Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Relative energy — the log-odds of OOD versus ID — makes LiDAR point-wise outlier detection state-of-the-art on STU and SemanticKITTI.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 23:13 UTC pith:RFA36WJZ

load-bearing objection Strong empirical LiDAR OOD detector with a clean relative-energy formulation; SOTA claim needs UEM on the test set and error bars before I'd trust the margin. the 4 major comments →

arxiv 2511.06720 v3 pith:RFA36WJZ submitted 2025-11-10 cs.CV

Relative Energy Learning for LiDAR Out-of-Distribution Detection

Zizhao Li , Zhengkang Xiang , Jiayang Ao , Joseph West , Kourosh Khoshelham This is my paper
classification cs.CV
keywords LiDARout-of-distribution detectionenergy-based modelrelative energypoint cloud segmentationsynthetic outlier generationautonomous drivinganomaly segmentation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that a simple change in score function — comparing out-of-distribution evidence against in-distribution evidence per point, rather than using absolute energy or maximum softmax probability — removes the overconfidence that makes LiDAR OOD detectors unreliable. It pairs this relative energy score with Point Raise, a geometry-aware synthesis that converts flat road patches into compact protrusions resembling real hazards, so training needs no external OOD datasets. On the STU benchmark the method reaches 97.85 AUROC and a 9.60 false-positive rate at 95% recall, roughly a quarter of the strongest baseline's error, while keeping closed-set panoptic quality nearly intact. The paper argues that this combination of a shift-invariant relative score and simple synthetic outliers gives autonomous driving systems a practical per-point OOD signal without sacrificing segmentation.

Core claim

The central discovery is that the relative energy margin ΔE = log( Σ_{i∈y⁻} exp(f_i) / Σ_{i∈y⁺} exp(f_i) ), which is exactly the log-odds of 'OOD versus ID' under the grouped softmax, is a trainable, shift-invariant OOD score. Trained with a logistic loss and a single imbalance weight, it outperforms both post-hoc scores (Max Logit, energy) and learned energy objectives (hinge, VOS, dual energy), cutting FPR@95 to 9.60 on the STU validation set versus 19.45 for the best energy baseline. The paper further shows that Point Raise — contracting a random road-point cluster radially toward the sensor with an adaptive decay and lifting it vertically — supplies sufficient auxiliary OOD data without

What carries the argument

The central object is the relative energy margin ΔE, the log ratio of summed negative-logit exponentials to summed positive-logit exponentials. Being a ratio of two aggregated energy terms, it is invariant to a constant shift in all logits, which directly addresses the calibration problem of raw energy scores. The companion machinery is Point Raise: its adaptive decay a = −log(d_min/d_max)/(γ(d_max−d_min)) maps a flat road patch into a compact, density-preserving protrusion by scaling each point by exp(−a·d_shift) and adding random heights, creating the negative samples that make ΔE's decision boundary learnable.

Load-bearing premise

The load-bearing premise is that Point Raise's radially contracted and lifted road patches adequately cover the geometric variety of real anomalies; the paper itself acknowledges residual failures on high-curvature surfaces and object boundaries, where sparse sampling makes reliable anomaly segmentation difficult.

What would settle it

Measure REL's point-wise AUROC on the STU test set separately for naturalistic anomalies (e.g., flat road debris, bins) versus controlled, protruding object placements; if FPR@95 on naturalistic anomalies is substantially above the reported 21.69, the Point Raise training signal does not cover non-protruding hazards. Alternatively, replace Point Raise with a different synthetic OOD generator (e.g., random-scaled external object clusters) and check whether the ΔE boundary transfers; a large performance drop would show the specific synthetic geometry, not the relative energy score itself, is ess

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • A per-point OOD score can be added as a lightweight auxiliary branch to a mask-transformer LiDAR segmentation model without substantial loss of panoptic quality (PQ within ~3.7 on STU and ~1.6 on SemanticKITTI).
  • Relative energy reduces FPR@95 to 9.60 versus 19.45 for the best prior energy objective, suggesting that ratio-based scoring is more robust than absolute-energy margin learning.
  • Point Raise removes the need for external OOD datasets and density post-processing; a single additional training sequence suffices for the STU fine-tuning, making the approach scalable to new deployments.
  • The method transfers across datasets: trained on STU-style geometry, it achieves 96.76 AUROC on SemanticKITTI outlier classes, outperforming likelihood-ratio and prototype-based baselines.
  • Even with a frozen backbone, the OOD projector alone yields 94.43 AUROC, so REL can serve as a cheap add-on to already-trained segmentation models.

