Pith. sign in

REVIEW 4 major objections 6 minor 34 references

Adaptive Margin Contrastive Learning for Ambiguity-aware 3D Semantic Segmentation

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper claims that per-point ambiguity—estimated from label disagreement among a point's nearest neighbours—should set the margin of a supervised contrastive loss, with negative margins for highly ambiguous points, and that this lifts…

desk verdict A sensible, modestly effective 3D segmentation method whose ambiguity proxy is plausible but unproven; worth a serious look but needs robustness checks. read the letter →

arxiv 2502.04111 v1 pith:EGGAGGA4 submitted 2025-02-06 cs.CV

Yang Chen , Yueqi Duan , Runzhong Zhang , Yap-Peng Tan This is my paper · ORCID
classification cs.CV
keywords 3Dsemanticsegmentationpointcloudcontrastivelearningadaptivemarginambiguityestimationtransitionregionssupervisedloss
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

AMContrast3D argues that 3D semantic segmentation fails when the training loss treats every point equally, because points near transitions between classes are intrinsically ambiguous and their labels are unreliable. The paper's proposal is to read each point's ambiguity from the labels of its Euclidean nearest neighbors and convert that ambiguity into a per-point margin in a supervised contrastive loss: low-ambiguity points get positive margins that force hard separation, borderline points get zero margins, and highly ambiguous points get negative margins that relax the objective. With this schedule, the model should concentrate learning on points it can actually get right while not being derailed by mislabeled or undecidable points. On S3DIS Area 5 and ScanNet, the method reports higher mIoU than the PointNeXt baseline and than prior contrastive-boundary methods. If correct, it would mean that decision boundaries in dense 3D prediction can be usefully per-point and data-dependent rather than uniform.

What carries the argument

The load-bearing object is the pair formed by the ambiguity estimator and the margin generator. The estimator uses position embeddings $p_i$ to separate the $K$-nearest neighbourhood into intra-points $N^+_i$ (same label) and inter-points $N^-_i$ (different label), computes closeness centralities $cc^+_i = |N^+_i|/d^+_i$ and $cc^-_i = |N^-_i|/d^-_i$, and maps their difference through an inverse sigmoid (with piecewise endpoints at 0 and 1) to get $a_i$. The margin generator then sets $m_i = \mu a_i + \nu$, and the contrastive objective in Eq. (10) uses $\exp((\mathrm{sim}(f_i,f_j)-m_i)/\tau)$ for intra-pairs, so the effective required separation between intra- and inter-similarity is $m_i$ instead of 0. With $\mu=-1$ and $\nu=0.5$ on S3DIS or $\nu=0.6$ on ScanNet, low-ambiguity points face a positive margin, semi-ambiguous points face zero, and the most ambiguous points face a negative margin, which is what lets the training difficulty vary point by point.

What would settle it

Train the same PointNeXt architecture with the AMContrast3D loss on S3DIS Area 5, but replace the estimated ambiguity $a_i$ with random per-point margins drawn from the same distribution of $m_i$; if random margins match the reported mIoU, the improvement is not driven by the ambiguity signal. A complementary check is to corrupt the neighbour labels used by the estimator while keeping the training labels intact: if performance does not drop, the estimator is not the source of the gain.

Watch

Extended reading notes

Core claim

The central claim is that the additive margin idea from classification can be transplanted into point-level supervised contrastive learning for 3D point clouds, with the margin made a linear function of an estimated per-point ambiguity: $m_i = \mu a_i + \nu$. The ambiguity $a_i \in [0,1]$ is computed from position embeddings by counting how many of the point's $K$ nearest neighbours disagree with its label and comparing their closeness centralities; points surrounded only by same-label neighbours get $a_i=0$, and points whose neighbourhood is fully conflicting get $a_i=1$. Injecting this margin into the exponent of the contrastive softmax, $\exp((\mathrm{sim}(f_i,f_j)-m_i)/\tau)$, shifts the decision boundary between intra-class and inter-class similarity so that the required separation shrinks or reverses as ambiguity grows. The paper reports that including this adaptive term, weighted by $1-\lambda$ alongside cross-entropy, improves mIoU to 71.8% on S3DIS Area 5 and 72.6% on ScanNet test, and its ablation shows the gain disappears when the margin is constant or clipped to be non-negative.

