Research map / Point clouds / Shape and surface completion
Unsupervised shape correspondence
Learn dense shape correspondence and implicit structure from raw point clouds.
102 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Shape and surface completion, keyword cluster “reconstruction · implicit · meshes”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
21 papers in 2023–24 and 22 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 23.6% to 19.5%.
No longer reported by papers here: ModelNet40 (accuracy, last in 2022).
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
- 2016 · Functional Faces: Groupwise Dense Correspondence Using Functional Maps (CVPR 2016 · 17 citations)
- 2017 · A Point Set Generation Network for 3D Object Reconstruction From a Single Image (CVPR 2017 · 2,524 citations)
- 2018 · Learning Representations and Generative Models for 3D Point Clouds (ICML 2018 · 530 citations)
- 2019 · PartNet: A Recursive Part Decomposition Network for Fine-Grained and Hierarchical Shape Segmentation (CVPR 2019 · 113 citations)
- 2020 · SSRNet: Scalable 3D Surface Reconstruction Network (CVPR 2020 · 66 citations)
- 2021 · CorrNet3D: Unsupervised End-to-End Learning of Dense Correspondence for 3D Point Clouds (CVPR 2021 · 85 citations)
- 2022 · Surface Representation for Point Clouds (CVPR 2022 · 247 citations)
- 2023 · HGNet: Learning Hierarchical Geometry From Points, Edges, and Surfaces (CVPR 2023 · 38 citations)
- 2024 · Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds (CVPR 2024 · 42 citations)
- 2025 · SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single Images (CVPR 2025 · 21 citations)
- 2026 · VGGTFace: Topologically Consistent Facial Geometry Reconstruction in the Wild (AAAI 2026 · 1 citation)
Earliest papers here
- Functional Faces: Groupwise Dense Correspondence Using Functional Maps (CVPR 2016 · 17 citations)
- Similarity Metric For Curved Shapes In Euclidean Space (CVPR 2016 · 12 citations)
- Piecewise-Planar 3D Approximation From Wide-Baseline Stereo (CVPR 2016 · 9 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.
ModelNet40 · accuracy (higher is better)
4 papers, 2018–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 84.5 | Learning Representations and Generative Models for 3D Point Clouds (ICML 2018 · 530 citations) |
| 2020 | 84.7 | Unsupervised Deep Shape Descriptor With Point Distribution Learning (CVPR 2020 · 19 citations) |
| 2021 | 93 | Unsupervised 3D Learning for Shape Analysis via Multiresolution Instance Discrimination (AAAI 2021 · 42 citations) |
What has been tried
86 papers here report 849 measured changes (a component added, removed or swapped, compared with a baseline): helped 77%, no clear effect 4%, hurt 16%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use the proposed mixed SiLU and sinusoidal activations instead of ReLU | ABC freeform surfaces with σ = 0.05 noise · P-coverage | 0.519 -> 0.585 (open); 0.509 -> 0.581 (closed) | Helped | Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds (CVPR 2024 · 42 citations) |
| add the point sampling stage | GSO · CD | 0.136 -> 0.120 | Helped | SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single Images (CVPR 2025 · 21 citations) |
| replace diffusion with one-step regression | D-FAUST unseen motion, monocular noisy depth scans · IoU | 71.1% -> 83.8% | Helped | Motion2VecSets: 4D Latent Vector Set Diffusion for Non-rigid Shape Reconstruction and Tracking (CVPR 2024 · 9 citations) |
| add template generation and correlation fusion | TOSCA · correspondence accuracy (0.01) | 52.3% -> 54.6% | Helped | Unsupervised Template-assisted Point Cloud Shape Correspondence Network (CVPR 2024 · 6 citations) |
| Add DLL with DSPT backbone to ALTO | Synthetic Rooms, 10K points · IoU | 0.895 -> 0.921 | Helped | DITTO: Dual and Integrated Latent Topologies for Implicit 3D Reconstruction (CVPR 2024 · 5 citations) |
| add diffusion learning to NU-MCC (adaptive denoised query points) | CO3D-v2 · F1 | +4.9 (80.9→85.8) | Helped | IPoD: Implicit Field Learning with Point Diffusion for Generalizable 3D Object Reconstruction from Single RGB-D Images (CVPR 2024 · 5 citations) |
| Add distortion loss Ld | 4D-Dress Pairs S/S · SAσ(×10) | 0.115 -> 0.018 | Helped | 4Deform: Neural Surface Deformation for Robust Shape Interpolation (CVPR 2025 · 4 citations) |
| Use multi-view attention instead of omitting it | Objaverse-LVIS · Top1 accuracy | 54.61 -> 55.24 (+0.63) | Helped | Duoduo CLIP: Efficient 3D Understanding with Multi-View Images (ICLR 2025 · 2 citations) |
| Remove self-reconstruction loss | KeypointNet motorcycle · PCK | Hurt | Learning SO(3)-Invariant Semantic Correspondence via Local Shape Transform (CVPR 2024 · 2 citations) | |
| Replace PointNet with a DGCNN encoder | Pancreas Point2SSM ablation · CD | 3.00 -> 2.87 | Helped | Point2SSM: Learning Morphological Variations of Anatomies from Point Clouds (ICLR 2024 · 2 citations) |
| add T-CDM, local cost, initial flow, optimal transport, and cross-consistent matching | TOSCA/TOSCA · accuracy | 54.58% -> 66.71% | Helped | DiffCorr: Conditional Diffusion Model with Reliable Pseudo-Label Guidance for Unsupervised Point Cloud Shape Correspondence (AAAI 2025 · 1 citation) |
| replace corner-based representation with edge-based representation | Structured3D · Room F1 | 96.4% to 98.2% | Helped | CAGE: Continuity-Aware edGE Network Unlocks Robust Floorplan Reconstruction (NeurIPS 2025 · 1 citation) |
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
- SMAL · 3 of 9 papers
- SURREAL · 2 of 9 papers
- SHREC'19 · 2 of 9 papers
- TOSCA · 2 of 9 papers
- ShapeNet · 2 of 9 papers
Baselines
- Diff-FMaps · 2 of 9 papers
- 3D-CODED · 2 of 9 papers
- Elementary · 2 of 9 papers
- CorrNet3D · 2 of 9 papers
- DPC · 2 of 9 papers
- SE-ORNet · 2 of 9 papers
- HSTR · 2 of 9 papers
- TANet · 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 43 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 4 papers
- Zhejiang University · 2 papers
- Deep Space Exploration Lab · 2 papers
- Alibaba Group · 2 papers
- Hong Kong University of Science and Technology · 2 papers
Use it
Download BibTeX of 40 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
- Point cloud upsampling and normals (64 papers)
- Distance-function surface reconstruction (50 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).