Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Accurate and efficient zero-shot 6D pose estimation with frozen foundation models

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Task-specific training is not necessary: frozen foundation models deliver state-of-the-art 6D pose estimation for unseen objects.

desk verdict A solid second-gen engineering paper whose BOP numbers are credible, but the 'training-free' label is softer than it looks and the SOTA claim rests on trained segmentation priors. read the letter →

arxiv 2506.09784 v1 pith:SHFQE5XZ submitted 2025-06-11 cs.CV

Andrea Caraffa , Davide Boscaini , Fabio Poiesi This is my paper
classification cs.CV
keywords 6Dobjectposeestimationzero-shottraining-freefoundationmodelsDINOv2GeDiBOPbenchmarkRGBD
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

FreeZeV2 sets out to answer a direct question: must a 6D pose estimator be trained on task-specific pose data to recognize and localize objects it has never seen? The paper's answer is no. It builds a pipeline entirely from frozen, off-the-shelf foundation models, DINOv2 for visual features and GeDi for geometric features, and shows that this training-free recipe achieves state-of-the-art results on the seven core BOP datasets, with 74.3 average recall in 1.4 seconds per image using SAM-6D masks, and 80.1 average recall when an ensemble of four segmentors is used. If correct, this means the expensive synthetic-data training regimes of current zero-shot methods are not a prerequisite for strong generalization, and pose estimation for novel objects can be both accurate and practical.

What carries the argument

The load-bearing machinery is the fusion of two frozen foundation models into one descriptor: DINOv2, a self-supervised vision transformer whose patch-level features are semantically rich and spatially precise, and GeDi, a geometric encoder that produces rotation-invariant local descriptors. They are L2-normalized, PCA-aligned, and concatenated to form a fused feature point cloud for the query model and for the scene. That fused space carries the whole argument: correspondences are nearest neighbors in it, RANSAC hypotheses are scored by feature similarity within it, and the final ranking combines its coarse and fine similarity with an ICP inlier ratio. The second mechanism is modularity: multiple zero-shot segmentation models propose candidate masks independently, and every mask is processed and ranked by the same pose score, turning segmentation uncertainty from a failure mode into a source of candidate poses.

What would settle it

Take FreeZeV2 through a set of test scenes where CNOS, SAM-6D, NIDS, and MUSE all fail to produce a mask covering the target object, for example a heavily occluded object or an appearance far from the rendered templates. If the method still recovers a correct pose, the stated segmentation dependence is wrong; if it fails exactly when masks fail, the dependence is confirmed. A cleaner quantitative version is to compare FreeZeV2's average recall using ground-truth masks against its average recall using zero-shot masks on the same BOP images, since the gap directly measures how much of the reported accuracy is carried by the segmentation prior rather than by the pose pipeline.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that a sparse-to-dense matching scheme over fused visual and geometric features reproduces or beats the accuracy of training-based competitors while being far faster. The query object's 3D model is densely encoded offline: surface points carry DINOv2 visual features aggregated over rendered views, concatenated with rotation-invariant GeDi geometric descriptors. At test time, only a sparse set of up to 256 points inside each candidate segmentation mask is encoded, and those sparse target features are matched to the dense query set by cosine similarity, with top-k nearest neighbors feeding RANSAC-based 3D registration. A feature-aware score that multiplies coarse feature similarity, refined feature similarity, and the ICP inlier ratio selects the final pose, and because this score is independent of the segmentor's confidence, accurate poses can be recovered even from low-confidence masks. The result is a new state of the art on the BOP benchmark in both 6D localization and 6D detection for unseen objects.

Load-bearing premise

The whole pipeline depends on at least one candidate segmentation mask actually covering the target object; if every zero-shot segmentor misses or badly under-segments it, the subsequent feature matching has no correct region to work with and no scoring can recover the pose.

Editorial extensions

If this is right

  • Large-scale task-specific synthetic training datasets are not required for strong zero-shot 6D pose estimation; frozen general-purpose features suffice when matched sparsely to dense model features.
  • Pose estimation for unseen objects can run at about 0.7 frames per second (1.4 seconds per image) on a single A40 GPU, bringing zero-shot pose estimation closer to real-time robotics and augmented-reality use.
  • Because the final pose score is segmentation-agnostic, the paper's results show accuracy rising from 74.3 to 80.1 average recall when four segmentors are ensembled, without task-specific fusion learning.
  • The same pipeline handles 6D detection, where instance count and identity are unknown, by only changing the number of candidate masks, so one architecture covers both settings.
  • The slower high-accuracy variant, at 82.1 average recall in 24.8 seconds, is still faster than the leading training-based baseline FoundationPose at 29.3 seconds, suggesting training-free methods no longer trade accuracy for speed.

