Pith. sign in

REVIEW 3 major objections 5 minor 63 references

Unified Category-Level Object Detection and Pose Estimation from RGB Images using 3D Prototypes

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A single RGB model now performs category-level object detection and 6D pose estimation together, reporting a 22.9% average gain over prior two-stage systems on REAL275.

desk verdict A plausible unified RGB detection+pose framework worth a close look, but the visible text leaves the headline 22.9% gain and the mean-prototype limitation unverified. read the letter →

arxiv 2508.02157 v1 pith:5MN2H33B submitted 2025-08-04 cs.CV

Tom Fischer , Xiaojie Zhang , Eddy Ilg This is my paper
classification cs.CV
keywords category-level6DobjectposeestimationRGB-onlyneuralmeshmodels3Dprototypesmulti-modelRANSACREAL275benchmarkunifieddetectionanddense2D-3Dcorrespondences
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper aims to show that category-level object detection and 6D pose estimation — recognizing an object as a member of a known category and finding its position and orientation, even for instances never seen before — can be done by one unified model working on ordinary RGB images. This is in contrast to existing RGB approaches, which first detect each object with one model and then estimate its pose with a separate model; the paper argues that this separation creates avoidable failure modes. It reports a 22.9% average improvement over the prior state of the art on the REAL275 benchmark across all scale-agnostic metrics, and additional experiments suggest the unified model degrades more gracefully under image corruption than single-stage baselines. If the claim holds, category-level 3D understanding could be obtained from any monocular camera, which matters for robotics, autonomous driving, and augmented reality.

What carries the argument

The load-bearing object is the neural mesh model: a category-level 3D prototype mesh with learned neural features attached to its surface, paired with a 2D image feature extractor trained on the same representation. Dense 2D-3D correspondences come from comparing image features with prototype surface features, and these correspondences simultaneously localize objects and constrain their pose. Multi-model RANSAC, a variant of random sample consensus that fits several geometric models at once, turns the noisy correspondence set into pose hypotheses for multiple instances, while per-instance scale prediction and foreground modeling refine which matches count. Because detection and pose are read from the same geometric matches, no separate detection stage or per-object crop is needed.

What would settle it

Evaluate a trained model on a category with large within-category shape variation, such as chairs, and group pose errors by how far each instance's shape deviates from the category mean prototype; if instances far from the prototype show a sharp accuracy drop, the single-prototype assumption is the limiting factor, and the central mechanism fails on such shapes.

Watch

Extended reading notes

Core claim

The central claim is that a category can be represented by a neural mesh prototype — a mean 3D shape whose surface carries learned neural features — and that matching image features to that surface produces dense 2D-3D correspondences from which both detection and pose can be read off in a single forward pass. The model predicts per-instance scale to adapt the mean prototype, filters noisy matches with foreground modeling, and uses multi-model RANSAC to fit pose hypotheses for all object instances in the scene at once. The authors state this is the first time an RGB-only method has unified the two tasks in a single model, and they claim the unification is not merely simpler but more accurate, beating all compared two-stage RGB pipelines on REAL275 with a 22.9% average gain across scale-agnostic metrics.

Load-bearing premise

The method assumes that a single mean-shape 3D prototype per category, adjusted only by per-instance scale, can generate reliable image-to-mesh correspondences for every object in that category.

