pith. sign in

arxiv: 2605.26949 · v1 · pith:AFHQSQNOnew · submitted 2026-05-26 · 💻 cs.CV · cs.GR

DinoComplete: 3D Shape Completion with Distilled Semantic Priors and State Space Models

Furkan Mert Algan , Eckehard Steinbach This is my paper

Pith reviewed 2026-06-29 17:46 UTC · model grok-4.3

classification 💻 cs.CV cs.GR
keywords 3D shape completionsemantic priorsDINO featuresstate space modelsvoxel Mambapartial scansgeneralizationunseen categories
0
0 comments X

The pith

Distilling DINO semantic features into a voxel Mamba network improves 3D shape completion from partial scans of unseen categories.

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

The paper introduces DinoComplete to complete 3D shapes from partial or noisy scans where geometry alone is often insufficient. It first builds multi-view DINO feature volumes from complete ShapeNet data and trains a student network to predict aligned semantic features from incomplete inputs. These features are then fused with geometric voxels inside a completion network that uses a multi-scale voxel Mamba module for efficient long-range reasoning. Experiments demonstrate higher completion quality than prior deterministic and generative methods on unseen ShapeNet categories and ScanNet objects, while using fewer parameters, less memory, and faster inference.

Core claim

DinoComplete augments geometric reconstruction with voxel-aligned semantic priors distilled from DINO features. A student network predicts dense semantic features directly from incomplete shapes; these are fused with geometric representations through voxel state-space modeling. A multi-scale voxel Mamba module refines the fused features by combining full-grid and chunk-wise sequence modeling, enabling better generalization and efficiency without sacrificing resolution.

What carries the argument

The multi-scale voxel Mamba module, which fuses geometric and distilled semantic voxel features and refines them via combined full-grid and chunk-wise sequence modeling for long-range reasoning.

If this is right

  • Stronger completion quality than prior deterministic and generative methods on unseen ShapeNet categories and ScanNet objects.
  • Lower parameter count, reduced memory footprint, and faster inference times.
  • Improved robustness when inputs are noisy real-world observations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same distillation step could be tested on other 3D tasks such as surface reconstruction or part segmentation to check transfer of the semantic priors.
  • Ablation studies that remove the semantic branch on progressively more distant categories would quantify how much the DINO features drive the reported gains.
  • The efficiency numbers suggest the method could run on-device for robotics or AR pipelines, though real-time latency on embedded hardware remains untested in the paper.

Load-bearing premise

DINO features extracted from complete multi-view ShapeNet data remain predictive and semantically useful when the input is a partial or noisy scan of an unseen category.

What would settle it

Evaluating DinoComplete on a fresh collection of unseen object categories with partial scans and finding that its completion metrics such as IoU or Chamfer distance no longer exceed those of prior methods would falsify the performance advantage.

Figures

Figures reproduced from arXiv: 2605.26949 by Eckehard Steinbach, Furkan Mert Algan.

Figure 1
Figure 1. Figure 1: Overview of the TSDF-DINO distillation pipeline. (a) DINO features extracted from [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our Multi-Scale Voxel Mamba architecture. Starting from a full voxel grid, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the shape completion pipeline. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: TSDF-DINO model qualitative results on unseen objects. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Shape completion results on both synthetic (blue) and real-world (yellow) objects from [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 11
Figure 11. Figure 11: Shape completion results on known categories on 3D-EPN dataset. [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: The effect of occlussion between synthetic and real data on our TSDF-DINO model. [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 6
Figure 6. Figure 6: The main architecture of TSDF-DINO model. D denotes feature dimension [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The detailed illustration of model blocks. [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The main overview of shape completion model. [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The generated ground truth data for TSDF-DINO model. [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Additional qualitative results of our TSDF-DINO model. [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
Figure 13
Figure 13. Figure 13: Additional shape completion results on both synthetic (blue) and real-world (yellow) [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗
read the original abstract

