Pith. sign in

REVIEW 3 major objections 4 minor 42 references

The paper claims that scaling NOC-supervised feature learning to 674K real and synthetic images and pairing it with geometrically consistent correspondence matching lets zero-shot CAD-to-image alignment surpass fully supervised accuracy on

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 00:15 UTC pith:XUDVWKBU

load-bearing objection A real step forward in zero-shot CAD alignment, but the 'surpasses supervised' headline leans on a benchmark-tuned depth estimator and a slim margin, so read the fine print. the 3 major comments →

arxiv 2607.15058 v1 pith:XUDVWKBU submitted 2026-07-16 cs.CV cs.RO

SUFLECA: Scaling Up Feature Learning for CAD-to-image Alignment

Saad Ejaz , Miguel Fernandez-Cortizas , Javier Civera , Holger Voos , Jose Luis Sanchez-Lopez This is my paper
classification cs.CV cs.RO
keywords CAD-to-image alignment9D pose estimationzero-shotNormalized Object Coordinatesgeometric consistencycorrespondence matchingweak supervisionScanNet25k
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

SUFLECA sets out to prove that accurate 9D pose alignment between a single photo and a CAD model does not require pose-annotated training data, iterative refinement, or high-dimensional appearance features. The paper's core move is to train a lightweight decoder on top of a frozen visual encoder with Normalized Object Coordinates (NOCs) over a 674K-image mixture of real and synthetic scenes, producing compact, geometry-aware descriptors that transfer across domains. It then replaces plain nearest-neighbour matching with mutual k-nearest neighbours plus a geometric-consistency filter that estimates anisotropic scale from pairwise distances and keeps only correspondences agreeing with a common 9D transform. On ScanNet25k this yields 33.4% category and 42.3% instance accuracy, beating the strongest zero-shot baseline by 10.3/12.2 points and, the paper reports, surpassing fully supervised methods for the first time on this benchmark. A sympathetic reader would care because it suggests the alignment bottleneck is the feature space and correspondence geometry, not supervision scale.

Core claim

The central claim is that CAD-to-image alignment can be made zero-shot and more accurate than supervised approaches by (i) supervising feature learning with Normalized Object Coordinates at scale—674K frames across 12 real and synthetic datasets—so that frozen encoder features become spatially smooth and geometrically grounded, and (ii) replacing one-to-many nearest-neighbour matching with mutual k-NN matching followed by an anisotropic-scale-aware geometric consensus filter and RANSAC-Procrustes registration. As a result, on the ScanNet25k validation split the method reaches 33.4% category-averaged and 42.3% instance-averaged accuracy at 20cm/20°/20% thresholds, exceeding ZeroCAD by 10.3/12

What carries the argument

The central object is the NOC-supervised feature model: multi-scale features from a frozen perception encoder are fused and upsampled by a dense-prediction transformer, then decoded by a binned classifier over Normalized Object Coordinates (a canonical coordinate frame per object category) trained with cross-entropy plus L1 loss; the NOC head is discarded at test time and the ℓ2-normalized dense features serve as correspondence descriptors. The matching mechanism is the second pillar: mutual k-NN matching to enforce one-to-one-ness, then anisotropic scale estimation from pairwise distance ratios (Eqs. 4-5) with IRLS refinement shrunk toward the robust isotropic mode, a consensus matrix whose

Load-bearing premise

The pipeline assumes the monocular metric depth estimates are accurate at object scale, so that the pairwise-distance equations (4-5) initialize the anisotropic scale correctly and the geometric-consistency filter separates inliers from outliers; if per-object depth noise is high, the 3D points feeding both scale estimation and Procrustes registration are corrupted before the learned features can compensate.

What would settle it

Run SUFLECA on the ScanNet25k validation split with ground-truth depth from the Scan2CAD annotations instead of the monocular estimator; a large accuracy jump would indicate the geometric machinery is largely inheriting depth quality, while near-identical accuracy would show the features and matching alone determine the result.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Zero-shot CAD alignment now beats fully supervised methods on ScanNet25k, removing pose annotations as a prerequisite for strong alignment.
  • The compact 384-dimensional descriptor cuts GPU memory roughly 2-5x versus baselines and enables sub-second per-instance alignment, which matters for on-robot perception.
  • The geometric-consistency filter is responsible for a large share of the gain: ablating it to plain nearest-neighbour drops accuracy from 33.4/42.3 to 26.3/34.2.
  • The S_fit alignment-quality score improves NMS ranking over using the CAD retrieval score, so registration residuals can serve as a more principled confidence signal.
  • On the stricter DiffCAD split (no NMS) SUFLECA more than doubles ZeroCAD and even exceeds DiffCAD with ground-truth hypothesis selection, indicating the gains are not an artifact of NMS.

