Pith. sign in

REVIEW 3 major objections 4 minor 56 references

Open-Set LiDAR Panoptic Segmentation Guided by Uncertainty-Aware Learning

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper claims that Dirichlet-based evidential uncertainty, sharpened by three specialized losses, enables LiDAR panoptic segmentation models to detect and segment object classes unseen at training time, consistently outperforming…

desk verdict A useful open-set LiDAR panoptic segmentation method whose central uncertainty equation is printed backwards; fix Eq. (1) and verify the code, and the empirical contribution holds. read the letter →

arxiv 2506.13265 v3 pith:W2P7GFC3 submitted 2025-06-16 cs.CV cs.AIcs.LGcs.RO

Rohit Mohan , Julia Hindel , Florian Drews , Claudius Gläser , Daniele Cattaneo , Abhinav Valada This is my paper
classification cs.CVcs.AIcs.LGcs.RO
keywords LiDARpanopticsegmentationopen-setrecognitionevidentialdeeplearningpredictiveuncertaintyunknownobjectdetectionautonomousdrivingpointcloudDirichletdistribution
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

This paper proposes that predictive uncertainty, quantified through Dirichlet-based evidential learning, can tell a LiDAR panoptic segmentation model which voxels belong to object classes it was never trained on. The ULOPS framework adds a semantic decoder that outputs evidence parameters for the known classes, derives a per-voxel uncertainty score, and then uses three new losses—Uniform Evidence, Adaptive Uncertainty Separation, and Contrastive Uncertainty—to force unknown regions to carry higher uncertainty than known regions. At inference, voxels above an adaptive uncertainty threshold are treated as unknown instances and clustered separately, while known instances are recovered through embedding-to-prototype association and instance-center prediction. The authors introduce stricter open-set evaluation vocabularies on KITTI-360 and a new panoptic nuScenes open-set setting, and report that ULOPS consistently outperforms existing open-set LiDAR panoptic segmentation methods, with unknown-quality gains of roughly 2 to 7 points over the OWL baseline.

What carries the argument

The central object is the Dirichlet-based evidential semantic decoder. Instead of softmax logits, it predicts strictly positive Dirichlet parameters $\alpha_k$ for the $K$ known classes via softplus, with class evidence defined as $e_k = \alpha_k - 1$; the paper's uncertainty score is $u = (\sum_{k=1}^K \alpha_k)/K$, intended to be high when evidence for every known class is weak. Three losses shape this map: Uniform Evidence Loss pulls $\alpha$ toward the all-ones vector for unknown voxels, Adaptive Uncertainty Separation Loss shrinks a decaying function of the batch-level gap between known and unknown mean uncertainties, and Contrastive Uncertainty Loss imposes a per-pair margin $u_{\mathrm{unknown}} > u_{\mathrm{known}} + \delta$. At inference a voxel is unknown when $u(v)$ exceeds $\mu_u + t\cdot\sigma_u$, the threshold computed from the mean and standard deviation of the uncertainty map.

What would settle it

On a trained ULOPS model, compute the mean of $u(v)$ separately over ground-truth known and unknown voxels on a held-out KITTI-360 sequence: if the known-voxel mean is not below the unknown-voxel mean, the threshold rule in Eq. (1) cannot be separating as claimed. Since the printed formula $u = (\sum \alpha_k)/K$ grows with total evidence under the standard evidential definition, a direct inspection of the released code should confirm whether the implemented score is inverted or otherwise corrected relative to the text.

Watch

Extended reading notes

Core claim

The central discovery is that a Dirichlet-distribution head over known-class probabilities produces a per-voxel uncertainty signal that separates known from unknown objects in LiDAR point clouds, and that this signal becomes strong enough to threshold when supervised by three complementary losses acting at voxel, batch, and pairwise levels. With that signal, open-set panoptic segmentation reduces to uncertainty thresholding: low-uncertainty voxels are assigned to known stuff and things via embedding prototypes and predicted instance centers, while high-uncertainty voxels are handed to DBSCAN clustering to become unknown instances. Across SemanticKITTI, KITTI-360, and nuScenes, the authors show this mechanism beats the fixed 'other'-category strategy of OWL and the K+1 supervision of proposal-based baselines, while also improving known-class Panoptic Quality in several settings.

