pith. sign in

arxiv: 2604.23604 · v1 · submitted 2026-04-26 · 💻 cs.CV · cs.RO

Learning to Identify Out-of-Distribution Objects for 3D LiDAR Anomaly Segmentation

Simone Mosco , Daniel Fusaro , Alberto Pretto This is my paper

Pith reviewed 2026-05-08 06:35 UTC · model grok-4.3

classification 💻 cs.CV cs.RO
keywords anomaly segmentation3D LiDARout-of-distribution detectionfeature space modelingsemantic segmentationautonomous drivingmixed datasetspoint cloud
0
0 comments X

The pith

Modeling the feature distribution of known classes in 3D LiDAR networks identifies out-of-distribution objects.

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

The paper proposes an efficient anomaly segmentation method for 3D LiDAR that operates directly in the network's feature space by modeling the distribution of features from inlier classes. This modeling step constrains samples that do not match the learned distributions, allowing detection of previously unseen objects. Existing 3D approaches mostly rely on post-processing borrowed from 2D vision and suffer from limited public datasets that feature only simple scenarios and sensor domain gaps. The authors also release new mixed real-synthetic datasets containing multiple out-of-distribution objects in diverse complex environments. Experiments show the method reaches state-of-the-art results on the existing real-world dataset and competitive performance on the new mixed datasets.

Core claim

The central claim is that directly modeling the feature distribution of inlier classes inside the network constrains anomalous samples and enables effective 3D LiDAR anomaly segmentation without relying on 2D post-processing techniques. The paper further claims that newly introduced mixed real-synthetic datasets, built on established semantic segmentation benchmarks with multiple out-of-distribution objects and varied environments, close the gap left by the only prior public dataset and provide a more realistic testbed, with the proposed method delivering strong results on both.

What carries the argument

A model of the feature distribution of inlier classes that operates inside the network feature space to constrain and identify out-of-distribution samples.

Load-bearing premise

That the learned feature distributions of known inlier classes will differ enough from those of unseen objects to separate them reliably in complex real-world 3D LiDAR scenes without many false positives.

What would settle it

A test set where the method produces high false-positive rates on known objects or fails to flag certain anomalous objects in the mixed datasets would show the inlier modeling does not reliably constrain anomalies.

Figures

Figures reproduced from arXiv: 2604.23604 by Alberto Pretto, Daniel Fusaro, Simone Mosco.

Figure 1
Figure 1. Figure 1: Overview of the LiDAR anomaly segmentation task. view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed LIDO approach. A backbone extracts per-point features, which are then processed by two different view at source ↗
Figure 3
Figure 3. Figure 3: Anomaly Segmentation results on STU and our proposed SemanticKITTI-OoD dataset (Multi split). Ground truth anomaly ob view at source ↗
Figure 4
Figure 4. Figure 4: Examples of ModelNet objects selected for the creation view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of anomaly points on the XY plane across all proposed OoD datasets. view at source ↗
Figure 6
Figure 6. Figure 6: Example of computed intensity values in the proposed view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison of 3D LiDAR anomaly segmentation results on STU validation set. view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison of 3D LiDAR anomaly segmentation results on SemanticPOSS-OoD view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparison of 3D LiDAR anomaly segmentation results on SemanticKITTI-OoD view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison of 3D LiDAR anomaly segmentation results on nuScenes-OoD view at source ↗
read the original abstract

Understanding the surrounding environment is fundamental in autonomous driving and robotic perception. Distinguishing between known classes and previously unseen objects is crucial in real-world environments, as done in Anomaly Segmentation. However, research in the 3D field remains limited, with most existing approaches applying post-processing techniques from 2D vision. To cover this lack, we propose a new efficient approach that directly operates in the feature space, modeling the feature distribution of inlier classes to constrain anomalous samples. Moreover, the only publicly available 3D LiDAR anomaly segmentation dataset contains simple scenarios, with few anomaly instances, and exhibits a severe domain gap due to its sensor resolution. To bridge this gap, we introduce a set of mixed real-synthetic datasets for 3D LiDAR anomaly segmentation, built upon established semantic segmentation benchmarks, with multiple out-of-distribution objects and diverse, complex environments. Extensive experiments demonstrate that our approach achieves state-of-the-art and competitive results on the existing real-world dataset and the newly introduced mixed datasets, respectively, validating the effectiveness of our method and the utility of the proposed datasets. Code and datasets are available at https://simom0.github.io/lido-page/.

