pith. sign in

arxiv: 2212.02011 · v3 · submitted 2022-12-05 · 💻 cs.CV

PointCaM: Cut-and-Mix for Open-Set Point Cloud Learning

Jie Hong , Shi Qiu , Weihao Li , Saeed Anwar , Mehrtash Harandi , Nick Barnes , Lars Petersson This is my paper

Pith reviewed 2026-05-24 10:34 UTC · model grok-4.3

classification 💻 cs.CV
keywords open-set learningpoint cloudcut-and-mixout-of-distribution detectionunknown object recognitionmulti-level features3D classificationgeometric augmentation
0
0 comments X

The pith

PointCaM trains open-set point cloud models by cutting and mixing known samples to simulate unknowns and discriminating them with multi-level features.

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

The paper addresses the inability of standard point cloud models to handle unknown objects they were never trained on. It introduces a cut-and-mix process that alters the geometry of partially known point clouds to create training examples of out-of-distribution data. An estimator module then learns to separate known from unknown inputs by examining feature contexts at several levels rather than only the final classifier layer. Experiments on indoor, synthetic, and real scanned datasets show gains over prior open-set baselines. A reader would care because many practical 3D sensing systems encounter objects outside their training distribution and need a reliable way to flag them.

Core claim

The paper claims that an Unknown-Point Simulator can generate representative out-of-distribution point clouds by manipulating geometric context through cut-and-mix operations on partially known data, allowing an Unknown-Point Estimator to learn a discriminator based on multi-level feature contexts that reliably identifies unknown objects at inference time without ever seeing them during training.

What carries the argument

The Point Cut-and-Mix mechanism, consisting of an Unknown-Point Simulator that creates synthetic out-of-distribution samples and an Unknown-Point Estimator that exploits multi-level feature contexts for discrimination.

If this is right

  • Open-set point cloud classifiers can be trained using only known-class data while still flagging unknowns at test time.
  • Using feature contexts from multiple network layers outperforms reliance on classifier features alone for unknown detection.
  • The same simulator-estimator pipeline improves results across indoor segmentation, synthetic object classification, and real scanned object datasets.
  • Unknown objects are identified during inference without requiring any exposure to unknown-class examples in the training set.

Where Pith is reading between the lines

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

  • The cut-and-mix simulation could be adapted to other 3D representations such as voxels or meshes if the geometric manipulation steps are adjusted accordingly.
  • In deployed systems the approach might lower the cost of collecting exhaustive class labels by allowing models to operate safely with partial class coverage.
  • Combining the multi-level estimator with existing uncertainty estimation techniques could further reduce false positives on borderline known samples.
  • Performance on very large outdoor scenes with sparse point density remains an open question that would require new test sets beyond the three datasets used here.

Load-bearing premise

Manipulating the geometric context of partially known point cloud data via cut-and-mix operations produces simulated out-of-distribution samples that are sufficiently representative for training an effective discriminator.

What would settle it

An experiment in which models trained with the cut-and-mix simulator show no improvement in unknown-class recall compared with baselines when evaluated on a test set containing real unknown objects never used in simulation.

Figures

Figures reproduced from arXiv: 2212.02011 by Jie Hong, Lars Petersson, Mehrtash Harandi, Nick Barnes, Saeed Anwar, Shi Qiu, Weihao Li.