Editorial extensions

If this is right

  • Category-level 6D pose estimation becomes possible from plain RGB video with a single network, removing the detector-crop bottleneck and the need for depth sensors.
  • The reported 22.9% average gain on REAL275 implies that two-stage RGB pipelines leave substantial accuracy on the table that can be recovered by sharing one representation across both tasks.
  • The corruption experiments — noise, blur, compression, fog, and similar degradations — suggest the unified model degrades more gracefully than single-stage baselines in real-world conditions.
  • One model must be trained and maintained instead of a detector plus a separate per-object pose estimator, which simplifies deployment in robotics and AR systems.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An untested next step would be conditioning the prototype on shape, since the single mean prototype could become the limiting factor for categories with large within-category shape variation, such as chairs or tables.
  • Because detection and pose share the same correspondences, the model could likely be extended to temporal tracking by propagating 2D-3D matches across video frames, a use case the paper does not address.
  • The corruption results suggest a testable extension for robot manipulation: measuring whether the unified model improves grasp success under lighting and sensor noise compared with two-stage pipelines, rather than measuring pose error alone.
  • The method's reliance on learned prototype features implies the same framework might transfer to categories without available CAD meshes if prototypes are generated from predicted shape reconstructions, though the paper does not explore this.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a unified model for category-level object detection and 6D pose estimation from RGB images, using neural mesh models as 3D prototypes with learned features and multi-model RANSAC. The authors claim that this is the first single-model RGB method for both tasks, report a 22.9% average improvement over prior state-of-the-art on REAL275 across all scale-agnostic metrics, and show greater robustness against image corruptions. The manuscript as provided contains only the abstract, a partial introduction, a reference list, and several figure captions; the method, experiments, and ablations are absent.

Significance. If the claims are substantiated, the unified single-model approach would be a meaningful advance for RGB-only category-level pose estimation, simplifying pipelines and potentially improving accuracy by avoiding error propagation between detection and pose stages. The reported 22.9% improvement on REAL275 is substantial, and the use of neural mesh models with multi-model RANSAC for joint detection and pose is an interesting direction. The paper also makes a clear novelty claim: no prior RGB method integrates detection and pose in a single model. However, the significance currently cannot be assessed because the manuscript omits the methodological and experimental core; moreover, the novelty claim appears to conflict with at least one cited prior work. The potential value is real, but verification is impossible without the missing sections.

major comments (3)
  1. [Overall structure; Section 1] The manuscript as submitted jumps from Section 1 directly to the references; there is no method section, no experimental protocol, no baseline tables, no ablations, and no error bars. The central empirical claim of a 22.9% average improvement on REAL275 appears only in the abstract and in the caption of Figure 8, with no quantitative results in the text. This is load-bearing: the claim cannot be verified or reproduced from the provided content. The authors must include the full method description, evaluation setup, and complete results.
  2. [Section 1] The statement that 'All current RGB methods fall into this category of two-stage methods and use separate models per task' is contradicted by reference [36], Lin et al., 'Single-stage keypoint-based category-level object pose estimation from an RGB image.' Since [36] is in the reference list, the novelty claim of 'for the first time' needs to be reconciled with this prior work, either by explaining why [36] does not constitute a unified detection-and-pose model or by revising the claim accordingly.
  3. [Section 1; Figure 7] The method appears to estimate only a per-instance scale for a category-level mean-shape prototype, with no mechanism for handling intra-class shape variation. REAL275 contains objects such as cameras, mugs, and bottles with substantial shape differences from a mean prototype. The paper does not describe any deformation mechanism or provide an analysis of how prototype bias affects dense 2D-3D correspondences. Without such an analysis or an explicit shape-deformation component, the claimed accuracy on instances far from the prototype is not well supported; an ablation or error-vs-shape-deviation study is needed.
minor comments (5)
  1. [Figure 1 caption] The caption contains a typo: 'object detection and and single-object pose estimation' repeats the word 'and.'
  2. [Section 1] The introduction text appears to be truncated: the sentence ending 'where a 2D feature extractor for images and the neural features on the' is cut off, and the rest of the manuscript is missing. The submission should contain the full text.
  3. [References] Several references in the list (e.g., [36]) are not cited in the visible portion of the text; the authors should ensure every reference is cited in the body and conversely every citation appears in the list.
  4. [Figure 8 caption] The caption states 'We show mean Average Precision (mAP) on REAL275 using scale-agnostic metrics,' while the abstract reports '22.9% averaged across all scale-agnostic metrics.' Please clarify whether Figure 8 shows the aggregated metric or the individual scale-agnostic metrics, and how the average is computed.
  5. [Abstract] The GitHub link is given as plain text; consider adding the repository as a footnote or in the references for accessibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the method is an application of existing neural mesh representations with external benchmark validation.