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 paper proposes an efficient feature-space method for 3D LiDAR anomaly segmentation that models the distribution of inlier-class features to constrain and identify out-of-distribution objects, avoiding 2D post-processing. It also introduces mixed real-synthetic datasets built from existing semantic segmentation benchmarks to address the limitations of the sole public real-world dataset (simple scenarios, few anomalies, sensor domain gap). Extensive experiments are reported to show state-of-the-art performance on the existing real-world benchmark and competitive results on the new mixed datasets.

Significance. If the empirical claims hold under rigorous scrutiny, the work would provide a practical advance for anomaly detection in autonomous driving and robotics by operating directly on learned 3D features and supplying new evaluation resources that better reflect complex environments. The datasets in particular could become a useful community benchmark if they are shown to close the domain gap without introducing artifacts.

major comments (2)
  1. [Method / Experiments] The central modeling assumption—that inlier feature distributions learned inside the network will reliably place all plausible real-world OOD geometries, intensities, and occlusion patterns outside the inlier region—is load-bearing for the method's validity. Given LiDAR sparsity and sensor-specific artifacts, the paper must demonstrate (e.g., via feature-space visualizations, nearest-neighbor analysis, or failure-case study in the experiments section) that the learned boundary does not admit real anomalous objects inside the convex hull of inlier features.
  2. [Abstract / Experiments] The abstract states SOTA and competitive results but supplies no quantitative metrics, ablation tables, or error analysis. The full manuscript must include clear tables (e.g., AUROC, AUPR, or mIoU for anomaly segmentation) with statistical significance, baseline comparisons, and per-scenario breakdowns on both the original and mixed datasets to substantiate the claims.
minor comments (2)
  1. [Datasets] Clarify the exact procedure used to construct the mixed real-synthetic datasets (sensor simulation parameters, anomaly insertion strategy, train/test splits) so that reproducibility is immediate from the text rather than only from the released code.
  2. [Method] Ensure all notation for feature distribution modeling (e.g., any density estimator or distance metric) is defined consistently and referenced to the relevant equation or algorithm box.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and describe the revisions planned for the next version of the manuscript.

read point-by-point responses

  1. Referee: [Method / Experiments] The central modeling assumption—that inlier feature distributions learned inside the network will reliably place all plausible real-world OOD geometries, intensities, and occlusion patterns outside the inlier region—is load-bearing for the method's validity. Given LiDAR sparsity and sensor-specific artifacts, the paper must demonstrate (e.g., via feature-space visualizations, nearest-neighbor analysis, or failure-case study in the experiments section) that the learned boundary does not admit real anomalous objects inside the convex hull of inlier features.

    Authors: We agree that direct validation of this assumption is important given LiDAR-specific challenges. The current manuscript supports the assumption indirectly through state-of-the-art quantitative performance on real and mixed datasets. In the revision we will add t-SNE and PCA visualizations of inlier versus OOD features, nearest-neighbor distance analysis between anomalous points and the inlier hull, and a failure-case study section that explicitly checks for any anomalous objects falling inside the learned boundary. revision: yes

  2. Referee: [Abstract / Experiments] The abstract states SOTA and competitive results but supplies no quantitative metrics, ablation tables, or error analysis. The full manuscript must include clear tables (e.g., AUROC, AUPR, or mIoU for anomaly segmentation) with statistical significance, baseline comparisons, and per-scenario breakdowns on both the original and mixed datasets to substantiate the claims.

    Authors: Abstracts are kept concise by convention and do not contain numerical tables. The full manuscript already reports AUROC, AUPR, and mIoU tables with baseline comparisons on both the original SemanticKITTI-based benchmark and the new mixed real-synthetic datasets. To strengthen the presentation we will add statistical significance (means and standard deviations over multiple runs), expanded ablation tables, and explicit per-scenario breakdowns in the revised experiments section. revision: partial

Circularity Check

0 steps flagged

No significant circularity; empirical ML method with measured results on held-out data

full rationale

The paper describes an empirical approach for 3D LiDAR anomaly segmentation that models the feature distribution of known inlier classes inside a neural network to constrain anomalous samples. All performance claims are presented as experimental measurements on the existing real-world dataset and newly introduced mixed real-synthetic datasets, with results reported as state-of-the-art or competitive on held-out test splits. No equations, derivations, or first-principles steps are provided that reduce any claimed prediction or separation capability to a fitted parameter or self-referential definition. No self-citation load-bearing arguments, uniqueness theorems, or ansatz smuggling appear in the method description. The core claim remains an independent modeling choice whose validity is assessed externally via benchmark performance rather than by construction from its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that inlier feature distributions form a compact, modelable region that can be used to flag outliers; no free parameters or invented entities are described in the abstract.

