Pith. sign in

REVIEW 4 major objections 5 minor 65 references

TeaMatch: Teachable Cross-Modal Representation Learning for 2D-3D Matching

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read TeaMatch claims that 2D-3D correspondences improve when the matcher's pair representations must remain recoverable by degraded weak learners on unseen patches, and reports consistent gains in inlier ratio and pose recall without changing…

desk verdict A clean training-only regularizer that improves 2D-3D matching by a few percent; the core teachability signal is partly self-referential, but the geometry-anchored losses and the occlusion experiments largely back the robustness claim. read the letter →

arxiv 2608.09590 v1 pith:ROQJP4E4 submitted 2026-08-10 cs.CV

classification cs.CV
keywords 2D-3Dmatchingcross-modalrepresentationlearningteachabilityweakprobesteacher-studentvisuallocalizationdetection-freeposeestimation
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

TeaMatch claims that the quality of image-to-point-cloud representations for 2D-3D matching can be improved by demanding that they remain recoverable under degraded conditions. The paper defines this property as teachability and enforces it with low-capacity probes that imitate the teacher's softened matching scores on one set of coarse patches and are evaluated on a disjoint set. All probes and auxiliary losses exist only during training, so the framework can be layered onto a standard detection-free coarse-to-fine matcher with no change at inference time. On the 7-Scenes and RGB-D Scenes V2 benchmarks, the authors report consistent improvements in inlier ratio, feature matching recall, and registration recall, including under severe point-cloud occlusion. If the claim is right, cross-modal matching robustness can be bought entirely through training-time representation learning rather than architectural changes.

What carries the argument

The load-bearing object is the pair-level multimodal representation $z_j = \phi(f^I_j, f^P_j, c^I_j, c^P_j, \hat{u}_j, \hat{x}_j)$, built from local image and point features, coarse contextual features, and normalized spatial coordinates. TeaMatch degrades $z_j$ through image weakening, geometry weakening, context removal, or a low-rank bottleneck, then fits a weak ridge-regression probe in closed form, $W^*_m = ((H^{(m)}_{\mathrm{teach}})^\top H^{(m)}_{\mathrm{teach}} + \lambda_r I)^{-1} (H^{(m)}_{\mathrm{teach}})^\top Y_{\mathrm{teach}}$, on a teach split and evaluates it on a patch-disjoint meta split. The meta-split probe predictions feed three losses, $L^{(m)}_{\mathrm{corr}}$, $L^{(m)}_{\mathrm{inlier}}$, and $L^{(m)}_{\mathrm{pose}}$, which are combined as $L_{\mathrm{teach}} = \frac{1}{|\mathcal{S}|}\sum_m (\beta_1 L^{(m)}_{\mathrm{corr}} + \beta_2 L^{(m)}_{\mathrm{inlier}} + \beta_3 L^{(m)}_{\mathrm{pose}})$ and added to the task loss as $L = L_{\mathrm{task}} + \lambda_t L_{\mathrm{teach}}$. Gradients are stopped through the closed-form probe solution and propagated only through the teacher representations, so the probes act as pure regularizers.

What would settle it

Train the same TeaMatch pipeline but replace the probe targets $y_j = \sigma(s_j/\tau)$ with ground-truth geometric inlier labels $q_j = \exp(-r_j^2/\delta^2)$ computed from the true pose, keeping the patch-disjoint split, degradation modes, and loss weighting unchanged. If that geometrically supervised variant does not equal or beat TeaMatch's inlier ratio and registration recall, the paper's claim that teacher-score teachability is the driving signal is not supported.

Watch

Extended reading notes

Core claim

TeaMatch's central claim is that a pair-level representation is good for 2D-3D matching only if a deliberately weak learner can recover the teacher's matching preference from it after structured degradation, and only on patch regions the learner did not imitate. The paper turns this criterion into a training objective: degrade candidate pair representations, fit ridge-regression probes in closed form on a teach split of coarse patch pairs, evaluate them on a disjoint meta split, and supervise the teacher with three losses—correspondence-score recovery, soft inlier ratio against ground-truth geometry, and differentiable PnP pose alignment. The authors report that this teachability regularization lifts mean inlier ratio from 50.1% to 54.6% and registration recall from 75.8% to 85.2% on 7-Scenes over the teacher baseline, and improves all metrics on RGB-D Scenes V2, with the auxiliary machinery removed at inference.

Load-bearing premise

The whole method rests on the assumption that 'a weak learner's ability to recover the teacher's own softened matching scores on degraded, unseen image-point patches' is a faithful measure of whether those representations will support geometrically correct correspondences and poses.