full rationale

The paper's central claim is a new architecture-level contribution: it couples a neural mesh prototype representation (from prior work [2,40]) with multi-model RANSAC to perform detection and pose estimation jointly from RGB. I examined the visible text for places where a stated output is fed back as an input or where a fitted parameter is renamed as a prediction. No equation in the provided text defines the pose or detection result in terms of the reported SOTA numbers; the 22.9% improvement is an externally measured comparison on REAL275 against OLD-Net, DMSR, and LaPose. The citations to NeMo ([2]) and iNeMo ([16]) supply the 3D representation and learned-feature machinery; they are used as building blocks, not as a uniqueness claim or as justification that the unified system must work. Figure 7 shows qualitative correspondences, and the benchmark numbers are independent of those citations. The weakest assumption (mean-shape prototype may induce biased correspondences for high shape variance) is a correctness and generalization risk, not circularity. Accordingly, no step reduces the claimed derivation to its inputs.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical or conceptual entities; it combines existing neural mesh models, learned features, and multi-model RANSAC. The main unstated burden is the shape-prototype assumption and the reliance on learned correspondences, both of which are domain assumptions rather than invented entities.

free parameters (1)
  • Multi-model RANSAC hyperparameters = not stated in available text
    Multi-model RANSAC requires choices such as inlier thresholds, iteration counts, and confidence settings. These are typically tuned on validation data and are not described in the reviewed excerpt, yet they directly affect detection and pose hypotheses.
assumptions (3)
  • domain assumption Category-level objects can be represented by a shared 3D prototype mesh with per-instance scale variation.
    The method 'leverages neural mesh models as a 3D representation of object categories.' If instances within a category deviate too much from the mean prototype shape, the dense correspondences and the resulting poses will be biased. This enters in Section 1.
  • domain assumption Learned visual features provide sufficiently accurate dense 2D-3D correspondences across object categories.
    The pipeline matches image features to neural features on the mesh. The success of both detection and pose depends on these correspondences being reliable. This is invoked throughout the method description, including the references to DINOv2 and prior neural mesh models.
  • standard math RANSAC combined with PnP or least-squares fitting recovers correct 6D poses from noisy correspondences.
    Multi-model RANSAC is a standard robust fitting technique; the paper relies on it to simultaneously find object instances and their poses. This is a standard mathematical tool with well-known assumptions about outlier ratios and initialization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unified Category-Level Object Detection and Pose Estimation from RGB Images using 3D Prototypes." pith.science (2026). https://pith.science/paper/5MN2H33B

@misc{pith2026250802157,
  author       = {Pith},
  title        = {Pith review of: Unified Category-Level Object Detection and Pose Estimation from RGB Images using 3D Prototypes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5MN2H33B}},
  note         = {Machine review of arXiv:2508.02157}
}
read the original abstract

Recognizing objects in images is a fundamental problem in computer vision. Although detecting objects in 2D images is common, many applications require determining their pose in 3D space. Traditional category-level methods rely on RGB-D inputs, which may not always be available, or employ two-stage approaches that use separate models and representations for detection and pose estimation. For the first time, we introduce a unified model that integrates detection and pose estimation into a single framework for RGB images by leveraging neural mesh models with learned features and multi-model RANSAC. Our approach achieves state-of-the-art results for RGB category-level pose estimation on REAL275, improving on the current state-of-the-art by 22.9% averaged across all scale-agnostic metrics. Finally, we demonstrate that our unified method exhibits greater robustness compared to single-stage baselines. Our code and models are available at https://github.com/Fischer-Tom/unified-detection-and-pose-estimation.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 49 canonical work pages

  1. [36]

    Single-stage keypoint-based category-level object pose estimation from an rgb image

    Yunzhi Lin, Jonathan Tremblay, Stephen Tyree, Patricio A Vela, and Stan Birchfield. Single-stage keypoint-based category-level object pose estimation from an rgb image. In 2022 International Conference on Robotics and Automation (ICRA), pages 1547–1553. IEEE, 2022. 1, 3

    work page 2022

  2. [1]