Reading between the lines

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

  • Editorial inference: the paper implies that the remaining bottleneck for zero-shot 6D pose is object localization, not pose estimation; improving zero-shot segmentation should translate almost directly into pose accuracy once the pose pipeline is fixed.
  • Editorial inference: the same sparse-to-dense fused-feature recipe may transfer to other correspondence tasks such as point-cloud registration or open-vocabulary detection, because neither the encoder choice nor the scoring function is pose-specific.
  • Editorial inference: a direct testable extension would swap DINOv2 and GeDi for other frozen encoders; if accuracy holds across encoder families, the method's generality comes from the matching-plus-scoring structure rather than from these two specific models.
  • Editorial inference: the reported speedup comes largely from extracting features at only 256 sparse points per mask, so scaling the sparse grid or the number of masks trades runtime against accuracy in a predictable way that could be tuned per application.
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. FreeZeV2 is a zero-shot RGB-D 6D pose estimation method that uses frozen foundation models (DINOv2 for visual features, GeDi for geometric features) and requires no task-specific training of the pose module. The pipeline precomputes dense fused features on the query 3D model, extracts sparse visual and geometric features from the masked image region, establishes top-k correspondences, generates coarse hypotheses by RANSAC-based 3D registration, refines with ICP, and ranks candidates with a product score combining feature similarity and ICP inlier ratio (Eq. 7). The method supports ensembles of off-the-shelf segmentors. On the seven core BOP datasets, the paper reports 71.1 AR at 1.5 s with CNOS masks, 74.3 AR at 1.4 s with SAM-6D masks, 74.7 AR with MUSE masks, and 80.1 AR with an ensemble of four segmentors, with a FreeZeV2-Accurate variant reaching 82.1 AR. The paper claims a new state-of-the-art in zero-shot 6D pose estimation of unseen objects and reports having won the Best Overall Method award at the BOP Challenge 2024.

Significance. The paper demonstrates that a pose-estimation module built entirely from frozen, general-purpose features can be competitive with task-trained methods, which is an important result for the zero-shot pose literature. The sparse-to-dense feature matching and the feature-aware scoring mechanism are clear, modular contributions; the evaluation is broad, covers all seven core BOP datasets, reports per-dataset accuracy and runtimes, and compares against FreeZe under identical segmentation masks. The explicit runtime measurements and the equal-mask FreeZe comparison make the engineering claims checkable. However, the headline 'training-free SOTA' claim is conditional: the strongest single-detector configuration and the ensemble configuration rely on task-trained segmentation components, and key hyperparameters are selected by ablations on the same benchmark test sets used for the final evaluation. These issues do not invalidate the method, but they do mean the reported numbers are not yet evidence of fully training-free state-of-the-art performance.

major comments (3)
  1. [Abstract; Table I; Table II; Section III-C] The 'training-free, new state-of-the-art' claim conflates the pose module with the full pipeline. In Table II, the fully training-free configuration using CNOS masks (row 11) gives 71.1 AR, which is below FoundationPose+SAM-6D (row 13, 73.4 AR). The headline 74.3 AR (row 15) and the ensemble 80.1 AR (row 18) use SAM-6D, NIDS, and MUSE masks; SAM-6D is listed as a training-based method in Table I and its mask confidence comes from a learned template matcher, while NIDS and MUSE are additional learned detectors. The claim that the method is 'training-free' should therefore be restricted to the pose module under a fixed external segmentation prior, or the CNOS-only configuration should be presented as the headline for the training-free claim. This matters because the central novelty is 'training-free SOTA', and the evidence for that claim is configuration-dependent.
  2. [Section IV-A; Section IV-F; Tables IV and V] The hyperparameters used in the final results are selected using the same benchmark test sets on which the method is evaluated. Section IV-A fixes alpha=beta=gamma=1, k=10, M=N+1, and tau_mask=0.4, and Section IV-F justifies these choices by ablations computed on LM-O, T-LESS, TUD-L, IC-BIN, and YCB-V, which are five of the seven datasets in the main BOP evaluation. Selecting k, alpha, beta, gamma, and M against the same AR numbers that are later reported as the method's performance creates a selection-on-test loop, so the reported 74.3 and 80.1 AR values are not unbiased estimates of zero-shot generalization. The authors should either move the ablation and hyperparameter selection to held-out validation splits or to datasets not used in the headline evaluation, or provide a sensitivity analysis over a pre-specified range and report results for a pre-registered configuration.
  3. [Section IV-A; Table II] The run-time numbers are not defined precisely enough to support the speed claims. Table II reports 'mean per-image run-time' but does not state whether the time includes generation of the segmentation masks. For FreeZeV2 with SAM-6D masks the reported time is 1.4 s; if the SAM-6D mask generation time is excluded, the comparison to the end-to-end times of SAM-6D (row 12, 4.4 s), FoundationPose (row 13, 29.3 s), and FreeZe (row 14, 11.5 s) is not apples-to-apples. Please specify exactly what is included in each reported runtime and report end-to-end latency, including the selected mask generator, for the configurations in rows 11, 15, and 18-19.
minor comments (5)
  1. [Section III-D, Eqs. (1)-(2)] The PCA operator is used without specifying the data on which it is fit. Please state whether the PCA is fit once per query object on the rendered query features and then applied to target features, or recomputed online; this affects reproducibility and the interpretation of the 'frozen' claim.
  2. [Section IV-D, Table II] The sentence 'All methods are training-based, with the exception of FreeZe and FreeZeV2' is imprecise because the table also includes training-free rows for FreeZe and FreeZeV2; reword to 'all comparison methods other than FreeZe and FreeZeV2 are training-based.'
  3. [Tables II and III] In the provided version, the table formatting makes it difficult to separate the detector name from the method name (e.g., the group labels and method names are visually entangled). A dedicated 'Segmentation' column would make the detector-specific claims easier to verify.
  4. [Section III-C] The description of mask confidence as based on 'similarity to rendered templates' applies most directly to CNOS; SAM-6D uses a learned matcher, and NIDS/MUSE use different score definitions. Please make the description of each mask generator explicit so that the reader can assess how much task-specific machinery enters the mask selection stage.
  5. [Section IV-A] The visibility threshold V=18 is introduced without any sensitivity analysis; given that this threshold controls the size of the query point cloud, a brief ablation or justification would be useful.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: FreeZeV2's pose outputs are computed by a self-contained pipeline of frozen feature extractors, RANSAC, scoring, and ICP; self-citations to FreeZe and GeDi are not load-bearing, and the benchmark and segmentation caveats are evaluation-validity issues, not derivation-level circularity.