Load-bearing premise

The method assumes that the trained uncertainty score is truly higher for unknown voxels than for known voxels, and that a single adaptive threshold can cleanly separate the two groups across diverse scenes.

Editorial extensions

If this is right

  • LiDAR perception systems can flag novel objects such as construction vehicles, trailers, and barriers instead of silently folding them into known classes, which directly improves safety-relevant failure modes in autonomous driving.
  • The three uncertainty losses only consume Dirichlet parameters and unknown-region labels, so they can be grafted onto other evidential segmentation decoders without architectural change.
  • The proposed 'Vocabulary Unseen' split for KITTI-360 gives the community a stricter protocol that excludes unknown classes shared with the training set, better isolating true open-set generalization.
  • The reported gains over the fixed 'other'-category baseline suggest that uncertainty supervision is a more transferable mechanism than K+1 classification for novel-class discovery in 3D point clouds.

Reading between the lines

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

  • A natural extension the authors leave implicit is applying the same evidential uncertainty head to 2D panoptic segmentation or to range-view LiDAR representations, where the same closed-set softmax overconfidence problem appears.
  • The uncertainty map is a by-product that could feed downstream open-set modules such as object detection, active learning, or online class discovery, beyond its role as a segmentation mask.
  • The printed Eq. (1) appears to make $u$ increase with total evidence rather than decrease; the reported open-set behavior therefore presumably relies on an unstated correction (such as an inverse or a negative transform) in the released implementation, which future readers should verify before reusing the formula.
  • Because the Contrastive Uncertainty Loss delivers the largest single ablation gain, a testable extension is to replace the fixed margin $\delta$ with a class-conditional or learned margin that adapts to scenes with many near-unknown objects.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes ULOPS, an uncertainty-guided open-set LiDAR panoptic segmentation framework. It uses Dirichlet-based evidential learning in the semantic decoder to estimate per-voxel uncertainty, and introduces three uncertainty-driven losses—Uniform Evidence Loss, Adaptive Uncertainty Separation Loss, and Contrastive Uncertainty Loss—to encourage high uncertainty in unknown regions during training. At inference, voxels whose uncertainty exceeds an adaptive threshold are labeled unknown and clustered with DBSCAN, while known objects are segmented via embedding prototypes and center heatmaps. The method is evaluated on SemanticKITTI, KITTI-360 (including a new 'Vocabulary Unseen' split), and nuScenes, reporting gains in Unknown Quality (UQ) and Panoptic Quality (PQ) over OWL, 4D-PLS, and PolarSeg-Panoptic. Ablations show that the Dirichlet uncertainty alone achieves UQ 27.9 on Vocabulary Unseen, and that each proposed loss adds further gains.

Significance. If the mechanism works as intended, the paper makes a useful empirical contribution: it transfers Dirichlet-based evidential uncertainty from classification to LiDAR panoptic segmentation, introduces two new open-set evaluation splits (KITTI-360 Vocabulary Unseen and a nuScenes open-set vocabulary), and provides an ablation that separates annotation-free uncertainty performance from performance obtained with explicit unknown-label supervision. The planned code release and the inclusion of multiple baselines strengthen reproducibility. The main conceptual value is in showing that a Dirichlet-derived uncertainty score can separate known and unknown voxels in a polar-BEV LiDAR representation, and that supervising that separation with unknown masks yields further improvement.

