mlsanityData to 7 Oct 2026

ConDor: Self-Supervised Canonicalization of 3D Pose for Partial Shapes

arXiv 2201.07788 ↗First posted 19 Jan 2022cs.CV cs.AI cs.LGcited by 31

Not on the map. Why?

Abstract (arXiv)

Progress in 3D object understanding has relied on manually canonicalized shape datasets that contain instances with consistent position and orientation (3D pose). This has made it hard to generalize these methods to in-the-wild shapes, eg., from internet model collections or depth sensors. ConDor is a self-supervised method that learns to Canonicalize the 3D orientation and position for full and partial 3D point clouds. We build on top of Tensor Field Networks (TFNs), a class of permutation- and rotation-equivariant, and translation-invariant 3D networks. During inference, our method takes an unseen full or partial 3D point cloud at an arbitrary pose and outputs an equivariant canonical pose. During training, this network uses self-supervision losses to learn the canonical pose from an un-canonicalized collection of full and partial 3D point clouds. ConDor can also learn to consistently co-segment object parts without any supervision. Extensive quantitative results on four new metrics show that our approach outperforms existing methods while enabling new applications such as operation on depth images and annotation transfer.

Key facts. Cited by 31 papers in the map (data through 7 Oct 2026). Builds on 38 of its 63 references that are in the map. First posted on arXiv 19 Jan 2022.

Builds on

38 of 63 references.

References · 38

  1. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  2. 2016PointNet: Deep Learning on Point Sets for 3D Classification and Segmentationcited by 5,223
  3. 2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,384
  4. 2015ShapeNet: An Information-Rich 3D Model Repositorycited by 2,880
  5. 2019DeepSDF: Learning Continuous Signed Distance Functions for Shape Representationcited by 2,682
  6. 2018Dynamic Graph CNN for Learning on Point Cloudscited by 2,220
  7. 2018Occupancy Networks: Learning 3D Reconstruction in Function Spacecited by 2,037
  8. 2018Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point cloudscited by 735
  9. 2021Common Objects in 3D: Large-Scale Learning and Evaluation of Real-life 3D Category Reconstructioncited by 646
  10. 20163D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstructioncited by 587
  11. 2015Multi-view Convolutional Neural Networks for 3D Shape Recognitioncited by 564
  12. 2018AtlasNet: A Papier-Mach'e Approach to Learning 3D Surface Generationcited by 539
26 more
  1. 2018PartNet: A Large-scale Benchmark for Fine-grained and Hierarchical Part-level 3D Object Understandingcited by 521
  2. 2019Normalized Object Coordinate Space for Category-Level 6D Object Pose and Size Estimationcited by 490
  3. 2019PointFlow: 3D Point Cloud Generation with Continuous Normalizing Flowscited by 487
  4. 2018Spherical CNNscited by 452
  5. 20183D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Datacited by 348
  6. 2019Deep Closest Point: Learning Representations for Point Cloud Registrationcited by 318
  7. 2019Cormorant: Covariant Molecular Neural Networkscited by 262
  8. 2019What Do Single-view 3D Reconstruction Networks Learn?cited by 259
  9. 2017Learning SO(3) Equivariant Representations with Spherical CNNscited by 244
  10. 2016Volumetric and Multi-View CNNs for Object Classification on 3D Datacited by 240
  11. 2019Learning elementary structures for 3D shape generation and matchingcited by 98
  12. 2020DeepGMR: Learning Latent Gaussian Mixture Models for Registrationcited by 91
  13. 2019BAE-NET: Branched Autoencoder for Shape Co-Segmentationcited by 89
  14. 2019Rotation Invariant Convolutions for 3D Point Clouds Deep Learningcited by 80
  15. 20196-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypointscited by 78
  16. 2018Learning to Reconstruct Shapes from Unseen Classescited by 60
  17. 2019C3DPO: Canonical 3D Pose Networks for Non-Rigid Structure From Motioncited by 60
  18. 2020A Wigner-Eckart Theorem for Group Equivariant Convolution Kernelscited by 46
  19. 2020Weakly-supervised 3D Shape Completion in the Wildcited by 33
  20. 2020Canonical Capsules: Self-Supervised Capsules in Canonical Posecited by 30
  21. 2020Learning to Orient Surfaces by Self-supervised Spherical CNNscited by 26
  22. 2020Pix2Surf: Learning Parametric 3D Surface Models of Objects from Imagescited by 22
  23. 2020FG-Net: Fast Large-Scale LiDAR Point Clouds Understanding Network Leveraging Correlated Feature Mining and Geometric-Aware Modellingcited by 17
  24. 2020Continuous Geodesic Convolutions for Learning on 3D Shapescited by 14
  25. 2019Multiview Aggregation for Learning Category-Specific Shape Reconstructioncited by 12
  26. 2020DRACO: Weakly Supervised Dense Reconstruction And Canonicalization of Objectscited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 6; 2023: 13; 2024: 6; 2025: 2; 2026: 4.