Load-bearing premise

The whole method rests on the premise that a point's ambiguity—and therefore the right amount of training pressure—can be read off from label disagreement among its nearest neighbours in 3D position space, even though the paper itself notes that labels near transition regions are questionable for human annotators.

Editorial extensions

If this is right

  • Training difficulty becomes a per-point quantity: the same loss formula applies at every point but with decision boundaries that depend on the local label configuration of the scene.
  • Points sitting on semantic boundaries are explicitly de-emphasized, so gradients concentrate on interior points whose labels are reliable, which should make training more stable than a uniform contrastive term.
  • The margin generator is decoupled from the backbone and can be attached to any point-level contrastive loss used in supervised 3D segmentation.
  • Negative margins are reported as essential: clipping them at zero (the last row of the ablation) drops mIoU from 71.8% to 70.5%, indicating that the relaxation, not merely the adaptivity, carries part of the gain.

Reading between the lines

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

  • The same ambiguity-to-margin map could be read as a self-derived label-noise weighting: it downweights exactly the points whose neighbour disagreement makes labels suspect, so the mechanism may be a general ambiguity-weighting principle rather than a specifically contrastive one.
  • One testable extension is to feed the estimated ambiguity $a_i$ into the cross-entropy term as well, for instance as instance-dependent label smoothing or loss weights; if the gains persist, the margin is a vehicle for a broader ambiguity-aware objective.
  • If the ambiguity proxy is sound, it could transfer to other dense 3D tasks such as instance segmentation or object detection, where transition regions between objects are also the hardest to annotate consistently.
  • A stronger test would replace the inverse-sigmoid curve with the raw count of disagreeing neighbours; the paper's ablations do not isolate the centrality weighting from the count, so the contribution of the closeness term remains open.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes AMContrast3D, a supervised contrastive learning method for 3D point cloud semantic segmentation with per-point adaptive margins. The method first estimates a scalar ambiguity per point from the label agreement of its K nearest neighbors in position space (Eqs. 1-4), then maps the ambiguity to a margin mi = mu*ai + nu (Eq. 7). The margin is inserted into the supervised contrastive objective (Eqs. 9-10), which is jointly trained with cross-entropy (Eq. 11). Experiments on S3DIS Area 5 and ScanNet report mIoU gains over the PointNeXt baseline of about 1.3 and 1.4 points, and an ablation on S3DIS shows that the best configuration uses negative margins for high-ambiguity points.

Significance. If the central mechanism is sound, the paper makes a modest but useful contribution: it introduces a simple, parameter-light way to make contrastive objectives adaptive to per-point difficulty in 3D segmentation, and the improvements are consistent across two widely used benchmarks. The formulation is internally coherent, and the ablation supports the claim that negative margins, in particular, are beneficial. The main strengths are the clean integration of a margin generator into an existing backbone and the explicit ablation of the margin mapping. The principal weakness is that the ambiguity estimator itself is not validated, so the mechanistic interpretation of the gains remains open. The significance is therefore conditional: the method is plausible and potentially reproducible, but its conceptual claim requires additional evidence.