Where Pith is reading between the lines

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

  • The synthetic protrusion bias may limit transfer to flat, non-raised anomalies such as road debris or overhead objects; a natural extension is to vary Point Raise with lateral or descending perturbations to test geometric coverage.
  • Because ΔE is exactly the softmax log-odds of a two-component mixture, the score can be interpreted as a log-likelihood ratio, suggesting principled conversion to probabilities or calibrated uncertainty estimates for downstream planning.
  • The reliance on road patches as the sole inlier surface assumes road geometry is always present; off-road or unstructured scenes (e.g., vegetation-heavy rural driving) might require using other planar inlier surfaces for synthesis.
  • A testable extension is to combine Point Raise with a small set of real hazard scans, using the synthetic augmentation to boost sample diversity and potentially close the gap to fully unsupervised OOD detection.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes Relative Energy Learning (REL), a point-wise OOD detection framework for LiDAR semantic/panoptic segmentation. REL defines a relative energy margin ΔE = log(p_neg/p_pos) between auxiliary negative logits and in-distribution positive logits, trains a lightweight OOD projector with a weighted logistic loss, and introduces Point Raise, a geometry-aware synthesis that contracts and lifts patches of road points to create pseudo-OOD training samples. The method is evaluated on the STU benchmark and SemanticKITTI, reporting state-of-the-art AUROC and FPR@95 over prior post-hoc and training-based baselines. The derivation in Eqs. (4)-(6) is correct, and the held-out STU test set is used for the main test-table results. However, the evidence for the headline claim is weakened by the omission of the strongest validation baseline from the test table, the absence of variance estimates, and under-reported per-class closed-set degradation.

Significance. If the reported results hold, REL is a simple and practical contribution: the relative-energy score is a natural log-odds formulation, avoids margin and temperature hyperparameters, and Point Raise is a lightweight alternative to external synthetic OOD sources. The reported gains are large (STU test AUROC 96.26 vs Deep Ensemble 86.74 and FPR@95 21.69 vs 58.05; SemanticKITTI AUROC 96.76 vs UEM 93.15), and the ablations isolate the contribution of the REL objective and the Point Raise augmentation. Strengths of the paper include the transparent derivation of ΔE, the use of a held-out STU test split, and a closed-set quality comparison. The main weaknesses are evidentiary: the STU test table omits UEM, the strongest reproduced baseline; no multi-seed or error-bar analysis is provided despite stochastic augmentation; and the per-class closed-set tables show severe drops on rare classes that are not discussed in the main text. These issues do not invalidate the method, but they need to be addressed before the state-of-the-art claim can be considered fully supported.