Editorial extensions

If this is right

  • Because all teachability components are dropped at inference, the framework can be added to any existing detection-free coarse-to-fine matcher with no added latency, memory, or model size.
  • On 7-Scenes the mean inlier ratio rises from 50.1% to 54.6% and registration recall from 75.8% to 85.2% relative to the teacher baseline.
  • On RGB-D Scenes V2 the method reaches 63.1% registration recall and 78.3% feature matching recall, showing the benefit carries across unseen scenes.
  • Under point-cloud occlusion up to 60%, the inlier ratio remains at 44.2% while the strongest prior baseline drops to 23.5%, indicating more graceful degradation.
  • Ablations show the gains depend on all three teachability losses and on the patch-disjoint teach/meta split; removing them or splitting at random lowers performance.

Reading between the lines

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

  • Because the training recipe is agnostic to the modality pair, the same teachability objective could be transferred to other correspondence problems, such as image-image or point-cloud-point-cloud matching, whenever candidate pair representations and soft matching targets exist; the paper does not test these settings.
  • The correspondence-level loss only needs the teacher's own soft scores, not ground-truth pose, so a semi-supervised variant could apply teachability to large unlabeled image-point collections and reserve the inlierness and pose losses for labeled subsets; the paper does not propose this.
  • A natural stress test would be to evaluate TeaMatch-trained representations under degradation types absent from the four training modes; if the gains are a genuine property of the representation, they should transfer to novel corruptions, but the paper only reports the trained modes.
  • Making the degradation family adaptive—selecting, per mini-batch, the mode that most hurts probe recovery—could push the representation harder than the fixed set of four modes studied here.
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

4 major / 5 minor

Summary. The paper proposes TeaMatch, a training-only framework for 2D-3D matching that adds a 'teachability' regularizer to a detection-free coarse-to-fine matcher. The teacher produces pair-level representations for candidate correspondences; during training, four degradation modes weaken these representations, low-capacity linear probes are fit on a patch-disjoint teach split via closed-form ridge regression, and the probes' ability to recover the teacher's soft correspondence scores, inlierness, and pose on the meta split is used as an additional loss. At inference all probes and losses are removed, so the pipeline is unchanged. Experiments on 7-Scenes and RGB-D Scenes V2 report improvements over 2D3D-MATR and several older baselines, with a claimed state-of-the-art result.

Significance. If the reported results hold, the contribution is a novel and inexpensive regularization scheme: it requires no architectural change or inference overhead, and the idea of using weak, degraded probes as training-time constraints is original in this domain. The paper also delivers consistent internal ablations showing that each loss component and each degradation mode contributes positively, and the patch-disjoint split is a sensible precaution against trivial memorization. The main open risk is that the primary teachability signal (the correspondence loss) is defined relative to the teacher's own soft labels, so the current experiments do not yet establish that the gains come from geometric robustness rather than from a generic linear-decodability regularizer; control experiments with ground-truth-anchored probe targets are needed to close this gap. The absence of error bars and the narrow baseline set also temper the 'state-of-the-art' claim.