Where Pith is reading between the lines

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

  • The large gap between nearest-neighbour and geometrically filtered matching suggests the method's advantage will grow with clutter, occlusion, and duplicated object parts, where appearance-only matches are systematically ambiguous; this can be tested by adding synthetic occluders at varying levels to ScanNet images.
  • Because the depth estimator is fine-tuned on the ScanNet25k training split, the 'zero-shot' headline is specific to the alignment stage; on truly unseen depth distributions, the pairwise-distance scale initialisation could become the weak link, and an extension would couple depth uncertainty into the consensus threshold.
  • The anisotropic pairwise-distance formulation is a general robust-registration tool: it could transfer to point-cloud alignment, RGB-D tracking, or multi-view CAD fitting, where scale varies per axis and correspondences contain many outliers.
  • The observed bottleneck in zero-shot CAD retrieval (3.8% instance retrieval accuracy) suggests SUFLECA's S_fit could be repurposed as a retrieval-re-ranking signal, closing the loop between alignment quality and CAD selection.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes SUFLECA, a weakly-supervised zero-shot CAD-to-image alignment method. It trains a DPT-based NOC head on 674K images from 12 real and synthetic datasets, then uses mutual k-NN matching with geometric consistency filtering to estimate 9D pose (rotation, translation, anisotropic scale) via RANSAC/Procrustes. On ScanNet25k it reports 33.4%/42.3% category/instance accuracy, surpassing the strongest zero-shot baseline by 10.3/12.2 points and, for the first time, fully supervised methods, while using a smaller computational footprint and no iterative refinement.

Significance. If confirmed, this is a significant result: it shows that NOC-supervised feature learning at scale plus geometric consensus can outperform supervised pose estimators without iterative refinement, with lower memory and compute. The paper is transparent about its zero-shot definition, provides per-category tables, and includes ablations isolating each contribution (Table IV). The geometric-consistency equations are derived from the 9D transformation model, and the method ships with a clear implementation description. However, the headline zero-shot claim is compromised by the use of a depth estimator fine-tuned on the ScanNet25k training split, and the anisotropic scale initialization has a potential technical flaw. These issues need to be addressed before the central claim is fully convincing.

major comments (3)
  1. [IV.C, III.B] The headline result in Table II is obtained with depth from a monocular metric depth estimator fine-tuned on the ScanNet25k training split (Section IV.C). This depth enters the alignment pipeline at Section III.B and is used to lift image pixels to q_i, which feed Eqs. (4)-(8) and the Procrustes solver. Thus the 'zero-shot' alignment stage has, through its depth model, seen the ScanNet distribution; the comparison against supervised methods is not a clean test of the paper's central claim. The DiffCAD split uses non-fine-tuned depth, but it is a different protocol (no NMS, six categories) and does not directly bound the effect on Table II. Please report results with a depth estimator not trained on ScanNet25k, or explicitly characterize the sensitivity of the reported 33.4/42.3 to depth quality.
  2. [III.D, Eqs. (4)-(6)] The isotropic log-scale histogram mode initialization assumes that inlier pairs 'coarsely concentrate in a common bin' (text after Eq. (5)). For anisotropic S, however, ψ_ij = 0.5 log(∥q_i−q_j∥²/∥Δ_ij∥²) is direction-dependent: for a pair separated along axis k, ψ_ij ≈ log(s_k). Thus inlier pairs from different spatial directions produce different ψ values, and the mode may be a poor initialization for the anisotropic scale unless the object scale is nearly isotropic or one axis dominates the pair distribution. The paper does not quantify the scale anisotropy in ScanNet25k or ablate this initialization. Please provide a synthetic experiment or a per-axis robust initialization to justify the current approach.
  3. [IV.E, Table IV] The ablation in Table IV isolates training data, correspondence estimation, and NMS scoring, but not the depth estimator that produces all 3D points q_i. Since the geometric consistency filter (Eqs. (6)-(8)) operates on q_i, depth noise can directly corrupt scale estimation and inlier selection. An ablation with ground-truth depth, or with a different depth model, is needed to understand how much of the reported accuracy relies on the fine-tuned depth. Without it, the contribution of the two named components (feature learning and matching) to the headline number is not fully separated from the upstream depth.
minor comments (4)
  1. [V (Conclusion)] The statement 'without pose-annotated training data' is inaccurate because the NOC supervision (Section III.C) is derived from object poses and CAD models (6D poses + depth). Please rephrase to 'without using test-time pose annotations' or 'without direct 9D pose supervision on the test benchmark'.
  2. [Table II] FoundationPose is marked with * (weak supervision), but FoundationPose [37] typically uses pose-labelled synthetic training; clarify its supervision status or adjust the footnote.
  3. [Eq. (5)] The definition of φ_ij divides by ∥Δ_ij∥²; if two source points spatially coincide, this is undefined. FPS avoids duplicates, but a guard would improve rigor.
  4. [IV.C (DiffCAD split)] State explicitly that this split uses non-fine-tuned depth (it does) and note that this is a different evaluation protocol from the main NMS results, to avoid confusion.