full rationale

The paper's claimed derivation chain is empirical rather than mathematical: given a query CAD model and an RGBD image, FreeZeV2 extracts frozen DINOv2 and GeDi features, matches them, runs RANSAC-based registration with the score in Eqs. (5)-(7), and refines with ICP. None of these equations contains the ground-truth pose or the reported AR; the predicted pose is a function of the input image and object model, not of a fitted constant. The self-citations to FreeZe [9] and GeDi [50] supply a baseline and a feature extractor, but the method is described in full and no uniqueness theorem is imported, so they are not load-bearing. The ablation in Section IV-F does select hyperparameters (M, k, alpha, beta, gamma) on the same BOP datasets later used for the headline results; this is potential benchmark overfitting and an evaluation-validity caveat, but it is not a circular reduction of the pose estimate, since the reported pose outputs are still computed from the inputs rather than being equal to the hyperparameter values. Similarly, the use of SAM-6D masks (which the paper's own Table I classifies as training-based) sits awkwardly with the 'training-free' label, but that is a scope/labeling concern: the pose module itself performs no task-specific training, and the accuracy is an external benchmark result, not an input-equivalent construction. Overall, the central claim has independent empirical content and the derivation is self-contained.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method is built on frozen foundation models and a set of hyperparameters. The main data-dependent choices are the score weights (alpha, beta, gamma), top-k parameter k, mask count M, and tau_mask, all selected using the BOP benchmark itself. The axiomatic assumptions are domain assumptions about the informativeness and complementarity of DINOv2 and GeDi features, and about segmentation mask reliability. No new physical or conceptual entities are introduced.

free parameters (5)
  • alpha, beta, gamma (Eq. 7 score weights) = 1, 1, 1
    Selected by ablation on five BOP datasets (Table IV); no held-out validation.
  • k (top-k nearest neighbors) = 10
    Chosen by ablation; k=1 yields 74.7 AR and k=10 yields 75.4 AR on a BOP subset (Table IV).
  • M (candidate masks per segmentor) = N+1 for localization, 100 for detection
    M=N+1 chosen by ablation (Table V); M=N+2 gains +0.2 AR but slows inference.
  • tau_mask (mask confidence threshold) = 0.4
    Used in 6D detection to discard low-confidence masks; no ablation is reported.
  • V (minimum visible views for query points) = 18
    Used to filter query points; fixed in experiments with no ablation.
assumptions (4)
  • domain assumption Frozen DINOv2 visual features and GeDi geometric features provide complementary information that supports reliable 3D-3D correspondence matching on unseen objects.
    Core premise of the feature matching stage (Section III-D).
  • domain assumption A 16x16 sparse patch grid sampled within the mask captures sufficient scene information for accurate pose estimation.
    Basis of sparse target extraction (Section III-D).
  • domain assumption Zero-shot segmentation models produce masks with sufficient coverage, and ensemble fusion does not hurt performance.
    Pipeline depends on masks; the conclusion admits sensitivity to segmentation quality.
  • standard math RANSAC with triplet sampling and geometric pruning can find correct coarse poses from noisy correspondences.
    Standard robust estimation technique used in Section III-E.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Accurate and efficient zero-shot 6D pose estimation with frozen foundation models." pith.science (2026). https://pith.science/paper/SHFQE5XZ

@misc{pith2026250609784,
  author       = {Pith},
  title        = {Pith review of: Accurate and efficient zero-shot 6D pose estimation with frozen foundation models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SHFQE5XZ}},
  note         = {Machine review of arXiv:2506.09784}
}
read the original abstract

Estimating the 6D pose of objects from RGBD data is a fundamental problem in computer vision, with applications in robotics and augmented reality. A key challenge is achieving generalization to novel objects that were not seen during training. Most existing approaches address this by scaling up training on synthetic data tailored to the task, a process that demands substantial computational resources. But is task-specific training really necessary for accurate and efficient 6D pose estimation of novel objects? To answer No!, we introduce FreeZeV2, the second generation of FreeZe: a training-free method that achieves strong generalization to unseen objects by leveraging geometric and vision foundation models pre-trained on unrelated data. FreeZeV2 improves both accuracy and efficiency over FreeZe through three key contributions: (i) a sparse feature extraction strategy that reduces inference-time computation without sacrificing accuracy; (ii) a feature-aware scoring mechanism that improves both pose selection during RANSAC-based 3D registration and the final ranking of pose candidates; and (iii) a modular design that supports ensembles of instance segmentation models, increasing robustness to segmentation masks errors. We evaluate FreeZeV2 on the seven core datasets of the BOP Benchmark, where it establishes a new state-of-the-art in 6D pose estimation of unseen objects. When using the same segmentation masks, FreeZeV2 achieves a remarkable 8x speedup over FreeZe while also improving accuracy by 5%. When using ensembles of segmentation models, FreeZeV2 gains an additional 8% in accuracy while still running 2.5x faster than FreeZe. FreeZeV2 was awarded Best Overall Method at the BOP Challenge 2024.