major comments (4)
  1. [Sec. 3.3.4 / Eq. (11), (15)] The correspondence teachability loss measures how well a linear probe recovers the teacher's own softened logits y_j = sigma(s_j/tau). This makes the core criterion self-referential: a representation can become easier for the probe to imitate without becoming more geometrically correct. The ablation in Table 4 shows that adding only L_corr raises RR from 75.8 to 79.6 (nearly half of the total gain to 85.2), so a large part of the reported improvement is driven by this self-referential term. To support the claim that teachability reflects robust, geometry-preserving representations, the authors should add a control experiment in which the probe targets are replaced by ground-truth correspondence labels or by the geometric inlierness q_j of Eq. (16), and compare the resulting IR/FMR/RR with the teacher-target version. If the teacher-target probe only acts as a regularizer, the current interpretation and the term 'teachability' should be recalibrated.
  2. [Sec. 3.3.5 / Eqs. (12)-(13)] The paper stops gradients through the closed-form ridge solution W* and propagates only through the representations and B_m. This means the teach split is not directly trained through the probe-fitting step; the training signal on the meta split uses a probe whose weights are treated as fixed. This is a non-trivial design choice with no ablations or analysis. The authors should either compare full differentiation through the ridge solution against the stop-gradient variant, or provide a clear justification for why the stop-gradient variant is preferred, and report any effect on the final metrics. Without this, it is unclear whether the closed-form adaptation is actually load-bearing or whether a fixed random projection would give similar results.
  3. [Tables 1-2 and abstract] The 'state-of-the-art' claim is supported by comparison with only five baselines, the most recent of which (Diff2I2P) is outperformed by margins of 2.2 and 2.6 percentage points in RR, with no error bars or significance tests. Run-to-run variance of this magnitude is often substantial in 2D-3D matching. The authors should report means and standard deviations over at least three seeds for the main comparisons, and include at least one recent competitive detection-free method (e.g., CorrI2P or VP2P-Match) to substantiate the state-of-the-art claim.
  4. [Introduction and Sec. 4.1] The contribution statement claims TeaMatch 'consistently improves strong detection-free baselines', and the abstract states that it can be 'seamlessly integrated into existing coarse-to-fine matching pipelines'. However, all experiments are run on a single teacher (the 2D3D-MATR architecture; see Table 3, where the teacher baseline matches the MATR row of Table 1). No experiment applies TeaMatch to Diff2I2P or another pipeline, so the generality claim is not demonstrated. Either add experiments on a second teacher or revise the claims to be specific to the 2D3D-MATR teacher.
minor comments (5)
  1. [Sec. 3.1 and 3.5] The citation 'PnP-RANSAC [14?]' contains a stray question mark; reference [14] is RANSAC (Fischler and Bolles), and the '?' suggests a broken cross-reference. Please fix the citation and also cite EPnP properly where the solver is described.
  2. [Figure 2] Several symbols in the figure (e.g., '?', '??', '??') are rendered as placeholder glyphs, making the architecture diagram hard to read. Provide a clean version with all notations matching the text.
  3. [Table 1] The RR row for Diff2I2P appears as '99.095.6 74.098.986.8...' without spacing; reformat the table so that all numbers are clearly separated.
  4. [Sec. 4.4] The paper does not report the values of the hyperparameters introduced in the method (rho, tau, delta, lambda_r, lambda_t, beta_1-3, lambda_f, lambda_p). Since TeaMatch has a large number of added coefficients, please include the chosen values and a sensitivity analysis for at least the most important ones (e.g., rho and lambda_t).
  5. [References] Many reference entries only provide a Semantic Scholar API URL instead of the original venue and DOI; please reformat the bibliography to the ACM style.

Circularity Check

1 steps flagged · score 5.0 of 10

The correspondence-level teachability loss reduces to predicting the teacher's own softened logits, though GT-anchored losses keep the overall method partially independent.

  1. self definitional [Sec. 3.3.4 (Eq. 11) and Sec. 3.4.1 (Eq. 15)]
    "The weak probes are trained to recover the teacher's soft preference over candidate correspondences. We define the scalar soft target of each candidate as y_j = σ(s_j/τ), (11) ... At the correspondence level, the probe is required to recover the teacher's soft preference on unseen local regions: L_corr = 1/n_m ||P_meta − Y_meta||^2. (15)"

    Y_meta (Eq. 11) is built from y_j = σ(s_j/τ), where s_j is the teacher-predicted matching logit of the very network being optimized (Eq. 4). Eq. (15) then minimizes ||P_meta − Y_meta||^2, so the correspondence-level teachability loss rewards representations from which a weak ridge probe linearly recovers the teacher's own softened scores. The target is the teacher's own output, not ground-truth correspondence labels or geometric inlierness; the patch-disjoint meta split changes which regions are predicted, but not the self-generated nature of the target. Minimizing Eq. (15) is therefore, by construction, optimizing linear self-decodability of the teacher's score geometry. Tab.

full rationale

TeaMatch's final evaluation is externally grounded: IR, FMR, RR, and PIR are measured against ground-truth geometry and pose, and the inlierness and pose teachability losses (Eqs. 19 and 21) use ground-truth residuals and ground-truth pose, so the central empirical comparisons are not forced by construction. No load-bearing self-citation is present: the author-authored references [15] and [39] are background related-work citations only. The circular component is the correspondence-level teachability criterion itself: Eq. 15 uses Eq. 11's teacher-generated soft targets as the quantity to be 'recovered', making that loss a self-distillation and linear-decodability regularizer rather than a ground-truth-anchored measure of robustness. The paper even acknowledges that 'Recovering teacher preference alone is insufficient' before adding the GT-anchored losses, and Tab. 4 shows the isolated L_corr gain (75.8 to 79.6 RR) is empirically load-bearing. Score 5 reflects partial circularity: one loss component reduces by construction to self-imitation, while the full pipeline retains independent GT-anchored supervision and external evaluation.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The method introduces no new physical or conceptual entities beyond the training procedure itself. The central claim rests on a set of hand-chosen hyperparameters and on the domain assumption that recoverability by a weak linear probe is a faithful proxy for representation robustness. These are reasonable but not independently justified in the paper.

