pith. sign in

arxiv: 2606.18429 · v2 · pith:XXCTSR3Znew · submitted 2026-06-16 · 💻 cs.CV · cs.AI· cs.LG

CAOA -- Completion-Assisted Object-CAD Alignment

Hiranya Garbha Kumar , Minhas Kamal , Balakrishnan Prabhakaran This is my paper

Pith reviewed 2026-06-27 01:08 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LG
keywords CAD alignmentpoint cloud completionRGB-D scans9-DoF pose estimationScan2CADsymmetry-aware poseindoor reconstructionS2C-Completion
0
0 comments X

The pith

Integrating point cloud completion with symmetry-aware pose estimation improves CAD model alignment to real scans by 17%.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents CAOA to align CAD models to objects in noisy and incomplete indoor RGB-D scans by first completing the object's point cloud before estimating a 9-DoF pose. A custom synthetic data generation strategy trains the completion module to bridge the gap to real scans, while symmetry information is used in the pose estimation to handle ambiguities. The authors also release the S2C-Completion dataset of over 8,500 expert-annotated real object-CAD pairs as a new benchmark. This pipeline produces a 17% accuracy gain over prior methods on the Scan2CAD benchmark.

Core claim

CAOA integrates a semantically and contextually aware point cloud completion module, trained on synthetic data generated to match indoor scene statistics, with a symmetry-aware relative pose estimation algorithm that incorporates symmetry information via a dedicated loss, enabling precise 9-DoF alignment of CAD models to objects in real RGB-D scans despite noise, incompleteness, and segmentation errors.

What carries the argument

The completion-assisted alignment pipeline that first completes partial scanned objects and then performs symmetry-aware relative pose estimation.

Load-bearing premise

The synthetic data generation strategy sufficiently reduces the synthetic-to-real domain gap for the completion module.

What would settle it

Running the full CAOA pipeline on Scan2CAD after removing the completion module and measuring whether the 17% accuracy gain disappears.

Figures

Figures reproduced from arXiv: 2606.18429 by Balakrishnan Prabhakaran, Hiranya Garbha Kumar, Minhas Kamal.

Figure 1
Figure 1. Figure 1: Overview of CAOA: The input to CAOA is a 3D room scan and its corresponding instance segmentation mask. Using this mask, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Annotations from Scan2CAD with CAD model (black) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of pose estimation on raw instance point [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Proposed architecture of OCAM using MinkowskiFCNN[7] as backbone. We use a shared back￾bone for extracting pose features from CAD and completed object point clouds. The extracted features are concatenated with the symmetry feature vector from SEM and forwarded to 3 different MLP heads, one each for estimating translation, rotation and scale. not—using the Scan2CAD dataset. We also experimented with trainin… view at source ↗
Figure 5
Figure 5. Figure 5: Annotations from S2C-Completion dataset with CAD [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Incomplete point cloud generation steps from synthetic [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: A qualitative comparison of alignment results using CIS2VR[ [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Examples from S2C-Completion dataset showing colored scan instance point clouds and aligned CAD models (gray). Each [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: A qualitative comparison of synthetic point cloud completion datasets. The last column is from real-world scans. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
read the original abstract