3D shape completion from partial scans remains challenging for unseen categories and noisy real-world observations, where geometry alone is often insufficient for inferring missing structure. We present DinoComplete, a deterministic and efficient shape completion framework that augments geometric reconstruction with voxel-aligned semantic priors distilled from DINO features. First, we construct multi-view DINO feature volumes aligned with ShapeNet data and train a student network to predict dense semantic features directly from incomplete shapes. These predicted features capture global structure and part-aware semantic context while remaining aligned with the underlying geometry. We then integrate these distilled features into a completion network, where geometric and semantic voxel representations are fused through voxel state-space modeling. To enable efficient long-range reasoning without sacrificing resolution, we introduce a multi-scale voxel Mamba module that refines the fused features by combining full-grid and chunk-wise sequence modeling. Experiments on unseen ShapeNet categories and ScanNet objects show that DinoComplete achieves stronger completion quality than prior deterministic and generative based completion methods while using fewer parameters, requiring lower memory, and achieving faster inference. Our results demonstrate that distilling semantic priors from visual foundation models improves generalization and robustness in 3D shape completion.

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 / 2 minor

Summary. The paper introduces DinoComplete, a deterministic 3D shape completion method that first constructs multi-view DINO feature volumes from complete ShapeNet data, trains a student network to regress these features from partial inputs, and then fuses the predicted semantic features with geometry inside a multi-scale voxel Mamba module for long-range reasoning. Experiments are reported to show improved completion quality over prior deterministic and generative baselines on unseen ShapeNet categories and ScanNet objects, together with lower parameter count, memory usage, and faster inference.

Significance. If the transfer of distilled DINO features to partial unseen inputs holds, the work would demonstrate a practical way to inject semantic context from 2D foundation models into efficient 3D completion pipelines, potentially reducing reliance on generative models while preserving resolution through state-space modeling. The efficiency claims (fewer parameters, lower memory, faster inference) would be a notable practical contribution if substantiated.

major comments (2)
  1. [Experiments section (unseen-category results)] The headline claim that distilled semantic priors improve generalization on unseen categories rests on the student network producing informative DINO features from partial inputs outside the training distribution. The manuscript should include an explicit ablation (e.g., in the experiments section) that isolates the semantic branch on the unseen ShapeNet and ScanNet test sets; without it, gains could be attributable solely to the voxel Mamba architecture.
  2. [§3 (method) and corresponding ablation table] The description of the multi-scale voxel Mamba states that geometric and semantic voxel representations are fused before refinement, yet no quantitative measure is provided of how much the semantic features actually contribute versus the Mamba layers alone on out-of-distribution data. A controlled comparison removing the semantic input on the same unseen splits would directly test the central hypothesis.
minor comments (2)
  1. [§3.2] Notation for the student network output and the voxel feature fusion step could be made more explicit to avoid ambiguity between the predicted DINO volume and the geometric occupancy grid.
  2. [Figure 4] Figure captions for the qualitative results on ScanNet should indicate whether the shown completions are from the full model or an ablated variant.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful review and the recommendation for major revision. The two major comments both correctly identify the absence of an explicit ablation that isolates the contribution of the distilled semantic features on unseen categories. We agree this controlled comparison is necessary to substantiate the central hypothesis and will add the requested experiments in the revised manuscript.

read point-by-point responses

  1. Referee: [Experiments section (unseen-category results)] The headline claim that distilled semantic priors improve generalization on unseen categories rests on the student network producing informative DINO features from partial inputs outside the training distribution. The manuscript should include an explicit ablation (e.g., in the experiments section) that isolates the semantic branch on the unseen ShapeNet and ScanNet test sets; without it, gains could be attributable solely to the voxel Mamba architecture.