Figure 1
Figure 1. Figure 1: The PointCaM mechanism consists of two main modules: the Unknown-Point Simulator [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The detailed structure of the Unknown-Point Estimator (UPE). The figure illustrates the [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The visualization examples: open-set point cloud semantic segmentation of PointNet on S3DIS-Split with the ‘Manual-10-3’ split. Under the ‘Manual-10-3’ split, ‘table,’ ‘chair’ and ‘sofa’ are taken as unknown classes. We visualize heatmaps of estimated unknown-class scores using different open-set methods, i.e., MSP [69], UPS+MSP, and our PointCaM (UPS+UPE). The ‘blue’ and ‘red’ points represent points from… view at source ↗
Figure 4
Figure 4. Figure 4: The visualization examples: open-set point cloud semantic segmentation of Point￾Net on S3DIS-Split with the ‘Manual-10-3’ split. We visualize the density distributions of the unknown-class score of MSP and UPS+MSP. that its performance improves when difference levels of feature context are fused. 5.5. Discussion Number of unknown classes. The results on S3DIS-Split (See Tab. 1) imply the negative impacts o… view at source ↗
read the original abstract

Point cloud learning is receiving increasing attention. However, most existing point cloud models lack the practical ability to deal with the unavoidable presence of unknown objects. This paper primarily discusses point cloud learning in open-set settings, where we train the model without data from unknown classes and identify them during the inference stage. In essence, we propose a novel Point Cut-and-Mix mechanism for solving open-set point cloud learning, comprising an Unknown-Point Simulator and an Unknown-Point Estimator module. Specifically, we use the Unknown-Point Simulator to simulate out-of-distribution data in the training stage by manipulating the geometric context of partially known data. Based on this, the Unknown-Point Estimator module learns to exploit the point cloud's feature context to discriminate between known and unknown data. Unlike existing methods that only consider classifier features, our proposed solution leverages multi-level feature contexts to recognize unknown point cloud objects more effectively. We test the proposed approach on several datasets, including customized S3DIS, ModelNet40, and ScanObjectNN. The improved open-set performances over comparative baselines show the effectiveness of our PointCaM method. Our code is available at https://github.com/JHome1/pointcam.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes PointCaM for open-set point cloud learning. It introduces an Unknown-Point Simulator that applies cut-and-mix operations to manipulate the geometric context of partially known point cloud data in order to simulate out-of-distribution samples during training, paired with an Unknown-Point Estimator that exploits multi-level feature contexts (rather than only classifier features) to discriminate known from unknown objects at inference. Experiments on customized S3DIS, ModelNet40, and ScanObjectNN report improved open-set performance over comparative baselines, with code released at the cited GitHub repository.

Significance. If the cut-and-mix simulation produces OOD training signals whose feature contexts overlap real unknown distributions, the approach could advance practical open-set recognition for point clouds by providing a data-driven way to train discriminators without access to true unknown samples and by using multi-level contexts. The public code release supports reproducibility and is a clear strength.

major comments (2)
  1. [§3.2] §3.2 (Unknown-Point Simulator): the simulation is performed exclusively by cutting and mixing points from the known training distribution. No quantitative verification (e.g., distributional overlap metrics or nearest-neighbor analysis between simulated and held-out unknown point statistics) is provided to confirm that the generated samples are representative of genuine OOD objects whose geometry lies outside convex combinations of the training classes; this assumption is load-bearing for the claim that the estimator generalizes to real unknowns.
  2. [Experiments section] Experiments section (results tables): while improved open-set metrics are reported on the held-out splits of S3DIS/ModelNet40/ScanObjectNN, the evaluation does not include ablation isolating the contribution of the multi-level estimator versus the simulator quality, nor failure-case analysis for unknowns whose local/global statistics are not mixable from known classes; this leaves the source of the gains and the scope of generalization unclear.
minor comments (2)
  1. [Abstract] Abstract: states that 'improved open-set performances' are observed but supplies no numerical values, dataset-specific metrics, or baseline names, reducing immediate readability.
  2. Notation: the cut-and-mix operation and the multi-level feature extraction lack explicit equations or pseudocode, making the precise implementation of the geometric manipulation harder to follow.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and indicate where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Unknown-Point Simulator): the simulation is performed exclusively by cutting and mixing points from the known training distribution. No quantitative verification (e.g., distributional overlap metrics or nearest-neighbor analysis between simulated and held-out unknown point statistics) is provided to confirm that the generated samples are representative of genuine OOD objects whose geometry lies outside convex combinations of the training classes; this assumption is load-bearing for the claim that the estimator generalizes to real unknowns.