Accurately aligning CAD models to their corresponding objects in indoor RGB-D scans is a central challenge in 3D semantic reconstruction. The task requires estimating a 9-Degree-of-Freedom (DoF) pose-position, rotation, and scale along three axes-but is hindered by noisy and incomplete scans, as well as segmentation errors that cause geometric distortions. We present Completion-Assisted Object-CAD Alignment (CAOA), a method that integrates a semantically and contextually aware point cloud completion module with a symmetry-aware relative pose estimation algorithm, enabling precise alignment of CAD models to scanned objects. Existing completion methods are typically trained and evaluated on synthetic datasets, which often fail to generalize to real-world scans. To bridge this gap, we introduce a synthetic data generation strategy tailored to indoor scenes, significantly reducing the synthetic-to-real domain gap-validated through quantitative comparisons with widely used completion datasets. In addition, we release S2C-Completion, an expert-annotated dataset of over 8,500 object-CAD pairs from Scan2CAD, created for real-world indoor single-object completion and intended as a new benchmark for this task. For object-CAD alignment, we incorporate symmetry information via a symmetry-aware loss, improving robustness to symmetric ambiguities. On the Scan2CAD benchmark, CAOA achieves a 17% accuracy improvement over state-of-the-art methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper introduces Completion-Assisted Object-CAD Alignment (CAOA), which combines a point cloud completion network trained via a new indoor-scene synthetic data generation pipeline with a symmetry-aware 9-DoF pose estimator. It releases the expert-annotated S2C-Completion dataset (>8500 real Scan2CAD object-CAD pairs) and reports a 17% accuracy gain over prior art on the Scan2CAD benchmark, attributing the gain to reduced synthetic-to-real domain gap in completion.

Significance. If the completion module demonstrably improves real-scan completion quality and that improvement drives the alignment gain, the work would provide a practical advance for 3D indoor reconstruction pipelines and a useful new benchmark dataset. The symmetry-aware loss is a standard but useful addition.

major comments (2)
  1. [Abstract] Abstract: the headline claim that the synthetic data strategy 'significantly reduc[es] the synthetic-to-real domain gap' and thereby produces the 17% Scan2CAD lift rests on 'quantitative comparisons with widely used completion datasets.' No completion metrics (e.g., Chamfer distance, F-score) or ablation tables are referenced on the real S2C-Completion objects or on Scan2CAD scans themselves; without these, the performance attribution cannot be verified and other factors (symmetry loss, pose estimator) cannot be ruled out.
  2. [Abstract] The central experimental claim (17% Scan2CAD improvement) is load-bearing for the paper; if the completion module does not generalize to real noisy scans, the entire pipeline contribution collapses. Direct real-world completion ablations and an ablation removing the completion module on the Scan2CAD test set are therefore required.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the constructive feedback on our manuscript. We appreciate the emphasis on strengthening the experimental evidence linking the completion module to the reported alignment gains. We address each major comment below and commit to incorporating the requested analyses and metrics in the revised version.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline claim that the synthetic data strategy 'significantly reduc[es] the synthetic-to-real domain gap' and thereby produces the 17% Scan2CAD lift rests on 'quantitative comparisons with widely used completion datasets.' No completion metrics (e.g., Chamfer distance, F-score) or ablation tables are referenced on the real S2C-Completion objects or on Scan2CAD scans themselves; without these, the performance attribution cannot be verified and other factors (symmetry loss, pose estimator) cannot be ruled out.

    Authors: We agree that the current validation relies on comparisons with standard (primarily synthetic) completion datasets and does not yet include direct metrics on the real S2C-Completion objects or Scan2CAD scans. To substantiate the domain-gap reduction claim and isolate contributions, the revised manuscript will add Chamfer distance and F-score results on S2C-Completion, corresponding metrics on Scan2CAD scans, and ablation tables separating the effects of the completion module, symmetry loss, and pose estimator on the 17% Scan2CAD improvement. revision: yes

  2. Referee: [Abstract] The central experimental claim (17% Scan2CAD improvement) is load-bearing for the paper; if the completion module does not generalize to real noisy scans, the entire pipeline contribution collapses. Direct real-world completion ablations and an ablation removing the completion module on the Scan2CAD test set are therefore required.