Circularity Check

0 steps flagged

No significant circularity; the derivation is self-contained and benchmark supervision is not fed back into the method's fitted components.

full rationale

SUFLECA's claimed derivation chain does not reduce to its inputs. The 9D model in Eq. (1) defines the transformation, and the pairwise scale equations (4)-(5) follow algebraically from that model; the scale is initialized from the histogram mode of pairwise distance ratios, not from ground-truth scale or benchmark accuracy. The geometric-consistency filter (Eqs. 6-8) uses estimated scales and relative thresholds, and the final alignment is obtained by RANSAC/Procrustes on the filtered correspondences. S_fit is computed from registration residuals (Eqs. 9-10) and is ablated against the ROCA retrieval score, so it is not fitted to benchmark labels. The feature model is trained with NOC supervision from external real and synthetic datasets, and the paper explicitly states that 'ScanNet [23] dataset is omitted from training when evaluating on ScanNet25K [1] to have a strict zero-shot setting.' The closest concern, the monocular depth estimator fine-tuned on the ScanNet25k training split, is disclosed as a shared upstream component ('Following [2], all methods use ... depth from a monocular metric depth estimator [32] fine-tuned on the ScanNet25k training split'). Since this depth model is not a fitted parameter of SUFLECA's alignment pipeline, is shared by all zero-shot baselines, and is not optimized against validation accuracy, it is an evaluation-protocol caveat rather than a circular derivation. There are no load-bearing self-citations and no imported uniqueness theorems. The paper's central claims have independent content and are benchmarked against external supervised and zero-shot methods.

Axiom & Free-Parameter Ledger

9 free parameters · 4 axioms · 0 invented entities

The central claim rests on trained features plus a matching algorithm with nine hand-set hyperparameters, and on the quality of the upstream depth and NOC pseudo-labels. No new physical or mathematical entities are postulated beyond the S_fit alignment-quality score, which is a method output rather than an invented entity.