major comments (3)
  1. [III-B1, Eq. (1)] Equation (1) defines u = (sum_k alpha_k)/K, which is the mean Dirichlet concentration and increases with evidence. This contradicts the sentence immediately after it ('which increases when a voxel lacks strong evidence') and the standard evidential uncertainty from the cited reference [45], which is u = K / sum_k alpha_k. Under the printed formula, the inference rule in Sec. III-C (u(v) >= mu_u + t*sigma_u) and the three losses in Sec. III-D would push in the wrong direction: a confident known voxel with alpha=[100,1,1] scores u=34, while an unknown voxel forced by L_uniform to alpha=[1,1,1] scores u=1, so the threshold would select known voxels as unknown and L_contrastive would penalize the intended behavior. The authors must correct the definition (or explicitly report the actual score used, e.g., K/sum(alpha)), because this quantity is load-bearing for the entire open-set mechanism and all reported results.
  2. [Tables I-III] No error bars, confidence intervals, or significance tests are reported for any of the benchmark tables. Several of the claimed improvements are small—for example, Table I (Vocabulary 2, SemanticKITTI) shows PQ 62.1 vs. 61.9 for OWL, and Table I (Vocabulary 1, KITTI-360) shows PQ 60.6 vs. 59.4 and UQ 38.6 vs. 36.3—so without multiple runs or statistical testing it is unclear whether the 'consistently outperforms' claim is robust. Please add standard deviations over at least three training seeds, or an appropriate significance test, for the main comparisons in Tables I-III.
  3. [IV-D2 and training protocol] The three uncertainty losses (Eqs. 6, 8, and 9) use ground-truth unknown masks during training, so the final ULOPS model reported in Tables I-III is partly a fitted unknown detector rather than a purely uncertainty-driven open-set system. The paper does disclose this in Sec. IV-D, and Table IV's annotation-free Dirichlet result (UQ 27.9 on Vocabulary Unseen) mitigates the concern, but the abstract and introduction should be more precise: the full method uses unknown-label supervision, whereas the annotation-free ablation is the only strictly open-set variant. Please reframe the claims and the comparison with OWL accordingly.
minor comments (4)
  1. [IV-D1] The sentence 'with a PQ of 60.4% and UQ of 27.9% on Vocabulary 1, and a PQ of 59.1% and UQ of 31.2% on Vocabulary Unseen' is reversed relative to Table IV; the table reports PQ 60.4 and UQ 27.9 for Vocabulary Unseen and PQ 59.1 and UQ 31.2 for Vocabulary 1.
  2. [Figures 2 and 3] The architecture overview is labeled Fig. 2 but is referenced as 'Fig. 3' in Sec. III, and the qualitative results figure is labeled Fig. 3 but referenced as 'Fig. 6' in Sec. IV-E. Please renumber consistently.
  3. [Table V] The table caption contains typos: 'IMAPACT' should be 'IMPACT' and 'UNCERTAINITY' should be 'UNCERTAINTY'. In addition, the table row label 'Proto-Unknowns [17]' is inconsistent with the text, which calls the same baseline 'Association-Based Unknowns'.
  4. [Throughout] The term 'V ocabulary' contains a spurious space in several places (e.g., 'V ocabulary 1', 'V ocabulary Unseen'). Please fix to 'Vocabulary'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the open-set pipeline is evaluated on disjoint vocabularies and the Dirichlet mechanism is externally grounded; Eq. (1) is an internal inconsistency, not a circular reduction.

full rationale

The central claim is that uncertainty estimates identify and segment unknown instances. This is not circular: (i) the Dirichlet evidential formulation is adopted from Sensoy et al. [45], an external, parameter-free source, and the paper's own ablation Tab. IV shows that the annotation-free Dirichlet uncertainty already reaches UQ 27.9/31.2, so the mechanism has independent grounding; (ii) the three uncertainty losses are training regularizers supervised by ground-truth unknown labels on SemanticKITTI, while the headline results are reported on KITTI-360 with Vocabulary 1/2 and on the newly proposed Vocabulary Unseen that excludes unknown classes shared between training and evaluation, and on nuScenes with disjoint training/evaluation unknown categories, so the predictions are genuinely out-of-sample and not identical to the training labels by construction; (iii) the inference threshold is an adaptive global threshold u >= mu_u + t sigma_u, with t a hand-set hyperparameter, and the final unknown segmentation further depends on DBSCAN clustering of embeddings, so the reported UQ is not determined solely by the loss definition. No load-bearing self-citation is present: the core evidential, prototype-association, and contrastive references [17,45,53] are external. The only flagged issue is Eq. (1): u = (sum_k alpha_k)/K is the mean Dirichlet concentration, which increases with evidence, opposite to the claimed 'increases when a voxel lacks strong evidence'; under the cited evidential framework the vacuous uncertainty is K/S. This is an internal consistency error that should be corrected or verified against the released code, but it is not an equivalence-by-construction between input and prediction, so it does not count as circularity under the given criteria.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests mainly on the correctness of the uncertainty measure, direct supervision from unknown labels during training, and a hand-set inference threshold. The uncertainty formula in Eq. (1) is questionable, and no code is provided to check the actual implementation.