    Authors: We concur that explicit evidence of generalization to real noisy scans is necessary to support the pipeline's contribution. The revision will include direct real-world completion ablations with quantitative metrics on Scan2CAD scans and a dedicated ablation that removes the completion module entirely, reporting the resulting 9-DoF alignment accuracy on the Scan2CAD test set. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical method with external benchmark results

full rationale

The paper presents CAOA as an empirical pipeline combining a completion module (trained on a new synthetic indoor-scene generation strategy) with symmetry-aware pose estimation, evaluated on the external Scan2CAD benchmark. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains appear in the abstract or described content. The 17% accuracy gain is reported as a measured outcome on an independent benchmark rather than a quantity forced by construction from the method's own inputs. The synthetic-to-real gap reduction is claimed via quantitative comparisons on standard datasets, but this is presented as validation evidence, not a self-referential loop. The work is self-contained against external benchmarks with no load-bearing steps that reduce to tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only view shows no explicit free parameters or invented entities; the method relies on standard assumptions in deep learning for point clouds (e.g., network architectures) and the claim that synthetic data matches real distribution.

axioms (1)
  • domain assumption The completion module trained on the new synthetic indoor data generalizes to real Scan2CAD scans.
    Stated as validated through quantitative comparisons, but no details on metrics or failure cases provided.

pith-pipeline@v0.9.1-grok · 5780 in / 1177 out tokens · 31532 ms · 2026-06-27T01:08:01.342832+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 8 canonical work pages · 3 internal anchors

  1. [1]

    Automatically annotating indoor images with cad models via rgb-d scans

    Stefan Ainetter, Sinisa Stekovic, Friedrich Fraundorfer, and Vincent Lepetit. Automatically annotating indoor images with cad models via rgb-d scans. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 3156–3164, 2023. 2, 3

    2023

  2. [2]

    Chang, and Matthias NieBner

    Armen Avetisyan, Manuel Dahnert, Angela Dai, Manolis Savva, Angel X. Chang, and Matthias NieBner. Scan2CAD: Learning CAD Model Alignment in RGB-D Scans. In 2019 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 2609–2618, Long Beach, CA, USA, 2019. IEEE. 1, 2, 3, 7, 8

    2019

  3. [3]

    End- to-End CAD Model Retrieval and 9DoF Alignment in 3D Scans

    Armen Avetisyan, Angela Dai, and Matthias Niessner. End- to-End CAD Model Retrieval and 9DoF Alignment in 3D Scans. In2019 IEEE/CVF International Conference on Computer Vision (ICCV), pages 2551–2560, Seoul, Korea (South), 2019. IEEE. 1, 2, 3, 7, 8

    2019

  4. [4]

    SceneCAD: Predicting Object Alignments and Layouts in RGB-D Scans

    Armen Avetisyan, Tatiana Khanova, Christopher Choy, Den- ver Dash, Angela Dai, and Matthias Nießner. SceneCAD: Predicting Object Alignments and Layouts in RGB-D Scans. InComputer Vision – ECCV 2020, pages 596–612, Cham,

    2020

  5. [5]

    1, 2, 3, 7, 8

    Springer International Publishing. 1, 2, 3, 7, 8

  6. [6]

    Orthogonal dictionary guided shape completion network for point cloud

    Pingping Cai, Deja Scott, Xiaoguang Li, and Song Wang. Orthogonal dictionary guided shape completion network for point cloud. InProceedings of the AAAI Conference on Ar- tificial Intelligence, pages 864–872, 2024. 4, 6

    2024

  7. [7]

    ShapeNet: An Information-Rich 3D Model Repository

    Angel X Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, et al. Shapenet: An information-rich 3d model repository.arXiv preprint arXiv:1512.03012, 2015. 2, 3, 4

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  8. [8]

    4d spatio-temporal convnets: Minkowski convolutional neural networks

    Christopher Choy, JunYoung Gwak, and Silvio Savarese. 4d spatio-temporal convnets: Minkowski convolutional neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3075–3084,

  9. [9]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes

    Angela Dai, Angel X Chang, Manolis Savva, Maciej Hal- ber, Thomas Funkhouser, and Matthias Nießner. Scannet: Richly-annotated 3d reconstructions of indoor scenes. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5828–5839, 2017. 11

    2017

  10. [10]

    BundleFusion: Real-time Globally Consistent 3D Reconstruction using On-the-fly Surface Re-integration