    Authors: We agree that the lack of quantitative verification (such as distributional overlap or nearest-neighbor analysis) leaves the representativeness of the simulated samples less substantiated. The Unknown-Point Simulator relies on the premise that cut-and-mix operations on known data can produce useful OOD training signals, and the reported gains on held-out unknowns provide indirect support. However, to directly address this concern, the revised manuscript will include additional quantitative analysis comparing simulated and real unknown distributions. revision: yes

  2. Referee: [Experiments section] Experiments section (results tables): while improved open-set metrics are reported on the held-out splits of S3DIS/ModelNet40/ScanObjectNN, the evaluation does not include ablation isolating the contribution of the multi-level estimator versus the simulator quality, nor failure-case analysis for unknowns whose local/global statistics are not mixable from known classes; this leaves the source of the gains and the scope of generalization unclear.

    Authors: We acknowledge that the current experiments report aggregate improvements without isolating the simulator versus the multi-level estimator or examining failure modes for non-mixable unknowns. This limits insight into the source of gains and generalization boundaries. The revised version will add the requested ablations (e.g., variants ablating each component) and a failure-case discussion or analysis for unknowns whose statistics fall outside convex combinations of known classes. revision: yes

Circularity Check

0 steps flagged

No circularity: method proposal and empirical evaluation are independent

full rationale

The paper proposes a new Point Cut-and-Mix mechanism (Unknown-Point Simulator + Estimator) that operates by direct manipulation of known point clouds and multi-level feature discrimination. This construction is defined in the method section without reference to fitted parameters or prior self-citations as load-bearing premises. Effectiveness is then measured by performance gains on held-out splits of external datasets (customized S3DIS, ModelNet40, ScanObjectNN) against comparative baselines. No equation or claim reduces the reported open-set improvements to a statistical artifact of the training distribution or to a self-referential definition. The derivation chain is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The approach rests on the domain assumption that geometric manipulation creates useful unknown simulations and on the introduction of two new modules whose effectiveness is asserted via empirical comparison.

axioms (1)
  • domain assumption Manipulating the geometric context of known point clouds can simulate realistic out-of-distribution data
    This premise underpins the Unknown-Point Simulator as stated in the abstract.
invented entities (2)
  • Unknown-Point Simulator no independent evidence
    purpose: Generate simulated unknown point clouds via cut-and-mix on known data
    Newly introduced component whose validity is not independently verified outside the paper.
  • Unknown-Point Estimator no independent evidence
    purpose: Discriminate known versus unknown using multi-level feature contexts
    Newly introduced component whose validity is not independently verified outside the paper.

pith-pipeline@v0.9.0 · 5752 in / 1127 out tokens · 25978 ms · 2026-05-24T10:34:27.771959+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

81 extracted references · 81 canonical work pages · 3 internal anchors

  1. [1]

    Endres, J

    F. Endres, J. Hess, J. Sturm, D. Cremers, W. Burgard, 3-d mapping with an rgb-d camera, Transactions on Robotics 30 (1) (2013) 177–187. 2

    work page 2013

  2. [2]

    Jaboyedoff, T

    M. Jaboyedoff, T. Oppikofer, A. Abell ´an, M.-H. Derron, A. Loye, R. Met- zger, A. Pedrazzini, Use of lidar in landslide investigations: a review, Natural hazards 61 (1) (2012) 5–28. 2

    work page 2012

  3. [3]

    Pomerleau, F

    F. Pomerleau, F. Colas, R. Siegwart, et al., A review of point cloud registra- tion algorithms for mobile robotics, Foundations and Trends® in Robotics 4 (1) (2015) 1–104. 2

    work page 2015

  4. [4]