    Authors: We agree that the current experimental section does not contain a direct ablation removing the semantic branch on the unseen splits. In the revision we will add this comparison: the full DinoComplete model versus an otherwise identical variant that receives only the geometric voxel input (i.e., the semantic student network and its features are disabled). Results will be reported on the same unseen ShapeNet categories and ScanNet objects using the same metrics, thereby isolating the contribution of the distilled DINO priors from the multi-scale voxel Mamba architecture. revision: yes

  2. Referee: [§3 (method) and corresponding ablation table] The description of the multi-scale voxel Mamba states that geometric and semantic voxel representations are fused before refinement, yet no quantitative measure is provided of how much the semantic features actually contribute versus the Mamba layers alone on out-of-distribution data. A controlled comparison removing the semantic input on the same unseen splits would directly test the central hypothesis.

    Authors: The referee is correct that no such controlled removal of the semantic input is currently quantified on out-of-distribution data. We will introduce the requested ablation table (or subsection) that reports performance of the geometric-only Mamba variant against the fused model on the identical unseen test sets. This will provide a direct quantitative measure of the semantic features' contribution under the exact conditions highlighted in the paper's claims. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation chain is self-contained and externally falsifiable

full rationale

The paper describes a pipeline that extracts DINO features from complete multi-view ShapeNet data, trains a student network to regress those features from partial inputs, and fuses the result with geometry inside a voxel Mamba module. None of these steps reduce to a self-definition, a fitted parameter renamed as a prediction, or a load-bearing self-citation. The central claim (superior completion on unseen categories) is presented as an empirical outcome measured against external benchmarks (ShapeNet unseen splits and ScanNet), not derived by algebraic identity from the inputs. No equations are shown that equate the output to the training targets by construction, and no uniqueness theorem or ansatz is imported from prior work by the same authors. The method therefore remains open to falsification by independent evaluation on the reported test sets.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on the untested transferability of DINO features to partial 3D inputs and on the effectiveness of the student network and Mamba fusion; no explicit free parameters or invented physical entities are named.

axioms (2)
  • domain assumption DINO features extracted from complete multi-view renders remain semantically meaningful when predicted from incomplete geometry.
    Central premise stated in the abstract for the distillation step.
  • domain assumption Voxel-aligned semantic and geometric features can be fused effectively by state-space sequence modeling.
    Assumed by the multi-scale voxel Mamba module description.

pith-pipeline@v0.9.1-grok · 5740 in / 1237 out tokens · 35963 ms · 2026-06-29T17:46:55.459691+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

49 extracted references · 4 canonical work pages · 3 internal anchors

  1. [1]

    Principal component analysis.Wiley interdisciplinary reviews: computational statistics, 2(4):433–459, 2010

    Hervé Abdi and Lynne J Williams. Principal component analysis.Wiley interdisciplinary reviews: computational statistics, 2(4):433–459, 2010

    2010

  2. [2]

    Scan2cad: Learning cad model alignment in rgb-d scans

    Armen Avetisyan, Manuel Dahnert, Angela Dai, Manolis Savva, Angel X Chang, and Matthias Nießner. Scan2cad: Learning cad model alignment in rgb-d scans. InProceedings of the IEEE/CVF Conference on computer vision and pattern recognition, pages 2614–2623, 2019

    2019

  3. [3]

    Dino-vo: A feature-based visual odometry leveraging a visual foundation model.IEEE Robotics and Automation Letters, 2025

    Maulana Bisyir Azhari and David Hyunchul Shim. Dino-vo: A feature-based visual odometry leveraging a visual foundation model.IEEE Robotics and Automation Letters, 2025

    2025

  4. [4]

    Layer Normalization

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization.arXiv preprint arXiv:1607.06450, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  5. [5]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 9650–9660, 2021

    2021

  6. [6]

    Chang, Angela Dai, Thomas Funkhouser, Maciej Halber, Matthias Niessner, Manolis Savva, Shuran Song, Andy Zeng, and Yinda Zhang

    Angel X. Chang, Angela Dai, Thomas Funkhouser, Maciej Halber, Matthias Niessner, Manolis Savva, Shuran Song, Andy Zeng, and Yinda Zhang. Matterport3d: Learning from rgb-d data in indoor environments. In3DV, 2017

    2017

  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

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  8. [8]