    Angela Dai, Matthias Nießner, Michael Zollh ¨ofer, Shahram Izadi, and Christian Theobalt. BundleFusion: Real-time Globally Consistent 3D Reconstruction using On-the-fly Surface Re-integration, 2017. arXiv:1604.01093 [cs]. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  11. [11]

    Shape Completion Using 3D-Encoder-Predictor CNNs and Shape Synthesis

    Angela Dai, Charles Ruizhongtai Qi, and Matthias NieBner. Shape Completion Using 3D-Encoder-Predictor CNNs and Shape Synthesis. In2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6545–6554, Honolulu, HI, 2017. IEEE. 3

    2017

  12. [12]

    U-RED: Unsuper- vised 3D Shape Retrieval and Deformation for Partial Point Clouds

    Yan Di, Chenyangguang Zhang, Ruida Zhang, Fabian Man- hardt, Yongzhi Su, Jason Rambach, Didier Stricker, Xi- angyang Ji, and Federico Tombari. U-RED: Unsuper- vised 3D Shape Retrieval and Deformation for Partial Point Clouds. In2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 8850–8861, Paris, France,

  13. [13]

    Vision meets robotics: The kitti dataset.The Inter- national Journal of Robotics Research, 32(11):1231–1237,

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The kitti dataset.The Inter- national Journal of Robotics Research, 32(11):1231–1237,

  14. [14]

    Nilesh Kulkarni, Justin Johnson, and David F. Fouhey. What’s Behind the Couch? Directed Ray Distance Functions (DRDF) for 3D Scene Reconstruction, 2022. arXiv:2112.04481 [cs]. 1

    work page arXiv 2022

  15. [15]

    Cis2vr: Cnn-based indoor scan to vr environment authoring framework

    Hiranya Kumar, Ninad Khargonkar, and Balakrishnan Prab- hakaran. Cis2vr: Cnn-based indoor scan to vr environment authoring framework. In2024 IEEE International Confer- ence on Artificial Intelligence and eXtended and Virtual Re- ality (AIxVR), pages 128–137. IEEE, 2024. 1, 2, 3, 7, 8, 11, 12

    2024

  16. [16]

    Fastcad: Real-time cad retrieval and alignment from scans and videos.ArXiv, abs/2403.15161, 2024

    Florian Langer, Jihong Ju, Georgi Dikov, Gerhard Reit- mayr, and Mohsen Ghafoorian. Fastcad: Real-time cad retrieval and alignment from scans and videos.ArXiv, abs/2403.15161, 2024. 2

    work page arXiv 2024

  17. [17]

    Database-assisted object retrieval for real-time 3d reconstruction

    Yangyan Li, Angela Dai, Leonidas Guibas, and Matthias Nießner. Database-assisted object retrieval for real-time 3d reconstruction. InComputer graphics forum, pages 435–

  18. [18]

    Wiley Online Library, 2015. 7

    2015

  19. [19]

    Database-assisted object retrieval for real-time 3d reconstruction.Computer Graphics Forum, 34:435–446,

    Yangyan Li, Angela Dai, Leonidas Guibas, and Matthias Nießner. Database-assisted object retrieval for real-time 3d reconstruction.Computer Graphics Forum, 34:435–446,

  20. [20]

    Lim, Hamed Pirsiavash, and Antonio Torralba

    Joseph J. Lim, Hamed Pirsiavash, and Antonio Torralba. Parsing ikea objects: Fine pose estimation. In2013 IEEE International Conference on Computer Vision, pages 2992– 2999, 2013. 3

    2013

  21. [21]

    InfoCD: A Contrastive Chamfer Distance Loss for Point Cloud Completion.Advances in Neural Informa- tion Processing Systems, 36:76960–76973, 2023

    Fangzhou Lin, Yun Yue, Ziming Zhang, Songlin Hou, Kazunori Yamada, Vijaya Kolachalama, and Venkatesh Saligrama. InfoCD: A Contrastive Chamfer Distance Loss for Point Cloud Completion.Advances in Neural Informa- tion Processing Systems, 36:76960–76973, 2023. 5