    Y . Ze, N. Hansen, Y . Chen, M. Jain, X. Wang, Visual reinforcement learn- ing with self-supervised 3d representations, IEEE Robotics and Automation Letters 8 (5) (2023) 2890–2897. 2

    work page 2023

  5. [5]

    W. Li, Q. Guo, M. K. Jakubowski, M. Kelly, A new method for segmenting individual trees from the lidar point cloud, Photogrammetric Engineering & Remote Sensing 78 (1) (2012) 75–84. 2

    work page 2012

  6. [6]

    Y . Chen, Q. Wang, H. Chen, X. Song, H. Tang, M. Tian, An overview of augmented reality technology, in: Journal of Physics: Conference Series, V ol. 1237, IOP Publishing, 2019, p. 022082. 2

    work page 2019

  7. [7]

    C. R. Qi, H. Su, K. Mo, L. J. Guibas, Pointnet: Deep learning on point sets for 3d classification and segmentation, in: CVPR, 2017, pp. 652–660. 2, 5, 8, 15, 29, 30, 32

    work page 2017

  8. [8]

    C. R. Qi, L. Yi, H. Su, L. J. Guibas, Pointnet++: Deep hierarchical feature learning on point sets in a metric space, in: NIPS, 2017, pp. 5099–5108. 2, 5, 8, 12, 15, 29, 30, 31, 33 20 ©2020. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/

    work page 2017

  9. [9]

    H. Zhao, L. Jiang, J. Jia, P. H. Torr, V . Koltun, Point transformer, in: Pro- ceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 16259–16268. 2, 5, 12, 15, 29, 31

    work page 2021

  10. [10]

    Y . Guo, H. Wang, Q. Hu, H. Liu, L. Liu, M. Bennamoun, Deep learning for 3d point clouds: A survey, TPAMI (2020). 2, 5

    work page 2020

  11. [11]

    Joint 2D-3D-Semantic Data for Indoor Scene Understanding

    I. Armeni, S. Sax, A. R. Zamir, S. Savarese, Joint 2d-3d-semantic data for indoor scene understanding, arXiv:1702.01105 (2017). 2, 8

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  12. [12]

    Behley, M

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stachniss, J. Gall, Semantickitti: A dataset for semantic scene understanding of lidar sequences, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 9297–9307. 2

    work page 2019

  13. [13]

    Z. Wu, S. Song, A. Khosla, F. Yu, L. Zhang, X. Tang, J. Xiao, 3d shapenets: A deep representation for volumetric shapes, in: CVPR, 2015, pp. 1912–

    work page 2015

  14. [14]

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

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  15. [15]

    S. Qiu, S. Anwar, N. Barnes, Geometric back-projection network for point cloud classification, IEEE Transactions on Multimedia 24 (2021) 1943–

    work page 2021

  16. [16]

    C. Choy, J. Gwak, S. Savarese, 4d spatio-temporal convnets: Minkowski convolutional neural networks, in: CVPR, 2019. 2, 12

    work page 2019

  17. [17]

    Q. Hu, B. Yang, L. Xie, S. Rosa, Y . Guo, Z. Wang, N. Trigoni, A. Markham, Randla-net: Efficient semantic segmentation of large-scale point clouds, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 11108–11117. 2, 9, 12

    work page 2020

  18. [18]

    S. Qiu, Y . Wu, S. Anwar, C. Li, Investigating attention mechanism in 3d point cloud object detection, in: 2021 International Conference on 3D Vision (3DV), IEEE, 2021, pp. 403–412. 2

    work page 2021

  19. [19]

    C. R. Qi, O. Litany, K. He, L. J. Guibas, Deep hough voting for 3d object detection in point clouds, in: proceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 9277–9286. 2 21 ©2020. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/

    work page 2019

  20. [20]

    W. Wang, R. Yu, Q. Huang, U. Neumann, Sgpn: Similarity group pro- posal network for 3d point cloud instance segmentation, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 2569–2578. 2

    work page 2018

  21. [21]