free parameters (9)
  • lambda (loss weight) = 0.33
    Weight between cross-entropy and L1 NOC loss in Eq. (2), hand-set.
  • m (NOC bins) = 64 (base), 50 (S)
    Number of uniform bins per NOC coordinate axis; architecture choice.
  • n (pre-rendered views) = 6
    Number of fixed viewpoints per CAD model for view selection and matching.
  • N (FPS sample count) = 512
    Number of representative 3D points sampled for correspondence estimation.
  • k (mutual k-NN) = 13
    Neighborhood size for mutual k-nearest-neighbor matching.
  • M (putative correspondences) = 256
    Cap on correspondence set before geometric filtering.
  • alpha (shrinkage) = 0.5
    Trust placed in anisotropic log-scale vs. isotropic estimate in Eq. (6).
  • beta (inlier threshold) = 0.01
    Relative threshold for pairwise distance consistency in Eq. (7).
  • delta (gating threshold) = 0.05
    Relative eigenvector threshold for retaining correspondences in Eq. (8).
axioms (4)
  • domain assumption NOC pseudo-labels derived by SAM2 segmentation plus depth backprojection are accurate enough for supervision after verification thresholding.
    Section III-C: NOCs are computed from backprojected depth and 'verified against the overlaid CAD NOCs' with instances rejected only if mean error exceeds a threshold; this assumes the derived supervision is reliable.
  • domain assumption Frozen DUNE-B features can be steered into a geometry-aware space by a lightweight binned NOC classification head, and the resulting features transfer to unseen images.
    The entire feature-learning contribution (Section III-C) rests on the trainability and transferability of this adapter; no theoretical guarantee is offered.
  • standard math Object geometry is well-modeled by an axis-aligned anisotropic scale transform (Eq. 1), making pairwise Euclidean distance ratios the correct invariant for geometric consensus filtering (Eqs. 4-8).
    The derivation of Eqs. (4)-(5) assumes the 9D transform R,S,t with diagonal S; if the CAD model or object deviates from this model (e.g., non-axis-aligned scale or deformation), the consistency test is misspecified.
  • domain assumption The evaluation protocol (shared ROCA boxes/retrievals, SAM2 masks, and fine-tuned MetricAnything depth) provides a fair common testbed for comparing the zero-shot alignment stage.
    Section IV-C: all methods use these upstream components, making the comparison fair for the alignment stage, but the 'zero-shot' label is only valid under this restricted definition.

pith-pipeline@v1.3.0-alltime-deepseek · 13147 in / 11906 out tokens · 113474 ms · 2026-08-02T00:15:00.353561+00:00 · methodology

0 comments
read the original abstract

CAD-to-image alignment aims to estimate an object's 9D pose (rotation, translation, and anisotropic scale) from a single RGB image, enabling applications in robotics and augmented reality. Recent zero-shot methods use visual foundation models to match image regions to CAD models, yet typically their correspondences are appearance-driven and degrade under occlusion or sim-to-real domain shift. To address these limitations, we introduce SUFLECA (Scaling Up Feature LEarning for CAD Alignment), a weakly-supervised framework for zero-shot CAD alignment with two key contributions. First, SUFLECA scales up geometry-grounded feature learning from pretrained visual representations through Normalized Object Coordinates (NOCs) supervision on 674K images spanning 12 real and synthetic datasets, learning compact geometry-aware features that generalize across domains. Second, we propose a geometrically consistent matching algorithm that establishes reliable one-to-one CAD-to-image correspondences. Together, these contributions enable accurate, sub-second alignment per object instance without iterative pose refinement. On ScanNet25k, SUFLECA achieves 33.4%/42.3% category/instance accuracy, outperforming, with a smaller computational footprint, the strongest zero-shot baseline by 10.3/12.2 percentage points and, for the first time on this benchmark, even surpassing fully supervised methods. Code is available at: https://github.com/snt-arg/SUFLECA

Figures

Figures reproduced from arXiv: 2607.15058 by Holger Voos, Javier Civera, Jose Luis Sanchez-Lopez, Miguel Fernandez-Cortizas, Saad Ejaz.

Figure 1
Figure 1. Figure 1: Accuracy, runtime, and GPU memory comparison of zero-shot CAD-to-image alignment methods on the Scan￾Net25k validation split [1]. ZeroCAD† denotes ZeroCAD [2] without iterative pose refinement. SUFLECA achieves the highest accuracy, surpassing even supervised methods, while attaining the lowest runtime and smallest GPU memory foot￾print among zero-shot methods. CAD alignment through pretrained backbones th… view at source ↗
Figure 2
Figure 2. Figure 2: Example frames from the multi-dataset validation set. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the feature model architecture. Multi-scale features from a frozen perception encoder are fused and [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of zero-shot methods in align [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

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

42 extracted references · 7 linked inside Pith

  1. [1]

    Roca: Robust cad model retrieval and alignment from a single image,

    C. G ¨umeli, A. Dai, and M. Nießner, “Roca: Robust cad model retrieval and alignment from a single image,” in Proc. IEEE/CVF conference on Computer Vision and Pattern Recognition, 2022

