Research map / 3D vision / Gaussian splatting
SGAD-SLAM: Splatting Gaussians at Adjusted Depth for Better Radiance Fields in RGBD SLAM
Pengchong Hu, Zhizhong Han
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 |
|---|---|---|---|
| let pixel-aligned Gaussians move along rays (adjusted depth) instead of being fixed at depth (ablation) | SGAD-SLAM · TUM-RGBD | PSNR: 37.16 -> 38.60 | Improved |
| pixel-aligned adjusted-depth Gaussians vs freely movable Gaussians in view frustum (ablation) | SGAD-SLAM · TUM-RGBD | PSNR: 37.67 -> 38.60 | Improved |
| pixel-aligned Gaussians vs all Gaussians movable in whole scene (ablation) | SGAD-SLAM · TUM-RGBD | PSNR: 20.90 -> 38.60 | Improved |
| add rendering-based pose initialization before geometric matching (ablation) | SGAD-SLAM · ScanNet++ | ATE RMSE (cm): 6.5 -> 0.59 | Improved |
| use point-to-surface distance instead of point-to-point (ablation) | SGAD-SLAM tracking · TUM-RGBD | ATE RMSE (cm): 17.0 -> 2.0 | Improved |
| use ellipse scale normalization instead of no scale normalization (ablation) | SGAD-SLAM tracking · TUM-RGBD | ATE RMSE (cm): 104.6 -> 2.0 | Improved |
| use ellipse scale normalization instead of flat-plane normalization (ablation) | SGAD-SLAM tracking · TUM-RGBD | ATE RMSE (cm): 27.0 -> 2.0 | Improved |
| learnable depth offset under 10-40% extra depth noise (ablation) | SGAD-SLAM · TUM-RGBD fr3/office | PSNR: 39.86 at 10%-40% noise | No clear effect |
| add noise to 10-40% of depth pixels (ablation) | SGAD-SLAM tracking · TUM-RGBD fr3/office | ATE RMSE (cm): 2.0 -> 2.1-2.2 | No clear effect |
| replace baseline tracking with SGAD geometry-similarity tracking | SGAD-SLAM · Replica | ATE RMSE (cm): 0.16 vs 0.16 (GS-ICP SLAM) | No clear effect |
| SGAD-SLAM vs loop-closure methods | SGAD-SLAM · ScanNet | ATE RMSE (cm): 7.9 vs 7.7 (LoopSplat) | Mixed |
| SGAD-SLAM vs VTGS-SLAM on perceptual quality | SGAD-SLAM · Replica | LPIPS: 0.012 -> 0.021 | Hurt |
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.
- Loopy-SLAM — Loopy-SLAM: Dense Neural SLAM with Loop Closures (CVPR 2024) · better on 2 of 2 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
- Gaussian Splatting SLAM (CVPR 2024 · similarity 0.92)
- GS-SLAM: Dense Visual SLAM with 3D Gaussian Splatting (CVPR 2024 · similarity 0.92)
- VTGaussian-SLAM: RGBD SLAM for Large Scale Scenes with Splatting View-Tied 3D Gaussians (ICML 2025 · similarity 0.91)
- SplaTAM: Splat Track & Map 3D Gaussians for Dense RGB-D SLAM (CVPR 2024 · similarity 0.90)
- DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization (NeurIPS 2024 · similarity 0.89)
- Textured Gaussians for Enhanced 3D Scene Appearance Modeling (CVPR 2025 · 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.