    A simple frame- work for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple frame- work for contrastive learning of visual representations. In Hal Daumé III and Aarti Singh, editors,Proceedings of the 37th International Conference on Machine Learning, volume 119 of Proceedings of Machine Learning Research, pages 1597–1607. PMLR, 13–18 Jul 2020

    2020

  9. [9]

    Implicit functions in feature space for 3d shape reconstruction and completion

    Julian Chibane, Thiemo Alldieck, and Gerard Pons-Moll. Implicit functions in feature space for 3d shape reconstruction and completion. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, jun 2020

    2020

  10. [10]

    Diffcomplete: Diffusion-based generative 3d shape completion

    Ruihang Chu, Enze Xie, Shentong Mo, Zhenguo Li, Matthias Nießner, Chi-Wing Fu, and Jiaya Jia. Diffcomplete: Diffusion-based generative 3d shape completion. InThirty-seventh Conference on Neural Information Processing Systems, 2023

    2023

  11. [11]

    Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, and Matthias Nießner

    Angela Dai, Angel X. Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, and Matthias Nießner. Scannet: Richly-annotated 3d reconstructions of indoor scenes. InCVPR, 2017

    2017

  12. [12]

    Bundlefusion: Real-time globally consistent 3d reconstruction using on-the-fly surface reinte- gration.ACM Transactions on Graphics (ToG), 36(4):1, 2017

    Angela Dai, Matthias Nießner, Michael Zollhöfer, Shahram Izadi, and Christian Theobalt. Bundlefusion: Real-time globally consistent 3d reconstruction using on-the-fly surface reinte- gration.ACM Transactions on Graphics (ToG), 36(4):1, 2017

    2017

  13. [13]

    3d shape completion using 3d-encoder-predictor cnns and shape synthesis

    Angela Dai, Daniel Ritchie, Martin Bokeloh, Scott Reed, Jürgen Sturm, and Matthias Nießner. 3d shape completion using 3d-encoder-predictor cnns and shape synthesis. InCVPR, 2017

    2017

  14. [14]

    Transformers are SSMs: Generalized models and efficient algorithms through structured state space duality

    Tri Dao and Albert Gu. Transformers are SSMs: Generalized models and efficient algorithms through structured state space duality. InForty-first International Conference on Machine Learning, 2024

    2024

  15. [15]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13142–13153, 2023

    2023

  16. [16]

    Bootstrap your own latent - a new approach to self-supervised learning

    Jean-Bastien Grill, Florian Strub, Florent Altché, Corentin Tallec, Pierre Richemond, Elena Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Guo, Mohammad Gheshlaghi Azar, Bilal Piot, koray kavukcuoglu, Remi Munos, and Michal Valko. Bootstrap your own latent - a new approach to self-supervised learning. In H. Larochelle, M. Ranzato, R. Hadsell, M....

    2020

  17. [17]

    Mamba: Linear-time sequence modeling with selective state spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. In First Conference on Language Modeling, 2024

    2024

  18. [18]

    Efficiently modeling long sequences with structured state spaces

    Albert Gu, Karan Goel, and Christopher Re. Efficiently modeling long sequences with structured state spaces. InInternational Conference on Learning Representations, 2022

    2022

  19. [19]

    Combining recurrent, convolutional, and continuous-time models with linear state space layers

    Albert Gu, Isys Johnson, Karan Goel, Khaled Kamal Saab, Tri Dao, Atri Rudra, and Christopher Re. Combining recurrent, convolutional, and continuous-time models with linear state space layers. In A. Beygelzimer, Y . Dauphin, P. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, 2021

    2021

  20. [20]

    Momentum Contrast for Unsupervised Visual Representation Learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum Contrast for Unsupervised Visual Representation Learning . In2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9726–9735, Los Alamitos, CA, USA, June 2020. IEEE Computer Society

    2020

  21. [21]