major comments (4)
  1. [Section III-B, Eqs. (1)-(4) and Table III] The ambiguity proxy ai is the sole driver of the adaptive margin, yet the paper never validates that position-space label disagreement actually measures the feature-space ambiguity that motivates the method. The ablation in Table III varies only the margin parameters mu and nu; it does not test whether the specific assignment of ai carries the signal. We request an experiment that breaks the link between ai and the labels/geometry while keeping the same marginal distribution, for example by permuting ai across points, or by using a constant or reversed assignment. In addition, report on a held-out set the correlation of ai with per-point prediction error or model confidence. If the proxy mainly tracks annotation noise in transition regions, as the paper itself argues those labels are unreliable, the adaptive margins may be set by noise rather than by meaningful ambiguity.
  2. [Tables I and II] All reported numbers are single runs, and the improvements over the PointNeXt baseline are 1.3 and 1.4 mIoU. For 3D semantic segmentation, these differences are small enough that run-to-run variance could change the ranking. Please report the mean and standard deviation over at least three random seeds for the main comparison, and if possible a paired significance test. This is necessary to support the claim that the improvement is not an artifact of a single run.
  3. [Section IV-A and IV-C] The margin parameters are set per dataset (mu=-1, nu=0.5 for S3DIS; mu=-1, nu=0.6 for ScanNet) but the paper gives no sensitivity analysis or protocol for selecting them. The ablation in Table III covers only five manually chosen settings on S3DIS, and the ScanNet value of nu=0.6 is not justified. Please provide a sensitivity sweep over nu (and, secondarily, mu and tau) and state how the final values were chosen, for example by a validation split. Otherwise it is unclear whether the reported gain comes from the adaptive mechanism or from per-dataset hyperparameter tuning.
  4. [Section III-B and III-C] The ambiguity estimate is computed from ground-truth labels of K nearest neighbors, which are exactly the labels that the paper argues are unreliable in transition regions. This raises a correctness risk: if label noise is concentrated at the same transition points, ai may encode noise rather than task difficulty, and the margin reweighting may be responding to that noise. We recommend a robustness experiment in which the training labels are synthetically perturbed near boundaries, and checking whether the method's advantage (and the ordering of margins) remains stable. This is a direct test of the mechanism proposed in Eqs. (7)-(10).
minor comments (6)
  1. [Section III-B, Eqs. (1)-(2)] The quantities cc+ and cc- are called closeness centrality, but the formula is the inverse of the mean squared distance, not the standard closeness centrality over shortest paths. Please align the terminology with the cited definition or use a different name such as 'inverse mean squared distance'.
  2. [Section III-B, Eq. (3)] The function G is described as an 'inverse sigmoid,' but it is in fact a decreasing logistic (sigmoid) function. The wording should be corrected to avoid confusion.
  3. [Section IV-A] The implementation details omit several items needed for reproduction: the optimizer and its settings, batch size, weight decay, and the label-embedding strategy borrowed from CBL. Please specify these.
  4. [Table II and Section IV-B] On ScanNet validation, AMContrast3D achieves 72.5 mIoU, which is lower than PointMetaBase's 72.8, yet the text says the method 'outperforms' without qualifying that this refers to the test set. Please clarify the comparison.
  5. [Section IV-C] The phrase 'significant improvements' is used without a statistical test. Given the single-run results, please replace it with 'reported improvements' or add significance testing.
  6. [Eq. (8)] The notation DB+ and DB- is used both for decision boundaries and for the inequalities defining the margin. Please separate the boundary plane from the margin region to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method is validated on external benchmarks and its components (ambiguity proxy, margin generator, contrastive objective) are not derived from the outcomes they predict.

full rationale

The paper's central claim is empirical: AMContrast3D improves mIoU over PointNeXt on S3DIS Area 5 and ScanNet (Tables I and II). These evaluations are performed on held-out benchmarks and are not constructed from the reported numbers. The ambiguity estimate ai in Eq. (4) is computed from ground-truth label disagreement among K Euclidean nearest neighbors in position space; this is a stated modeling assumption, not a quantity fitted to the evaluation outcome. The margin generator mi = mu*ai + nu (Eq. 7) uses two scalar hyperparameters (mu, nu) that are fixed per dataset and ablated in Table III on S3DIS; although the choice of mu = -1 and nu = 0.5 may have been selected with knowledge of benchmark performance, the paper does not present these values as predictions derived from the benchmarks, and the ablation shows a range of settings with lower performance rather than a forced equivalence. The contrastive objective Ls_AM in Eq. (10) and total loss in Eq. (11) are standard supervised contrastive forms with a per-point margin; no equation reduces by construction to the reported mIoU. The paper cites prior work by one of its authors ([5], [14], [22]), but these citations are contextual references to point-cloud methods and are not load-bearing for the AMContrast3D mechanism, nor do they import a uniqueness theorem or ansatz that defines the method. The weakest point is the unvalidated link between the position-space label-disagreement proxy ai and the feature-space ambiguity that motivates relaxing constraints; this is a legitimate correctness or generalization concern, but it is not circularity because the proxy is an input assumption, not the output of the derivation. Under the stated rules, the absence of a quoted equation-to-equation reduction or a fitted-parameter-renamed-as-prediction means the appropriate finding is no significant circularity.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical or architectural entities; the per-point ambiguity ai and margin mi are functions of existing data and hyperparameters. The ledger lists the hand-chosen hyperparameters and the domain assumptions that carry the method's intuition.