axioms (1)
  • domain assumption Feature distributions of in-distribution classes can be modeled to constrain and identify anomalous samples in 3D LiDAR feature space
    Core premise of the proposed method stated in the abstract.

pith-pipeline@v0.9.0 · 5510 in / 1126 out tokens · 23748 ms · 2026-05-08T06:35:14.976172+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

78 extracted references · 78 canonical work pages

  1. [1]

    Salsanet: Fast road and vehicle segmentation in lidar point clouds for autonomous driving

    Eren Erdal Aksoy, Saimir Baci, and Selcuk Cavdar. Salsanet: Fast road and vehicle segmentation in lidar point clouds for autonomous driving. InIV, pages 926–932, 2020. 2

    work page 2020

  2. [2]

    Rangevit: Towards vision transformers for 3d semantic segmentation in au- tonomous driving

    Angelika Ando, Spyros Gidaris, Andrei Bursuc, Gilles Puy, Alexandre Boulch, and Renaud Marlet. Rangevit: Towards vision transformers for 3d semantic segmentation in au- tonomous driving. InCVPR, pages 5240–5250, 2023. 2

    work page 2023

  3. [3]

    Sood-imagenet: a large-scale dataset for semantic out-of-distribution image classification and se- mantic segmentation

    Alberto Bacchin, Davide Allegro, Stefano Ghidoni, and Emanuele Menegatti. Sood-imagenet: a large-scale dataset for semantic out-of-distribution image classification and se- mantic segmentation. InECCV, pages 80–97, 2024. 2

    work page 2024

  4. [4]

    Behley, M

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stachniss, and J. Gall. SemanticKITTI: A Dataset for Se- mantic Scene Understanding of LiDAR Sequences. InICCV, pages 9297–9307, 2019. 1, 5, 6

    work page 2019

  5. [5]

    The lov ´asz-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks

    Maxim Berman, Amal Rannen Triki, and Matthew B Blaschko. The lov ´asz-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks. InCVPR, pages 4413–4421, 2018. 4

    work page 2018

  6. [6]

    The fishyscapes benchmark: Measuring blind spots in semantic segmentation.IJCV, 129 (11):3119–3135, 2021

    Hermann Blum, Paul-Edouard Sarlin, Juan Nieto, Roland Siegwart, and Cesar Cadena. The fishyscapes benchmark: Measuring blind spots in semantic segmentation.IJCV, 129 (11):3119–3135, 2021. 1, 2, 6, 4

    work page 2021

  7. [7]

    Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom

    Holger Caesar, Varun Bankiti, Alex H. Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom. nuscenes: A multi- modal dataset for autonomous driving. InCVPR, 2020. 1, 5, 6

    work page 2020

  8. [8]

    Lidar panoptic segmentation in an open world.IJCV, 133(3):1153–1174, 2025

    Anirudh S Chakravarthy, Meghana Reddy Ganesina, Peiyun Hu, Laura Leal-Taix´e, Shu Kong, Deva Ramanan, and Aljosa Osep. Lidar panoptic segmentation in an open world.IJCV, 133(3):1153–1174, 2025. 1, 2

    work page 2025

  9. [9]

    Segmentmeifyoucan: A benchmark for anomaly segmentation.NeurIPS, 2021

    Robin Chan, Krzysztof Lis, Svenja Uhlemeyer, Hermann Blum, Sina Honari, Roland Siegwart, Pascal Fua, Mathieu Salzmann, and Matthias Rottmann. Segmentmeifyoucan: A benchmark for anomaly segmentation.NeurIPS, 2021. 1, 2

    work page 2021

  10. [10]

    Entropy maximization and meta classification for out-of- distribution detection in semantic segmentation

    Robin Chan, Matthias Rottmann, and Hanno Gottschalk. Entropy maximization and meta classification for out-of- distribution detection in semantic segmentation. InICCV, pages 5128–5137, 2021. 2

    work page 2021

  11. [11]

    A simple framework for contrastive learn- ing of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton. A simple framework for contrastive learn- ing of visual representations. InICML, pages 1597–1607. PmLR, 2020. 4

    work page 2020

  12. [12]