    Über die stetige abbildung einer linie auf ein flächenstück

    David Hilbert. Über die stetige abbildung einer linie auf ein flächenstück. InDritter Band: Analysis· Grundlagen der Mathematik· Physik Verschiedenes: Nebst Einer Lebensgeschichte, pages 1–2. Springer, 1935

    1935

  22. [22]

    Mu Hu, Wei Yin, Chi Zhang, Zhipeng Cai, Xiaoxiao Long, Hao Chen, Kaixuan Wang, Gang Yu, Chunhua Shen, and Shaojie Shen. Metric3d v2: A versatile monocular geometric foundation model for zero-shot metric depth and surface normal estimation.IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12):10579–10596, 2024

    2024

  23. [23]

    3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42(4), July 2023

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42(4), July 2023

    2023

  24. [24]

    Open- VLA: An open-source vision-language-action model

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan P Foster, Pannag R Sanketi, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. Open- VLA: An open-source vision-language-action model. In8th Annual Conference on Robot Learn...

    2024

  25. [25]

    DINO in the room: Leveraging 2D foundation models for 3D segmentation

    Karim Knaebel, Kadir Yilmaz, Daan de Geus, Alexander Hermans, David Adrian, Timm Linder, and Bastian Leibe. DINO in the room: Leveraging 2D foundation models for 3D segmentation. In2026 International Conference on 3D Vision (3DV), 2026

    2026

  26. [26]

    Vision mamba: A comprehensive survey and taxonomy.IEEE Transactions on Neural Networks and Learning Systems, 2025

    Xiao Liu, Chenxu Zhang, Fuxiang Huang, Shuyin Xia, Guoyin Wang, and Lei Zhang. Vision mamba: A comprehensive survey and taxonomy.IEEE Transactions on Neural Networks and Learning Systems, 2025

    2025

  27. [27]

    Repaint: Inpainting using denoising diffusion probabilistic models

    Andreas Lugmayr, Martin Danelljan, Andres Romero, Fisher Yu, Radu Timofte, and Luc Van Gool. Repaint: Inpainting using denoising diffusion probabilistic models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11461–11471, 2022

    2022

  28. [28]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoor- thi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 65(1):99–106, 2021

    2021

  29. [29]

    Autosdf: Shape priors for 3d completion, reconstruction and generation

    Paritosh Mittal, Yen-Chi Cheng, Maneesh Singh, and Shubham Tulsiani. Autosdf: Shape priors for 3d completion, reconstruction and generation. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 306–315. IEEE, 2022

    2022

  30. [30]

    Newcombe, Shahram Izadi, Otmar Hilliges, David Molyneaux, David Kim, An- drew J

    Richard A. Newcombe, Shahram Izadi, Otmar Hilliges, David Molyneaux, David Kim, An- drew J. Davison, Pushmeet Kohli, Jamie Shotton, Steve Hodges, and Andrew Fitzgibbon. Kinectfusion: Real-time dense surface mapping and tracking. InIEEE International Symposium on Mixed and Augmented Reality (ISMAR), 2011. 11

    2011

  31. [31]

    Real-time 3d reconstruction at scale using voxel hashing.ACM Transactions on Graphics (ToG), 32(6):1–11, 2013

    Matthias Nießner, Michael Zollhöfer, Shahram Izadi, and Marc Stamminger. Real-time 3d reconstruction at scale using voxel hashing.ACM Transactions on Graphics (ToG), 32(6):1–11, 2013

    2013

  32. [32]

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herve Jegou, Julien Mairal, Patrick L...