free parameters (6)
  • loss weights lambda_f, lambda_p, lambda_t = not reported
    Balance the task losses in Eq. (6) and the teachability loss in Eq. (23).
  • teachability weights beta_1, beta_2, beta_3 = not reported
    Balance correspondence, inlierness, and pose losses in Eq. (22).
  • teach/meta split ratio rho = not reported
    Controls the proportion of coarse patch pairs in the teach and meta splits (Sec. 3.3.3).
  • temperature tau = not reported
    Softens teacher logits to produce soft targets in Eq. (11).
  • inlier scale delta = not reported
    Converts geometric residuals into soft inlier targets in Eq. (16).
  • ridge regularization lambda_r = not reported
    Regularizes the closed-form probe fit in Eq. (12).
assumptions (5)
  • standard math The matrix (H_teach^T H_teach + lambda_r I) is invertible for every training sample and mini-batch
    Required for the closed-form probe solution in Eq. (13).
  • domain assumption A linear ridge probe operating on degraded features is a sufficient learner to expose the quality of the teacher's representations
    Core premise of the teachability criterion (Sec. 3.3).
  • domain assumption A patch-disjoint split within a single image-point cloud pair approximates cross-region generalization
    Justifies the teach/meta split in Sec. 3.3.3 and the claim that the probe cannot exploit local redundancy.
  • ad hoc to paper Stopping gradients through the closed-form W* and propagating only through representations and B_m yields a valid training signal
    Stated in Sec. 3.3.5 without formal justification or empirical validation of the approximation.
  • domain assumption Ground-truth pose T_gt is available for all training pairs
    Used to define residuals in Eq. (16) and the pose target in Eq. (20).

how reviews work

0 comments
Cite this review

Pith. "Pith review of TeaMatch: Teachable Cross-Modal Representation Learning for 2D-3D Matching." pith.science (2026). https://pith.science/paper/ROQJP4E4

@misc{pith2026260809590,
  author       = {Pith},
  title        = {Pith review of: TeaMatch: Teachable Cross-Modal Representation Learning for 2D-3D Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ROQJP4E4}},
  note         = {Machine review of arXiv:2608.09590}
}
read the original abstract

Learning reliable correspondences between images and point clouds is fundamental for 2D-3D matching. Despite recent progress in detection-free methods, existing approaches primarily optimize matching within a single model and often struggle to maintain reliable correspondences under challenging conditions such as noisy inputs, low overlap, and ambiguous structures. In this work, we propose TeaMatch, a novel framework that introduces teachability as a criterion for cross-modal representation learning. We define teachability as the ability of a representation to be effectively recovered by weak learners under degraded inputs, reflecting its structural consistency and robustness. To this end, we construct a set of task-specific weak students that simulate common failure modes and train them to imitate the teacher on a training split while evaluating their recoverability on a disjoint meta split. The teacher is then optimized to improve the students' ability to recover reliable correspondences, guided by correspondence-level and geometry-aware constraints. Our framework can be seamlessly integrated into existing coarse-to-fine matching pipelines without additional inference cost. Extensive experiments demonstrate that TeaMatch improves matching robustness and achieves state-of-the-art performance on challenging 2D-3D matching benchmarks.

Figures

Figures reproduced from arXiv: 2608.09590 by the authors.

Figure 1
Figure 1. Qualitative comparison of correspondence local [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of TeaMatch. A detection-free coarse-to-fine teacher first produces candidate 2D–3D correspondences [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Qualitative results on the 7-Scenes dataset. The red lines indicate erroneous correspondences (3D distance greater [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Robustness under local point cloud degradation. From left to right, the occlusion ratio increases from [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 56 canonical work pages

  1. [1]

    Guillaume Alain and Yoshua Bengio. 2016. Understanding intermediate lay- ers using linear classifier probes.ArXivabs/1610.01644 (2016). https://api. semanticscholar.org/CorpusID:9794990

  2. [2]

    Markham, and Yulan Guo

    Sheng Ao, Qingyong Hu, Bo Yang, A. Markham, and Yulan Guo. 2020. SpinNet: Learning a General Surface Descriptor for 3D Point Cloud Registration.2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2020), 11748–11757. https://api.semanticscholar.org/CorpusID:227151932

  3. [3]

    Ronald T. Azuma. 1997. A Survey of Augmented Reality.Presence: Teleopera- tors & Virtual Environments6 (1997), 355–385. https://api.semanticscholar.org/ CorpusID:469744

  4. [4]

    Xuyang Bai, Zixin Luo, Lei Zhou, Hongkai Chen, Lei Li, Zeyu Hu, Hongbo Fu, and Chiew-Lan Tai. 2021. PointDSC: Robust Point Cloud Registration using Deep Spatial Consistency.2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2021), 15854–15864. https://api.semanticscholar.org/ CorpusID:232168842

