Research map / Point clouds / Shape and surface completion
DiffCD: A Symmetric Differentiable Chamfer Distance for Neural Implicit Surface Fitting
Linus Härenstam-Nielsen, Lu Sang, Abhishek Saroha, Nikita Araslanov, Daniel Cremers
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 |
|---|---|---|---|
| add surface-to-points Chamfer term (symmetric DiffCD loss) to the IGR points-to-surface loss | IGR · FAMOUS No Noise | CD squared: 3.092 -> 0.542 | Improved |
| use DiffCD instead of IGR loss under medium noise | IGR · FAMOUS Medium Noise | Chamfer distance (CD): 1.454 -> 0.690 | Improved |
| use DiffCD instead of SIREN off-surface regularizer (mu=0.33) | SIREN · Thingi10k No Noise | CD squared: 0.894 -> 0.580 | Improved |
| use DiffCD instead of SIREN (mu=0.33) on Chamfer angle | SIREN · Thingi10k No Noise | Chamfer angle (deg): 12.7 -> 13.0 | Hurt |
| increase SIREN SSA weight mu from 0.033 to 0.33 (ablation) | SIREN · FAMOUS Medium Noise | Chamfer distance (CD): 0.963 -> 0.739 | Improved |
| increase SIREN SSA weight mu from 0.033 to 0.33 on noise-free data (ablation) | SIREN · FAMOUS No Noise | CD squared: 1.151 -> 1.525 | Hurt |
| increase SIREN weight mu (surface area regularization) | SIREN · Dragon and Serapis shapes | reconstruction error | Mixed |
| compare DiffCD to supervised methods in medium noise | DiffCD · FAMOUS Medium Noise | Chamfer distance (CD): NKSR 0.610 vs DiffCD 0.690 | Hurt |
| increase eikonal weight lambda to 0.5 and 1 for higher noise settings | DiffCD · FAMOUS Medium/Max Noise | surface quality | Improved |
| use DiffCD instead of IGR in compute cost | DiffCD · runtime | time per iteration: roughly 6x longer | 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.
- Neural-Pull — Neural-Pull: Learning Signed Distance Function from Point clouds by Learning to Pull Space onto Surface (ICML 2021) · better on 1 of 1 shared result
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
- Iso-Points: Optimizing Neural Implicit Surfaces With Hybrid Representations (CVPR 2021 · similarity 0.89)
- GeoUDF: Surface Reconstruction from 3D Point Clouds via Geometry-guided Distance Representation (ICCV 2023 · similarity 0.89)
- Neural Unsigned Distance Fields for Implicit Function Learning (NeurIPS 2020 · similarity 0.89)
- Small Steps and Level Sets: Fitting Neural Surface Models with Point Guidance (CVPR 2024 · similarity 0.89)
- Diffeomorphic Mesh Deformation via Efficient Optimal Transport for Cortical Surface Reconstruction (ICLR 2024 · similarity 0.88)
- OffsetOPT: Explicit Surface Reconstruction without Normals (CVPR 2025 · 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.
