSample vs gradient-only PSLD) (ablation) | Latent diffusion · Medical CT noise-free (40 angles) | reconstruction coherence | Improved |
| add explicit data-consistency optimization steps to latent diffusion under noise (ablation) | ReSample · Medical CT noisy (80 angles) | image quality | Hurt |
| use supervised training (SwinIR) instead of diffusion priors | Diffusion methods · Medical CT config iv | PSNR (dB): 29.70 -> 30.79 | Improved |
| use diffusion priors instead of classical FBP/SIRT | Classical reconstruction · Medical CT config iv | PSNR (dB): 25.86 -> 29.70 | Improved |
| use pixel-space instead of latent-space diffusion | Diffusion methods · CT reconstruction efficiency | time and memory | Improved |
| use diffusion priors under sparse/high-noise measurements | Classical and MBIR methods · Medical and industrial CT | PSNR | Improved |
| apply empirical linear value-range mapping to raw projections | MCG · Medical CT raw projections | PSNR (dB): 19.48 -> 29.12 | Improved |
Closest papers by meaning
- DiffusionBlend: Learning 3D Image Prior through Position-aware Diffusion Score Blending for 3D Computed Tomography Reconstruction (NeurIPS 2024 · similarity 0.90)
- InverseBench: Benchmarking Plug-and-Play Diffusion Priors for Inverse Problems in Physical Sciences (ICLR 2025 · similarity 0.89)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.88)
- Principled Probabilistic Imaging using Diffusion Models as Plug-and-Play Priors (NeurIPS 2024 · similarity 0.88)
- SITCOM: Step-wise Triple-Consistent Diffusion Sampling For Inverse Problems (ICML 2025 · similarity 0.87)
- Learning Image Priors Through Patch-Based Diffusion Models for Solving Inverse Problems (NeurIPS 2024 · 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.