    Cenet: Toward concise and efficient lidar semantic segmen- tation for autonomous driving

    Hui Xian Cheng, Xian Feng Han, and Guo Qiang Xiao. Cenet: Toward concise and efficient lidar semantic segmen- tation for autonomous driving. InICME, pages 01–06, 2022. 2

    work page 2022

  13. [13]

    Af2-s3net: Attentive feature fusion with adap- tive feature selection for sparse semantic segmentation net- work

    Ran Cheng, Ryan Razani, Ehsan Taghavi, Enxu Li, and Bingbing Liu. Af2-s3net: Attentive feature fusion with adap- tive feature selection for sparse semantic segmentation net- work. InCVPR, pages 12547–12556, 2021. 2

    work page 2021

  14. [14]

    4d spatio-temporal convnets: Minkowski convolutional neural networks

    Christopher Choy, JunYoung Gwak, and Silvio Savarese. 4d spatio-temporal convnets: Minkowski convolutional neural networks. InCVPR, pages 3075–3084, 2019. 2, 3, 5

    work page 2019

  15. [15]

    Salsanext: Fast, uncertainty-aware semantic segmentation of lidar point clouds

    Tiago Cortinhal, George Tzelepis, and Eren Erdal Aksoy. Salsanext: Fast, uncertainty-aware semantic segmentation of lidar point clouds. InInternational Symposium on Visual Computing (ISVC), pages 207–222, 2020. 2

    work page 2020

  16. [16]

    Outlier detec- tion by ensembling uncertainty with negative objectness

    Anja Deli ´c, Matej Grcic, and Sini ˇsa ˇSegvi´c. Outlier detec- tion by ensembling uncertainty with negative objectness. In BMVC, 2024. 1, 2

    work page 2024

  17. [17]

    Reducing network agnostophobia.NeurIPS, 31, 2018

    Akshay Raj Dhamija, Manuel G ¨unther, and Terrance Boult. Reducing network agnostophobia.NeurIPS, 31, 2018. 4

    work page 2018

  18. [18]

    Pixel-wise anomaly detection in complex driving scenes

    Giancarlo Di Biase, Hermann Blum, Roland Siegwart, and Cesar Cadena. Pixel-wise anomaly detection in complex driving scenes. InCVPR, pages 16918–16927, 2021. 2

    work page 2021

  19. [19]

    Carla: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. Carla: An open urban driving simulator. InCoRL, pages 1–16, 2017. 5

    work page 2017

  20. [20]

    Lsk3dnet: Towards effective and efficient 3d perception with large sparse kernels

    Tuo Feng, Wenguan Wang, Fan Ma, and Yi Yang. Lsk3dnet: Towards effective and efficient 3d perception with large sparse kernels. InCVPR, pages 14916–14927, 2024. 1, 2

    work page 2024

  21. [21]

    Exploiting local features and range images for small data real-time point cloud semantic segmentation

    Daniel Fusaro, Simone Mosco, Emanuele Menegatti, and Al- berto Pretto. Exploiting local features and range images for small data real-time point cloud semantic segmentation. In IROS, pages 4980–4987, 2024. 2

    work page 2024

  22. [22]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. InICML, pages 1050–1059. PMLR, 2016. 2

    work page 2016

  23. [23]

    Deep learning for 3d point clouds: A survey.IEEE TPAMI, 43(12):4338–4364, 2020

    Yulan Guo, Hanyun Wang, Qingyong Hu, Hao Liu, Li Liu, and Mohammed Bennamoun. Deep learning for 3d point clouds: A survey.IEEE TPAMI, 43(12):4338–4364, 2020. 1

    work page 2020

  24. [24]

    A baseline for detect- ing misclassified and out-of-distribution examples in neural networks.ICLR, 2017

    Dan Hendrycks and Kevin Gimpel. A baseline for detect- ing misclassified and out-of-distribution examples in neural networks.ICLR, 2017. 2, 6, 7, 4

    work page 2017

  25. [25]

    Point-to-voxel knowledge distillation for lidar semantic segmentation

    Yuenan Hou, Xinge Zhu, Yuexin Ma, Chen Change Loy, and Yikang Li. Point-to-voxel knowledge distillation for lidar semantic segmentation. InCVPR, pages 8479–8488, 2022. 2

    work page 2022

  26. [26]

    Generalized odin: Detecting out-of-distribution image with- out learning from out-of-distribution data