    Deep vit features as dense visual descriptors

    Shir Amir, Yossi Gandelsman, Shai Bagon, and Tali Dekel. Deep vit features as dense visual descriptors. arXiv preprint arXiv:2112.05814, 2(3):4, 2021. 4

    arXiv 2021

  3. [2]

    Nemo: Neural mesh models of contrastive features for robust 3d pose estimation

    Wang Angtian, Adam Kortylewski, and Alan Yuille. Nemo: Neural mesh models of contrastive features for robust 3d pose estimation. In Proceedings International Conference on Learning Representations (ICLR), 2021. 1, 2, 3, 4

    work page 2021

  4. [3]

    Coke: Contrastive learning for robust keypoint detec- tion

    Yutong Bai, Angtian Wang, Adam Kortylewski, and Alan Yuille. Coke: Contrastive learning for robust keypoint detec- tion. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 65–74, 2023. 4

    work page 2023

  5. [4]

    Graph-cut RANSAC

    Daniel Barath and Jiri Matas. Graph-cut RANSAC. In CVPR, 2018. 5

    work page 2018

  6. [5]

    Progressive-x: Efficient, anytime, multi-model fitting algorithm

    Daniel Barath and Jiri Matas. Progressive-x: Efficient, anytime, multi-model fitting algorithm. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3780–3788, 2019. 2, 5, 7

    work page 2019

  7. [6]

    Point pair features based object detection and pose estimation revisited

    Tolga Birdal and Slobodan Ilic. Point pair features based object detection and pose estimation revisited. In 2015 In- ternational conference on 3D vision, pages 527–535. IEEE,

    work page 2015

  8. [7]

    Efficientpose: An efficient, accurate and scalable end-to-end 6d multi object pose estimation approach

    Yannick Bukschat and Marcus Vetter. Efficientpose: An efficient, accurate and scalable end-to-end 6d multi object pose estimation approach. arXiv preprint arXiv:2011.04307,

    arXiv 2011

Show all 63 references
  1. [8]

    End-to- end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to- end object detection with transformers. In European confer- ence on computer vision, pages 213–229. Springer, 2020. 2

    2020

  2. [9]

    Sgpa: Structure-guided prior adapta- tion for category-level 6d object pose estimation

    Kai Chen and Qi Dou. Sgpa: Structure-guided prior adapta- tion for category-level 6d object pose estimation. InProceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 2773–2782, 2021. 2

    2021

  3. [10]

    Adaptformer: Adapting vision transformers for scalable visual recognition

    Shoufa Chen, Chongjian Ge, Zhan Tong, Jiangliu Wang, Yibing Song, Jue Wang, and Ping Luo. Adaptformer: Adapting vision transformers for scalable visual recognition. NeurIPS, 35:16664–16678, 2022. 4

    2022

  4. [11]

    Pointposenet: Point pose network for robust 6d object pose estimation

    Wei Chen, Jinming Duan, Hector Basevi, Hyung Jin Chang, and Ales Leonardis. Pointposenet: Point pose network for robust 6d object pose estimation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 2824–2833, 2020. 3

    2020

  5. [12]

    Secondpose: Se (3)- consistent dual-stream feature fusion for category-level pose estimation

    Yamei Chen, Yan Di, Guangyao Zhai, Fabian Manhardt, Chenyangguang Zhang, Ruida Zhang, Federico Tombari, Nassir Navab, and Benjamin Busam. Secondpose: Se (3)- consistent dual-stream feature fusion for category-level pose estimation. In Proceedings of the IEEE/CVF Conference on ...

