Research map / 3D vision / Face and shape
Phase Transitions, Distance Functions, and Implicit Neural Representations
Yaron Lipman
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 |
|---|---|---|---|
| use PHASE instead of the strongest listed baseline | surface reconstruction benchmark with point clouds and normals · Anchor, Daratech, DC, Gargoyle, and Lord Quas | d_C: Anchor: 0.22 -> 0.21; Daratech: 0.20 -> 0.18; DC: 0.14 -> 0.15; Gargoyle: 0.16 -> 0.16; Lord Quas: 0.12 -> 0.11 | Mixed |
| use PHASE instead of the strongest listed baseline | surface reconstruction benchmark with point clouds and normals · Anchor, Daratech, DC, Gargoyle, and Lord Quas | d_H: Anchor: 4.49 -> 4.29; Daratech: 3.14 -> 2.92; DC: 1.35 -> 2.52; Gargoyle: 3.20 -> 3.14; Lord Quas: 0.69 -> 0.96 | Mixed |
| add Fourier features to PHASE and train on points only | PHASE · Anchor, Daratech, DC, Gargoyle, Lord Quas | d_C: Anchor: 0.72 -> 0.29; Daratech: 2.48 -> 0.35; DC: 0.86 -> 0.19; Gargoyle: 0.26 -> 0.17; Lord Quas: 0.49 -> 0.11 | Improved |
| add the unit-gradient loss at input points (ablation) | PHASE · 2D point-cloud reconstructions | surface regularity at data points | Improved |
| decrease epsilon (ablation) | PHASE · 2D point-cloud reconstructions | distance-function approximation | Improved |
| add Fourier features to PHASE (ablation) | PHASE with normal data · surface reconstruction benchmark | d_C: Anchor: 0.21 -> 0.24; Daratech: 0.18 -> 0.18; DC: 0.15 -> 0.15; Gargoyle: 0.16 -> 0.16; Lord Quas: 0.11 -> 0.12 | Mixed |
| remove the gradient loss (mu=0 instead of mu=0.1) (ablation) | PHASE · 2D point-cloud reconstructions | surface regularity at data points | Hurt |
| use Fourier features with IGR rather than PHASE | implicit surface reconstruction from point clouds · surface reconstruction benchmark, points only | extraneous surface parts | Hurt |
Closest papers by meaning
- StEik: Stabilizing the Optimization of Neural Signed Distance Functions and Finer Shape Representation (NeurIPS 2023 · similarity 0.88)
- VisCo Grids: Surface Reconstruction with Viscosity and Coarea Grids (NeurIPS 2022 · similarity 0.88)
- Iso-Points: Optimizing Neural Implicit Surfaces With Hybrid Representations (CVPR 2021 · similarity 0.87)
- VI^3NR: Variance Informed Initialization for Implicit Neural Representations (CVPR 2025 · similarity 0.86)
- Implicit Representations via Operator Learning (ICML 2024 · similarity 0.85)
- Grids Often Outperform Implicit Neural Representation at Compressing Dense Signals (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.
