Research map / 3D vision / Gaussian splatting
Exact-GS: Mathematically Rigorous and Accurate 3D Gaussian Splatting for 3D X-ray Reconstruction
Guangpu Yang, Steffen Kieß, Hanxiang Luo, Xingyu Liu, Sven Simon
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 affine-approximated projection with exact orthogonal pixel-plane splatting | R2-GS · single Gaussian projection | PSNR-2D (dB): 55.56 -> 149.64 | Improved |
| use exact splatting in convergence study | R2-GS · one-Gaussian convergence | PSNR-3D (dB): 88.27 -> 155.26 | Improved |
| use exact splatting on synthetic data, 75 views | R2-GS · synthetic CT 75 views | PSNR-3D (dB): 57.17 -> 59.17 | Improved |
| use exact splatting on synthetic data, 25 views | R2-GS · synthetic CT 25 views | PSNR-3D (dB): 45.54 -> 45.83 | Improved |
| use exact splatting on real FIPS data | R2-GS · real-world FIPS 75 views | PSNR-3D (dB): 44.02 -> 44.11 | No clear effect |
| use exact splatting on real FIPS data | R2-GS · real-world FIPS 25 views | PSNR-3D (dB): 38.10 -> 38.11 | No clear effect |
| use exact splatting with smaller initialization error (ablation) | R2-GS · synthetic CT initialization ablation | PSNR-3D (dB): 57.08 -> 59.13 (Case 4); 43.88 -> 43.95 (Case 1) | Improved |
| use exact splatting versus ray tracing for forward kernel | ray tracing method · CUDA kernel runtime | forward runtime (ms): 16.746 -> 8.695 | Improved |
| use exact splatting instead of R2-GS projection step | R2-GS · reconstruction runtime | total time: 13m 35s -> 28m 26s | Hurt |
| use Gaussian splatting instead of NeRF on real data | NAF · real-world FIPS 75 views | PSNR-3D (dB): 44.14 vs 44.11 | Hurt |
Closest papers by meaning
- R$^2$-Gaussian: Rectifying Radiative Gaussian Splatting for Tomographic Reconstruction (NeurIPS 2024 · similarity 0.92)
- Volumetrically Consistent 3D Gaussian Rasterization (CVPR 2025 · similarity 0.91)
- GS-2DGS: Geometrically Supervised 2DGS for Reflective Object Reconstruction (CVPR 2025 · similarity 0.90)
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.90)
- Sort-free Gaussian Splatting via Weighted Sum Rendering (ICLR 2025 · similarity 0.90)
- DiffGS: Functional Gaussian Splatting Diffusion (NeurIPS 2024 · 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.