major comments (4)
  1. [Table 2 / Fig. 1] The STU test-set table and Figure 1 omit UEM [37], which is the strongest training-based baseline on the validation set (Table 1: UEM* AUROC 95.80, FPR@95 26.37). The caption states test results are taken from [39], but UEM is not from [39]; it was reproduced by the authors only for validation. Since the abstract and Fig. 1 make a global state-of-the-art claim, leaving out the strongest competitor on the held-out test split is a direct evidence gap. Please run UEM on the STU test set with the same protocol and report it. If UEM's test performance is close to its validation numbers, the margin over the strongest baseline is much narrower than the paper's 'large margin' wording suggests.
  2. [Tables 1-3 and 6-7] No error bars or multi-seed variance are reported in any table. Point Raise involves random sampling of road points, radii, heights, and the training procedure is stochastic, so single-run differences such as FPR@95 9.60 vs 21.69 may not be significant. Please provide mean and standard deviation over at least three seeds for the main results and ablations, or otherwise justify that the reported differences are stable.
  3. [Supplementary Tables 8 and 9] The claim in Section 5.2 that REL preserves closed-set segmentation with 'minimal impact' is contradicted by the per-class results: bicycle PQ drops from 47.65 to 0.0 on STU, and motorcyclist PQ drops from 25.01 to 0.0 on SemanticKITTI. Overall PQ is comparable, but the collapse of these rare classes is a substantial degradation. Please either report and discuss this trade-off in the main text, or provide an explanation (e.g., class-frequency interaction with fine-tuning) and adjust the claim accordingly.
  4. [Section 3.4 / Eq. (8)] Point Raise assumes that radially contracted and lifted road patches are a sufficient synthetic surrogate for real OOD objects. The paper offers cross-dataset AUROC as indirect evidence but no mechanistic argument that the synthetic geometry covers real anomalies (e.g., non-protruding, off-road, or overhead objects). The decision boundary learned from these synthetic protrusions may not transfer. A concrete test would be to stratify STU test anomalies by geometric properties (point count, protrusion, off-road placement) and report per-category performance, or to compare against an alternative synthetic OOD source such as LiON's ShapeNet-based outliers. This is a correctness-risk concern, not a circularity issue, and it directly affects the generalizability claim.
minor comments (6)
  1. [Algorithm 1 / Eq. (8)] The notation d_min/d_max is unclear; use d_min and d_max. Also clarify that the scaling s applies to the x,y coordinates only, and that the height perturbation h has per-point dimensionality matching the cluster size.
  2. [Section 3.3] The number of negative logits K is not explicitly defined in the method. It appears to equal the number of ID classes, but this should be stated clearly.
  3. [References] Reference [16] has a typo: 'kuefeng Du' should be 'Kufeng Du'. There is also a typo in Section 2: 'effectively seperate' should be 'effectively separate'.
  4. [Table 3] REAL, APF, and LiON use the Cylinder3D architecture while the other methods use Mask4Former. Since the comparison is across backbone architectures, this should be explicitly noted as a potential confound in the text.
  5. [Table 6] The 'None*' row means unlabeled points are used as auxiliary OOD data. This is important and should be explained in the main text, not only in the caption.
  6. [Supplementary, Sec. 7] The phrase 'our proposed baseline' in the supplementary qualitative section is ambiguous; use 'our method' for consistency.

Circularity Check

0 steps flagged

No significant circularity: REL's score is a defined training objective, and its central claims are evaluated on held-out data independent of the fitting process.

full rationale

The paper's derivation chain is not circular. The relative energy score ΔE(x;f) = log(Σ_{i∈y−} exp(f_i)/Σ_{i∈y+} exp(f_i)) is a definition, and the learning objective L_REL trains a binary logistic classifier on that score using synthetic OOD points generated by Point Raise. This is a standard supervised training setup, not a case where a fitted parameter is renamed as a prediction. The headline numbers (STU validation AUROC 97.85, STU test AUROC 96.26, SemanticKITTI AUROC 96.76) are measured on data that did not participate in fitting the OOD branch. Hyperparameters ω=100 and γ=2 are tuned on the STU validation split and validation results are reported, but a held-out STU test set is also reported, so the central claim does not reduce to the training objective. The omission of UEM from the STU test table (Table 2) is a legitimate completeness concern about the 'state-of-the-art' claim, but it is an evidence gap, not circularity: UEM's validation numbers are not an input to REL's derivation or fitting. The only self-citation, reference [30], is in related work and is not load-bearing for any technical claim. No uniqueness theorem, no ansatz smuggled via citation, and no renaming of a known result is used as a substitute for evidence. Overall, the paper's derivation is self-contained with respect to its own claims, and any concerns are about benchmark completeness or generalization of the synthetic outlier proxy, not circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 1 invented entities

The load-bearing choices are ω=100 (imbalance weight), γ=2 (pull factor, tuned via Table 6), the radius/height ranges [0.25,0.75], and the unablated K-negative-logit design. Domain assumptions: road points are flat radial layers whose contraction yields object-like clusters; the synthetic OOD distribution transfers to real anomalies across scenes and sensors; one STU closed-set sequence is sufficient for fine-tuning. The 'RAISED CLASS' label is a training construct with no independent evidence beyond the benchmark results.