free parameters (8)
  • beta (inverse sigmoid width) = 0.04
    Controls how sharply cc+ minus cc- maps to ambiguity in Eq. 3; chosen by hand.
  • margin scale mu (S3DIS) = -1
    Selected via ablation in Table III on S3DIS Area 5; determines negative and positive margin range.
  • margin bias nu (S3DIS) = 0.5
    Selected via ablation in Table III; sets the ambiguity level where the margin crosses zero.
  • margin scale mu (ScanNet) = -1
    Per-dataset setting; no ablation shown for ScanNet.
  • margin bias nu (ScanNet) = 0.6
    Per-dataset setting; shifts the zero-crossing to ai=0.6.
  • temperature tau = 0.3 (S3DIS), 0.5 (ScanNet)
    Contrastive temperature in Eq. 9; set per dataset.
  • loss balance lambda = 0.1
    Weights cross-entropy versus contrastive loss in Eq. 11.
  • neighborhood size K = 24
    Determines the N+ and N- sets in the ambiguity estimation framework.
assumptions (4)
  • domain assumption Closeness centrality in a local graph captures per-point ambiguity when computed separately for intra-label and inter-label neighbors.
    Central to Section III-B; no validation that this proxy correlates with label noise or feature ambiguity beyond the final mIoU.
  • domain assumption Point labels in transition regions are unreliable, yet those same labels are used to define N+ and N- and hence the ambiguity estimate.
    Stated in Section I and used in Section III-B; if the labels are unreliable, the supervision signal for ambiguity is also uncertain.
  • ad hoc to paper A linear margin mapping mi = mu times ai plus nu, including negative margins, is an appropriate way to translate ambiguity into contrastive decision boundaries.
    Introduced in Eq. 7; supported only by the S3DIS ablation in Table III with per-dataset parameters.
  • domain assumption Supervised contrastive learning improves 3D semantic segmentation.
    Taken from CBL [7] and used as the base objective in Eq. 10.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Margin Contrastive Learning for Ambiguity-aware 3D Semantic Segmentation." pith.science (2026). https://pith.science/paper/EGGAGGA4

@misc{pith2026250204111,
  author       = {Pith},
  title        = {Pith review of: Adaptive Margin Contrastive Learning for Ambiguity-aware 3D Semantic Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EGGAGGA4}},
  note         = {Machine review of arXiv:2502.04111}
}
read the original abstract

In this paper, we propose an adaptive margin contrastive learning method for 3D point cloud semantic segmentation, namely AMContrast3D. Most existing methods use equally penalized objectives, which ignore per-point ambiguities and less discriminated features stemming from transition regions. However, as highly ambiguous points may be indistinguishable even for humans, their manually annotated labels are less reliable, and hard constraints over these points would lead to sub-optimal models. To address this, we design adaptive objectives for individual points based on their ambiguity levels, aiming to ensure the correctness of low-ambiguity points while allowing mistakes for high-ambiguity points. Specifically, we first estimate ambiguities based on position embeddings. Then, we develop a margin generator to shift decision boundaries for contrastive feature embeddings, so margins are narrowed due to increasing ambiguities with even negative margins for extremely high-ambiguity points. Experimental results on large-scale datasets, S3DIS and ScanNet, demonstrate that our method outperforms state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2502.04111 by the authors.

Figure 1
Figure 1. Adaptive margin from ambiguity. An illustration among (a) position [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The AMContrast3D with encoder-decoder network architecture. In the ambiguity estimation framework following the [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Ambiguity visualization. A 3D point cloud scene is categorized by [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visualization results on S3DIS (Area 5). The images from left to [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Visualization results on ScanNet. The images from left to right are the [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 32 canonical work pages

  1. [1]

    Qi, Hao Su, Kaichun Mo, and Leonidas J

    Charles R. Qi, Hao Su, Kaichun Mo, and Leonidas J. Guibas, ”PointNet: Deep learning on point sets for 3D classification and segmentation,” in CVPR, 2017, pp. 652–660

    work page 2017

  2. [2]

    Guibas, ”PointNet++: Deep hierarchical feature learning on point sets in a metric space,” NeurIPS, vol

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J. Guibas, ”PointNet++: Deep hierarchical feature learning on point sets in a metric space,” NeurIPS, vol. 30, 2017

    work page 2017

  3. [3]

    Zeyu Hu, Mingmin Zhen, Xuyang Bai, Hongbo Fu, and Chiew-lan Tai, ”JSENet: Joint semantic segmentation and edge detection network for 3D point clouds,” in ECCV, 2020, pp. 222–239

    work page 2020

  4. [4]