    S. Qiu, S. Anwar, N. Barnes, Semantic segmentation for real point cloud scenes via bilateral augmentation and adaptive fusion, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 1757–1767. 2

    work page 2021

  22. [22]

    S. Qiu, S. Anwar, N. Barnes, Pu-transformer: Point cloud upsampling trans- former, in: Proceedings of the Asian Conference on Computer Vision, 2022, pp. 2475–2493. 2

    work page 2022

  23. [23]

    Zhang, A

    H. Zhang, A. Li, J. Guo, Y . Guo, Hybrid models for open set recognition, in: European Conference on Computer Vision, Springer, 2020, pp. 102–117. 2, 5

    work page 2020

  24. [24]

    Dhamija, M

    A. Dhamija, M. Gunther, J. Ventura, T. Boult, The overlooked elephant of object detection: Open set, in: Proceedings of the IEEE/CVF Winter Con- ference on Applications of Computer Vision, 2020, pp. 1021–1030. 2, 18

    work page 2020

  25. [25]

    S. Kong, D. Ramanan, Opengan: Open-set recognition via open data gener- ation, in: Proceedings of the IEEE/CVF International Conference on Com- puter Vision (ICCV), 2021, pp. 813–822. 2

    work page 2021

  26. [26]

    Zheng, W

    J. Zheng, W. Li, J. Hong, L. Petersson, N. Barnes, Towards open-set object detection and discovery, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2022, pp. 3961–3970. 2, 5

    work page 2022

  27. [27]

    H. Ma, R. Xiong, Y . Wang, S. Kodagoda, L. Shi, Towards open-set semantic labeling in 3d point clouds: Analysis on the unknown class, Neurocomputing 275 (2018) 1282–1294. 3, 6, 18

    work page 2018

  28. [28]

    Bhardwaj, S

    A. Bhardwaj, S. Pimpale, S. Kumar, B. Banerjee, Empowering knowledge distillation via open set recognition for robust 3d point cloud classification, Pattern Recognition Letters 151 (2021) 172–179. 3, 6 22 ©2020. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/

    work page 2021

  29. [29]

    Zhao, T.-S

    N. Zhao, T.-S. Chua, G. H. Lee, Few-shot 3d point cloud semantic segmen- tation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 8873–8882. 3, 6

    work page 2021

  30. [30]

    K. Wong, S. Wang, M. Ren, M. Liang, R. Urtasun, Identifying unknown in- stances for autonomous driving, in: Conference on Robot Learning, PMLR, 2020, pp. 384–393. 3, 6

    work page 2020

  31. [31]

    J. Cen, P. Yun, J. Cai, M. Y . Wang, M. Liu, Open-set 3d object detection, in: 2021 International Conference on 3D Vision (3DV), IEEE, 2021, pp. 869–878. 3, 6, 15, 29

    work page 2021

  32. [32]

    J. Cen, P. Yun, S. Zhang, J. Cai, D. Luan, M. Y . Wang, M. Liu, M. Tang, Open-world semantic segmentation for lidar point clouds, arXiv preprint arXiv:2207.01452 (2022). 3, 6, 16, 18, 19, 29, 32

    work page arXiv 2022

  33. [33]

    H. Dong, Z. Chen, M. Yuan, Y . Xie, J. Zhao, F. Yu, B. Dong, L. Zhang, Region-aware metric learning for open world semantic segmentation via meta-channel aggregation, in: 31th International Joint Conference on Ar- tificial Intelligence (IJCAI-22), 2022. 3, 6

    work page 2022

  34. [34]

    W. J. Scheirer, A. de Rezende Rocha, A. Sapkota, T. E. Boult, Toward open set recognition, IEEE transactions on pattern analysis and machine intelli- gence 35 (7) (2012) 1757–1772. 4

    work page 2012

  35. [35]