free parameters (5)
  • ω (imbalance weight in L_REL) = 100
    Hand-set in §4.3 to balance scarce auxiliary OOD points against millions of ID points; no ablation is reported for ω, and it is chosen on the STU validation set.
  • γ (Point Raise pull factor) = 2
    Selected by the ablation in Table 6 (γ=2 gives the best AUROC/FPR@95/AP on the STU validation set).
  • Point Raise radius range [rmin, rmax] = [0.25, 0.75]
    Stated in §4.3; controls cluster size; no ablation is reported.
  • Point Raise height range [hmin, hmax] = [0.25, 0.75]
    Stated in §4.3; controls vertical extent of pseudo-OOD; no ablation is reported.
  • Number of negative logits K = K (one per ID class)
    Architectural choice in §3.2 (2K logits with K negative channels); the pairing is not ablated or justified.
axioms (4)
  • standard math Softmax/energy identities (Eq. 2-6): ΔE = log(Σ_neg e^f / Σ_pos e^f) = log(p_neg/p_pos).
    Unproved background algebra; standard and correct.
  • domain assumption Road points form flat radial layers with locally uniform sampling, so radial contraction toward the sensor produces compact object-like clusters (Eq. 8, Algo 1).
    Basis of Point Raise; verified only qualitatively via Figure 3.
  • domain assumption The synthetic Point-Raise OOD distribution is representative of test-time anomalies; the boundary ΔE=0 transfers across scenes and sensors.
    Supported indirectly by STU test and SemanticKITTI transfer results, but no mechanism or external validation is given.
  • ad hoc to paper Auxiliary OOD availability (a single STU closed-set sequence) is sufficient for the OOD projector to learn transferable boundaries.
    Fine-tuning on one closed-set sequence is a protocol choice specific to this paper.
invented entities (1)
  • 'RAISED CLASS' pseudo-OOD label no independent evidence
    purpose: Training signal for the OOD projector during Point Raise augmentation
    A training construct; its validity rests entirely on downstream benchmark performance, with no independent handle outside this paper's experiments.

pith-pipeline@v1.3.0-alltime-deepseek · 13519 in / 20912 out tokens · 214867 ms · 2026-08-03T23:13:28.627861+00:00 · methodology

0 comments
read the original abstract

Out-of-distribution (OOD) detection is a critical requirement for reliable autonomous driving, where safety depends on recognizing road obstacles and unexpected objects beyond the training distribution. Despite extensive research on OOD detection in 2D images, direct transfer to 3D LiDAR point clouds has been proven ineffective. Current LiDAR OOD methods struggle to distinguish rare anomalies from common classes, leading to high false-positive rates and overconfident errors in safety-critical settings. We propose Relative Energy Learning (REL), a simple yet effective framework for OOD detection in LiDAR point clouds. REL leverages the energy gap between positive (in-distribution) and negative logits as a relative scoring function, mitigating calibration issues in raw energy values and improving robustness across various scenes. To address the absence of OOD samples during training, we propose a lightweight data synthesis strategy called Point Raise, which perturbs existing point clouds to generate auxiliary anomalies without altering the inlier semantics. Evaluated on SemanticKITTI and the Spotting the Unexpected (STU) benchmark, REL consistently outperforms existing methods by a large margin. Our results highlight that modeling relative energy, combined with simple synthetic outliers, provides a principled and scalable solution for reliable OOD detection in open-world autonomous driving.

Figures

Figures reproduced from arXiv: 2511.06720 by Jiayang Ao, Joseph West, Kourosh Khoshelham, Zhengkang Xiang, Zizhao Li.

Figure 1
Figure 1. Figure 1: Point-level OOD detection performance on the STU test [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the Relative Energy Learning (REL) framework. We adopt the Mask4Former [ [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of Raise Cluster with Different [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of anomaly segmentation results on the STU benchmark. Points are categorized into three types: [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of anomaly segmentation results on the [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Neural Distribution Prior for LiDAR Out-of-Distribution Detection

    cs.CV 2026-04 unverdicted novelty 5.0

    NDP models prediction distributions and uses Perlin noise OOD synthesis to reach 61.31% point-level AP on STU LiDAR benchmark, over 10x prior best.