free parameters (3)
  • Uncertainty threshold t = 3
    Inference marks voxels as unknown when u(v) >= mu_u + t*sigma_u (Sec. III-C). The choice t=3 is stated but no sensitivity analysis is given, and UQ is directly sensitive to this threshold.
  • Contrastive margin delta = not reported
    Eq. (9) requires unknown uncertainty to exceed known uncertainty by margin delta, but the value used in experiments is not reported.
  • Loss weights = Lcenter=200, Luniform=0.1, Ladaptive=0.1, Lcontrastive=0.7
    Loss weights in Sec. IV-B are hand-set; no ablation of the weights is provided, so the reported balance is not justified by sensitivity analysis.
assumptions (4)
  • ad hoc to paper Total uncertainty is defined by Eq. (1) as u = (sum_k alpha_k)/K and is high when evidence is lacking.
    This definition conflicts with the standard evidential uncertainty K/(sum alpha_k) from cited [45]; the paper provides no derivation or justification for the inverted formula.
  • domain assumption Ground-truth unknown masks are available for training voxels.
    The three uncertainty losses in Sec. III-D operate on voxels labeled as unknown in the training data, so the setup is not fully unsupervised and assumes access to open-set labels.
  • domain assumption Voxel labels from majority voting of LiDAR points are a reliable supervision signal.
    Used for semantic, embedding, and uncertainty supervision (Sec. III-B); can be noisy at voxel boundaries but is standard in polar BEV pipelines.
  • domain assumption The adaptive threshold mu_u + t*sigma_u generalizes across datasets with a fixed t=3.
    The same threshold is applied on SemanticKITTI, KITTI-360, and nuScenes without per-dataset tuning or reported sensitivity (Sec. III-C, Sec. IV-B).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Open-Set LiDAR Panoptic Segmentation Guided by Uncertainty-Aware Learning." pith.science (2026). https://pith.science/paper/W2P7GFC3

@misc{pith2026250613265,
  author       = {Pith},
  title        = {Pith review of: Open-Set LiDAR Panoptic Segmentation Guided by Uncertainty-Aware Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W2P7GFC3}},
  note         = {Machine review of arXiv:2506.13265}
}
read the original abstract

Autonomous vehicles that navigate in open-world environments may encounter previously unseen object classes. However, most existing LiDAR panoptic segmentation models rely on closed-set assumptions, failing to detect unknown object instances. In this work, we propose ULOPS, an uncertainty-guided open-set panoptic segmentation framework that leverages Dirichlet-based evidential learning to model predictive uncertainty. Our architecture incorporates separate decoders for semantic segmentation with uncertainty estimation, embedding with prototype association, and instance center prediction. During inference, we leverage uncertainty estimates to identify and segment unknown instances. To strengthen the model's ability to differentiate between known and unknown objects, we introduce three uncertainty-driven loss functions. Uniform Evidence Loss to encourage high uncertainty in unknown regions. Adaptive Uncertainty Separation Loss ensures a consistent difference in uncertainty estimates between known and unknown objects at a global scale. Contrastive Uncertainty Loss refines this separation at the fine-grained level. To evaluate open-set performance, we extend benchmark settings on KITTI-360 and introduce a new open-set evaluation for nuScenes. Extensive experiments demonstrate that ULOPS consistently outperforms existing open-set LiDAR panoptic segmentation methods.

Figures

Figures reproduced from arXiv: 2506.13265 by the authors.