    2023

  22. [22]

    Cocca: Point cloud completion through cad cross-attention

    Adam Misik, Driton Salihu, Heike Brock, and Eckehard Steinbach. Cocca: Point cloud completion through cad cross-attention. In2023 IEEE International Conference on Image Processing (ICIP), pages 580–584. IEEE, 2023. 3

    2023

  23. [23]

    AutoSDF: Shape Priors for 3D Completion, Reconstruction and Generation, 2023

    Paritosh Mittal, Yen-Chi Cheng, Maneesh Singh, and Shub- ham Tulsiani. AutoSDF: Shape Priors for 3D Completion, Reconstruction and Generation, 2023. arXiv:2203.09516 [cs] version: 3. 3

    work page arXiv 2023

  24. [24]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space.Advances in neural information processing systems, 30, 2017

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space.Advances in neural information processing systems, 30, 2017. 4

    2017

  25. [25]

    Fast point feature histograms (fpfh) for 3d registration

    Radu Bogdan Rusu, Nico Blodow, and Michael Beetz. Fast point feature histograms (fpfh) for 3d registration. In2009 IEEE international conference on robotics and automation, pages 3212–3217. IEEE, 2009. 7

    2009

  26. [26]

    Deepspf: Spherical so (3)-equivariant patches for scan-to-cad estimation

    Driton Salihu, Adam Misik, Yuankai Wu, Constantin Patsch, Fabian Seguel, and Eckehard Steinbach. Deepspf: Spherical so (3)-equivariant patches for scan-to-cad estimation. InThe Twelfth International Conference on Learning Representa- tions, 2024. 3 9

    2024

  27. [27]

    Spherical mask: Coarse-to- fine 3d point cloud instance segmentation with spherical rep- resentation

    Sangyun Shin, Kaichen Zhou, Madhu Vankadari, Andrew Markham, and Niki Trigoni. Spherical mask: Coarse-to- fine 3d point cloud instance segmentation with spherical rep- resentation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4060– 4069, 2024. 3, 7

    2024

  28. [28]

    RetrievalFuse: Neu- ral 3D Scene Reconstruction with a Database, 2021

    Yawar Siddiqui, Justus Thies, Fangchang Ma, Qi Shan, Matthias Nießner, and Angela Dai. RetrievalFuse: Neu- ral 3D Scene Reconstruction with a Database, 2021. arXiv:2104.00024 [cs]. 1

    work page arXiv 2021

  29. [29]

    Chang, Manolis Savva, and Thomas Funkhouser

    Shuran Song, Fisher Yu, Andy Zeng, Angel X. Chang, Manolis Savva, and Thomas Funkhouser. Semantic Scene Completion from a Single Depth Image. In2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 190–198, Honolulu, HI, 2017. IEEE. 3

    2017

  30. [30]

    Learning 3D Shape Completion under Weak Supervision

    David Stutz and Andreas Geiger. Learning 3D Shape Completion under Weak Supervision.International Journal of Computer Vision, 128(5):1162–1181, 2020. arXiv:1805.07290 [cs]. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  31. [31]

    Pix3d: Dataset and methods for single-image 3d shape modeling

    Xingyuan Sun, Jiajun Wu, Xiuming Zhang, Zhoutong Zhang, Chengkai Zhang, Tianfan Xue, Joshua B Tenenbaum, and William T Freeman. Pix3d: Dataset and methods for single-image 3d shape modeling. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. 3

    2018

  32. [32]

    Dynamic fo- cused masking for autoregressive embodied occupancy pre- diction

    Yuan Sun, Julio Contreras, and Jorge Ortiz. Dynamic fo- cused masking for autoregressive embodied occupancy pre- diction. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems. 3

  33. [33]