Figures

Figures reproduced from arXiv: 2506.09784 by the authors.

Figure 1
Figure 1. (Top) Comparison of 6D pose estimation performance across various [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed FreeZeV2 pipeline. Given a query object Q represented as a 3D model and an RGBD test image I, our method estimates the 6D pose of Q through three stages: (1) feature extraction, (2) feature matching, and (3) pose refinement. Top row: Candidate masks are generated by applying multiple zero-shot instance segmentation models independently. All masks are retained and processed in parallel. Stage… view at source ↗
Figure 3
Figure 3. Qualitative results on several sample images from the BOP benchmark. The top row shows input images, while the bottom row shows FreeZeV2’s [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Semantic Prior Guided One-View 6D Pose Estimation for Novel Objects

    cs.CV 2026-05 unverdicted novelty 6.0 of 10

    OneViewAll reports 92.5% ADD-0.1 pose accuracy on LINEMOD from a single real reference RGB-D view, using projection-based refinement with mirror-fusion symmetry priors rather than CAD rendering.

  2. xperception -- Making Robotic Grasping Easier

    cs.CV 2026-07 reject novelty 3.0 of 10

    xperception reports industrial deployment of the authors' existing FreeZe algorithm, claiming zero-shot millimeter-accurate 6D poses from CAD models without presenting new experimental evidence.

Reference graph

Works this paper leans on

86 extracted references · 72 canonical work pages · cited by 2 Pith papers

  1. [1]

    BOP: Benchmark for 6D object pose estimation,

    T. Hodan, F. Michel, E. Brachmann, W. Kehl, A. GlentBuch, D. Kraft, B. Drost, J. Vidal, S. Ihrke, X. Zabulis,et al., “BOP: Benchmark for 6D object pose estimation,” inECCV, 2018. 1, 2

    2018

  2. [2]

    Robotic Continuous Grasping System by Shape Transformer-Guided Multi-Object Category- Level 6D Pose Estimation,

    J. Liu, W. Sun, C. Liu, X. Zhang, and Q. Fu, “Robotic Continuous Grasping System by Shape Transformer-Guided Multi-Object Category- Level 6D Pose Estimation,”IEEE TII, 2023. 1

    2023

  3. [3]

    Instance segmentation based 6D pose estimation of industrial objects using point clouds for robotic bin-picking,

    C. Zhuang, S. Li, and H. Ding, “Instance segmentation based 6D pose estimation of industrial objects using point clouds for robotic bin-picking,” RCIM, 2023. 1

    2023

  4. [4]

    Towards More Effective Human- Robot Collaboration via Accurate Pose Estimation,

    M. ´A. Mateo-Casal´ı, L. Moya-Ruiz, A. Caraffa, D. Boscaini, A. Hamza, P. Chippendale, F. Poiesi, and F. Fraile, “Towards More Effective Human- Robot Collaboration via Accurate Pose Estimation,” inERF, 2024, pp. 157–161. 1

    2024

  5. [5]

    Deep multi-state object pose estimation for augmented reality assembly,

    Y . Su, J. Rambach, N. Minaskan, P. Lesur, A. Pagani, and D. Stricker, “Deep multi-state object pose estimation for augmented reality assembly,” inISMAR-Adjunct, 2019. 1

    2019

  6. [6]

    Model globally, match locally: Efficient and robust 3D object recognition,

    B. Drost, M. Ulrich, N. Navab, and S. Ilic, “Model globally, match locally: Efficient and robust 3D object recognition,” inCVPR, 2010. 1, 2, 3

    2010

  7. [7]

    PVNet: Pixel-Wise V oting Network for 6DoF Pose Estimation,

    S. Peng, Y . Liu, Q.-X. Huang, H. Bao, and X. Zhou, “PVNet: Pixel-Wise V oting Network for 6DoF Pose Estimation,” inCVPR, 2019. 1

    work page 2019

  8. [8]

    DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion,

    C. Wang, D. Xu, Y . Zhu, R. Martin-Martin, C. Lu, L. Fei-Fei, and S. Savarese, “DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion,” inCVPR, 2019. 1

    work page 2019

Show all 86 references
  1. [9]

    FreeZe: Training- free zero-shot 6D pose estimation with geometric and vision foundation models,

    A. Caraffa, D. Boscaini, A. Hamza, and F. Poiesi, “FreeZe: Training- free zero-shot 6D pose estimation with geometric and vision foundation models,” inECCV, 2024. 1, 2, 7, 8

    2024

  2. [10]

    FoundationPose: Unified 6D pose estimation and tracking of novel objects,

    B. Wen, W. Yang, J. Kautz, and S. Birchfield, “FoundationPose: Unified 6D pose estimation and tracking of novel objects,” inCVPR, 2024. 1, 2, 7

    2024

  3. [11]

    SAM-6D: Segment anything model meets zero-shot 6D object pose estimation,

    J. Lin, L. Liu, D. Lu, and K. Jia, “SAM-6D: Segment anything model meets zero-shot 6D object pose estimation,” inCVPR, 2024. 1, 2, 3, 4, 6, 7

    2024

  4. [12]

    Zero123-6d: Zero-shot novel view synthesis for rgb category-level 6d pose estimation,

