Research map / 3D vision / 3D generation
DiffNR: Diffusion-Enhanced Neural Representation Optimization for Sparse-View 3D Tomographic Reconstruction
Shiyan Su, Ruyi Zha, Danli Shi, Hongdong Li, Xuelian Cheng
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 DiffNR to R2-Gaussian | R2-Gaussian · LUNA16 (36-view) | PSNR: 24.11 -> 28.82 | Improved |
| add DiffNR to NAF | NAF · ToothFairy and LUNA16 (all view counts) | PSNR: +2.19 dB average | Improved |
| add DiffNR to R2-Gaussian with SliceFixer trained on another dataset | R2-Gaussian · OOD real-projection dataset (12-view) | PSNR: 29.71 -> 31.04 | Improved |
| add DiffNR to R2-Gaussian for lung segmentation | R2-Gaussian · LUNA16 (12-view) | Dice: 59.73 -> 84.93 | Improved |
| use DiffNR on NAF versus diffusion iterative DiffusionMBIR at 36-view on LUNA16 | NAF · LUNA16 (36-view) | PSNR: NAF+DiffNR 26.27 vs DiffusionMBIR 29.35; time 8m41s vs 11h15m | Mixed |
| finetune SliceFixer at 512 resolution with up-downsampling (ablation) | SliceFixer · LUNA16 (36-view) | PSNR: 27.65 -> 27.91 | Improved |
| add SSIM loss to SliceFixer finetuning (ablation) | SliceFixer · LUNA16 (36-view) | PSNR: 27.91 -> 28.21 | Improved |
| add biplanar projection conditioning (ablation) | SliceFixer · LUNA16 (36-view) | PSNR: 28.21 -> 28.82 | Improved |
| augment with Difix3D+ novel-view projections instead of slice supervision (ablation) | R2-Gaussian · LUNA16 (36-view) | PSNR: 24.11 -> 23.23 | Hurt |
| integrate SliceFixer into optimization rather than post-processing only (ablation) | R2-Gaussian · LUNA16 (36-view) | PSNR: 26.70 -> 28.82 | Improved |
| use 3D SSIM loss instead of voxel-wise L1 on pseudo-reference (ablation) | DiffNR · LUNA16 (36-view) | PSNR: 26.42 -> 28.82 | Improved |
| supervise every 5 steps instead of 10 (ablation) | DiffNR · LUNA16 (36-view) | PSNR: 28.82 -> 28.76; time 12m56s -> 27m35s | No clear effect |
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)
- DIFIX3D+: Improving 3D Reconstructions with Single-Step Diffusion Models (CVPR 2025 · similarity 0.89)
- Sparse3D: Distilling Multiview-Consistent Diffusion for Object Reconstruction from Sparse Views (AAAI 2024 · similarity 0.87)
- Learning Image Priors Through Patch-Based Diffusion Models for Solving Inverse Problems (NeurIPS 2024 · similarity 0.87)
- Diffusion State-Guided Projected Gradient for Inverse Problems (ICLR 2025 · similarity 0.87)
- QN-Mixer: A Quasi-Newton MLP-Mixer Model for Sparse-View CT Reconstruction (CVPR 2024 · similarity 0.86)
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.