    Yen-Chang Hsu, Yilin Shen, Hongxia Jin, and Zsolt Kira. Generalized odin: Detecting out-of-distribution image with- out learning from out-of-distribution data. InCVPR, pages 10951–10960, 2020. 7

    work page 2020

  27. [27]

    Rethinking range view representation for lidar segmentation

    Lingdong Kong, Youquan Liu, Runnan Chen, Yuexin Ma, Xinge Zhu, Yikang Li, Yuenan Hou, Yu Qiao, and Ziwei Liu. Rethinking range view representation for lidar segmentation. InICCV, pages 228–240, 2023. 1, 2

    work page 2023

  28. [28]

    Calib3d: Calibrating model preferences for reliable 3d scene understanding

    Lingdong Kong, Xiang Xu, Jun Cen, Wenwei Zhang, Liang Pan, Kai Chen, and Ziwei Liu. Calib3d: Calibrating model preferences for reliable 3d scene understanding. InWACV, pages 1965–1978, 2025. 7, 8, 5

    work page 1965

  29. [29]

    Spherical transformer for lidar-based 3d recognition

    Xin Lai, Yukang Chen, Fanbin Lu, Jianhui Liu, and Jiaya Jia. Spherical transformer for lidar-based 3d recognition. In CVPR, pages 17545–17555, 2023. 2

    work page 2023

  30. [30]

    Simple and scalable predictive uncertainty esti- mation using deep ensembles.NeurIPS, 30, 2017

    Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive uncertainty esti- mation using deep ensembles.NeurIPS, 30, 2017. 1, 2, 6, 7, 4

    work page 2017

  31. [31]

    Coda: A real-world road corner 9 case dataset for object detection in autonomous driving

    Kaican Li, Kai Chen, Haoyu Wang, Lanqing Hong, Chao- qiang Ye, Jianhua Han, Yukuai Chen, Wei Zhang, Chunjing Xu, Dit-Yan Yeung, et al. Coda: A real-world road corner 9 case dataset for object detection in autonomous driving. In ECCV, pages 406–423. Springer, 2022. 1, 2, 5

    work page 2022

  32. [32]

    Rapid-seg: Range-aware pointwise distance distribution networks for 3d lidar segmentation

    Li Li, Hubert PH Shum, and Toby P Breckon. Rapid-seg: Range-aware pointwise distance distribution networks for 3d lidar segmentation. InECCV, pages 222–241, 2025. 2, 7

    work page 2025

  33. [33]

    Dpgla: Bridging the gap between synthetic and real data for unsupervised domain adaptation in 3d lidar semantic segmentation

    Wanmeng Li, Simone Mosco, Daniel Fusaro, and Alberto Pretto. Dpgla: Bridging the gap between synthetic and real data for unsupervised domain adaptation in 3d lidar semantic segmentation. InIROS, pages 11553–11560, 2025. 8

    work page 2025

  34. [34]

    Diversity-measurable anomaly detection

    Wenrui Liu, Hong Chang, Bingpeng Ma, Shiguang Shan, and Xilin Chen. Diversity-measurable anomaly detection. InCVPR, pages 12147–12156, 2023. 2

    work page 2023

  35. [35]

    Uniseg: A unified multi-modal lidar segmentation network and the openpcseg codebase

    Youquan Liu, Runnan Chen, Xin Li, Lingdong Kong, Yuchen Yang, Zhaoyang Xia, Yeqi Bai, Xinge Zhu, Yuexin Ma, Yikang Li, et al. Uniseg: A unified multi-modal lidar segmentation network and the openpcseg codebase. InICCV, pages 21662–21673, 2023. 2

    work page 2023

  36. [36]

    Calibrated and efficient sampling-free confidence esti- mation for lidar scene semantic segmentation.arXiv, 2024

    Hanieh Shojaei Miandashti, Qianqian Zou, and Claus Bren- ner. Calibrated and efficient sampling-free confidence esti- mation for lidar scene semantic segmentation.arXiv, 2024. 8

    work page 2024

  37. [37]

    Rangenet++: Fast and accurate lidar semantic segmentation

    Andres Milioto, Ignacio Vizzo, Jens Behley, and Cyrill Stachniss. Rangenet++: Fast and accurate lidar semantic segmentation. InIROS, pages 4213–4220, 2019. 2, 5

    work page 2019

  38. [38]

    Confidence prediction for lexicon-free ocr

    Noam Mor and Lior Wolf. Confidence prediction for lexicon-free ocr. InWACV, pages 218–225, 2018. 2

    work page 2018

  39. [39]