    Bendale, T

    A. Bendale, T. E. Boult, Towards open set deep networks, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR),

    work page

  36. [36]

    Y . Yang, C. Hou, Y . Lang, D. Guan, D. Huang, J. Xu, Open-set human activity recognition based on micro-doppler signatures, Pattern Recognition 85 (2019) 60–69. 4

    work page 2019

  37. [37]

    C. Geng, L. Tao, S. Chen, Guided cnn for generalized zero-shot and open-set recognition using visual and semantic prototypes, Pattern Recognition 102 (2020) 107263. 4

    work page 2020

  38. [38]

    Dhamija, M

    A. Dhamija, M. Gunther, J. Ventura, T. Boult, The overlooked elephant of object detection: Open set, in: Proceedings of the IEEE/CVF Winter Con- ference on Applications of Computer Vision (W ACV), 2020. 5 23 ©2020. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/

    work page 2020

  39. [39]

    Joseph, S

    K. Joseph, S. Khan, F. S. Khan, V . N. Balasubramanian, Towards open world object detection, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 5830–5840. 5

    work page 2021

  40. [40]

    Hendrycks, S

    D. Hendrycks, S. Basart, M. Mazeika, A. Zou, J. Kwon, M. Mostajabi, J. Steinhardt, D. Song, Scaling out-of-distribution detection for real-world settings, in: Proceedings of the 39th International Conference on Machine Learning (ICML), PMLR, 2022, pp. 8759–8773. 5, 9, 16, 18, 29, 32

    work page 2022

  41. [41]

    J. Hong, W. Li, J. Han, J. Zheng, P. Fang, M. Harandi, L. Petersson, Goss: Towards generalized open-set semantic segmentation, arXiv preprint arXiv:2203.12116 (2022). 5

    work page arXiv 2022

  42. [42]

    Enhancing the reliability of out-of-distribution image detection in neural networks

    S. Liang, Y . Li, R. Srikant, Enhancing the reliability of out-of-distribution image detection in neural networks, arXiv preprint arXiv:1706.02690 (2017). 5

    work page arXiv 2017

  43. [43]

    Pidhorskyi, R

    S. Pidhorskyi, R. Almohsen, G. Doretto, Generative probabilistic novelty de- tection with adversarial autoencoders, Advances in neural information pro- cessing systems 31 (2018). 5

    work page 2018

  44. [44]

    Radford, J

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sas- try, A. Askell, P. Mishkin, J. Clark, et al., Learning transferable visual mod- els from natural language supervision, in: International conference on ma- chine learning, PMLR, 2021, pp. 8748–8763. 5

    work page 2021

  45. [45]

    H. Pham, Z. Dai, G. Ghiasi, K. Kawaguchi, H. Liu, A. W. Yu, J. Yu, Y .-T. Chen, M.-T. Luong, Y . Wu, et al., Combined scaling for open-vocabulary image classification, arXiv e-prints (2021) arXiv–2111. 5

    work page 2021

  46. [46]

    Veeramacheneni, M

    L. Veeramacheneni, M. Valdenegro-Toro, A benchmark for out of distribu- tion detection in point cloud 3d semantic segmentation, in: NeurIPS 2022 Workshop on Robot Learning: Trustworthy Robotics, 2022. 5

    work page 2022

  47. [47]

    Salehi, H

    M. Salehi, H. Mirzaei, D. Hendrycks, Y . Li, M. H. Rohban, M. Sabokrou, A unified survey on anomaly, novelty, open-set, and out of-distribution de- tection: Solutions and future challenges, Transactions of Machine Learning Research (2022). 5 24 ©2020. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/li...

    work page 2022

  48. [48]

    H. Su, S. Maji, E. Kalogerakis, E. Learned-Miller, Multi-view convolutional neural networks for 3d shape recognition, in: Proceedings of the IEEE inter- national conference on computer vision, 2015, pp. 945–953. 5

    work page 2015

  49. [49]