Figure 1
Figure 1. Illustration of the comparison between closed-set and open-set [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the ULOPS architecture for open-set LiDAR panoptic segmentation. The input LiDAR point cloud is processed by a shared [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Qualitative results of ULOPS compared to the OWL baseline for [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 6
Figure 6. Figure 6: Both models successfully segment barriers in (a). [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

56 extracted references · 50 canonical work pages

  1. [45]

    Evidential deep learning to quantify classification uncertainty,

    M. Sensoy, L. Kaplan, and M. Kandemir, “Evidential deep learning to quantify classification uncertainty,” Adv. Neural Inform. Process. Syst. , vol. 31, 2018

    work page 2018

  2. [1]

    Neural architecture search for dense prediction tasks in computer vision,

    R. Mohan, T. Elsken, A. Zela, J. H. Metzen, B. Staffler, T. Brox, A. Valada, and F. Hutter, “Neural architecture search for dense prediction tasks in computer vision,” International Journal of Computer Vision , vol. 131, no. 7, pp. 1784–1807, 2023

    work page 2023

  3. [2]

    Amodal optical flow,

    M. Luz, R. Mohan, A. R. Sekkat, O. Sawade, E. Matthes, T. Brox, and A. Valada, “Amodal optical flow,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 14 677–14 684

    work page 2024

  4. [3]

    Codeps: Online continual learning for depth estimation and panoptic segmentation,

    N. V ¨odisch, K. Petek, W. Burgard, and A. Valada, “Codeps: Online continual learning for depth estimation and panoptic segmentation,” Robotics: Science and Systems , 2023

    work page 2023

  5. [4]

    Cmrnext: Camera to lidar matching in the wild for localization and extrinsic calibration,

    D. Cattaneo and A. Valada, “Cmrnext: Camera to lidar matching in the wild for localization and extrinsic calibration,” IEEE Transactions on Robotics, 2025

    work page 2025

  6. [5]

    Vehicle localization using 3d building models and point cloud matching,

    A. L. Ballardini, S. Fontana, D. Cattaneo, M. Matteucci, and D. G. Sorrenti, “Vehicle localization using 3d building models and point cloud matching,” Sensors, vol. 21, no. 16, p. 5356, 2021

    work page 2021

  7. [6]

    Effi- cientlps: Efficient lidar panoptic segmentation,

    K. Sirohi, R. Mohan, D. B ¨uscher, W. Burgard, and A. Valada, “Effi- cientlps: Efficient lidar panoptic segmentation,” IEEE Trans. on Robotics, vol. 38, no. 3, pp. 1894–1914, 2021

    work page 1914

  8. [7]

    Taxonomy-aware continual semantic segmentation in hyperbolic spaces for open-world perception,

    J. Hindel, D. Cattaneo, and A. Valada, “Taxonomy-aware continual semantic segmentation in hyperbolic spaces for open-world perception,” IEEE Rob. and Auto. Let. , vol. 10, no. 2, pp. 1904–1911, 2025

    work page 1904

Show all 56 references
  1. [8]

    Panoptic out-of-distribution segmentation,

    R. Mohan, K. Kumaraswamy, J. V . Hurtado, K. Petek, and A. Val- ada, “Panoptic out-of-distribution segmentation,” IEEE Robotics and Automation Letters, 2024

    2024

  2. [9]

    Few- shot panoptic segmentation with foundation models,

    M. K ¨appeler, K. Petek, N. V ¨odisch, W. Burgard, and A. Valada, “Few- shot panoptic segmentation with foundation models,” in Proc. IEEE Int. Conf. on Rob. and Auto. , 2024, pp. 7718–7724

    2024

  3. [10]

    Sky- eye: Self-supervised bird’s-eye-view semantic mapping using monocular frontal view images,

    N. Gosala, K. Petek, P. L. Drews-Jr, W. Burgard, and A. Valada, “Sky- eye: Self-supervised bird’s-eye-view semantic mapping using monocular frontal view images,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog. , 2023, pp. 14 901–14 910

    2023

  4. [11]

    Perceiving the invisible: Proposal-free amodal panoptic segmentation,

    R. Mohan and A. Valada, “Perceiving the invisible: Proposal-free amodal panoptic segmentation,” IEEE Rob. and Auto. Let. , vol. 7, no. 4, pp. 9302–9309, 2022

    2022

  5. [12]

    Unsupervised domain adaptation for lidar panoptic segmentation,

    B. Beˇsi´c, N. Gosala, D. Cattaneo, and A. Valada, “Unsupervised domain adaptation for lidar panoptic segmentation,” IEEE Rob. and Auto. Let. , vol. 7, no. 2, pp. 3404–3411, 2022

    2022

  6. [13]

    Progressive multi- modal fusion for robust 3d object detection,

    R. Mohan, D. Cattaneo, F. Drews, and A. Valada, “Progressive multi- modal fusion for robust 3d object detection,” in 8th Annual Conference on Robot Learning , 2024

    2024

  7. [14]

    Open-set recognition: A good closed-set classifier is all you need?