    Point-plane projections for accurate lidar semantic segmentation in small data scenarios

    Simone Mosco, Daniel Fusaro, Wanmeng Li, Emanuele Menegatti, and Alberto Pretto. Point-plane projections for accurate lidar semantic segmentation in small data scenarios. arXiv, 2025. 2

    work page 2025

  40. [40]

    Revisiting retentive networks for fast range-view 3d lidar semantic segmentation

    Simone Mosco, Daniel Fusaro, Wanmeng Li, and Alberto Pretto. Revisiting retentive networks for fast range-view 3d lidar semantic segmentation. InWACV, pages 2499–2509,

    work page

  41. [41]

    Rba: Segmenting unknown regions rejected by all

    Nazir Nayal, Misra Yavuz, Joao F Henriques, and Fatma G¨uney. Rba: Segmenting unknown regions rejected by all. InICCV, pages 711–722, 2023. 1, 2, 6, 7, 4

    work page 2023

  42. [42]

    A likelihood ratio-based approach to segmenting unknown objects.IJCV, pages 1–13, 2025

    Nazir Nayal, Youssef Shoeb, and Fatma G¨uney. A likelihood ratio-based approach to segmenting unknown objects.IJCV, pages 1–13, 2025. 1, 2

    work page 2025

  43. [43]

    Spotting the unexpected (stu): A 3d lidar dataset for anomaly segmenta- tion in autonomous driving

    Alexey Nekrasov, Malcolm Burdorf, Stewart Worrall, Bas- tian Leibe, and Julie Stephany Berrio Perez. Spotting the unexpected (stu): A 3d lidar dataset for anomaly segmenta- tion in autonomous driving. InCVPR, pages 11875–11885,

    work page

  44. [44]

    Generalization of the lam- bertian model and implications for machine vision.IJCV, 14 (3):227–251, 1995

    Michael Oren and Shree K Nayar. Generalization of the lam- bertian model and implications for machine vision.IJCV, 14 (3):227–251, 1995. 5, 1

    work page 1995

  45. [45]

    Semanticposs: A point cloud dataset with large quantity of dynamic instances

    Yancheng Pan, Biao Gao, Jilin Mei, Sibo Geng, Chengkun Li, and Huijing Zhao. Semanticposs: A point cloud dataset with large quantity of dynamic instances. InIV, pages 687–

    work page

  46. [46]

    Using a waffle iron for automotive point cloud semantic segmenta- tion

    Gilles Puy, Alexandre Boulch, and Renaud Marlet. Using a waffle iron for automotive point cloud semantic segmenta- tion. InICCV, pages 3379–3389, 2023. 2

    work page 2023

  47. [47]

    Qi, Hao Su, Kaichun Mo, and Leonidas J

    C. Qi, Hao Su, Kaichun Mo, and Leonidas J. Guibas. Point- net: Deep learning on point sets for 3d classification and seg- mentation.CVPR, pages 77–85, 2017. 2

    work page 2017

  48. [48]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space.NeurIPS, 30, 2017

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space.NeurIPS, 30, 2017. 2

    work page 2017

  49. [49]

    Unmasking anomalies in road- scene segmentation

    Shyam Nandan Rai, Fabio Cermelli, Dario Fontanel, Carlo Masone, and Barbara Caputo. Unmasking anomalies in road- scene segmentation. InICCV, pages 4037–4046, 2023. 2

    work page 2023

  50. [50]

    Denseclip: Language-guided dense prediction with context- aware prompting

    Yongming Rao, Wenliang Zhao, Guangyi Chen, Yansong Tang, Zheng Zhu, Guan Huang, Jie Zhou, and Jiwen Lu. Denseclip: Language-guided dense prediction with context- aware prompting. InCVPR, pages 18082–18091, 2022. 2

    work page 2022

  51. [51]

    Towards to- tal recall in industrial anomaly detection

    Karsten Roth, Latha Pemula, Joaquin Zepeda, Bernhard Sch¨olkopf, Thomas Brox, and Peter Gehler. Towards to- tal recall in industrial anomaly detection. InCVPR, pages 14318–14328, 2022. 2

    work page 2022

  52. [52]

    Cosmix: Compositional semantic mix for domain adaptation in 3d lidar segmenta- tion