    F. Di Felice, A. Remus, S. Gasperini, B. Busam, L. Ott, F. Tombari, R. Siegwart, and C. A. Avizzano, “Zero123-6d: Zero-shot novel view synthesis for rgb category-level 6d pose estimation,” inIROS, 2024. 1

    2024

  5. [13]

    GCE-Pose: Global Context Enhancement for Category-level Object Pose Estimation,

    W. Li, H. Xu, J. Huang, H. Jung, P. K. Yu, N. Navab, and B. Busam, “GCE-Pose: Global Context Enhancement for Category-level Object Pose Estimation,”arXiv preprint arXiv:2502.04293, 2025. 1

    2025 arXiv

  6. [14]

    GS-Pose: Category-level object pose estimation via geometric and semantic correspondence,

    P. Wang, T. Ikeda, R. Lee, and K. Nishiwaki, “GS-Pose: Category-level object pose estimation via geometric and semantic correspondence,” in ECCV, 2024. 1

    2024

  7. [15]

    ZeroPose: CAD-Prompted Zero-shot Object 6D Pose Estimation in Cluttered Scenes,

    J. Chen, Z. Zhou, M. Sun, R. Zhao, L. Wu, T. Bao, and Z. He, “ZeroPose: CAD-Prompted Zero-shot Object 6D Pose Estimation in Cluttered Scenes,” IEEE TCSVT, 2024. 1, 2, 3, 7

    2024

  8. [16]

    GenFlow: Generalizable recurrent flow for 6D pose refinement of novel objects,

    S. Moon, H. Son, D. Hur, and S. Kim, “GenFlow: Generalizable recurrent flow for 6D pose refinement of novel objects,” inCVPR, 2024. 1, 2, 7, 8

    2024

  9. [17]

    MatchU: Matching unseen objects for 6D pose estimation from RGB-D images,

    J. Huang, H. Yu, K.-T. Yu, N. Navab, S. Ilic, and B. Busam, “MatchU: Matching unseen objects for 6D pose estimation from RGB-D images,” inCVPR, 2024. 1, 2, 3, 7

    2024

  10. [18]

    MegaPose: 6D pose estimation of novel objects via render & compare,

    Y . Labb´e, L. Manuelli, A. Mousavian, S. Tyree, S. Birchfield, J. Tremblay, J. Carpentier, M. Aubry, D. Fox, and J. Sivic, “MegaPose: 6D pose estimation of novel objects via render & compare,” inCoRL, 2022. 1, 2, 7

    2022

  11. [19]

    GigaPose: Fast and robust novel object pose estimation via one correspondence,

    V . N. Nguyen, T. Groueix, M. Salzmann, and V . Lepetit, “GigaPose: Fast and robust novel object pose estimation via one correspondence,” inCVPR, 2024. 1, 2, 3

    2024

  12. [20]

    Distilling 3D distinctive local descriptors for 6D pose estimation,

    A. Hamza, A. Caraffa, D. Boscaini, and F. Poiesi, “Distilling 3D distinctive local descriptors for 6D pose estimation,”arXiv preprint arXiv:2503.15106, 2025. 1

    2025

  13. [21]

    FoundPose: Unseen object pose estimation with foundation features,

    E. P. Ornek, Y . Labb ´e, B. Tekin, L. Ma, C. Keskin, C. Forster, and T. Hodan, “FoundPose: Unseen object pose estimation with foundation features,” inECCV, 2024. 1, 2, 3, 6

    2024

  14. [22]

    ZS6D: Zero-shot 6D object pose estimation using vision transformers,

    P. Ausserlechner, D. Haberger, S. Thalhammer, J.-B. Weibel, and M. Vincze, “ZS6D: Zero-shot 6D object pose estimation using vision transformers,” inICRA, 2024. 1, 2, 3

    2024

  15. [23]

    Onepose: One-shot object pose estimation without cad models,

    J. Sun, Z. Wang, S. Zhang, X. He, H. Zhao, G. Zhang, and X. Zhou, “Onepose: One-shot object pose estimation without cad models,” inCVPR,

  16. [24]

    Onepose++: Keypoint-free one-shot object pose estimation without CAD models,

    X. He, J. Sun, Y . Wang, D. Huang, H. Bao, and X. Zhou, “Onepose++: Keypoint-free one-shot object pose estimation without CAD models,” NeurIPS, 2022. 2

    2022

  17. [25]

    Gen6d: Generalizable model-free 6-dof object pose estimation from rgb images,

    Y . Liu, Y . Wen, S. Peng, C. Lin, X. Long, T. Komura, and W. Wang, “Gen6d: Generalizable model-free 6-dof object pose estimation from rgb images,” inECCV, 2022. 2

    2022

  18. [26]

    Novel object 6d pose estimation with a single reference view,

    J. Liu, W. Sun, K. Zeng, J. Zheng, H. Yang, L. Wang, H. Rahmani, and A. Mian, “Novel object 6d pose estimation with a single reference view,” arXiv preprint arXiv:2503.05578, 2025. 2

    2025 arXiv

  19. [27]

    InstantPose: Zero-Shot Instance-Level 6D Pose Estimation from a Single View,

    F. Di Felice, A. Remus, S. Gasperini, B. Busam, L. Ott, S. Thalhammer, F. Tombari, and C. A. Avizzano, “InstantPose: Zero-Shot Instance-Level 6D Pose Estimation from a Single View,”RA-L, 2025. 2

    2025

  20. [28]