    S. Vaze, K. Han, A. Vedaldi, and A. Zisserman, “Open-set recognition: A good closed-set classifier is all you need?” 2021

    2021

  8. [15]

    Conditional variational capsule network for open set recognition,

    Y . Guo, G. Camporese, W. Yang, A. Sperduti, and L. Ballan, “Conditional variational capsule network for open set recognition,” in Proc. of the IEEE/CVF Int. Conf. on Computer Vision , 2021, pp. 103– 111

    2021

  9. [16]

    A survey on open-set image recognition,

    J. Sun and Q. Dong, “A survey on open-set image recognition,” arXiv preprint arXiv:2312.15571, 2023

    2023 arXiv

  10. [17]

    Identifying unknown instances for autonomous driving,

    K. Wong, S. Wang, M. Ren, M. Liang, and R. Urtasun, “Identifying unknown instances for autonomous driving,” in Conf. on Robot Learning, 2020, pp. 384–393

    2020

  11. [18]

    Lidar panoptic segmentation in an open world,

    A. S. Chakravarthy, M. R. Ganesina, P. Hu, L. Leal-Taix ´e, S. Kong, D. Ramanan, and A. Osep, “Lidar panoptic segmentation in an open world,” Int. Journal of Computer Vision , pp. 1–22, 2024

    2024

  12. [19]

    Opengan: Open-set recognition via open data generation,

    S. Kong and D. Ramanan, “Opengan: Open-set recognition via open data generation,” in Proc. Int. Conf. Comput. Vis. , 2021, pp. 813–822

    2021

  13. [20]

    Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d,

    Y . Liao, J. Xie, and A. Geiger, “Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d,” IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 45, no. 3, pp. 3292–3310, 2022

    2022

  14. [21]

    Panoptic nuscenes: A large-scale benchmark for lidar panoptic segmentation and tracking,

    W. K. Fong, R. Mohan, J. V . Hurtado, L. Zhou, H. Caesar, O. Beijbom, and A. Valada, “Panoptic nuscenes: A large-scale benchmark for lidar panoptic segmentation and tracking,” IEEE Rob. and Auto. Let. , vol. 7, no. 2, pp. 3795–3802, 2022

    2022

  15. [22]

    Position- guided point cloud panoptic segmentation transformer,

    Z. Xiao, W. Zhang, T. Wang, C. C. Loy, D. Lin, and J. Pang, “Position- guided point cloud panoptic segmentation transformer,” Int. Journal of Computer Vision, vol. 133, no. 1, pp. 275–290, 2025

    2025

  16. [23]

    Cylindrical and asymmetrical 3d convolution networks for lidar segmentation,

    X. Zhu, H. Zhou, T. Wang, F. Hong, Y . Ma, W. Li, H. Li, and D. Lin, “Cylindrical and asymmetrical 3d convolution networks for lidar segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog. , 2021, pp. 9939–9948

    2021

  17. [24]

    Polarnet: An improved grid representation for online lidar point clouds semantic segmentation,

    Y . Zhang, Z. Zhou, P. David, X. Yue, Z. Xi, B. Gong, and H. Foroosh, “Polarnet: An improved grid representation for online lidar point clouds semantic segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog., 2020, pp. 9601–9610

    2020

  18. [25]

    Uncertainty-aware lidar panoptic segmentation,

    K. Sirohi, S. Marvi, D. B ¨uscher, and W. Burgard, “Uncertainty-aware lidar panoptic segmentation,” in 2023 IEEE international conference on robotics and automation (ICRA) . IEEE, 2023, pp. 8277–8283

    2023

  19. [26]

    Mopt: Multi- object panoptic tracking,

    J. V . Hurtado, R. Mohan, W. Burgard, and A. Valada, “Mopt: Multi- object panoptic tracking,” arXiv preprint arXiv:2004.08189 , 2020

    2004 arXiv

  20. [27]

