Research map / Probabilistic and scientific modeling / Differential equations and PDEs
StEik: Stabilizing the Optimization of Neural Signed Distance Functions and Finer Shape Representation
Huizong Yang, Yuxin Sun, Ganesh Sundaramoorthi, Anthony Yezzi
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 divergence loss with Laplacian normal loss, linear layers (ablation) | DiGS (linear) · ShapeNet | IoU mean: 0.9390 -> 0.9586 | Improved |
| replace divergence loss with Laplacian normal loss, linear layers (ablation) | DiGS (linear) · ShapeNet | squared Chamfer mean: 1.32e-4 -> 1.71e-4 | Mixed |
| replace linear with quadratic layers, keeping divergence loss (ablation) | DiGS · ShapeNet | squared Chamfer mean: 1.32e-4 -> 5.45e-5 | Improved |
| replace linear with quadratic layers, keeping divergence loss (ablation) | DiGS · ShapeNet | IoU mean: 0.9390 -> 0.9593 | Improved |
| replace divergence loss with Laplacian normal loss, quadratic layers (ablation) | Qua+L_div · ShapeNet | IoU mean: 0.9593 -> 0.9671 | Improved |
| replace divergence loss with Laplacian normal loss, quadratic layers (ablation) | Qua+L_div · ShapeNet | squared Chamfer mean: 5.45e-5 -> 6.86e-5 | Hurt |
| add quadratic layers on top of Laplacian normal loss (ablation) | Lin+L_L.n. · ShapeNet | IoU mean: 0.9586 -> 0.9671 | Improved |
| add quadratic layers on top of Laplacian normal loss (ablation) | Lin+L_L.n. · ShapeNet | squared Chamfer mean: 1.71e-4 -> 6.86e-5 | Improved |
| use StEik instead of DiGS, no normals (ablation) | DiGS · SRB ground truth | Hausdorff distance d_H: 3.52 -> 2.800 | Improved |
| use StEik instead of DiGS, no normals (ablation) | DiGS · SRB scans | one-sided Chamfer d_C: 0.08 -> 0.096 | Hurt |
| use Laplacian normal loss instead of divergence loss in runtime (ablation) | DiGS 5x256 · Runtime per iteration | runtime: 37.86ms -> 32.52ms | Improved |
| use quadratic layers (5x128) instead of linear (5x256) (ablation) | DiGS · Runtime per iteration | runtime: 37.86ms -> 42.20ms | Hurt |
Closest papers by meaning
- Implicit Geometric Regularization for Learning Shapes (ICML 2020 · similarity 0.89)
- Few-Shot Unsupervised Implicit Neural Shape Representation Learning with Spatial Adversaries (ICML 2024 · similarity 0.88)
- Small Steps and Level Sets: Fitting Neural Surface Models with Point Guidance (CVPR 2024 · similarity 0.88)
- DUDF: Differentiable Unsigned Distance Fields with Hyperbolic Scaling (CVPR 2024 · similarity 0.87)
- CoFie: Learning Compact Neural Surface Representations with Coordinate Fields (NeurIPS 2024 · similarity 0.87)
- Implicit-ARAP: Efficient Handle-Guided Neural Field Deformation via Local Patch Meshing (NeurIPS 2025 · similarity 0.86)
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.