    2024

  6. [13]

    Masked-attention mask transformer for universal image segmentation

    Bowen Cheng, Ishan Misra, Alexander G Schwing, Alexan- der Kirillov, and Rohit Girdhar. Masked-attention mask transformer for universal image segmentation. In Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1290–1299, 2022. 5

    2022

  7. [14]

    3d pose esti- mation of daily objects using an rgb-d camera

    Changhyun Choi and Henrik I Christensen. 3d pose esti- mation of daily objects using an rgb-d camera. In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 3342–3349. IEEE, 2012. 2

    2012

  8. [15]

    Object level depth recon- struction for category level 6d object pose estimation from monocular rgb image

    Zhaoxin Fan, Zhenbo Song, Jian Xu, Zhicheng Wang, Kejian Wu, Hongyan Liu, and Jun He. Object level depth recon- struction for category level 6d object pose estimation from monocular rgb image. In European Conference on Computer Vision, pages 220–236. Springer, 2022. 1, 2, 6, 7...

    2022

  9. [16]

    inemo: Incremental neural mesh models for robust class-incremental learning

    Tom Fischer, Yaoyao Liu, Artur Jesslen, Noor Ahmed, Prakhar Kaushik, Angtian Wang, Alan Yuille, Adam Ko- rtylewski, and Eddy Ilg. inemo: Incremental neural mesh models for robust class-incremental learning. In ECCV,

  10. [17]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography.Communications of the ACM, 24(6):381–395, 1981

    Martin A Fischler and Robert C Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography.Communications of the ACM, 24(6):381–395, 1981. 5

    1981

  11. [18]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Gir- shick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017. 2, 5, 7

    2017

  12. [19]

    Pvn3d: A deep point-wise 3d keypoints voting network for 6dof pose estimation

    Yisheng He, Wei Sun, Haibin Huang, Jianran Liu, Haoqiang Fan, and Jian Sun. Pvn3d: A deep point-wise 3d keypoints voting network for 6dof pose estimation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11632–11641, 2020. 1, 2

    2020

  13. [20]

    Benchmarking neural network robustness to common corruptions and sur- face variations

    Dan Hendrycks and Thomas G Dietterich. Benchmarking neural network robustness to common corruptions and sur- face variations. arXiv preprint arXiv:1807.01697, 2018. 7, 13

    2018 arXiv

  14. [21]

    A direct least- squares (dls) method for pnp

    Joel A Hesch and Stergios I Roumeliotis. A direct least- squares (dls) method for pnp. In 2011 International Confer- ence on Computer Vision, pages 383–390. IEEE, 2011. 2

    2011

  15. [22]

    Segmentation-driven 6d object pose estimation

    Yinlin Hu, Joachim Hugonot, Pascal Fua, and Mathieu Salz- mann. Segmentation-driven 6d object pose estimation. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 3385–3394, 2019. 3

    2019

  16. [23]

    Single-stage 6d object pose estimation

    Yinlin Hu, Pascal Fua, Wei Wang, and Mathieu Salzmann. Single-stage 6d object pose estimation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2930–2939, 2020. 3

    2020

  17. [24]

    Centersnap: Single-shot multi-object 3d shape reconstruction and categorical 6d pose and size estimation

    Muhammad Zubair Irshad, Thomas Kollar, Michael Laskey, Kevin Stone, and Zsolt Kira. Centersnap: Single-shot multi-object 3d shape reconstruction and categorical 6d pose and size estimation. In 2022 International Conference on Robotics and Automation (ICRA) , pages 10632–10640....

    2022

  18. [25]

    Shapo: Im- plicit representations for multi-object shape, appearance, and pose optimization

    Muhammad Zubair Irshad, Sergey Zakharov, Rares Ambrus, Thomas Kollar, Zsolt Kira, and Adrien Gaidon. Shapo: Im- plicit representations for multi-object shape, appearance, and pose optimization. In European Conference on Computer Vi- sion, pages 275–292. Springer, 2022. 3

    2022

  19. [26]

    Novum: Neural object volumes for robust object classification

    Artur Jesslen, Guofeng Zhang, Angtian Wang, Wufei Ma, Alan Yuille, and Adam Kortylewski. Novum: Neural object volumes for robust object classification. In European Con- ference on Computer Vision, pages 264–281. Springer, 2025. 2, 3

    2025

  20. [27]

    Real-time perception meets reactive motion gener- ation

    Daniel Kappler, Franziska Meier, Jan Issac, Jim Main- price, Cristina Garcia Cifuentes, Manuel W ¨uthrich, Vin- cent Berenz, Stefan Schaal, Nathan Ratliff, and Jeannette Bohg. Real-time perception meets reactive motion gener- ation. IEEE Robotics and Automation Letters , 3(3):...