  5. [5]

    Hong Chen, Pei Yan, Sihe Xiang, and Yihua Tan. 2024. Dynamic Cues-Assisted Transformer for Robust Point Cloud Registration.2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2024), 21698–21707. https: //api.semanticscholar.org/CorpusID:272722265

  6. [6]

    Zhi Chen, Kun Sun, Fan Yang, and Wenbing Tao. 2022. SC2-PCR: A Second Order Spatial Compatibility for Efficient and Robust Point Cloud Registration.2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2022), 13211–13221. https://api.semanticscholar.org/CorpusID:252571579

  7. [7]

    Christopher Bongsoo Choy, Jaesik Park, and Vladlen Koltun. 2019. Fully Convo- lutional Geometric Features.2019 IEEE/CVF International Conference on Computer Vision (ICCV)(2019), 8957–8965. https://api.semanticscholar.org/CorpusID: 201686687

  8. [8]

    Haowen Deng, Tolga Birdal, and Slobodan Ilic. 2018. PPFNet: Global Con- text Aware Local Features for Robust 3D Point Matching.2018 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition(2018), 195–205. https: //api.semanticscholar.org/CorpusID:3703761

Show all 65 references
  1. [9]

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. 2017. SuperPoint: Self-Supervised Interest Point Detection and Description.2018 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition Workshops (CVPRW)(2017), 337–33712. https://api.semanticscholar.org/Corpus...

  2. [10]

    Zhen Dong, Zhen Dong, Bisheng Yang, Yuan Liu, Fuxun Liang, Bi jun Li, and Yufu Zang. 2017. A novel binary shape context for 3D local surface description. Isprs Journal of Photogrammetry and Remote Sensing130 (2017), 431–452. https: //api.semanticscholar.org/CorpusID:125337993

  3. [11]

    Durrant-Whyte and Tim Bailey

    Hugh F. Durrant-Whyte and Tim Bailey. 2006. Simultaneous localization and mapping: part I.IEEE Robotics & Automation Magazine13 (2006), 99–110. https: //api.semanticscholar.org/CorpusID:8061430

  4. [12]

    Mihai Dusmanu, Ignacio Rocco, Tomás Pajdla, Marc Pollefeys, Josef Sivic, Akihiko Torii, and Torsten Sattler. 2019. D2-Net: A Trainable CNN for Joint Description and Detection of Local Features.2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2019), 80...

  5. [13]

    Ang, and Gim Hee Lee

    Mengdan Feng, Sixing Hu, Marcelo H. Ang, and Gim Hee Lee. 2019. 2D3D- Matchnet: Learning To Match Keypoints Across 2D Image And 3D Point Cloud. 2019 International Conference on Robotics and Automation (ICRA)(2019), 4790–

  6. [14]

    Fischler and Robert C

    Martin A. Fischler and Robert C. Bolles. 1981. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography.Commun. ACM24 (1981), 381–395. https://api.semanticscholar. org/CorpusID:972888

  7. [15]

    Junjie Gao, Qiujie Dong, Ruian Wang, Shuangmin Chen, Shiqing Xin, Changhe Tu, and Wenping Wang. 2023. OAAFormer: Robust and Efficient Point Cloud Registration Through Overlapping-Aware Attention in Transformer.Journal of Computer Science and Technology39 (2023), 755 – 770. htt...

  8. [16]

    Ben Glocker, Shahram Izadi, Jamie Shotton, and Antonio Criminisi. 2013. Real- time RGB-D camera relocalization.2013 IEEE International Symposium on Mixed and Augmented Reality (ISMAR)(2013), 173–179. https://api.semanticscholar. org/CorpusID:45781849

  9. [17]

    Zan Gojcic, Caifa Zhou, Jan Dirk Wegner, and Andreas Wieser. 2018. The Perfect Match: 3D Point Cloud Matching With Smoothed Densities.2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2018), 5540–5549. https://api.semanticscholar.org/CorpusID:53670229

  10. [18]

    Yulan Guo, Ferdous Sohel, Bennamoun, Jianwei Wan, and Min Lu. 2013. RoPS: A local feature descriptor for 3D rigid objects based on rotational projection statistics.2013 1st International Conference on Communications, Signal Processing, and their Applications (ICCSPA)(2013), 1–...