    2024

  33. [33]

    Patchcomplete: Learning multi-resolution patch priors for 3d shape completion on unseen categories

    Yuchen Rao, Yinyu Nie, and Angela Dai. Patchcomplete: Learning multi-resolution patch priors for 3d shape completion on unseen categories. In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho, editors,Advances in Neural Information Processing Systems, 2022

    2022

  34. [34]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InInternational Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer, 2015

    2015

  35. [35]

    Sc-diff: 3d shape completion with latent diffusion models.arXiv preprint arXiv:2403.12470, 2024

    Simon Schaefer, Juan D Galvis, Xingxing Zuo, and Stefan Leutengger. Sc-diff: 3d shape completion with latent diffusion models.arXiv preprint arXiv:2403.12470, 2024

    work page arXiv 2024

  36. [36]

    Structure-from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure-from-motion revisited. InProceed- ings of the IEEE conference on computer vision and pattern recognition, pages 4104–4113, 2016

    2016

  37. [37]

    Oriane Siméoni, Huy V V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Michaël Ramamonjisoa, et al. Dinov3. arXiv preprint arXiv:2508.10104, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  38. [38]

    Lichtenberg, and Jianxiong Xiao

    Shuran Song, Samuel P. Lichtenberg, and Jianxiong Xiao. Sun rgb-d: A rgb-d scene understand- ing benchmark suite. InCVPR, 2015

    2015

  39. [39]

    Vggt: Visual geometry grounded transformer

    Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. Vggt: Visual geometry grounded transformer. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 5294–5306, 2025

    2025

  40. [40]

    Autorecon: Automated 3d object discovery and reconstruction

    Peng Wang et al. Autorecon: Automated 3d object discovery and reconstruction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  41. [41]

    Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision

    Ruicheng Wang, Sicheng Xu, Cassie Dai, Jianfeng Xiang, Yu Deng, Xin Tong, and Jiaolong Yang. Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5261–5271, 2025

    2025

  42. [42]

    Moge-2: Accurate monocular geometry with metric scale and sharp details

    Ruicheng Wang, Sicheng Xu, Yue Dong, Yu Deng, Jianfeng Xiang, Zelong Lv, Guangzhong Sun, Xin Tong, and Jiaolong Yang. Moge-2: Accurate monocular geometry with metric scale and sharp details. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2026

    2026

  43. [43]

    Few-shot 3d shape completion

    Yifan Wang, Dragomir Anguelov, Xin Tong, and Angela Dai. Few-shot 3d shape completion. InECCV, 2020

    2020

  44. [44]

    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. InProceedings of the IEEE/CVF international conference on computer vision, pages 12498–12507, 2021

    2021

  45. [45]

    AdaPoinTr: Diverse Point Cloud Completion With Adaptive Geometry-Aware Transformers .IEEE Transactions on Pattern Analysis & Machine Intelligence, 45(12):14114–14130, December 2023

    Xumin Yu, Yongming Rao, Ziyi Wang, Jiwen Lu, and Jie Zhou. AdaPoinTr: Diverse Point Cloud Completion With Adaptive Geometry-Aware Transformers .IEEE Transactions on Pattern Analysis & Machine Intelligence, 45(12):14114–14130, December 2023. 12

    2023

  46. [46]

    Pcn: Point completion network

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

    2018

  47. [47]

    Latent uncertainty-aware multi-view sdf scan completion

    Faezeh Zakeri, Lukas Ruppert, Raphael Braun, and Hendrik Lensch. Latent uncertainty-aware multi-view sdf scan completion. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 3556–3566, 2026

    2026

  48. [48]

    V oxel mamba: Group-free state space models for point cloud based 3d object detection.Advances in Neural Information Processing Systems, 37:81489–81509, 2024

    Guowen Zhang, Lue Fan, Chenhang He, Zhen Lei, Zhaoxiang Zhang, and Lei Zhang. V oxel mamba: Group-free state space models for point cloud based 3d object detection.Advances in Neural Information Processing Systems, 37:81489–81509, 2024

    2024

  49. [49]

    Sdf-stylegan: implicit sdf-based stylegan for 3d shape generation

    Xinyang Zheng, Yang Liu, Pengshuai Wang, and Xin Tong. Sdf-stylegan: implicit sdf-based stylegan for 3d shape generation. InComputer Graphics Forum, volume 41, pages 52–63. Wiley Online Library, 2022. 13 A Technical appendices and supplementary material A.1 TSDF-DINO Model In this section, we provide additional details on our TSDF-DINO model, including th...

    2022