Research map / 3D vision / Face and shape
Gradient-SDF: A Semi-Implicit Surface Representation for 3D Reconstruction
Christiane Sommer, Lu Sang, David Schubert, Daniel Cremers
Check 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 |
|---|---|---|---|
| Store per-voxel SDF gradients and use Taylor expansion rather than trilinear interpolation | Gradient-SDF depth tracker · TUM RGB-D sequences | pose estimation runtime: 100--120 ms -> 30--40 ms | Improved |
| Store per-voxel gradients alongside SDF values | Gradient-SDF · Synthetic spheres, voxels within 10v_s of surface | mean angular deviation (degrees): 9.49° -> 5.07° | Improved |
| Use Gradient-SDF hash-map tracking rather than SDF tracking with a hash map | Depth tracking · TUM RGB-D | pose estimation runtime: 100--120 ms -> 30--40 ms | Improved |
| Use Gradient-SDF hash-map tracking rather than dense voxel storage | Depth-based reconstruction · 00577_sofa | memory: more than 20× less memory | Improved |
| Apply implicit photometric bundle adjustment | Gradient-SDF · TUM RGB-D fr1/xyz | pose translation RMSE (cm): 2.8 -> 1.0 | Improved |
| Apply implicit photometric bundle adjustment | Gradient-SDF · TUM RGB-D fr1/teddy | pose translation RMSE (cm): 6.8 -> 4.4 | Improved |
| Apply implicit photometric bundle adjustment | Gradient-SDF · TUM RGB-D fr2/xyz | pose translation RMSE (cm): 1.8 -> 1.0 | Improved |
| Apply implicit photometric bundle adjustment | Gradient-SDF · TUM RGB-D fr2/rpy | pose translation RMSE (cm): 2.7 -> 2.5 | Improved |
| Apply implicit photometric bundle adjustment | Gradient-SDF · TUM RGB-D fr3/household | pose translation RMSE (cm): 2.1 -> 0.6 | Improved |
| Optimize poses only rather than optimizing poses and geometry (ablation) | Gradient-SDF bundle adjustment · TUM RGB-D sequences | pose translation RMSE (cm): 1.0 -> 1.0; 4.4 -> 4.3; 1.0 -> 1.1; 2.5 -> 2.4; 0.6 -> 0.7 | No clear effect |
| Decouple pose optimization across frames (ablation) | Gradient-SDF bundle adjustment · TUM RGB-D sequences | pose translation RMSE (cm): 1.0 -> 1.2; 4.4 -> 4.6; 1.0 -> 1.0; 2.5 -> 2.5; 0.6 -> 0.6 | No clear effect |
| Decouple pose optimization and optimize poses only (ablation) | Gradient-SDF bundle adjustment · TUM RGB-D sequences | pose translation RMSE (cm): 1.0 -> 1.1; 4.4 -> 4.3; 1.0 -> 1.0; 2.5 -> 2.4; 0.6 -> 0.6 | No clear effect |
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.
- BAD SLAM — BAD SLAM: Bundle Adjusted Direct RGB-D SLAM (CVPR 2019) · better on 3 of 5 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
- SDFDiff: Differentiable Rendering of Signed Distance Fields for 3D Shape Optimization (CVPR 2020 · similarity 0.88)
- Differentiable Gradient Sampling for Learning Implicit 3D Scene Reconstructions from a Single Image (ICLR 2022 · similarity 0.88)
- Geo-Neus: Geometry-Consistent Neural Implicit Surfaces Learning for Multi-view Reconstruction (NeurIPS 2022 · similarity 0.88)
- Learning Scene-Level Signed Directional Distance Function With Ellipsoidal Priors and Neural Residuals (TPAMI 2026 · similarity 0.88)
- ProbeSDF: Light Field Probes For Neural Surface Reconstruction (CVPR 2025 · similarity 0.87)
- GSDF: 3DGS Meets SDF for Improved Neural Rendering and Reconstruction (NeurIPS 2024 · similarity 0.87)
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.