    Cristiano Saltori, Fabio Galasso, Giuseppe Fiameni, Nicu Sebe, Elisa Ricci, and Fabio Poiesi. Cosmix: Compositional semantic mix for domain adaptation in 3d lidar segmenta- tion. InECCV, pages 586–602. Springer, 2022. 8

    work page 2022

  53. [53]

    C. E. Shannon. A mathematical theory of communication. The Bell System Technical Journal, 27(3):379–423, 1948. 4

    work page 1948

  54. [54]

    Lidar guided small obstacle segmenta- tion

    Aasheesh Singh, Aditya Kamireddypalli, Vineet Gandhi, and K Madhava Krishna. Lidar guided small obstacle segmenta- tion. InIROS, pages 8513–8520, 2020. 2, 5

    work page 2020

  55. [55]

    Open-World Semantic Seg- mentation Including Class Similarity

    Matteo Sodano, Federico Magistri, Lucas Nunes, Jens Behley, and Cyrill Stachniss. Open-World Semantic Seg- mentation Including Class Similarity. InCVPR, 2024. 1, 2, 3, 4

    work page 2024

  56. [56]

    Dropout: A simple way to prevent neural networks from overfitting.JMLR, 15 (56):1929–1958, 2014

    Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: A simple way to prevent neural networks from overfitting.JMLR, 15 (56):1929–1958, 2014. 6, 4

    work page 1929

  57. [57]

    Searching efficient 3d architec- tures with sparse point-voxel convolution

    Haotian Tang, Zhijian Liu, Shengyu Zhao, Yujun Lin, Ji Lin, Hanrui Wang, and Song Han. Searching efficient 3d architec- tures with sparse point-voxel convolution. InECCV, pages 685–702, 2020. 2

    work page 2020

  58. [58]

    Kpconv: Flexible and deformable convolution for point clouds

    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. InICCV, pages 6411–6420, 2019. 2

    work page 2019

  59. [59]

    Multi-scale patch-based representation learning for image anomaly detection and segmentation

    Chin-Chia Tsai, Tsung-Hsuan Wu, and Shang-Hong Lai. Multi-scale patch-based representation learning for image anomaly detection and segmentation. InWACV, pages 3065– 3073, 2022. 2

    work page 2022

  60. [60]

    Dynamic graph cnn for learning on point clouds.ACM TOG, 38(5): 1–12, 2019

    Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon. Dynamic graph cnn for learning on point clouds.ACM TOG, 38(5): 1–12, 2019. 2

    work page 2019

  61. [61]

    Identifying unknown instances for au- tonomous driving

    Kelvin Wong, Shenlong Wang, Mengye Ren, Ming Liang, and Raquel Urtasun. Identifying unknown instances for au- tonomous driving. InCoRL, pages 384–393, 2020. 1, 2, 5

    work page 2020

  62. [62]

    Squeezeseg: Convolutional neural nets with recurrent crf for real-time road-object segmentation from 3d lidar point cloud

    Bichen Wu, Alvin Wan, Xiangyu Yue, and Kurt Keutzer. Squeezeseg: Convolutional neural nets with recurrent crf for real-time road-object segmentation from 3d lidar point cloud. InICRA, pages 1887–1893, 2018. 2

    work page 2018

  63. [63]

    Squeezesegv2: Improved model structure and 10 unsupervised domain adaptation for road-object segmenta- tion from a lidar point cloud

    Bichen Wu, Xuanyu Zhou, Sicheng Zhao, Xiangyu Yue, and Kurt Keutzer. Squeezesegv2: Improved model structure and 10 unsupervised domain adaptation for road-object segmenta- tion from a lidar point cloud. InICRA, pages 4376–4382,

    work page

  64. [64]

    Pointconv: Deep convolutional networks on 3d point clouds

    Wenxuan Wu, Zhongang Qi, and Li Fuxin. Pointconv: Deep convolutional networks on 3d point clouds. InCVPR, pages 9621–9630, 2019. 2

    work page 2019

  65. [65]

    Point transformer v2: Grouped vector atten- tion and partition-based pooling.NeurIPS, 35:33330–33342,

    Xiaoyang Wu, Yixing Lao, Li Jiang, Xihui Liu, and Heng- shuang Zhao. Point transformer v2: Grouped vector atten- tion and partition-based pooling.NeurIPS, 35:33330–33342,

    work page

  66. [66]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xi- hui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. In CVPR, pages 4840–4851, 2024. 1, 2

    work page 2024

  67. [67]