    UNOPose: Unseen Object Pose Estimation with an Unposed RGB-D Reference Image,

    X. Liu, G. Wang, R. Zhang, C. Zhang, F. Tombari, and X. Ji, “UNOPose: Unseen Object Pose Estimation with an Unposed RGB-D Reference Image,”arXiv preprint arXiv:2411.16106, 2024. 2

    2024 arXiv

  21. [29]

    One2Any: One-Reference 6D Pose Estimation for Any Object,

    M. Liu, S. Li, A. Chhatkuli, P. Truong, L. Van Gool, and F. Tombari, “One2Any: One-Reference 6D Pose Estimation for Any Object,”arXiv preprint arXiv:2505.04109, 2025. 2

    2025 arXiv

  22. [30]

    Open- vocabulary object 6D pose estimation,

    J. Corsetti, D. Boscaini, C. Oh, A. Cavallaro, and F. Poiesi, “Open- vocabulary object 6D pose estimation,” inCVPR, 2024. 2

    2024

  23. [31]

    High-resolution open-vocabulary object 6D pose estimation,

    J. Corsetti, D. Boscaini, F. Giuliari, C. Oh, A. Cavallaro, and F. Poiesi, “High-resolution open-vocabulary object 6D pose estimation,”arXiv preprint arXiv:2406.16384, 2024. 2

    2024 arXiv

  24. [32]

    Least-squares fitting of two 3-D point sets,

    K. S. Arun, T. S. Huang, and S. D. Blostein, “Least-squares fitting of two 3-D point sets,”IEEE TPAMI, 1987. 2

    1987

  25. [33]

    BOP Challenge 2024 on Model- Based and Model-Free 6D Object Pose Estimation,

    V . N. Nguyen, S. Tyree, A. Guo, M. Fourmy, A. Gouda, T. Lee, S. Moon, H. Son, L. Ranftl, J. Tremblay,et al., “BOP Challenge 2024 on Model- Based and Model-Free 6D Object Pose Estimation,” inCVPR-W, 2025. 2, 3, 8

    2024

  26. [34]

    OVE6D: Object viewpoint encoding for depth-based 6D object pose estimation,

    D. Cai, J. Heikkil ¨a, and E. Rahtu, “OVE6D: Object viewpoint encoding for depth-based 6D object pose estimation,” inCVPR, 2022. 2

    2022

  27. [35]

    Learning symmetry-aware geometry correspondences for 6D object pose estimation,

    H. Zhao, S. Wei, D. Shi, W. Tan, Z. Li, Y . Ren, X. Wei, Y . Yang, and S. Pu, “Learning symmetry-aware geometry correspondences for 6D object pose estimation,” inICCV, 2023. 2, 3

    2023

  28. [36]

    Templates for 3D object pose estimation revisited: Generalization to new objects and robustness to occlusions,

    V . N. Nguyen, Y . Hu, Y . Xiao, M. Salzmann, and V . Lepetit, “Templates for 3D object pose estimation revisited: Generalization to new objects and robustness to occlusions,” inCVPR, 2022. 2, 3

    2022

  29. [37]

    Object pose estimation via the aggregation of diffusion features,

    T. Wang, G. Hu, and H. Wang, “Object pose estimation via the aggregation of diffusion features,” inCVPR, 2024. 2, 3

    2024

  30. [38]

    OSOP: A multi-stage one shot object pose estimation framework,

    I. Shugurov, F. Li, B. Busam, and S. Ilic, “OSOP: A multi-stage one shot object pose estimation framework,” inCVPR, 2022. 2

    2022

  31. [39]

    Diffusion Features for Zero-Shot 6DoF Object Pose Estimation,

    B. V on Gimborn, P. Ausserlechner, M. Vincze, and S. Thalhammer, “Diffusion Features for Zero-Shot 6DoF Object Pose Estimation,”CVPR,

  32. [40]

    Zephyr: Zero-shot pose hypothesis rating,

    B. Okorn, Q. Gu, M. Hebert, and D. Held, “Zephyr: Zero-shot pose hypothesis rating,” inICRA, 2021. 2

    2021

  33. [41]

    Deep Learning-Based Object Pose Estimation: A Comprehensive Survey,

    J. Liu, W. Sun, H. Yang, Z. Zeng, C. Liu, J. Zheng, X. Liu, H. Rahmani, N. Sebe, and A. Mian, “Deep Learning-Based Object Pose Estimation: A Comprehensive Survey,”arXiv preprint arXiv:2405.07801, 2024. 2

    2024

  34. [42]

    ShapeNet: An information- rich 3D model repository,

    A. X. Chang, T. Funkhouser, L. Guibas, P. Hanrahan, Q. Huang, Z. Li, S. Savarese, M. Savva, S. Song, H. Su,et al., “ShapeNet: An information- rich 3D model repository,”arXiv preprint arXiv:1512.03012, 2015. 2

    2015 arXiv

  35. [43]

    Google Scanned Objects: A high- quality dataset of 3D scanned household items,

    L. Downs, A. Francis, N. Koenig, B. Kinman, R. Hickman, K. Reymann, T. B. McHugh, and V . Vanhoucke, “Google Scanned Objects: A high- quality dataset of 3D scanned household items,” inICRA, 2022. 2

    2022

  36. [44]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,” inICLR, 2021. 3

    2021

  37. [45]

    KPConv: Flexible and deformable convolution for point clouds,

    H. Thomas, C. R. Qi, J.-E. Deschaud, B. Marcotegui, F. Goulette, and L. J. Guibas, “KPConv: Flexible and deformable convolution for point clouds,” inICCV, 2019. 3