  11. [19]

    Zhang, Shaoqing Ren, and Jian Sun

    Kaiming He, X. Zhang, Shaoqing Ren, and Jian Sun. 2015. Deep Residual Learning for Image Recognition.2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)(2015), 770–778. https://api.semanticscholar.org/CorpusID: 206594692

  12. [20]

    John Hewitt and Percy Liang. 2019. Designing and Interpreting Probes with Control Tasks.ArXivabs/1909.03368 (2019). https://api.semanticscholar.org/ CorpusID:202538609

  13. [21]

    Hinton, Oriol Vinyals, and Jeffrey Dean

    Geoffrey E. Hinton, Oriol Vinyals, and Jeffrey Dean. 2015. Distilling the Knowledge in a Neural Network.ArXivabs/1503.02531 (2015). https://api. semanticscholar.org/CorpusID:7200347

  14. [22]

    Yicong Hong, Kai Zhang, Jiuxiang Gu, Sai Bi, Yang Zhou, Difan Liu, Feng Liu, Kalyan Sunkavalli, Trung Bui, and Hao Tan. 2023. LRM: Large Recon- struction Model for Single Image to 3D.ArXivabs/2311.04400 (2023). https: //api.semanticscholar.org/CorpusID:265050698

  15. [23]

    Chen Huang, Oier Mees, Andy Zeng, and Wolfram Burgard. 2022. Visual Lan- guage Maps for Robot Navigation.2023 IEEE International Conference on Robotics and Automation (ICRA)(2022), 10608–10615. https://api.semanticscholar.org/ CorpusID:252846548

  16. [24]

    Shengyu Huang, Zan Gojcic, Mikhail (Misha) Usvyatsov, Andreas Wieser, and Konrad Schindler. 2020. PREDATOR: Registration of 3D Point Clouds with Low Overlap.2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2020), 4265–4274. https://api.semanticscholar...

  17. [25]

    Haobo Jiang, Zheng Dang, Zhen Wei, Jin Xie, Jian Yang, and Mathieu Salzmann

  18. [26]

    Kevin Lai, Liefeng Bo, and Dieter Fox. 2014. Unsupervised feature learning for 3D scene labeling.2014 IEEE International Conference on Robotics and Automation (ICRA)(2014), 3050–3057. https://api.semanticscholar.org/CorpusID:15015476

  19. [27]

    Jae Yong Lee, Joseph DeGol, Victor Fragoso, and Sudipta N. Sinha. 2021. PatchMatch-Based Neighborhood Consensus for Semantic Correspondence.2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2021), 13148–13158. https://api.semanticscholar.org/CorpusID:235719539

  20. [28]

    Vincent Lepetit, Francesc Moreno-Noguer, and Pascal Fua. 2009. EPnP: An Accurate𝑂(𝑛)Solution to the PnP Problem.International Journal of Computer Vision81, 2 (2009), 155–166. doi:10.1007/s11263-008-0152-6

  21. [29]

    Vincent Leroy, Yohann Cabon, and Jérôme Revaud. 2024. Grounding Image Matching in 3D with MASt3R. InEuropean Conference on Computer Vision. https: //api.semanticscholar.org/CorpusID:270521424

  22. [30]

    Jiayuan Li, Qingwu Hu, and Mingyao Ai. 2020. GESAC: Robust graph enhanced sample consensus for point cloud registration.ISPRS Journal of Photogrammetry and Remote Sensing(2020). https://api.semanticscholar.org/CorpusID:225238083

  23. [31]

    Jiaxin Li and Gim Hee Lee. 2021. DeepI2P: Image-to-Point Cloud Registration via Deep Classification.2021 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR)(2021), 15955–15964. https://api.semanticscholar.org/ CorpusID:233181563

  24. [32]

    Minhao Li, Zheng Qin, Zhirui Gao, Renjiao Yi, Chenyang Zhu, Yulan Guo, and Kaiyang Xu. 2023. 2D3D-MATR: 2D-3D Matching Transformer for Detection- free Registration between Images and Point Clouds.2023 IEEE/CVF Inter- national Conference on Computer Vision (ICCV)(2023), 14082–1...

  25. [33]

    Costain, Henry Howard-Jenkins, and Victor Adrian Prisacariu

    Shuda Li, Kai Han, Theo W. Costain, Henry Howard-Jenkins, and Victor Adrian Prisacariu. 2020. Correspondence Networks With Adaptive Neighbourhood Consensus.2020 IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR)(2020), 10193–10202. https://api.semanticscho...

  26. [34]

    David G. Lowe. 1999. Object recognition from local scale-invariant features. Proceedings of the Seventh IEEE International Conference on Computer Vision2 (1999), 1150–1157 vol.2. https://api.semanticscholar.org/CorpusID:5258236

  27. [35]

    Zixin Luo, Lei Zhou, Xuyang Bai, Hongkai Chen, Jiahui Zhang, Yao Yao, Shiwei Li, Tian Fang, and Long Quan. 2020. ASLFeat: Learning Local Features of Accurate Shape and Localization.2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2020), 6588–6597. htt...