    Maturana, S

    D. Maturana, S. Scherer, V oxnet: A 3d convolutional neural network for real-time object recognition, in: 2015 IEEE/RSJ international conference on intelligent robots and systems (IROS), IEEE, 2015, pp. 922–928. 5

    work page 2015

  50. [50]

    Y . Wang, Y . Sun, Z. Liu, S. E. Sarma, M. M. Bronstein, J. M. Solomon, Dy- namic graph cnn for learning on point clouds, Acm Transactions On Graph- ics (tog) 38 (5) (2019) 1–12. 5, 10, 16, 30

    work page 2019

  51. [51]

    Y . Liu, B. Fan, S. Xiang, C. Pan, Relation-shape convolutional neural net- work for point cloud analysis, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 8895–8904. 5

    work page 2019

  52. [52]

    Guo, J.-X

    M.-H. Guo, J.-X. Cai, Z.-N. Liu, T.-J. Mu, R. R. Martin, S.-M. Hu, Pct: Point cloud transformer, Computational Visual Media 7 (2) (2021) 187–199. 5

    work page 2021

  53. [53]

    Alliegro, F

    A. Alliegro, F. Cappio Borlino, T. Tommasi, 3dos: Towards 3d open set learning-benchmarking and understanding semantic novelty detection on point clouds, Advances in Neural Information Processing Systems 35 (2022) 21228–21240. 6, 19, 33

    work page 2022

  54. [54]

    Ghiasi, Y

    G. Ghiasi, Y . Cui, A. Srinivas, R. Qian, T.-Y . Lin, E. D. Cubuk, Q. V . Le, B. Zoph, Simple copy-paste is a strong data augmentation method for in- stance segmentation, in: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), 2021, pp. 2918–2928. 7

    work page 2021

  55. [55]

    J. Han, L. Petersson, H. Li, I. Reid, Cropmix: Sampling a rich input dis- tribution via multi-scale cropping, arXiv preprint arXiv:2205.15955 (2022). 7

    work page arXiv 2022

  56. [56]

    S. Yun, D. Han, S. J. Oh, S. Chun, J. Choe, Y . Yoo, Cutmix: Regulariza- tion strategy to train strong classifiers with localizable features, in: Pro- ceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019. 7

    work page 2019

  57. [57]

    C.-L. Li, K. Sohn, J. Yoon, T. Pfister, Cutpaste: Self-supervised learning for anomaly detection and localization, in: Proceedings of the IEEE/CVF 25 ©2020. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/ Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 96...

    work page 2020

  58. [58]

    Dwibedi, I

    D. Dwibedi, I. Misra, M. Hebert, Cut, paste and learn: Surprisingly easy synthesis for instance detection, in: Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017. 7

    work page 2017

  59. [59]

    , author Marcu, A

    E. Harris, A. Marcu, M. Painter, M. Niranjan, A. Pr ¨ugel-Bennett, J. Hare, Fmix: Enhancing mixed sample data augmentation, arXiv preprint arXiv:2002.12047 (2020). 7

    work page arXiv 2002

  60. [60]

    D. Lee, J. Lee, J. Lee, H. Lee, M. Lee, S. Woo, S. Lee, Regularization strategy for point cloud via rigidly mixed sample, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 15900–15909. 7

    work page 2021

  61. [61]

    Zhang, L

    J. Zhang, L. Chen, B. Ouyang, B. Liu, J. Zhu, Y . Chen, Y . Meng, D. Wu, Pointcutmix: Regularization strategy for point cloud classification, Neuro- computing (2022). 7

    work page 2022

  62. [62]

    L. Neal, M. Olson, X. Fern, W.-K. Wong, F. Li, Open set learning with coun- terfactual images, in: Proceedings of the European Conference on Computer Vision (ECCV), 2018, pp. 613–628. 7

    work page 2018

  63. [63]

