Research map / Point clouds / Shape and surface completion
Distance-function surface reconstruction
Reconstruct surfaces by learning signed or unsigned distance functions from point clouds.
50 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Shape and surface completion, keyword cluster “reconstruction · implicit · meshes”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Leveling off Several papers from 2025–26 still report on 1 tracked benchmark, but the best result has barely moved since 2024.
18 papers in 2023–24 and 12 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 20.2% to 10.6%.
Open questions the data points to
- Papers report POCO on ShapeNet 13 classes (NC) anywhere from 0.858 to 0.95: comparisons against it depend on whose number you take.
What looks settled
- ShapeNet (NC): best up to 2024 0.974, best in 2025–26 0.909 from 2 papers, at most 0.2% better, if at all.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2018 · Deep Marching Cubes: Learning Explicit Surface Representations (CVPR 2018 · 274 citations)
- 2019 · Deep Geometric Prior for Surface Reconstruction (CVPR 2019 · 222 citations)
- 2020 · SAL: Sign Agnostic Learning of Shapes From Raw Data (CVPR 2020 · 453 citations)
- 2021 · Neural-Pull: Learning Signed Distance Function from Point clouds by Learning to Pull Space onto Surface (ICML 2021 · 11 citations)
- 2022 · POCO: Point Convolution for Surface Reconstruction (CVPR 2022 · 146 citations)
- 2023 · Neural Kernel Surface Reconstruction (CVPR 2023 · 99 citations)
- 2024 · GridFormer: Point-Grid Transformer for Surface Reconstruction (AAAI 2024 · 19 citations)
- 2025 · A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · 8 citations)
- 2026 · Metric—Phase Fields: Decoupling Distance and Sign for Thin-Structure Reconstruction from Unoriented Point Clouds (ICML 2026 · 0 citations)
Earliest papers here
- Deep Marching Cubes: Learning Explicit Surface Representations (CVPR 2018 · 274 citations)
- Deep Geometric Prior for Surface Reconstruction (CVPR 2019 · 222 citations)
- Convolutional Neural Networks on Non-uniform Geometrical Signals Using Euclidean Spectral Transformation (ICLR 2019 · 8 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
ShapeNet · NC (higher is better)
5 papers, 2024–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 0.974 | MultiPull: Detailing Signed Distance Functions by Pulling Multi-Level Queries at Multi-Step (NeurIPS 2024 · 2 citations) |
What has been tried
45 papers here report 496 measured changes (a component added, removed or swapped, compared with a baseline): helped 68%, no clear effect 5%, hurt 23%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| use point-grid attention with a volume grid ($64^3$) instead of triplane grid ($3 * 128^2$) | Synthetic Rooms · IoU | 0.835 -> 0.918 | Helped | GridFormer: Point-Grid Transformer for Surface Reconstruction (AAAI 2024 · 19 citations) |
| add entropy regularization to margin uncertainty sampling | ShapeNet Tables · CD1 | 1.02 -> 0.77 | Helped | Unsupervised Occupancy Learning from Sparse Point Cloud (CVPR 2024 · 15 citations) |
| add Points-Net and Cross-Attn | Noisy point clouds · L1-Chamfer distance (×100) | Helped | A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · 8 citations) | |
| Add high exterior level set initialization instead of spherical initialization | ShapeNet, 20 highest-SION shapes · IoU (mean) | 0.2795 -> 0.5694 | Helped | Small Steps and Level Sets: Fitting Neural Surface Models with Point Guidance (CVPR 2024 · 8 citations) |
| Use BSP parameterized supervision rather than sparse input alone | ShapeNet lamp · CDL1 × 10 | 0.873 -> 0.077 | Helped | Learning Bijective Surface Parameterization for Inferring Signed Distance Functions from Sparse Point Clouds with Grid Deformation (CVPR 2025 · 5 citations) |
| add offset optimization | FAUST · CD1 (×102) | 0.221 -> 0.217 | Helped | OffsetOPT: Explicit Surface Reconstruction without Normals (CVPR 2025 · 3 citations) |
| Add rotation parameters | Newer College Math Institute · F-score | 82.5 -> 84.7 | Helped | 3D-SLNR: A Super Lightweight Neural Representation for Large-scale 3D Mapping (CVPR 2025 · 2 citations) |
| replace the Wasserstein distance with the Sinkhorn distance | ShapeNet · CD1 | 0.77 -> 0.63 | Helped | Toward Robust Neural Reconstruction from Sparse Point Sets (CVPR 2025 · 2 citations) |
| Add frequency features instead of linear layers | FAMOUS · CD_L2 ×100 | 0.042 -> 0.036 | Helped | MultiPull: Detailing Signed Distance Functions by Pulling Multi-Level Queries at Multi-Step (NeurIPS 2024 · 2 citations) |
| Use MLSE instead of the original level-set equation (OLSE) | Intermediate shapes · signed distance field stability | Helped | Implicit Neural Surface Deformation with Explicit Velocity Fields (ICLR 2025 · 1 citation) | |
| Use compactly supported CSRBF basis instead of Tri-Harmonic basis | Single-shape learning · rank of VVᵀ | 954 -> 3000 | Helped | LISR: Learning Linear 3D Implicit Surface Representation Using Compactly Supported Radial Basis Functions (AAAI 2024 · 1 citation) |
| use embedding size 256 rather than 128 or 512 | ABC · CD_L2 ×100 | 0.102 -> 0.096; 0.114 -> 0.096 | Helped | Inferring Neural Signed Distance Functions by Overfitting on Single Noisy Point Clouds through Finetuning Data-Driven based Priors (NeurIPS 2024 · 1 citation) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| POCO | ShapeNet 13 classes | NC | 0.858 – 0.95 | 2 papers, e.g. AAAI 2025 and ICML 2023 |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ShapeNet · 3 of 9 papers
- ScanNet · 3 of 9 papers
- Faust · 3 of 9 papers
- ABC · 2 of 9 papers
- Matterport3D · 2 of 9 papers
- DeepFashion3D · 2 of 9 papers
- 3DScene · 2 of 9 papers
Baselines
- OnSurf · 3 of 9 papers
- SPSR · 3 of 9 papers
- NKSR · 3 of 9 papers
- POCO · 2 of 9 papers
- Nspline · 2 of 9 papers
- NP · 2 of 9 papers
- NDrop · 2 of 9 papers
- SparseOcc · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 30 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 10 papers
- Wayne State University · 9 papers
- Nanyang Technological University · 3 papers
- Texas A&M University - College Station · 3 papers
- Inria · 2 papers
- Chinese Academy of Sciences · 2 papers
- City University of Hong Kong · 2 papers
Use it
Download BibTeX of 37 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Unsupervised shape correspondence (102 papers)
- Point cloud upsampling and normals (64 papers)
- Point cloud diffusion generation (44 papers)
Elsewhere in the topic
- Point cloud completion networks (50 papers)
- Structure-guided completion (22 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).