  28. [36]

    Juncheng Mu, Chengwei Ren, Weixiang Zhang, Liang Pan, Xiao-Ping Zhang, and Yue Gao. 2025. Diff2I2P: Differentiable Image-to-Point Cloud Registration with Diffusion Prior. InProceedings of the IEEE/CVF International Conference on Computer Vision. 25777–25787

  29. [37]

    B. K. Patle, Ganesh Babu L., Anish Pandey, Dayal Ramakrushna Parhi, and Anne Jagadeesh. 2019. A review: On path planning strategies for navigation of mobile MM ’26, November 10–14, 2026, Rio de Janeiro, Brazil Chongjian Wang and Junjie Gao robot.Defence Technology(2019). https...

  30. [38]

    Quang-Hieu Pham, Mikaela Angelina Uy, Binh-Son Hua, Duc Thanh Nguyen, Gemma Roig, and Sai-Kit Yeung. 2019. LCD: Learned Cross-Domain Descriptors for 2D-3D Matching. InAAAI Conference on Artificial Intelligence. https://api. semanticscholar.org/CorpusID:208202088

  31. [39]

    Zheng Qin, Hao Yu, Changjian Wang, Yulan Guo, Yuxing Peng, Slobodan Ilic, Dewen Hu, and Kai Xu. 2023. GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer.IEEE Transactions on Pattern Analysis and Machine Intelligence45 (2023), 9806–9821. https:/...

  32. [40]

    Siyu Ren, Yiming Zeng, Junhui Hou, and Xiaodong Chen. 2022. CorrI2P: Deep Image-to-Point Cloud Registration via Dense Correspondence.IEEE Transactions on Circuits and Systems for Video Technology33 (2022), 1198–1208. https://api. semanticscholar.org/CorpusID:250451486

  33. [41]

    Ignacio Rocco, Relja Arandjelovi’c, and Josef Sivic. 2020. Efficient Neighbourhood Consensus Networks via Submanifold Sparse Convolutions. InEuropean Confer- ence on Computer Vision. https://api.semanticscholar.org/CorpusID:216056155

  34. [42]

    Ignacio Rocco, Mircea Cimpoi, Relja Arandjelović, Akihiko Torii, Tomás Pajdla, and Josef Sivic. 2018. Neighbourhood Consensus Networks. InNeural Information Processing Systems. https://api.semanticscholar.org/CorpusID:53084192

  35. [43]

    Adriana Romero, Nicolas Ballas, Samira Ebrahimi Kahou, Antoine Chassang, Carlo Gatta, and Yoshua Bengio. 2014. FitNets: Hints for Thin Deep Nets.CoRR abs/1412.6550 (2014). https://api.semanticscholar.org/CorpusID:2723173

  36. [44]

    Ethan Rublee, Vincent Rabaud, Kurt Konolige, and Gary R. Bradski. 2011. ORB: An efficient alternative to SIFT or SURF.2011 International Conference on Computer Vision(2011), 2564–2571. https://api.semanticscholar.org/CorpusID:206769866

  37. [45]

    Radu Bogdan Rusu, Nico Blodow, and Michael Beetz. 2009. Fast Point Feature Histograms (FPFH) for 3D registration.2009 IEEE International Conference on Robotics and Automation(2009), 3212–3217. https://api.semanticscholar.org/ CorpusID:15022990

  38. [46]

    Samuele Salti, Federico Tombari, and Luigi di Stefano. 2014. SHOT: Unique signatures of histograms for surface and texture description.Comput. Vis. Image Underst.125 (2014), 251–264. https://api.semanticscholar.org/CorpusID:17799553

  39. [47]

    Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabi- novich. 2019. SuperGlue: Learning Feature Matching With Graph Neural Net- works.2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2019), 4937–4946. https://api.semanticscholar.org...

  40. [48]

    Selvaraju, Abhishek Das, Ramakrishna Vedantam, Michael Cogswell, Devi Parikh, and Dhruv Batra

    Ramprasaath R. Selvaraju, Abhishek Das, Ramakrishna Vedantam, Michael Cogswell, Devi Parikh, and Dhruv Batra. 2016. Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization.International Journal of Computer Vision128 (2016), 336 – 359. https://api.seman...

  41. [49]

    Jiaming Sun, Zehong Shen, Yuang Wang, Hujun Bao, and Xiaowei Zhou. 2021. LoFTR: Detector-Free Local Feature Matching with Transformers.2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2021), 8918–8927. https://api.semanticscholar.org/CorpusID:232478646

  42. [50]

    Qi, Jean-Emmanuel Deschaud, Beatriz Marcotegui, François Goulette, and Leonidas J

    Hugues Thomas, C. Qi, Jean-Emmanuel Deschaud, Beatriz Marcotegui, François Goulette, and Leonidas J. Guibas. 2019. KPConv: Flexible and Deformable Con- volution for Point Clouds.2019 IEEE/CVF International Conference on Computer Vision (ICCV)(2019), 6410–6419. https://api.sema...

