mlsanityData to 14 Oct 2026

Unified Primitive Proxies for Structured Shape Completion

arXiv 2601.00759 ↗First posted 2 Jan 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Structured shape completion recovers missing geometry as primitives rather than as unstructured points, which enables primitive-based surface reconstruction. Instead of following the prevailing cascade, we rethink how primitives and points should interact, and find it more effective to decode primitives in a dedicated pathway that attends to shared shape features. Following this principle, we present UniCo, which in a single feed-forward pass predicts a set of primitives with complete geometry, semantics, and inlier membership. To drive this unified representation, we introduce primitive proxies, learnable queries that are contextualized to produce assembly-ready outputs. To ensure consistent optimization, our training strategy couples primitives and points with online target updates. Across synthetic and real-world benchmarks with four independent assembly solvers, UniCo consistently outperforms recent baselines, lowering Chamfer distance by up to 50% and improving normal consistency by up to 7%. These results establish an attractive recipe for structured 3D understanding from incomplete data. Project page: https://unico-completion.github.io.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 50 of its 73 references that are in the map. First posted on arXiv 2 Jan 2026.

Builds on

50 of 73 references.

References · 50

  1. 2016PointNet: Deep Learning on Point Sets for 3D Classification and Segmentationcited by 5,248
  2. 2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,401
  3. 2018Dynamic Graph CNN for Learning on Point Cloudscited by 2,223
  4. 2019KPConv: Flexible and Deformable Convolution for Point Cloudscited by 1,205
  5. 2021Per-Pixel Classification is Not All You Need for Semantic Segmentationcited by 1,152
  6. 2023Point Transformer V3: Simpler, Faster, Strongercited by 589
  7. 2018PCN: Point Completion Networkcited by 384
  8. 2018ABC: A Big CAD Model Dataset For Geometric Deep Learningcited by 341
  9. 2023OmniObject3D: Large-Vocabulary 3D Object Dataset for Realistic Perception, Reconstruction and Generationcited by 326
  10. 2020PointGroup: Dual-Set Point Grouping for 3D Instance Segmentationcited by 321
  11. 2022Mask3D: Mask Transformer for 3D Semantic Instance Segmentationcited by 303
  12. 2021PoinTr: Diverse Point Cloud Completion with Geometry-Aware Transformerscited by 286
38 more
  1. 2016Shape Completion using 3D-Encoder-Predictor CNNs and Shape Synthesiscited by 219
  2. 2019BSP-Net: Generating Compact Meshes via Binary Space Partitioningcited by 206
  3. 2022SoftGroup for 3D Instance Segmentation on Point Cloudscited by 200
  4. 2016Learning Shape Abstractions by Assembling Volumetric Primitivescited by 181
  5. 2020GRNet: Gridding Residual Network for Dense Point Cloud Completioncited by 174
  6. 2021SnowflakeNet: Point Cloud Completion by Snowflake Point Deconvolution with Skip-Transformercited by 164
  7. 2021Hierarchical Aggregation for 3D Instance Segmentationcited by 127
  8. 2020Points2Surf: Learning Implicit Surfaces from Point Cloud Patchescited by 122
  9. 2022Superpoint Transformer for 3D Scene Instance Segmentationcited by 116
  10. 2021Density-aware Chamfer Distance as a Comprehensive Metric for Point Cloud Completioncited by 112
  11. 2021Instance Segmentation in 3D Scenes using Semantic Superpoint Tree Networkscited by 99
  12. 2020ParSeNet: A Parametric Surface Fitting Network for 3D Point Cloudscited by 96
  13. 2018Supervised Fitting of Geometric Primitives to 3D Point Cloudscited by 91
  14. 2017High-Resolution Shape Completion Using Deep Neural Networks for Global Structure and Local Geometry Inferencecited by 87
  15. 2018Learning to predict crisp boundariescited by 82
  16. 20193D Instance Segmentation via Multi-Task Metric Learningcited by 80
  17. 2016Shape Completion Enabled Robotic Graspingcited by 79
  18. 2023AdaPoinTr: Diverse Point Cloud Completion with Adaptive Geometry-Aware Transformerscited by 67
  19. 2021CAPRI-Net: Learning Compact CAD Shapes with Adaptive Primitive Assemblycited by 62
  20. 2021CSG-Stump: A Learning Friendly CSG-Like Representation for Interpretable Shape Parsingcited by 51
  21. 2023SECAD-Net: Self-Supervised CAD Reconstruction by Learning Sketch-Extrude Operationscited by 46
  22. 2023Mask-Attention-Free Transformer for 3D Instance Segmentationcited by 46
  23. 2022Surface Reconstruction from Point Clouds: A Survey and a Benchmarkcited by 41
  24. 2021HPNet: Deep Primitive Segmentation Using Hybrid Representationscited by 36
  25. 2023Point2CAD: Reverse Engineering CAD Models from 3D Point Cloudscited by 36
  26. 2022LAKe-Net: Topology-Aware Point Cloud Completion by Localizing Aligned Keypointscited by 27
  27. 2021CPFN: Cascaded Primitive Fitting Networks for High-Resolution Point Cloudscited by 24
  28. 2022CSDN: Cross-modal Shape-transfer Dual-refinement Network for Point Cloud Completioncited by 22
  29. 2021Reconstructing Compact Building Models from Point Clouds Using Deep Implicit Fieldscited by 20
  30. 2022FBNet: Feedback Network for Point Cloud Completioncited by 20
  31. 2024Re-assembling the past: The RePAIR dataset and benchmark for real world 2D and 3D puzzle solvingcited by 19
  32. 2025SymmCompletion: High-Fidelity and High-Consistency Point Cloud Completion with Symmetry Guidancecited by 15
  33. 2023PolyGNN: Polyhedron-based Graph Neural Network for 3D Building Reconstruction from Point Cloudscited by 10
  34. 2025Parametric Point Cloud Completion for Polygonal Surface Reconstructioncited by 6
  35. 2024Building-PCC: Building Point Cloud Completion Benchmarkscited by 5
  36. 2024Concise Plane Arrangements for Low-Poly Surface and Volume Modellingcited by 4
  37. 2024Towards Flexible 3D Perception: Object-Centric Occupancy Completion Augments 3D Object Detectioncited by 3
  38. 2025Learning Generalizable Shape Completion with SIM(3) Equivariancecited by 2

Built on by

Not cited yet.