    Panoptic-polarnet: Proposal-free lidar point cloud panoptic segmentation,

    Z. Zhou, Y . Zhang, and H. Foroosh, “Panoptic-polarnet: Proposal-free lidar point cloud panoptic segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog., 2021, pp. 13 194–13 203

    2021

  21. [28]

    4d panoptic lidar segmentation,

    M. Aygun, A. Osep, M. Weber, M. Maximov, C. Stachniss, J. Behley, and L. Leal-Taix ´e, “4d panoptic lidar segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog. , 2021, pp. 5527–5537

    2021

  22. [29]

    Smac-seg: Lidar panoptic segmentation via sparse multi-directional attention clustering,

    E. Li, R. Razani, Y . Xu, and B. Liu, “Smac-seg: Lidar panoptic segmentation via sparse multi-directional attention clustering,” in Proc. IEEE Int. Conf. on Rob. and Auto. , 2022, pp. 9207–9213

    2022

  23. [30]

    Pups: Point cloud unified panoptic segmentation,

    S. Su, J. Xu, H. Wang, Z. Miao, X. Zhan, D. Hao, and X. Li, “Pups: Point cloud unified panoptic segmentation,” in Proc. of the AAAI Conf. on Artificial Intelligence , vol. 37, no. 2, 2023, pp. 2339–2347

    2023

  24. [31]

    Masked-attention mask transformer for universal image segmentation,

    B. Cheng, I. Misra, A. G. Schwing, A. Kirillov, and R. Girdhar, “Masked-attention mask transformer for universal image segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog. , 2022, pp. 1290–1299

    2022

  25. [32]

    Maskrange: A mask- classification model for range-view based lidar segmentation,

    Y . Gu, Y . Huang, C. Xu, and H. Kong, “Maskrange: A mask- classification model for range-view based lidar segmentation,” arXiv preprint arXiv:2206.12073, 2022

    2022 arXiv

  26. [33]

    Towards open set deep networks,

    A. Bendale and T. E. Boult, “Towards open set deep networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog. , 2016, pp. 1563–1572

    2016

  27. [34]

    Syn- mediverse: A multimodal synthetic dataset for intelligent scene under- standing of healthcare facilities,

    R. Mohan, J. Arce, S. Mokhtar, D. Cattaneo, and A. Valada, “Syn- mediverse: A multimodal synthetic dataset for intelligent scene under- standing of healthcare facilities,” IEEE Robotics and Automation Letters , 2024

    2024

  28. [35]

    Amodalsynthdrive: A synthetic amodal perception dataset for au- tonomous driving,

    A. R. Sekkat, R. Mohan, O. Sawade, E. Matthes, and A. Valada, “Amodalsynthdrive: A synthetic amodal perception dataset for au- tonomous driving,” IEEE Robotics and Automation Letters , 2024

    2024

  29. [36]

    Generative openmax for multi-class open set classification,

    Z. Ge, S. Demyanov, Z. Chen, and R. Garnavi, “Generative openmax for multi-class open set classification,” arXiv preprint arXiv:1707.07418, 2017

    2017 arXiv

  30. [37]

    Open set learning with counterfactual images,

    L. Neal, M. Olson, X. Fern, W.-K. Wong, and F. Li, “Open set learning with counterfactual images,” in Proc. Springer Eur. Conf. Comput. Vis. , 2018, pp. 613–628

    2018

  31. [38]

    Opengan: Open set generative adversarial networks,

    L. Ditria, B. J. Meyer, and T. Drummond, “Opengan: Open set generative adversarial networks,” in Proc. of the Asian Conf. on computer vision, 2020

    2020

  32. [39]

    Conditional gaussian distribution learning for open set recognition,

    X. Sun, Z. Yang, C. Zhang, K.-V . Ling, and G. Peng, “Conditional gaussian distribution learning for open set recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog. , 2020, pp. 13 480–13 489

    2020

  33. [40]

    Classification-reconstruction learning for open-set recognition,

    R. Yoshihashi, W. Shao, R. Kawakami, S. You, M. Iida, and T. Naemura, “Classification-reconstruction learning for open-set recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog. , 2019, pp. 4016–4025

    2019

  34. [41]

    Bar: Bayesian activity recog- nition using variational inference,

    R. Krishnan, M. Subedar, and O. Tickoo, “Bar: Bayesian activity recog- nition using variational inference,” arXiv preprint arXiv:1811.03305 , 2018