Reference graph

Works this paper leans on

51 extracted references · 1 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Se- manticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences

    Jens Behley, Martin Garbade, Andres Milioto, Jan Quen- zel, Sven Behnke, Cyrill Stachniss, and Juergen Gall. Se- manticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences. InInternational Conference on Com- puter Vision (ICCV), 2019. 2, 5, 6, 7, 8

    2019

  2. [2]

    The Fishyscapes Benchmark: Measuring Blind Spots in Semantic Segmentation.Interna- tional Journal on Computer Vision (IJCV), 2021

    Hermann Blum, Paul-Edouard Sarlin, Juan Nieto, Roland Siegwart, and Cesar Cadena. The Fishyscapes Benchmark: Measuring Blind Spots in Semantic Segmentation.Interna- tional Journal on Computer Vision (IJCV), 2021. 1, 2, 6, 7

    2021

  3. [3]

    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. InConference on Computer Vision and Pattern Recognition (CVPR), 2020. 2

    2020

  4. [4]

    Open- world semantic segmentation for lidar point clouds

    Jun Cen, Peng Yun, Shiwei Zhang, Junhao Cai, Di Luan, Mingqian Tang, Ming Liu, and Michael Yu Wang. Open- world semantic segmentation for lidar point clouds. InCom- puter Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XXXVIII, page 318–334, Berlin, Heidelberg, 2022. Springer- Verlag. 2, 5, 6

    2022

  5. [5]

    SegmentMeIfYou- Can: A Benchmark for Anomaly Segmentation

    Robin Chan, Krzysztof Lis, Svenja Uhlemeyer, Hermann Blum, Sina Honari, Roland Siegwart, Pascal Fua, Math- ieu Salzmann, and Matthias Rottmann. SegmentMeIfYou- Can: A Benchmark for Anomaly Segmentation. InProceed- ings of the Neural Information Processing Systems Track on Datasets and Benchmarks, 2021. 2

    2021

  6. [6]

    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. InInterna- tional Conference on Computer Vision (ICCV), 2021. 2, 5

    2021

  7. [7]

    Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Mano- lis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu

    Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Mano- lis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. ShapeNet: An Information-Rich 3D Model Repository. Technical Report arXiv:1512.03012 [cs.GR], Stanford University — Princeton University — Toyota Tech- nological Institute at ...

    Pith/arXiv arXiv 2015

  8. [8]

    Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation. In European Conference on Computer Vision (ECCV), 2018. 2

    2018

  9. [9]

    Dual energy-based model with open- world uncertainty estimation for out-of-distribution detec- tion

    Qi Chen and Hu Ding. Dual energy-based model with open- world uncertainty estimation for out-of-distribution detec- tion. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 25728– 25737, June 2025. 3, 8

    2025

  10. [10]

    Schwing, Alexan- der Kirillov, and Rohit Girdhar

    Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexan- der Kirillov, and Rohit Girdhar. Masked-attention Mask Transformer for Universal Image Segmentation. InConfer- ence on Computer Vision and Pattern Recognition (CVPR),

  11. [11]

    Schwing, and Alexander Kir- illov

    Bowen Cheng, Alexander G. Schwing, and Alexander Kir- illov. Per-pixel classification is not all you need for seman- tic segmentation. InNeural Information Processing Systems (NeurIPS), 2021. 2

    2021

  12. [12]

    Bal- anced energy regularization loss for out-of-distribution de- tection

    Hyunjun Choi, Hawook Jeong, and Jin Young Choi. Bal- anced energy regularization loss for out-of-distribution de- tection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1–9, 2023. 1, 3

    2023

  13. [13]

    4D Spatio-Temporal ConvNets: Minkowski Convolutional Neu- ral Networks

    Christopher Choy, JunYoung Gwak, and Silvio Savarese. 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neu- ral Networks. InConference on Computer Vision and Pattern Recognition (CVPR), 2019. 4

    2019

  14. [14]

    The cityscapes dataset for semantic urban scene understanding

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. InConfer- ence on Computer Vision and Pattern Recognition (CVPR),

