Research map / 3D vision / Gaussian splatting
Revisiting Pose Sensitivity in Splat-based Computed Tomography under Sparse-view Reconstruction
Kiseok Choi, Hyeongjun Cho, Inchul Kim, Min H. Kim
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 joint pose self-calibration to splat CT | R2-Gaussian · Broccoli, noisy dataset | PSNR (dB): 22.21 -> 30.20 | Improved |
| add joint pose self-calibration to splat CT | R2-Gaussian · Broccoli, noise-free dataset | PSNR (dB): 36.54 -> 34.70 | Hurt |
| add joint pose self-calibration to splat CT | R2-Gaussian · Beetle, no pose noise | PSNR (dB): 43.18 -> 43.22 | No clear effect |
| add joint pose self-calibration at moderate pose noise (ablation) | R2-Gaussian · Beetle, pose noise rot 0.02 trans 1.0 | PSNR (dB): 33.90 -> 40.65 | Improved |
| add joint pose self-calibration at the strongest pose noise (ablation) | R2-Gaussian · Beetle, pose noise rot 0.10 trans 5.0 | PSNR (dB): 30.80 -> 32.32 | Improved |
| add joint pose self-calibration at the strongest pose noise (ablation) | R2-Gaussian · Head, pose noise rot 0.10 trans 5.0 | PSNR (dB): 26.41 -> 28.17 | Improved |
| add joint pose self-calibration with 75 views | R2-Gaussian · Synthetic CT with pose noise, 75 views | PSNR (dB): 28.50 -> 33.42 | Improved |
| add joint pose self-calibration with 25 views | R2-Gaussian · Synthetic CT with pose noise, 25 views | PSNR (dB): 26.44 -> 29.02 | Improved |
| reduce input views from 75 to 25 | Ours (self-calibrating splat CT) · Synthetic CT with pose noise, 25 views | PSNR (dB): 33.42 -> 29.02 | Hurt |
| self-calibration instead of NeAT for pose estimation | Ours (self-calibrating splat CT) · Synthetic CT, 15 scenes mean (rot 0.03, trans 1.0) | translation RMSE (AU): 1.437 -> 0.726 | Improved |
| self-calibration instead of Thies et al. for pose estimation | Ours (self-calibrating splat CT) · Synthetic CT, 15 scenes mean (rot 0.03, trans 1.0) | orientation RMSE (deg): 4.076 -> 0.627 | Improved |
| remove TV regularization and add pose calibration | Ours (self-calibrating splat CT) · Synthetic CT, mean computation time | computation time (min): 23.19 -> 20.89 | 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.
- NeAT — NeAT: Learning Neural Implicit Surfaces With Arbitrary Topologies From Multi-View Images (CVPR 2023) · better on 3 of 3 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
- R$^2$-Gaussian: Rectifying Radiative Gaussian Splatting for Tomographic Reconstruction (NeurIPS 2024 · similarity 0.89)
- DDGS-CT: Direction-Disentangled Gaussian Splatting for Realistic Volume Rendering (NeurIPS 2024 · similarity 0.86)
- SelfSplat: Pose-Free and 3D Prior-Free Generalizable 3D Gaussian Splatting (CVPR 2025 · similarity 0.86)
- C^2RV: Cross-Regional and Cross-View Learning for Sparse-View CBCT Reconstruction (CVPR 2024 · similarity 0.86)
- FatesGS: Fast and Accurate Sparse-View Surface Reconstruction Using Gaussian Splatting with Depth-Feature Consistency (AAAI 2025 · similarity 0.86)
- Learning Efficient Fuse-and-Refine for Feed-Forward 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.85)
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.