    2022

  2. [2]

    Zero-shot inexact cad model alignment from a single image,

    P. Arsomngern, S. Khwanmuang, M. Nießner, and S. Suwajanakorn, “Zero-shot inexact cad model alignment from a single image,” in Proc. IEEE International Conference on Computer Vision, 2025

    2025

  3. [3]

    Optimal and robust category-level perception: Object pose and shape estimation from 2-d and 3-d semantic keypoints,

    J. Shi, H. Yang, and L. Carlone, “Optimal and robust category-level perception: Object pose and shape estimation from 2-d and 3-d semantic keypoints,” IEEE Transactions on Robotics, 2023

    2023

  4. [4]

    Object recog- nition and full pose registration from a single image for robotic manip- ulation,

    A. Collet, D. Berenson, S. S. Srinivasa, and D. Ferguson, “Object recog- nition and full pose registration from a single image for robotic manip- ulation,” in IEEE international conference on robotics and automation, 2009

    2009

  5. [5]

    Mask2cad: 3d shape prediction by learning to segment and retrieve,

    W. Kuo, A. Angelova, T.-Y . Lin, and A. Dai, “Mask2cad: 3d shape prediction by learning to segment and retrieve,” in Proc. European Conference on Computer Vision, 2020

    2020

  6. [6]

    Sparc: Sparse render- and-compare for cad model alignment in a single rgb image,

    F. Langer, G. Bae, I. Budvytis, and R. Cipolla, “Sparc: Sparse render- and-compare for cad model alignment in a single rgb image,” in Proc. British Machine Vision Conference, 2022

    2022

  7. [7]

    Sparse multi-object render-and- compare,

    F. Langer, R. Cipolla, and I. Budvytis, “Sparse multi-object render-and- compare,” in Proc. British Machine Vision Conference, 2023

    2023

  8. [8]

    Diorama: Unleashing zero-shot single-view 3d indoor scene modeling,

    Q. Wu, D. Iliash, D. Ritchie, M. Savva, and A. X. Chang, “Diorama: Unleashing zero-shot single-view 3d indoor scene modeling,” in Proc. IEEE/CVF International Conference on Computer Vision, 2025

    2025

  9. [9]

    SDFit: 3D object pose and shape by fitting a morphable SDF to a single image,

    D. Anti ´c, G. Paschalidis, S. Tripathi, T. Gevers, S. K. Dwivedi, and D. Tzionas, “SDFit: 3D object pose and shape by fitting a morphable SDF to a single image,” in Proc. IEEE/CVF International Conference on Computer Vision, 2025

    2025

  10. [10]

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

    H. Wang, S. Sridhar, J. Huang, J. Valentin, S. Song, and L. J. Guibas, “Normalized object coordinate space for category-level 6d object pose and size estimation,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019

    2019

  11. [11]

    Omninocs: A unified nocs dataset and model for 3d lifting of 2d objects,

    A. Krishnan, A. Kundu, K.-K. Maninis, J. Hays, and M. Brown, “Omninocs: A unified nocs dataset and model for 3d lifting of 2d objects,” in Proc. European Conference on Computer Vision, 2024

    2024

  12. [12]

    Objectron: A large scale dataset of object-centric videos in the wild with pose annotations,

    A. Ahmadyan, L. Zhang, A. Ablavatski, J. Wei, and M. Grundmann, “Objectron: A large scale dataset of object-centric videos in the wild with pose annotations,” Proc. IEEE Conference on Computer Vision and Pattern Recognition, 2021

    2021

  13. [13]

    Dune: Distilling a universal encoder from heterogeneous 2d and 3d teachers,

    M. B. Sarıyıldız, P. Weinzaepfel, T. Lucas, P. De Jorge, D. Larlus, and Y . Kalantidis, “Dune: Distilling a universal encoder from heterogeneous 2d and 3d teachers,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025

    2025

  14. [14]

    To- tal3dunderstanding: Joint layout, object pose and mesh reconstruction for indoor scenes from a single image,

    Y . Nie, X. Han, S. Guo, Y . Zheng, J. Chang, and J. J. Zhang, “To- tal3dunderstanding: Joint layout, object pose and mesh reconstruction for indoor scenes from a single image,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020