    2018

  21. [28]

    Pose estimation for an autonomous vehicle using monocular vision

    Nikunj Kothari, Misha Gupta, Leena Vachhani, and Hemen- dra Arya. Pose estimation for an autonomous vehicle using monocular vision. In 2017 Indian control conference (ICC), pages 424–431. IEEE, 2017. 1

    2017

  22. [29]

    Cosypose: Consistent multi-view multi-object 6d pose estimation

    Yann Labb ´e, Justin Carpentier, Mathieu Aubry, and Josef Sivic. Cosypose: Consistent multi-view multi-object 6d pose estimation. In Computer Vision–ECCV 2020: 16th Euro- pean Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part XVII 16 , pages 574–591. Springer, 2...

    2020

  23. [30]

    Category-level metric scale object shape and pose estimation

    Taeyeop Lee, Byeong-Uk Lee, Myungchul Kim, and In So Kweon. Category-level metric scale object shape and pose estimation. IEEE Robotics and Automation Letters , 6(4): 8575–8582, 2021. 1, 2, 6, 7, 14

    2021

  24. [31]

    Epnp: An accurate o(n) solution to the pnp problem

    Vincent Lepetit, Francesc Moreno-Noguer, and Pascal Fua. Epnp: An accurate o(n) solution to the pnp problem. Inter- national journal of computer vision , 81:155–166, 2009. 2, 5

    2009

  25. [32]

    Deepim: Deep iterative matching for 6d pose estimation

    Yi Li, Gu Wang, Xiangyang Ji, Yu Xiang, and Dieter Fox. Deepim: Deep iterative matching for 6d pose estimation. In Proceedings of the European conference on computer vision (ECCV), pages 683–698, 2018. 1, 2

    2018

  26. [33]

    Dualposenet: Category-level 6d object pose and size estimation using dual pose network with re- fined learning of pose consistency

    Jiehong Lin, Zewei Wei, Zhihao Li, Songcen Xu, Kui Jia, and Yuanqing Li. Dualposenet: Category-level 6d object pose and size estimation using dual pose network with re- fined learning of pose consistency. In Proceedings of the IEEE/CVF International Conference on Computer Visi...

    2021

  27. [34]

    Category-level 6d object pose and size estimation using self- supervised deep prior deformation networks

    Jiehong Lin, Zewei Wei, Changxing Ding, and Kui Jia. Category-level 6d object pose and size estimation using self- supervised deep prior deformation networks. In European Conference on Computer Vision , pages 19–34. Springer,

  28. [35]

    Instance-adaptive and geometric-aware keypoint learning for category-level 6d object pose estimation

    Xiao Lin, Wenfei Yang, Yuan Gao, and Tianzhu Zhang. Instance-adaptive and geometric-aware keypoint learning for category-level 6d object pose estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21040–21049, 2024. 1, 2

    2024

  29. [37]

    Deep learning-based object pose estimation: A comprehensive survey

    Jian Liu, Wei Sun, Hui Yang, Zhiwen Zeng, Chongpei Liu, Jin Zheng, Xingyu Liu, Hossein Rahmani, Nicu Sebe, and Ajmal Mian. Deep learning-based object pose estimation: A comprehensive survey. arXiv preprint arXiv:2405.07801,

  30. [38]

    Sgdr: Stochas- tic gradient descent with warm restarts

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochas- tic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983, 2016. 12

    2016 arXiv

  31. [39]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 12

    2017 arXiv

  32. [40]

    Robust category-level 6d pose estimation with coarse-to-fine rendering of neural features