    2019

  38. [46]

    Geometric Transformer for Fast and Robust Point Cloud Registration,

    Z. Qin, H. Yu, C. Wang, Y . Guo, Y . Peng, and K. Xu, “Geometric Transformer for Fast and Robust Point Cloud Registration,” inCVPR,

  39. [47]

    Rotation- invariant transformer for point cloud matching,

    H. Yu, Z. Qin, J. Hou, M. Saleh, D. Li, B. Busam, and S. Ilic, “Rotation- invariant transformer for point cloud matching,” inCVPR, 2023. 3

    2023

  40. [48]

    T-LESS: An RGB-D dataset for 6D pose estimation of texture-less objects,

    T. Hodan, P. Haluza, ˇS. Obdrˇz´alek, J. Matas, M. Lourakis, and X. Zabulis, “T-LESS: An RGB-D dataset for 6D pose estimation of texture-less objects,” inWACV, 2017. 3, 6, 8

    2017

  41. [49]

    BOP: Benchmark for 6D object pose estimation,

    T. Hodan, F. Michel, E. Brachmann, W. Kehl, A. G. Buch, D. Kraft, B. Drost, J. Vial, S. Ihrke, X. Zabulis, C. Sahin, F. Manhardt, F. Tombari, T.-K. Kim, J. Matas, and C. Rother, “BOP: Benchmark for 6D object pose estimation,” inECCV, 2018. 3, 6, 8

    2018

  42. [50]

    Learning general and distinctive 3D local deep descriptors for point cloud registration,

    F. Poiesi and D. Boscaini, “Learning general and distinctive 3D local deep descriptors for point cloud registration,”IEEE TPAMI, 2023. 3, 5, 6

    2023

  43. [51]

    DINOv2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby,et al., “DINOv2: Learning robust visual features without supervision,”arXiv preprint arXiv:2304.07193, 2023. 3, 5, 6

    2023 arXiv

  44. [52]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark,et al., “Learning transferable visual models from natural language supervision,” inICML, 2021. 3

    2021

  45. [53]

    Emerging Properties in Self-Supervised Vision Transformers,

    M. Caron, H. Touvron, I. Misra, H. J ´egou, J. Mairal, P. Bojanowski, and A. Joulin, “Emerging Properties in Self-Supervised Vision Transformers,” inICCV, 2021. 3

    2021

  46. [54]

    4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks,

    C. Choy, J. Gwak, and S. Savarese, “4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks,” inCVPR, 2019. 3

    2019

  47. [55]

    Fully Convolutional Geometric Features,

    C. Choy, J. Park, and V . Koltun, “Fully Convolutional Geometric Features,” inICCV, 2019. 3

    2019

  48. [56]

    Revisiting fully convolutional geometric features for object 6D pose estimation,

    J. Corsetti, D. Boscaini, and F. Poiesi, “Revisiting fully convolutional geometric features for object 6D pose estimation,” inICCV-W, 2023. 3

    2023

  49. [57]

    PointNet: Deep learning on point sets for 3d classification and segmentation,

    C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “PointNet: Deep learning on point sets for 3d classification and segmentation,” inCVPR, 2017. 3

    2017

  50. [58]

    PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space,

    C. R. Qi, L. Yi, H. Su, and L. J. Guibas, “PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space,”NeurIPS, 2017. 3

    2017

  51. [59]

    PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies,

    G. Qian, Y . Li, H. Peng, J. Mai, H. Hammoud, M. Elhoseiny, and B. Ghanem, “PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies,” inNeurIPS, 2022. 3

    2022

  52. [60]

    Geodesic Convolutional Neural Networks on Riemannian Manifolds,

    J. Masci, D. Boscaini, M. M. Bronstein, and P. Vandergheynst, “Geodesic Convolutional Neural Networks on Riemannian Manifolds,” inICCV-W,

  53. [61]

    Geometric Deep Learning on Graphs and Manifolds using Mixture Model CNNs,

    F. Monti, D. Boscaini, J. Masci, E. Rodol `a, J. Svoboda, and M. M. Bronstein, “Geometric Deep Learning on Graphs and Manifolds using Mixture Model CNNs,” inCVPR, 2017. 3

    2017

  54. [62]

    SplineCNN: Fast Geometric Deep Learning with Continuous B-Spline Kernels,

    M. Fey, J. E. Lenssen, F. Weichert, and H. M ¨uller, “SplineCNN: Fast Geometric Deep Learning with Continuous B-Spline Kernels,” inCVPR,

  55. [63]

    Dynamic graph CNN for learning on point clouds,

    Y . Wang, Y . Sun, Z. Liu, S. E. Sarma, M. M. Bronstein, and J. M. Solomon, “Dynamic graph CNN for learning on point clouds,”ACM TOG, 2019. 3

    2019

  56. [64]

    Point Transformer,

    H. Zhao, L. Jiang, J. Jia, P. H. Torr, and V . Koltun, “Point Transformer,” inICCV, 2021. 3

    2021

  57. [65]

    Point Transformer V2: Grouped Vector Attention and Partition-based Pooling,

    X. Wu, Y . Lao, L. Jiang, X. Liu, and H. Zhao, “Point Transformer V2: Grouped Vector Attention and Partition-based Pooling,” inNeurIPS,

  58. [66]