    2020

  15. [15]

    3d-rcnn: Instance-level 3d object reconstruction via render-and-compare,

    A. Kundu, Y . Li, and J. M. Rehg, “3d-rcnn: Instance-level 3d object reconstruction via render-and-compare,” in Proc. IEEE conference on computer vision and pattern recognition, 2018

    2018

  16. [16]

    Coscad: Cross- modal cad model retrieval and pose alignment from a single image,

    Z. Wen, H. Chen, Z. Zhu, Z. Wei, L. Nan, and M. Wei, “Coscad: Cross- modal cad model retrieval and pose alignment from a single image,” in International Conf. on Computational Visual Media, 2025

    2025

  17. [17]

    Diffcad: Weakly- supervised probabilistic cad model retrieval and alignment from an rgb image,

    D. Gao, D. Rozenberszki, S. Leutenegger, and A. Dai, “Diffcad: Weakly- supervised probabilistic cad model retrieval and alignment from an rgb image,” ACM Trans. Graph., 2024

    2024

  18. [18]

    Category-level object shape and pose estimation in less than a millisecond,

    L. Shaikewitz, T. Nguyen, and L. Carlone, “Category-level object shape and pose estimation in less than a millisecond,” IEEE International Conference on Robotics and Automation, 2026

    2026

  19. [19]

    The pascal visual object classes (voc) challenge,

    M. Everingham, L. Van Gool, C. K. Williams, J. Winn, and A. Zisser- man, “The pascal visual object classes (voc) challenge,” International journal of computer vision, 2010

    2010

  20. [20]

    ARKitscenes - a diverse real-world dataset for 3d indoor scene understanding using mobile RGB-d data,

    G. Baruch, Z. Chen et al., “ARKitscenes - a diverse real-world dataset for 3d indoor scene understanding using mobile RGB-d data,” in Conference on Neural Information Processing Systems, 2021

    2021

  21. [21]

    Stereo magnification: Learning view synthesis using multiplane images,

    T. Zhou, R. Tucker, J. Flynn, G. Fyffe, and N. Snavely, “Stereo magnification: Learning view synthesis using multiplane images,” in SIGGRAPH, 2018

    2018

  22. [22]

    Cad-estate: Large-scale cad model annotation in rgb videos,

    K.-K. Maninis, S. Popov, M. Nießner, and V . Ferrari, “Cad-estate: Large-scale cad model annotation in rgb videos,” in Proc. IEEE/CVF International Conference on Computer Vision, 2023

    2023

  23. [23]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes,

    A. Dai, A. X. Chang, M. Savva, M. Halber, T. Funkhouser, and M. Nießner, “Scannet: Richly-annotated 3d reconstructions of indoor scenes,” in IEEE Computer Vision and Pattern Recognition, 2017

    2017

  24. [24]

    Lever- aging automatic cad annotations for supervised learning in 3d scene understanding,

    Y . Rao, S. Ainetter, S. Stekovic, V . Lepetit, and F. Fraundorfer, “Lever- aging automatic cad annotations for supervised learning in 3d scene understanding,” arXiv preprint arXiv:2504.13580, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    Scannet++: A high- fidelity dataset of 3d indoor scenes,

    C. Yeshwanth, Y .-C. Liu, M. Nießner, and A. Dai, “Scannet++: A high- fidelity dataset of 3d indoor scenes,” in Proc. IEEE/CVF International Conference on Computer Vision, 2023

    2023

  26. [26]

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

    X. Sun, J. Wu, X. Zhang, Z. Zhang, C. Zhang, T. Xue, J. B. Tenenbaum, and W. T. Freeman, “Pix3d: Dataset and methods for single-image 3d shape modeling,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018

    2018

  27. [27]

    Objectnet3d: A large scale database for 3d object recognition,

    Y . Xiang, W. Kim, W. Chen, J. Ji, C. Choy, H. Su, R. Mottaghi, L. Guibas, and S. Savarese, “Objectnet3d: A large scale database for 3d object recognition,” in European Conference Computer Vision, 2016

    2016

  28. [28]

    3d-front: 3d furnished rooms with layouts and semantics,

    H. Fu et al., “3d-front: 3d furnished rooms with layouts and semantics,” in Proc. International Conference on Computer Vision, 2021