31 papers.

Sort:

By area:

  1. 2023LEGO-Net: Learning Regular Rearrangements of Objects in Roomscited by 75
  2. 2022Zero-Shot Category-Level Object Pose Estimationcited by 57
  3. 2024Open-Universe Indoor Scene Generation using LLM Program Synthesis and Uncurated Object Databasescited by 43
  4. 2024A Canonicalization Perspective on Invariant and Equivariant Learningcited by 25
  5. 2023Laplacian Canonization: A Minimalist Approach to Sign and Basis Invariant Spectral Embeddingcited by 21
  6. 20234D Panoptic Segmentation as Invariant and Equivariant Field Predictioncited by 19
  7. 2023Banana: Banach Fixed-Point Network for Pointcloud Segmentation with Inter-Part Equivariancecited by 19
  8. 2023EFEM: Equivariant Neural Field Expectation Maximization for 3D Object Segmentation Without Scene Supervisioncited by 18
  9. 2022Shape-Pose Disentanglement using SE(3)-equivariant Vector Neuronscited by 16
  10. 2023HyP-NeRF: Learning Improved NeRF Priors using a HyperNetworkcited by 14
  11. 2024OV9D: Open-Vocabulary Category-Level 9D Object Pose and Size Estimationcited by 10
  12. 2022Canonical Fields: Self-Supervised Learning of Pose-Canonicalized Neural Fieldscited by 9
19 more
  1. 2023SC3K: Self-supervised and Coherent 3D Keypoints Estimation from Rotated, Noisy, and Decimated Point Cloud Datacited by 9
  2. 2023SCARP: 3D Shape Completion in ARbitrary Poses for Improved Graspingcited by 8
  3. 2023PaRot: Patch-Wise Rotation-Invariant Network via Feature Disentanglement and Pose Restorationcited by 7
  4. 2022Vitruvio: 3D Building Meshes via Single Perspective Sketchescited by 5
  5. 2024Approximately Piecewise E(3) Equivariant Point Networkscited by 4
  6. 2024ESCAPE: Equivariant Shape Completion via Anchor Point Encodingcited by 4
  7. 2022Category-Level Global Camera Pose Estimation with Multi-Hypothesis Point Cloud Correspondencescited by 3
  8. 2023Stable and Consistent Prediction of 3D Characteristic Orientation via Invariant Residual Learningcited by 3
  9. 2023Improving Equivariance in State-of-the-Art Supervised Depth and Normal Predictorscited by 3
  10. 2022A Simple Strategy to Provable Invariance via Orbit Mappingcited by 2
  11. 20233D Surface Reconstruction in the Wild by Deforming Shape Priors from Synthetic Datacited by 2
  12. 2023Grasp Transfer based on Self-Aligning Implicit Representations of Local Surfacescited by 2
  13. 2025Learning Generalizable Shape Completion with SIM(3) Equivariancecited by 2
  14. 2024Symmetry-Robust 3D Orientation Estimationcited by 1
  15. 2026CanoVerse: 3D Object Scalable Canonicalization and Dataset for Generation and Posecited by 1
  16. 2025Towards Foundation Models for Cryo-ET Subtomogram Analysiscited by 0
  17. 2026Every9D-21M: Large-Scale Real-World 9D Canonicalization of Everyday Objectscited by 0
  18. 2026Annotation-Free Furniture Codes: What They Encode, and How Far They Transfercited by 0
  19. 2026CANIS: Generation-Assisted 3D Canonicalization via an Image-Semantic Bridgecited by 0