  43. [51]

    McLauchlan, Richard I

    Bill Triggs, Philip F. McLauchlan, Richard I. Hartley, and Andrew W. Fitzgibbon

  44. [52]

    Bing Wang, Changhao Chen, Zhaopeng Cui, Jie Qin, Chris Xiaoxuan Lu, Zhewen Yu, Peijun Zhao, Zhenchao Dong, Fan Zhu, Niki Trigoni, and A. Markham. 2021. P2-Net: Joint Description and Detection of Local Features for Pixel and Point Matching.2021 IEEE/CVF International Conference...

  45. [53]

    Haiping Wang, Yuan Liu, Bing Wang, Yujing Sun, Zhen Dong, Wenping Wang, and Bisheng Yang. 2023. FreeReg: Image-to-Point Cloud Registration Leveraging Pretrained Diffusion Models and Monocular Depth Estimators.arXiv preprint arXiv:2310.03420(2023)

  46. [54]

    Xuejun Xing, Zhengda Lu, Yiqun Wang, and Jun Xiao. 2024. Efficient Single Correspondence Voting for Point Cloud Registration.IEEE Transactions on Im- age Processing33 (2024), 2116–2130. https://api.semanticscholar.org/CorpusID: 268373790

  47. [55]

    Runzhao Yao, Shaoyi Du, Wenting Cui, Canhui Tang, and Chengwu Yang. 2024. PARE-Net: Position-Aware Rotation-Equivariant Networks for Robust Point Cloud Registration. InEuropean Conference on Computer Vision. https://api. semanticscholar.org/CorpusID:271212761

  48. [56]

    Zi Jian Yew and Gim Hee Lee. 2018. 3DFeat-Net: Weakly Supervised Local 3D Features for Point Cloud Registration.ArXivabs/1807.09413 (2018). https: //api.semanticscholar.org/CorpusID:50770050

  49. [57]

    Hao Yu, Fu Li, Mahdi Saleh, Benjamin Busam, and Slobodan Ilic. 2021. CoFiNet: Reliable Coarse-to-fine Correspondences for Robust Point Cloud Registration. In Neural Information Processing Systems. https://api.semanticscholar.org/CorpusID: 239998094

  50. [58]

    Hao Yu, Zheng Qin, Jiafan Hou, Mahdi Saleh, Dongsheng Li, Benjamin Busam, and Slobodan Ilic. 2023. Rotation-Invariant Transformer for Point Cloud Matching. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023), 5384–5393. https://api.semanticscholar...

  51. [59]

    Sergey Zagoruyko and Nikos Komodakis. 2016. Paying More Attention to Atten- tion: Improving the Performance of Convolutional Neural Networks via Attention Transfer.ArXivabs/1612.03928 (2016). https://api.semanticscholar.org/CorpusID: 829159

  52. [60]

    Junsheng Zhou, Baorui Ma, Wenyuan Zhang, Yi Fang, Yu-Shen Liu, and Zhizhong Han. 2023. Differentiable Registration of Images and LiDAR Point Clouds with VoxelPoint-to-Pixel Matching.ArXivabs/2312.04060 (2023). https: //api.semanticscholar.org/CorpusID:266051996

  53. [61]

    Qunjie Zhou, Maxim Maximov, Or Litany, and Laura Leal-Taix’e. 2024. The NeR- Fect Match: Exploring NeRF Features for Visual Localization. InEuropean Con- ference on Computer Vision. https://api.semanticscholar.org/CorpusID:268384911

  54. [62]

    Qunjie Zhou, Torsten Sattler, and Laura Leal-Taixé. 2020. Patch2Pix: Epipolar- Guided Pixel-Level Correspondences.2021 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR)(2020), 4667–4676. https://api. semanticscholar.org/CorpusID:227254594

  55. [1999]

    InWorkshop on Vision Algorithms

    Bundle Adjustment - A Modern Synthesis. InWorkshop on Vision Algorithms. https://api.semanticscholar.org/CorpusID:1354186

  56. [2023]

    https://api.semanticscholar.org/CorpusID:257921545

    Robust Outlier Rejection for 3D Registration with Variational Bayes.2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2023), 1148–1157. https://api.semanticscholar.org/CorpusID:257921545

  57. [4796]

    https://api.semanticscholar.org/CorpusID:128341753

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.