    2021

  29. [29]

    Hypersim: A photore- alistic synthetic dataset for holistic indoor scene understanding,

    M. Roberts, J. Ramapuram, A. Ranjan, A. Kumar, M. A. Bautista, N. Paczan, R. Webb, and J. M. Susskind, “Hypersim: A photore- alistic synthetic dataset for holistic indoor scene understanding,” in International Conference on Computer Vision, 2021

    2021

  30. [30]

    Shapenet: An information-rich 3d model repository,

    A. X. Chang, T. Funkhouser et al., “Shapenet: An information-rich 3d model repository,” arXiv preprint arXiv:1512.03012, 2015

    Pith/arXiv arXiv 2015

  31. [31]

    Sam 2: Segment anything in images and videos,

    N. Ravi, V . Gabeur et al., “Sam 2: Segment anything in images and videos,” arXiv preprint arXiv:2408.00714, 2024

    Pith/arXiv arXiv 2024

  32. [32]

    Metricanything: Scaling metric depth pretraining with noisy heteroge- neous sources,

    B. Ma, J. Yang, D. Di, X. Zhang, J. Cui, H. Li, Y . Xie, and W. Chen, “Metricanything: Scaling metric depth pretraining with noisy heteroge- neous sources,” arXiv preprint arXiv:2601.22054, 2026

    Pith/arXiv arXiv 2026

  33. [33]

    Vision transformers for dense prediction,

    R. Ranftl, A. Bochkovskiy, and V . Koltun, “Vision transformers for dense prediction,” in Proc. IEEE/CVF international conference on computer vision, 2021

    2021

  34. [34]

    Superansac: One ransac to rule them all,

    D. Barath, “Superansac: One ransac to rule them all,” arXiv preprint arXiv:2506.04803, 2025

    Pith/arXiv arXiv 2025

  35. [35]

    Vid2CAD: CAD Model Alignment Using Multi-View Constraints From Videos ,

    K.-K. Maninis, S. Popov, M. Niesner, and V . Ferrari, “ Vid2CAD: CAD Model Alignment Using Multi-View Constraints From Videos ,” IEEE Transactions on Pattern Analysis & Machine Intelligence, 2023

    2023

  36. [36]

    Sim ´eoni, H

    O. Sim ´eoni, H. V . V oet al., “Dinov3,” arXiv preprint arXiv:2508.10104, 2025

    Pith/arXiv arXiv 2025

  37. [37]

    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,” in Proc. IEEE conference on Computer Vision and Pattern Recognition, 2024

    2024

  38. [38]

    Common objects in 3d: Large-scale learning and evalua- tion of real-life 3d category reconstruction,

    J. Reizenstein, R. Shapovalov, P. Henzler, L. Sbordone, P. Labatut, and D. Novotny, “Common objects in 3d: Large-scale learning and evalua- tion of real-life 3d category reconstruction,” in International Conference on Computer Vision, 2021

    2021

  39. [39]

    Scan2cad: Learning cad model alignment in rgb-d scans,

    A. Avetisyan, M. Dahnert, A. Dai, M. Savva, A. X. Chang, and M. Nießner, “Scan2cad: Learning cad model alignment in rgb-d scans,” in Proc. Conference on computer vision and pattern recognition, 2019

    2019

  40. [40]

    Grounded sam: Assembling open-world models for diverse visual tasks,

    T. Ren, S. Liu, A. Zeng, J. Lin, K. Li, H. Cao, J. Chen, X. Huang, Y . Chen, F. Yanet al., “Grounded sam: Assembling open-world models for diverse visual tasks,” arXiv preprint arXiv:2401.14159, 2024

    Pith/arXiv arXiv 2024

  41. [41]

    Sam 3d: 3dfy anything in images,

    X. Chen et al., “Sam 3d: 3dfy anything in images,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026

    2026

  42. [42]

    Oscar: Open-set cad retrieval from a language prompt and a single image,

    T. Pulli, J.-B. Weibel, P. H ¨onig, M. Hirschmanner, M. Vincze, and A. Holzinger, “Oscar: Open-set cad retrieval from a language prompt and a single image,” arXiv preprint arXiv:2601.07333, 2026

    arXiv 2026