    23192–23204, 2022

    Guocheng Qian, Yuchen Li, Houwen Peng, Jinjie Mai, Hasan Hammoud, Mohamed Elhoseiny, and Bernard Ghanem, ”PointNext: Revisiting PointNet++ with improved training and scaling strategies,” NeurIPS, pp. 23192–23204, 2022

    work page 2022

  5. [5]

    Yu Zheng, Yueqi Duan, Zongtai Li, Jie Zhou, and Jiwen Lu, ”Learning dynamic scene-conditioned 3D object detectors,” IEEE TPAMI, 2023

    work page 2023

  6. [6]

    Qi, Leonidas Guibas, and Or Litany, ”PointContrast: Unsupervised pre-training for 3D point cloud understanding,” in ECCV, 2020, pp

    Saining Xie, Jiatao Gu, Demi Guo, Charles R. Qi, Leonidas Guibas, and Or Litany, ”PointContrast: Unsupervised pre-training for 3D point cloud understanding,” in ECCV, 2020, pp. 574–591

    work page 2020

  7. [7]

    8489–8499

    Liyao Tang, Yibing Zhan, Zhe Chen, Baosheng Yu, and Dacheng Tao, ”Contrastive boundary learning for point cloud segmentation,” in CVPR, 2022, pp. 8489–8499

    work page 2022

  8. [8]

    14930–14939

    Mengtian Li, Yuan Xie, Yunhang Shen, Bo Ke, Ruizhi Qiao, Bo Ren, Shaohui Lin, and Lizhuang Ma, ”HybridCR: Weakly-supervised 3D point cloud semantic segmentation via hybrid contrastive regularization,” in CVPR, 2022, pp. 14930–14939

    work page 2022

Show all 34 references
  1. [9]

    6423–6432

    Li Jiang, Shaoshuai Shi, Zhuotao Tian, Xin Lai, Shu Liu, Chi-Wing Fu, and Jiaya Jia, ”Guided point contrastive learning for semi-supervised point cloud semantic segmentation,” in ICCV, 2021, pp. 6423–6432

    2021

  2. [10]

    4690–4699

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou, ”Arc- Face: Additive angular margin loss for deep face recognition,” in CVPR, 2019, pp. 4690–4699

    2019

  3. [11]

    5265–5274

    Hao Wang, Yitong Wang, Zheng Zhou, Xing Ji, Dihong Gong, Jingchao Zhou, Zhifeng Li, and Wei Liu, ”CosFace: Large margin cosine loss for deep face recognition,” in CVPR, 2018, pp. 5265–5274

    2018

  4. [12]

    4166–4175

    Hangyu Li, Nannan Wang, Xi Yang, Xiaoyu Wang, and Xinbo Gao, ”Towards semi-supervised deep facial expression recognition with an adaptive confidence margin,” in CVPR, 2022, pp. 4166–4175

    2022

  5. [13]

    12576–12584

    Aoxue Li, Weiran Huang, Xu Lan, Jiashi Feng, Zhenguo Li, and Liwei Wang, ”Boosting few-shot learning with adaptive margin loss,” inCVPR, 2020, pp. 12576–12584

    2020

  6. [14]

    Shengjun Zhang, Xin Fei, and Yueqi Duan, ”GeoAuxNet: Towards universal 3D representation learning for multi-sensor point clouds,” arXiv preprint arXiv:2403.19220 , 2024

    2024 arXiv

  7. [15]

    Ze Liu, Han Hu, Yue Cao, Zheng Zhang, and Xin Tong, ”A closer look at local aggregation operators in point cloud analysis,” in ECCV, 2020, pp. 326–342

    2020

  8. [16]

    17682–17691

    Haojia Lin, Xiawu Zheng, Lijiang Li, Fei Chao, Shanshan Wang, Yan Wang, Yonghong Tian, and Rongrong Ji, ”Meta architecture for point cloud analysis,” in CVPR, 2023, pp. 17682–17691

    2023

  9. [17]

    Qi, Jean-Emmanuel Deschaud, Beatriz Marcotegui, Franc ¸ois Goulette, and Leonidas J

    Hugues Thomas, Charles R. Qi, Jean-Emmanuel Deschaud, Beatriz Marcotegui, Franc ¸ois Goulette, and Leonidas J. Guibas, ”KPConv: Flexible and deformable convolution for point clouds,” in ICCV, 2019, pp. 6411–6420

    2019

  10. [18]