  15. [15]

    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 British Machine Vision Conference (BMVC), 2024. 2

    2024

  16. [16]

    V os: Learning what you don’t know by virtual outlier synthe- sis

    kuefeng Du, Zhaoning Wang, Mu Cai, and Yixuan Li. V os: Learning what you don’t know by virtual outlier synthe- sis. InInternational Conference on Learning Representa- tions (ICLR), 2021. 3, 4, 8

    2021

  17. [17]

    A density-based algorithm for discovering clusters in large spatial databases with noise

    Martin Ester, Hans-Peter Kriegel, J ¨org Sander, and Xiaowei Xu. A density-based algorithm for discovering clusters in large spatial databases with noise. InProceedings of the Sec- ond International Conference on Knowledge Discovery and Data Mining, KDD’96, page 226–231. AAAI Press, 1996. 6

    1996

  18. [18]

    Vision meets robotics: The KITTI dataset

    A Geiger, P Lenz, C Stiller, and R Urtasun. Vision meets robotics: The KITTI dataset. InThe International Journal of Robotics Research, 2013. 2

    2013

  19. [19]

    Densehy- brid: Hybrid anomaly detection for dense open-set recogni- tion

    Matej Grci ´c, Petra Bevandi ´c, and Sini ˇsa ˇSegvi´c. Densehy- brid: Hybrid anomaly detection for dense open-set recogni- tion. InEuropean Conference on Computer Vision (ECCV),

  20. [20]

    Scaling Out-of-Distribution Detection for Real- World Settings

    Dan Hendrycks, Steven Basart, Mantas Mazeika, Andy Zou, Joe Kwon, Mohammadreza Mostajabi, Jacob Steinhardt, and Dawn Song. Scaling Out-of-Distribution Detection for Real- World Settings. InInternational Conference on Machine Learning (ICML), 2022. 1, 7

    2022

  21. [21]

    A Baseline for Detect- ing Misclassified and Out-of-Distribution Examples in Neu- ral Networks

    Dan Hendrycks and Kevin Gimpel. A Baseline for Detect- ing Misclassified and Out-of-Distribution Examples in Neu- ral Networks. InInternational Conference on Learning Rep- resentations (ICLR), 2018. 1, 2, 6

    2018

  22. [22]

    Deep Anomaly Detection with Outlier Exposure

    Dan Hendrycks, Mantas Mazeika, and Thomas Dietterich. Deep Anomaly Detection with Outlier Exposure. InInter- national Conference on Learning Representations (ICLR),

  23. [23]

    Czarnecki

    Chengjie Huang, Van Duong Nguyen, Vahdat Abdelzad, Christopher Gus Mannes, Luke Rowe, Benjamin Therien, Rick Salay, and K. Czarnecki. Out-of-distribution detection for lidar-based 3d object detection.IEEE Intelligent Trans- portation Systems Conference (ITSC), 2022. 2

    2022

  24. [24]

    H. W. Kuhn. The hungarian method for the assignment problem.Naval Research Logistics Quarterly, 2(1-2):83–97,

  25. [25]

    Revisiting Out-of-Distribution Detection in LiDAR-based 3D Object Detection

    Michael K ¨osel, Marcel Schreiber, Michael Ulrich, Claudius 9 Gl¨aser, and Klaus Dietmayer. Revisiting Out-of-Distribution Detection in LiDAR-based 3D Object Detection. InIntelli- gent Vehicles Symposium (IV), 2024. 2

    2024

  26. [26]

    Simple and Scalable Predictive Uncertainty Es- timation using Deep Ensembles

    Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and Scalable Predictive Uncertainty Es- timation using Deep Ensembles. InNeural Information Pro- cessing Systems (NeurIPS), 2017. 2, 6, 7

    2017

  27. [27]

    MIT Press, 2006

    Yann Lecun, Sumit Chopra, Raia Hadsell, Marc Aurelio Ranzato, and Fu Jie Huang.A tutorial on energy-based learning. MIT Press, 2006. 3

    2006

  28. [28]

    A simple unified framework for detecting out-of-distribution samples and adversarial attacks