    Wufei Ma, Angtian Wang, Alan Yuille, and Adam Ko- rtylewski. Robust category-level 6d pose estimation with coarse-to-fine rendering of neural features. In European Conference on Computer Vision , pages 492–508. Springer,

  33. [41]

    Pose estimation for augmented reality: a hands-on survey

    Eric Marchand, Hideaki Uchiyama, and Fabien Spindler. Pose estimation for augmented reality: a hands-on survey. IEEE transactions on visualization and computer graphics , 22(12):2633–2651, 2015. 1

    2015

  34. [42]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 4, 12

    2023 arXiv

  35. [43]

    Accelerating 3d deep learning with pytorch3d

    Nikhila Ravi, Jeremy Reizenstein, David Novotny, Tay- lor Gordon, Wan-Yen Lo, Justin Johnson, and Georgia Gkioxari. Accelerating 3d deep learning with pytorch3d. arXiv preprint arXiv:2007.08501, 2020. 12, 13

    2007 arXiv

  36. [44]

    Faster r-cnn: towards real-time object detection with re- gion proposal networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: towards real-time object detection with re- gion proposal networks. In Proceedings of the 28th In- ternational Conference on Neural Information Processing Systems-Volume 1, pages 91–99, 2015. 2

    2015

  37. [45]

    Fast 3d recognition and pose using the viewpoint feature histogram

    Radu Bogdan Rusu, Gary Bradski, Romain Thibaux, and John Hsu. Fast 3d recognition and pose using the viewpoint feature histogram. In 2010 IEEE/RSJ international confer- ence on intelligent robots and systems , pages 2155–2162. IEEE, 2010. 2

    2010

  38. [46]

    Deep multi-state object pose estimation for augmented reality assembly

    Yongzhi Su, Jason Rambach, Nareg Minaskan, Paul Lesur, Alain Pagani, and Didier Stricker. Deep multi-state object pose estimation for augmented reality assembly. In 2019 IEEE International Symposium on Mixed and Augmented Reality Adjunct (ISMAR-Adjunct) , pages 222–227. IEEE,

    2019

  39. [47]

    Zebrapose: Coarse to fine surface encod- ing for 6dof object pose estimation

    Yongzhi Su, Mahdi Saleh, Torben Fetzer, Jason Rambach, Nassir Navab, Benjamin Busam, Didier Stricker, and Fed- erico Tombari. Zebrapose: Coarse to fine surface encod- ing for 6dof object pose estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern ...

    2022

  40. [48]

    Generalised dice overlap as a deep learning loss function for highly unbalanced segmen- tations

    Carole H Sudre, Wenqi Li, Tom Vercauteren, Sebastien Ourselin, and M Jorge Cardoso. Generalised dice overlap as a deep learning loss function for highly unbalanced segmen- tations. In Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support...

    2017

  41. [49]

    Real-time seamless single shot 6d object pose prediction

    Bugra Tekin, Sudipta N Sinha, and Pascal Fua. Real-time seamless single shot 6d object pose prediction. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 292–301, 2018. 3

    2018

  42. [50]

    Cope: End-to-end trainable constant runtime object pose es- timation

    Stefan Thalhammer, Timothy Patten, and Markus Vincze. Cope: End-to-end trainable constant runtime object pose es- timation. In Proceedings of the IEEE/CVF Winter Confer- ence on Applications of Computer Vision, pages 2860–2870,

  43. [51]

    Shape prior deformation for categorical 6d object pose and size estima- tion

    Meng Tian, Marcelo H Ang, and Gim Hee Lee. Shape prior deformation for categorical 6d object pose and size estima- tion. In Computer Vision–ECCV 2020: 16th European Con- ference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXI 16, pages 530–546. Springer, 2020. 2, 6, 7

    2020

  44. [52]

    Least-squares estimation of transforma- tion parameters between two point patterns

    Shinji Umeyama. Least-squares estimation of transforma- tion parameters between two point patterns. IEEE Transac- tions on Pattern Analysis & Machine Intelligence , 13(04): 376–380, 1991. 2