    Point Transformer V3: Simpler Faster Stronger,

    X. Wu, L. Jiang, P.-S. Wang, Z. Liu, X. Liu, Y . Qiao, W. Ouyang, T. He, and H. Zhao, “Point Transformer V3: Simpler Faster Stronger,” inCVPR,

  59. [67]

    Rethinking network design and local geometry in point cloud: A simple residual MLP framework,

    X. Ma, C. Qin, H. You, H. Ran, and Y . Fu, “Rethinking network design and local geometry in point cloud: A simple residual MLP framework,” inICLR, 2022. 3

    2022

  60. [68]

    PointMixer: MLP-Mixer for point cloud understanding,

    J. Choe, C. Park, F. Rameau, J. Park, and I. S. Kweon, “PointMixer: MLP-Mixer for point cloud understanding,” inECCV, 2022. 3

    2022

  61. [69]

    PatchMixer: Rethinking network design to boost generalization for 3D point cloud understanding,

    D. Boscaini and F. Poiesi, “PatchMixer: Rethinking network design to boost generalization for 3D point cloud understanding,”IMAVIS, 2023. 3

    2023

  62. [70]

    MLP-Mixer: An all-MLP architecture for vision,

    I. O. Tolstikhin, N. Houlsby, A. Kolesnikov, L. Beyer, X. Zhai, T. Unterthiner, J. Yung, A. Steiner, D. Keysers, J. Uszkoreit,et al., “MLP-Mixer: An all-MLP architecture for vision,” inNeurIPS, 2021. 3

    2021

  63. [71]

    Patches are all you need?

    A. Trockman and J. Z. Kolter, “Patches are all you need?”arXiv preprint arXiv:2201.09792, 2022. 3

    2022 arXiv

  64. [72]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” inNeurIPS,

  65. [73]

    CNOS: A Strong Baseline for CAD-based Novel Object Segmentation,

    V . N. Nguyen, T. Hodan, G. Ponimatkin, T. Groueix, and V . Lepetit, “CNOS: A Strong Baseline for CAD-based Novel Object Segmentation,” inICCV-W, 2023. 4, 6, 7

    2023

  66. [74]

    Adapting Pre-Trained Vision Models for Novel Instance Detection and Segmentation,

    Y . Lu, J. J. P., Y . Guo, N. Ruozzi, and Y . Xiang, “Adapting Pre-Trained Vision Models for Novel Instance Detection and Segmentation,”arXiv preprint arXiv:2405.17859, 2024. 4, 6, 7

    2024 arXiv

  67. [75]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo,et al., “Segment anything,” inICCV, 2023. 4

    2023

  68. [76]

    Fast segment anything,

    X. Zhao, W. Ding, Y . An, Y . Du, T. Yu, M. Li, M. Tang, and J. Wang, “Fast segment anything,”arXiv preprint arXiv:2306.12156, 2023. 4

    2023 arXiv

  69. [77]

    Fast Poisson disk sampling in arbitrary dimensions,

    R. Bridson, “Fast Poisson disk sampling in arbitrary dimensions,” in SIGGRAPH, 2007. 5

    2007

  70. [78]

    I. T. Jolliffe,Principal component analysis for special types of data. Springer, 2002. 5

    2002

  71. [79]

    L. N. Trefethen and D. Bau,Numerical linear algebra. SIAM, 2022. 5

    2022

  72. [80]

    A method for registration of 3-D shapes,

    P. J. Besl and N. D. McKay, “A method for registration of 3-D shapes,” IEEE TPAMI, vol. 14, no. 2, pp. 239–256, 1992. 6

    1992

  73. [81]

    MUSE: Model-agnostic Unseen 2D Object Recognition via 3D-aware Similarity of Multi-Embeddings,

    “MUSE: Model-agnostic Unseen 2D Object Recognition via 3D-aware Similarity of Multi-Embeddings,” https://bop.felk.cvut.cz/method info/ 873/, 2024, accessed: 1st June 2025. 6, 7

    2024

  74. [82]

    Learning 6D object pose estimation using 3D object coordinates,

    E. Brachmann, A. Krull, F. Michel, S. Gumhold, J. Shotton, and C. Rother, “Learning 6D object pose estimation using 3D object coordinates,” in ECCV, 2014. 6, 8

    2014

  75. [83]

    Recovering 6D object pose and predicting next-best-view in the crowd,

    A. Doumanoglou, R. Kouskouridas, S. Malassiotis, and T.-K. Kim, “Recovering 6D object pose and predicting next-best-view in the crowd,” inCVPR, 2016. 6, 8

    2016

  76. [84]

    Introduc- ing MVTec ITODD – A dataset for 3D object recognition in industry,

    B. Drost, M. Ulrich, P. Bergmann, P. Hartinger, and C. Steger, “Introduc- ing MVTec ITODD – A dataset for 3D object recognition in industry,” inICCV-W, 2017. 6

    2017

  77. [85]

    HomebrewedDB: RGB-D dataset for 6D pose estimation of 3D objects,

    R. Kaskman, S. Zakharov, I. Shugurov, and S. Ilic, “HomebrewedDB: RGB-D dataset for 6D pose estimation of 3D objects,” inICCV-W, 2019. 6

    2019

  78. [86]

    PoseCNN: A Convolutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes,

    Y . Xiang, T. Schmidt, V . Narayanan, and D. Fox, “PoseCNN: A Convolutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes,”RSS, 2018. 6, 8

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.