    Kimin Lee, Kibok Lee, Honglak Lee, and Jinwoo Shin. A simple unified framework for detecting out-of-distribution samples and adversarial attacks. InNeural Information Pro- cessing Systems (NeurIPS), 2018. 1, 2

    2018

  29. [29]

    Open-set semantic segmenta- tion for point clouds via adversarial prototype framework

    Jianan Li and Qiulei Dong. Open-set semantic segmenta- tion for point clouds via adversarial prototype framework. In 2023 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 9425–9434, 2023. 2, 6

    2023

  30. [30]

    Con- trastive class anchor learning for open set object recognition in driving scenes.Transactions on Machine Learning Re- search, 2024

    Zizhao Li, Kourosh Khoshelham, and Joseph West. Con- trastive class anchor learning for open set object recognition in driving scenes.Transactions on Machine Learning Re- search, 2024. 2

    2024

  31. [31]

    GMMSeg: Gaussian Mixture based Generative Semantic Segmentation Models

    Chen Liang, Wenguan Wang, Jiaxu Miao, and Yi Yang. GMMSeg: Gaussian Mixture based Generative Semantic Segmentation Models. InNeural Information Processing Systems (NeurIPS), 2022. 2

    2022

  32. [32]

    Shiyu Liang, Yixuan Li, and R. Srikant. Enhancing the re- liability of out-of-distribution image detection in neural net- works. InInternational Conference on Learning Represen- tations, 2018. 2

    2018

  33. [33]

    Microsoft COCO: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft COCO: Common objects in context. In European Conference on Computer Vision (ECCV), 2014. 2

    2014

  34. [34]

    Energy-based out-of-distribution detection

    Weitang Liu, Xiaoyun Wang, John Owens, and Yixuan Li. Energy-based out-of-distribution detection. InNeural Infor- mation Processing Systems (NeurIPS), 2020. 1, 2, 3, 4, 8

    2020

  35. [35]

    Reid, and Gustavo Carneiro

    Yuyuan Liu, Choubo Ding, Yu Tian, Guansong Pang, Vasileios Belagiannis, Ian D. Reid, and Gustavo Carneiro. Residual pattern learning for pixel-wise out-of-distribution detection in semantic segmentation. InICCV, pages 1151– 1161, 2023. 1, 2, 3, 5

    2023

  36. [36]

    Mask-Based Panoptic LiDAR Segmentation for Autonomous Driving

    Rodrigo Marcuzzi, Lucas Nunes, Louis Wiesmann, Jens Behley, and Cyrill Stachniss. Mask-Based Panoptic LiDAR Segmentation for Autonomous Driving. InIEEE Robotics And Automation Letters (RAL), 2023. 7, 2

    2023

  37. [37]

    A likeli- hood ratio-based approach to segmenting unknown objects

    Nazir Nayal, Youssef Shoeb, and Fatma G ¨uney. A likeli- hood ratio-based approach to segmenting unknown objects. International Journal of Computer Vision, 2025. 1, 2, 6, 7

    2025

  38. [38]

    Henriques, and Fatma G¨uney

    Nazir Nayal, Mısra Yavuz, Jo ˜ao F. Henriques, and Fatma G¨uney. RbA: Segmenting Unknown Regions Rejected by All. InInternational Conference on Computer Vision (ICCV), 2023. 1, 2, 6

    2023

  39. [39]

    Spotting the Unexpected (STU): A 3D LiDAR Dataset for Anomaly Seg- mentation 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 Seg- mentation in Autonomous Driving. In”Conference on Com- puter Vision and Pattern Recognition (CVPR)”, 2025. 1, 2, 5, 6, 7, 8

    2025

  40. [40]

    Qi, Li Yi, Hao Su, and Leonidas J

    Charles R. Qi, Li Yi, Hao Su, and Leonidas J. Guibas. Point- net++: Deep hierarchical feature learning on point sets in a metric space, 2017. 4

    2017

  41. [41]

    Unmasking Anomalies in Road-Scene Segmentation

    Shyam Nandan Rai, Fabio Cermelli, Dario Fontanel, Carlo Masone, and Barbara Caputo. Unmasking Anomalies in Road-Scene Segmentation. InInternational Conference on Computer Vision (ICCV), 2023. 1, 2