    2018 arXiv

  35. [42]

    Multifaceted uncertainty estimation for label-efficient deep learning,

    W. Shi, X. Zhao, F. Chen, and Q. Yu, “Multifaceted uncertainty estimation for label-efficient deep learning,” Adv. Neural Inform. Process. Syst., vol. 33, pp. 17 247–17 257, 2020

    2020

  36. [43]

    Predictive uncertainty estimation via prior networks,

    A. Malinin and M. Gales, “Predictive uncertainty estimation via prior networks,” Adv. Neural Inform. Process. Syst. , vol. 31, 2018

    2018

  37. [44]

    Posterior network: Uncertainty estimation without ood samples via density-based pseudo- counts,

    B. Charpentier, D. Z ¨ugner, and S. G ¨unnemann, “Posterior network: Uncertainty estimation without ood samples via density-based pseudo- counts,” Adv. Neural Inform. Process. Syst. , vol. 33, pp. 1356–1367, 2020

    2020

  38. [46]

    Dropout sampling for robust object detection in open-set conditions,

    D. Miller, L. Nicholson, F. Dayoub, and N. S ¨underhauf, “Dropout sampling for robust object detection in open-set conditions,” in Proc. IEEE Int. Conf. on Rob. and Auto. , 2018, pp. 3243–3249

    2018

  39. [47]

    Open-world semantic segmentation for lidar point clouds,

    J. Cen, P. Yun, S. Zhang, J. Cai, D. Luan, M. Tang, M. Liu, and M. Yu Wang, “Open-world semantic segmentation for lidar point clouds,” in Proc. Springer Eur. Conf. Comput. Vis. , 2022, pp. 318–334

    2022

  40. [48]

    Exemplar-based open- set panoptic segmentation network,

    J. Hwang, S. W. Oh, J.-Y . Lee, and B. Han, “Exemplar-based open- set panoptic segmentation network,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog., 2021, pp. 1175–1184

    2021

  41. [49]

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

    C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “Pointnet: Deep learning on point sets for 3d classification and segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog., 2017, pp. 652–660

    2017

  42. [50]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in Proc. of the Int. Conf. on Medical image computing and computer-assisted intervention , 2015, pp. 234–241

    2015

  43. [51]

    Semantic instance segmentation with a discriminative loss function,

    B. De Brabandere, D. Neven, and L. Van Gool, “Semantic instance segmentation with a discriminative loss function,” arXiv preprint arXiv:1708.02551, 2017

    2017 arXiv

  44. [52]

    Density-based spatial clustering of applications with noise

    S. Learn, “Density-based spatial clustering of applications with noise.”

  45. [53]

    Contrastive training for improved out-of-distribution detection,

    J. Winkens, R. Bunel, A. G. Roy, R. Stanforth, V . Natarajan, J. R. Ledsam, P. MacWilliams, P. Kohli, A. Karthikesalingam, S. Kohl et al., “Contrastive training for improved out-of-distribution detection,” arXiv preprint arXiv:2007.05566, 2020

    2007 arXiv

  46. [54]

    Semantickitti: A dataset for semantic scene understanding of lidar sequences,

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stachniss, and J. Gall, “Semantickitti: A dataset for semantic scene understanding of lidar sequences,” in Proc. Int. Conf. Comput. Vis. , 2019, pp. 9297–9307

    2019

  47. [55]

    Uncertainty estimation using a single deep deterministic neural network,

    J. Van Amersfoort, L. Smith, Y . W. Teh, and Y . Gal, “Uncertainty estimation using a single deep deterministic neural network,” in Int. Conf. on machine learning , 2020, pp. 9690–9700

    2020

  48. [56]

    Simple and principled uncertainty estimation with deterministic deep learning via distance awareness,

    J. Liu, Z. Lin, S. Padhy, D. Tran, T. Bedrax Weiss, and B. Lak- shminarayanan, “Simple and principled uncertainty estimation with deterministic deep learning via distance awareness,” Adv. Neural Inform. Process. Syst., vol. 33, pp. 7498–7512, 2020. Open-Set LiDAR Panoptic Segm...

    2020

Pith tools

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