    3173–3182

    Mutian Xu, Runyu Ding, Hengshuang Zhao, and Xiaojuan Qi, ”PAConv: Position adaptive convolution with dynamic kernel assembling on point clouds,” in CVPR, 2021, pp. 3173–3182

    2021

  11. [19]

    Hao Liu, Yulan Guo, Yanni Ma, Yinjie Lei, and Gongjian Wen, ”Seman- tic context encoding for accurate 3D point cloud segmentation,” IEEE TMM, vol. 23, pp. 2045–2055, 2020

    2020

  12. [20]

    4558– 4567

    Loic Landrieu and Martin Simonovsky, ”Large-scale point cloud seman- tic segmentation with superpoint graphs,” in CVPR, 2018, pp. 4558– 4567

    2018

  13. [21]

    11683–11692

    Guocheng Qian, Abdulellah Abualshour, Guohao Li, Ali Thabet, and Bernard Ghanem, ”PU-GCN: Point cloud upsampling using graph convolutional networks,” in CVPR, 2021, pp. 11683–11692

    2021

  14. [22]

    An Tao, Yueqi Duan, Yi Wei, Jiwen Lu, and Jie Zhou, ”SegGroup: Seg- level supervision for 3D instance and semantic segmentation,”IEEE TIP, vol. 31, pp. 4952–4965, 2022

    2022

  15. [23]

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip H. S. Torr, and Vladlen Koltun, ”Point Transformer,” in ICCV, 2021, pp. 16259–16268

    2021

  16. [24]

    16949–16958

    Chunghyun Park, Yoonwoo Jeong, Minsu Cho, and Jaesik Park, ”Fast Point Transformer,” in CVPR, 2022, pp. 16949–16958

    2022

  17. [25]

    Michael Gutmann and Aapo Hyv ¨arinen, ”Noise-contrastive estimation: A new estimation principle for unnormalized statistical models,” in AISTATS, 2010, pp. 297–304

    2010

  18. [26]

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals, ”Represen- tation learning with contrastive predictive coding,” arXiv preprint arXiv:1807.03748, 2018

    2018 arXiv

  19. [27]

    18661–18673, 2020

    Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan, ”Supervised contrastive learning,” NeurIPS, pp. 18661–18673, 2020

    2020

  20. [28]

    Rongqing Kenneth Ong, Wei Qiu, and Andy W. H. Khong, ”Quad- Tier Entity Fusion Contrastive Representation Learning for Knowledge Aware Recommendation System,” in CIKM, 2023, pp. 1949–1959

    2023

  21. [29]

    Prokopyev, and Eduardo L

    Alexander Veremyev, Oleg A. Prokopyev, and Eduardo L. Pasiliao, ”Finding critical links for closeness centrality,” INFORMS Journal on Computing, vol. 31, no. 2, pp. 367–389, 2019

    2019

  22. [30]

    9729–9738

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick, ”Momentum contrast for unsupervised visual representation learning,” in CVPR, 2020, pp. 9729–9738

    2020

  23. [31]

    Martin, and Shi-Min Hu, ”PCT: Point cloud transformer,” CVM, pp

    Meng-Hao Guo, Jun-Xiong Cai, Zheng-Ning Liu, Tai-Jiang Mu, Ralph R. Martin, and Shi-Min Hu, ”PCT: Point cloud transformer,” CVM, pp. 187–199, 2021

    2021

  24. [32]

    9621–9630

    Wenxuan Wu, Zhongang Qi, and Li Fuxin, ”PointConv: Deep convolu- tional networks on 3D point clouds,” in CVPR, 2019, pp. 9621–9630

    2019

  25. [33]

    Zamir, and Silvio Savarese, ”Joint 2D-3D-semantic data for indoor scene understanding,” arXiv preprint arXiv:1702.01105, 2017

    Iro Armeni, Sasha Sax, Amir R. Zamir, and Silvio Savarese, ”Joint 2D-3D-semantic data for indoor scene understanding,” arXiv preprint arXiv:1702.01105, 2017

    2017 arXiv

  26. [34]

    Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, and Matthias Nießner, ”ScanNet: Richly-annotated 3D reconstructions of indoor scenes,” in CVPR, 2017, pp

    Angela Dai, Angel X. Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, and Matthias Nießner, ”ScanNet: Richly-annotated 3D reconstructions of indoor scenes,” in CVPR, 2017, pp. 5828–5839

    2017

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.