    2023

  42. [42]

    Mask3D: Mask Trans- former for 3D Semantic Instance Segmentation

    Jonas Schult, Francis Engelmann, Alexander Hermans, Or Litany, Siyu Tang, and Bastian Leibe. Mask3D: Mask Trans- former for 3D Semantic Instance Segmentation. InInter- national Conference on Robotics and Automation (ICRA),

  43. [43]

    Dropout: a simple way to prevent neural networks from overfitting

    Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. InNeural Information Processing Systems (NeurIPS), 2014. 1, 6

    2014

  44. [44]

    Pixel-wise Energy-biased Abstention Learning for Anomaly Segmentation on Com- plex Urban Driving Scenes

    Yu Tian, Yuyuan Liu, Guansong Pang, Fengbei Liu, Yuan- hong Chen, and Gustavo Carneiro. Pixel-wise Energy-biased Abstention Learning for Anomaly Segmentation on Com- plex Urban Driving Scenes. InEuropean Conference on Computer Vision (ECCV), 2022. 1, 2, 3, 5

    2022

  45. [45]

    Panoptic-CUDAL Technical Report: Ru- ral Australia Point Cloud Dataset in Rainy Conditions.arXiv preprint arXiv:2503.16378, 2025

    Tzu-Yun Tseng, Alexey Nekrasov, Malcolm Burdorf, Bas- tian Leibe, Julie Stephany Berrio Perez, Mao Shan, and Stewart Worrall. Panoptic-CUDAL Technical Report: Ru- ral Australia Point Cloud Dataset in Rainy Conditions.arXiv preprint arXiv:2503.16378, 2025. 6

    arXiv 2025

  46. [46]

    Lion: learning point-wise abstaining penalty for li- dar outlier detection using diverse synthetic data

    Shaocong Xu, Pengfei Li, Qianpu Sun, Xinyu Liu, Yang Li, Shihui Guo, Zhen Wang, Bo Jiang, Rui Wang, Ke- hua Sheng, Bo Zhang, Li Jiang, Hao Zhao, and Yilun Chen. Lion: learning point-wise abstaining penalty for li- dar outlier detection using diverse synthetic data. InPro- ceedings of the AAAI Conference on Artificial Intelligence, AAAI’25/IAAI’25/EAAI’25....

    2025

  47. [47]

    Generalized out-of-distribution detection: A survey.Inter- national Journal of Computer Vision, pages 1–28, 2024

    Jingkang Yang, Kaiyang Zhou, Yixuan Li, and Ziwei Liu. Generalized out-of-distribution detection: A survey.Inter- national Journal of Computer Vision, pages 1–28, 2024. 1, 2

    2024

  48. [48]

    Mask4Former: Mask Transformer for 4D Panoptic Segmentation

    Kadir Yilmaz, Jonas Schult, Alexey Nekrasov, and Bastian Leibe. Mask4Former: Mask Transformer for 4D Panoptic Segmentation. InInternational Conference on Robotics and Automation (ICRA), 2024. 2, 3, 4, 6

    2024

  49. [49]

    Cylinder3D: An Effective 3D Framework for Driving-scene LiDAR Semantic Segmenta- tion

    Hui Zhou, Xinge Zhu, Xiao Song, Yuexin Ma, Zhe Wang, Hongsheng Li, and Dahua Lin. Cylinder3D: An Effective 3D Framework for Driving-scene LiDAR Semantic Segmenta- tion. InConference on Computer Vision and Pattern Recog- nition (CVPR), 2020. 6 10 Relative Energy Learning for LiDAR Out-of-Distribution Detection Supplementary Material

    2020

  50. [50]

    Additional Qualitative Results We provide additional visualization results in Figure 5 and

  51. [51]

    Although our method yields more apparent false positives, these cases in fact correspond to unlabeled ground-truth points

    In both the driver’s view and the bird’s-eye view, our proposed baseline consistently outperforms the baselines. Although our method yields more apparent false positives, these cases in fact correspond to unlabeled ground-truth points. Such detections are still valuable and essential for autonomous driving, particularly when objects are unla- beled or mis...