    M. A. Uy, Q.-H. Pham, B.-S. Hua, T. Nguyen, S.-K. Yeung, Revisiting point cloud classification: A new benchmark dataset and classification model on real-world data, in: Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 1588–1597. 8

    work page 2019

  64. [64]

    J. Cen, P. Yun, J. Cai, M. Y . Wang, M. Liu, Deep metric learning for open world semantic segmentation, in: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, 2021, pp. 15333–15342. 8

    work page 2021

  65. [65]

    Davis, M

    J. Davis, M. Goadrich, The relationship between precision-recall and roc curves, in: Proceedings of the 23rd international conference on Machine learning, 2006, pp. 233–240. 9

    work page 2006

  66. [66]

    Fawcett, An introduction to roc analysis, Pattern recognition letters 27 (8) (2006) 861–874

    T. Fawcett, An introduction to roc analysis, Pattern recognition letters 27 (8) (2006) 861–874. 9 26 ©2020. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/

    work page 2006

  67. [67]

    Manning, H

    C. Manning, H. Schutze, Foundations of statistical natural language process- ing, MIT press, 1999. 9

    work page 1999

  68. [68]

    Saito, M

    T. Saito, M. Rehmsmeier, The precision-recall plot is more informative than the roc plot when evaluating binary classifiers on imbalanced datasets, PloS one 10 (3) (2015) e0118432. 9

    work page 2015

  69. [69]

    Hendrycks, K

    D. Hendrycks, K. Gimpel, A baseline for detecting misclassified and out-of- distribution examples in neural networks, in: ICLR, 2017. 9, 17, 18, 29, 30, 32

    work page 2017

  70. [70]

    X. Yan, C. Zheng, Z. Li, S. Wang, S. Cui, Pointasnl: Robust point clouds pro- cessing using nonlocal neural networks with adaptive sampling, in: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, 2020, pp. 5589–5598. 9

    work page 2020

  71. [71]

    S. Qiu, S. Anwar, N. Barnes, Pnp-3d: A plug-and-play for 3d point clouds, IEEE Transactions on Pattern Analysis and Machine Intelligence (2021). 9

    work page 2021

  72. [72]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, in: International Conference on Medical image computing and computer-assisted intervention, Springer, 2015, pp. 234–241. 12, 14

    work page 2015

  73. [73]

    J. Long, E. Shelhamer, T. Darrell, Fully convolutional networks for semantic segmentation, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 3431–3440. 14

    work page 2015

  74. [74]

    C.-Y . Lee, S. Xie, P. Gallagher, Z. Zhang, Z. Tu, Deeply-supervised nets, in: Artificial intelligence and statistics, PMLR, 2015, pp. 562–570. 14

    work page 2015

  75. [75]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980 (2014). 14

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  76. [76]

    S. Vaze, K. Han, A. Vedaldi, A. Zisserman, Open-set recognition: A good closed-set classifier is all you need, in: International Conference on Learning Representations, 2021. 30

    work page 2021

  77. [77]

    Huang, A

    R. Huang, A. Geng, Y . Li, On the importance of gradients for detecting distributional shifts in the wild, Advances in Neural Information Processing Systems 34 (2021) 677–689. 30 27 ©2020. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/

    work page 2021

  78. [78]

    H. Zhou, Y . Feng, M. Fang, M. Wei, J. Qin, T. Lu, Adaptive graph convolu- tion for point cloud analysis, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 4965–4974. 16, 30

    work page 2021

  79. [79]

    Behley, M

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stachniss, J. Gall, SemanticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences, in: Proc. of the IEEE/CVF International Conf. on Com- puter Vision (ICCV), 2019. 19, 32

    work page 2019

  80. [80]

    Y . Gal, Z. Ghahramani, Dropout as a bayesian approximation: Representing model uncertainty in deep learning, in: international conference on machine learning, PMLR, 2016, pp. 1050–1059. 32

    work page 2016

Showing first 80 references.