    1991

  45. [53]

    Socs: Semantically- aware object coordinate space for category-level 6d object pose estimation under large shape variations

    Boyan Wan, Yifei Shi, and Kai Xu. Socs: Semantically- aware object coordinate space for category-level 6d object pose estimation under large shape variations. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 14065–14074, 2023. 2

    2023

  46. [54]

    Normalized object coordinate space for category-level 6d object pose and size estimation

    He Wang, Srinath Sridhar, Jingwei Huang, Julien Valentin, Shuran Song, and Leonidas J Guibas. Normalized object coordinate space for category-level 6d object pose and size estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages ...

    2019

  47. [55]

    Rgb-based category-level object pose estimation via decoupled metric scale recovery

    Jiaxin Wei, Xibin Song, Weizhe Liu, Laurent Kneip, Hong- dong Li, and Pan Ji. Rgb-based category-level object pose estimation via decoupled metric scale recovery. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 2036–2042. IEEE, 2024. 1, 2, 6, 7, ...

    2024

  48. [56]

    Catgrasp: Learning category-level task-relevant grasping in clutter from simulation

    Bowen Wen, Wenzhao Lian, Kostas Bekris, and Stefan Schaal. Catgrasp: Learning category-level task-relevant grasping in clutter from simulation. In 2022 International Conference on Robotics and Automation (ICRA) , pages 6401–6408. IEEE, 2022. 1

    2022

  49. [57]

    You only demonstrate once: Category-level ma- nipulation from single visual demonstration

    Bowen Wen, Wenzhao Lian, Kostas Bekris, and Stefan Schaal. You only demonstrate once: Category-level ma- nipulation from single visual demonstration. arXiv preprint arXiv:2201.12716, 2022. 1

    2022 arXiv

  50. [58]

    Posecnn: A convolutional neural network for 6d object pose estimation in cluttered scenes

    Yu Xiang, Tanner Schmidt, Venkatraman Narayanan, and Dieter Fox. Posecnn: A convolutional neural network for 6d object pose estimation in cluttered scenes. arXiv preprint arXiv:1711.00199, 2017. 3

    2017 arXiv

  51. [59]

    Parameter-efficient fine-tuning for pre-trained vision models: A survey

    Yi Xin, Siqi Luo, Haodi Zhou, Junlong Du, Xiaohong Liu, Yue Fan, Qing Li, and Yuntao Du. Parameter-efficient fine-tuning for pre-trained vision models: A survey. arXiv preprint arXiv:2402.02242, 2024. 4

    2024

  52. [60]

    Dpod: 6d pose object detector and refiner

    Sergey Zakharov, Ivan Shugurov, and Slobodan Ilic. Dpod: 6d pose object detector and refiner. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1941–1950, 2019. 3

    1941

  53. [61]

    Genpose: gen- erative category-level object pose estimation via diffusion models

    Jiyao Zhang, Mingdong Wu, and Hao Dong. Genpose: gen- erative category-level object pose estimation via diffusion models. In Proceedings of the 37th International Conference on Neural Information Processing Systems , pages 54627– 54644, 2023. 1, 2

    2023

  54. [62]

    Lapose: Laplacian mixture shape modeling for rgb-based category-level object pose estimation

    Ruida Zhang, Ziqin Huang, Gu Wang, Chenyangguang Zhang, Yan Di, Xingxing Zuo, Jiwen Tang, and Xiangyang Ji. Lapose: Laplacian mixture shape modeling for rgb-based category-level object pose estimation. In European Confer- ence on Computer Vision, pages 467–484. Springer, 2024....

    2024

  55. [63]

    Deep fusion transformer network with weighted vector-wise keypoints voting for robust 6d object pose estimation

    Jun Zhou, Kai Chen, Linlin Xu, Qi Dou, and Jing Qin. Deep fusion transformer network with weighted vector-wise keypoints voting for robust 6d object pose estimation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13967–13977, 2023. 1, 2 Gauss...

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.