    Unique signatures of histograms for local surface descrip- tion

    Federico Tombari, Samuele Salti, and Luigi Di Stefano. Unique signatures of histograms for local surface descrip- tion. InComputer Vision–ECCV 2010: 11th European Conference on Computer Vision, Heraklion, Crete, Greece, September 5-11, 2010, Proceedings, Part III 11, pages 356–

    2010

  34. [34]

    Joint learning of 3d shape retrieval and deformation

    Mikaela Angelina Uy, Vladimir G Kim, Minhyuk Sung, Noam Aigerman, Siddhartha Chaudhuri, and Leonidas J Guibas. Joint learning of 3d shape retrieval and deformation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11713–11722, 2021. 2

    2021

  35. [35]

    Softgroup for 3d instance segmentation on point clouds

    Thang Vu, Kookhoi Kim, Tung M Luu, Thanh Nguyen, and Chang D Yoo. Softgroup for 3d instance segmentation on point clouds. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2708– 2717, 2022. 7, 11

    2022

  36. [36]

    Deep Octree- based CNNs with Output-Guided Skip Connections for 3D Shape and Scene Completion, 2020

    Peng-Shuai Wang, Yang Liu, and Xin Tong. Deep Octree- based CNNs with Output-Guided Skip Connections for 3D Shape and Scene Completion, 2020. arXiv:2006.03762 [cs] version: 1. 3

    work page arXiv 2020

  37. [37]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xi- hui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4840–4851, 2024. 2, 4, 7

    2024

  38. [38]

    Beyond pascal: A benchmark for 3d object detection in the wild

    Yu Xiang, Roozbeh Mottaghi, and Silvio Savarese. Beyond pascal: A benchmark for 3d object detection in the wild. In IEEE Winter Conference on Applications of Computer Vi- sion, pages 75–82, 2014. 3

    2014

  39. [39]

    Pointr: Diverse point cloud completion with geometry-aware transformers

    Xumin Yu, Yongming Rao, Ziyi Wang, Zuyan Liu, Jiwen Lu, and Jie Zhou. Pointr: Diverse point cloud completion with geometry-aware transformers. InICCV, 2021. 3, 6, 7

    2021

  40. [40]

    Pcn: Point completion network

    Wentao Yuan, Tejas Khot, David Held, Christoph Mertz, and Martial Hebert. Pcn: Point completion network. In2018 in- ternational conference on 3D vision (3DV), pages 728–737. IEEE, 2018. 3, 6, 7

    2018

  41. [41]

    3DMatch: Learning Local Geometric Descriptors from RGB-D Re- constructions

    Andy Zeng, Shuran Song, Matthias Niessner, Matthew Fisher, Jianxiong Xiao, and Thomas Funkhouser. 3DMatch: Learning Local Geometric Descriptors from RGB-D Re- constructions. In2017 IEEE Conference on Computer Vi- sion and Pattern Recognition (CVPR), pages 199–208. IEEE,

  42. [42]

    KP- RED: Exploiting Semantic Keypoints for Joint 3D Shape Retrieval and Deformation

    Ruida Zhang, Chenyangguang Zhang, Yan Di, Fabian Man- hardt, Xingyu Liu, Federico Tombari, and Xiangyang Ji. KP- RED: Exploiting Semantic Keypoints for Joint 3D Shape Retrieval and Deformation. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 20540–20550, Seattle, W A, USA, 2024. IEEE. 2 10

    2024

  43. [43]

    Qualitative Analysis Figure 7 presents qualitative CAD alignment results, com- paring CAOA (c) and CIS2VR [14] (a) with ground truth annotations (b)

    Appendix Section 8.1. Qualitative Analysis Figure 7 presents qualitative CAD alignment results, com- paring CAOA (c) and CIS2VR [14] (a) with ground truth annotations (b). The input object point clouds are instance predictions generated by SoftGroup [33] for indoor scenes from the ScanNet dataset [8]. For each set of visualiza- tions, CIS2VR results are s...