Research map / Point clouds / Shape and surface completion
3D Field of Junctions: A Noise-Robust, Training-Free Structural Prior for Volumetric Inverse Problems
Namhoon Kim, NARGES MOEINI, Justin Romberg, Sara Fridovich-Keil
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 |
|---|---|---|---|
| increase patch size from 4 to 12 (stride 2) (ablation) | 3D FoJ · Engine CT P50 | 3D PSNR: 11.33 -> 17.70 | Improved |
| increase patch size from 4 to 12 (stride 2) (ablation) | 3D FoJ · Engine CT P50 | runtime: 7.2 min -> 24.7 min | Hurt |
| use stride 1 instead of stride 2 (patch size 8) (ablation) | 3D FoJ · Engine CT P50 | MS-SSIM: 0.796 -> 0.722 | Hurt |
| use stride 1 instead of stride 2 (patch size 8) (ablation) | 3D FoJ · Engine CT P50 | runtime: 13.1 min -> 47.3 min | Hurt |
| increase batch size from 1 to 6 (ablation) | 3D FoJ · Engine CT P50 | runtime: 37.6 min -> 17.2 min | Improved |
| increase batch size from 1 to 6 (ablation) | 3D FoJ · Engine CT P50 | 3D PSNR: 17.17 -> 17.17 | No clear effect |
| use 3 regions per junction instead of 2 (ablation) | 3D FoJ · Engine CT P50 | 3D PSNR: 17.03 -> 17.17 | Mixed |
| increase regions per junction from 3 to 8 (ablation) | 3D FoJ · Engine CT P50 | 3D PSNR: 17.17 -> 16.12 | Hurt |
| use 3D FoJ prior instead of 3D-TV | low-dose CT reconstruction · Low-dose CT P50 | MS-SSIM: 0.703 -> 0.754 | Improved |
| use 3D FoJ instead of R2-Gaussian | low-dose CT reconstruction · Low-dose CT P1000 | 3D PSNR: 15.87 -> 16.72 | Hurt |
| use 3D FoJ instead of PointCleanNet | point cloud denoising · Outlier noise 60% | Chamfer-L2: 12.98 -> 0.22 | Improved |
| use 3D FoJ instead of PointCVaR | point cloud denoising · Spread noise 100k | Chamfer-L2: 0.44 -> 0.17 | Improved |
Closest papers by meaning
- V2V3D: View-to-View Denoised 3D Reconstruction for Light Field Microscopy (CVPR 2025 · similarity 0.86)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.84)
- Learning Cocoercive Conservative Denoisers via Helmholtz Decomposition for Poisson Imaging Inverse Problems (NeurIPS 2025 · similarity 0.84)
- Neural Field Thermal Tomography: A Differentiable Physics Framework for Non-Destructive Evaluation (NeurIPS 2026 · similarity 0.83)
- Virtual Scanning: Unsupervised Non-line-of-sight Imaging from Irregularly Undersampled Transients (NeurIPS 2024 · similarity 0.83)
- Provable Posterior Sampling with Denoising Oracles via Tilted Transport (NeurIPS 2024 · similarity 0.83)
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.