    3d shapenets: A deep representation for volumetric shapes

    Zhirong Wu, Shuran Song, Aditya Khosla, Fisher Yu, Lin- guang Zhang, Xiaoou Tang, and Jianxiong Xiao. 3d shapenets: A deep representation for volumetric shapes. In CVPR, pages 1912–1920, 2015. 2, 5, 1

    work page 1912

  68. [68]

    Synthesize then compare: Detecting failures and anomalies for semantic segmentation

    Yingda Xia, Yi Zhang, Fengze Liu, Wei Shen, and Alan L Yuille. Synthesize then compare: Detecting failures and anomalies for semantic segmentation. InECCV, pages 145– 161, 2020. 2

    work page 2020

  69. [69]

    Squeeze- segv3: Spatially-adaptive convolution for efficient point- cloud segmentation

    Chenfeng Xu, Bichen Wu, Zining Wang, Wei Zhan, Peter Vajda, Kurt Keutzer, and Masayoshi Tomizuka. Squeeze- segv3: Spatially-adaptive convolution for efficient point- cloud segmentation. InECCV, pages 1–19, 2020. 2

    work page 2020

  70. [70]

    Rpvnet: A deep and efficient range-point- voxel fusion network for lidar point cloud segmentation

    Jianyun Xu, Ruixiang Zhang, Jian Dou, Yushi Zhu, Jie Sun, and Shiliang Pu. Rpvnet: A deep and efficient range-point- voxel fusion network for lidar point cloud segmentation. In ICCV, pages 16024–16033, 2021. 2, 7

    work page 2021

  71. [71]

    2dpass: 2d priors as- sisted semantic segmentation on lidar point clouds

    Xu Yan, Jiantao Gao, Chaoda Zheng, Chao Zheng, Ruimao Zhang, Shuguang Cui, and Zhen Li. 2dpass: 2d priors as- sisted semantic segmentation on lidar point clouds. InECCV, pages 677–695, 2022. 2

    work page 2022

  72. [72]

    Mask4former: Mask transformer for 4d panoptic seg- mentation

    Kadir Yilmaz, Jonas Schult, Alexey Nekrasov, and Bastian Leibe. Mask4former: Mask transformer for 4d panoptic seg- mentation. InICRA, pages 9418–9425, 2024. 6, 7, 4, 5

    work page 2024

  73. [73]

    Dino in the room: Leveraging 2d foundation models for 3d segmentation.arXiv, 2025

    Karim Abou Zeid, Kadir Yilmaz, Daan de Geus, Alexander Hermans, David Adrian, Timm Linder, and Bastian Leibe. Dino in the room: Leveraging 2d foundation models for 3d segmentation.arXiv, 2025. 2

    work page 2025

  74. [74]

    Polarnet: An improved grid representation for online lidar point clouds se- mantic segmentation

    Yang Zhang, Zixiang Zhou, Philip David, Xiangyu Yue, Ze- rong Xi, Boqing Gong, and Hassan Foroosh. Polarnet: An improved grid representation for online lidar point clouds se- mantic segmentation. InCVPR, pages 9601–9610, 2020. 2

    work page 2020

  75. [75]

    Point transformer

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. InICCV, pages 16259– 16268, 2021. 2

    work page 2021

  76. [76]

    Omnial: A unified cnn framework for unsu- pervised anomaly localization

    Ying Zhao. Omnial: A unified cnn framework for unsu- pervised anomaly localization. InCVPR, pages 3924–3933,

    work page

  77. [77]

    Regionclip: Region-based language-image pretraining

    Yiwu Zhong, Jianwei Yang, Pengchuan Zhang, Chunyuan Li, Noel Codella, Liunian Harold Li, Luowei Zhou, Xiyang Dai, Lu Yuan, Yin Li, et al. Regionclip: Region-based language-image pretraining. InCVPR, pages 16793–16803,

    work page

  78. [78]

    Cylindrical and asymmetrical 3d convolution networks for lidar segmenta- tion

    Xinge Zhu, Hui Zhou, Tai Wang, Fangzhou Hong, Yuexin Ma, Wei Li, Hongsheng Li, and Dahua Lin. Cylindrical and asymmetrical 3d convolution networks for lidar segmenta- tion. InCVPR, pages 9939–9948, 2021. 2 11 Learning to Identify Out-of-Distribution Objects for 3D LiDAR Anomaly Segmentation Supplementary Material A. Further Details on OoD Datasets In this...

    work page arXiv 2021