Pith. sign in

REVIEW 3 major objections 5 minor 34 references

Can we Trust Unreliable Voxels? Exploring 3D Semantic Occupancy Prediction under Label Noise

T0 review · 3 major / 5 minor · reviewed 2026-07-15 · grok-4.5

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

Pith's one-line read 3D occupancy models can keep usable geometry and semantics even when 90% of voxel labels are wrong, if candidates are built from model memory and feature prototypes rather than hard noisy labels.

desk verdict Solid first stress-test for occupancy under label noise: OccNL shows 2D robust methods collapse on sparse voxels, and DPR-Occ keeps geometry usable at 90% noise even if minority classes stay weak. read the letter →

arxiv 2603.06279 v2 pith:C6IU2A2F submitted 2026-03-06 cs.CV cs.ROeess.IV

Wenxin Li , Kunyu Peng , Di Wen , Junwei Zheng , Jiale Wei , Mengfei Duan , Yuheng Zhang , Rui Fan
show 1 more author
Kailun Yang
This is my paper · ORCID
classification cs.CVcs.ROeess.IV
keywords 3DsemanticoccupancylabelnoisepartiallearningKITTIdynamictrailingprototypeaffinityEMAteacherroboticperception
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

Real robot maps are built from voxel labels that are often wrong: category flips from sparse or misaligned sensors, and ghost trails left by moving cars and people. This paper asks whether autonomous systems can still trust that supervision, and shows that the usual 2D noisy-label tricks fail badly once the data become sparse 3D grids. It releases OccNL, a benchmark that injects both synthetic occupancy-asymmetric flips and real dynamic trailing artifacts into SemanticKITTI, then measures how far models collapse. The authors' answer is DPR-Occ: after a short warm-up that lets the network memorize clean patterns, it builds a small candidate label set for every voxel by uniting the top-K predictions of a slow-moving teacher with the top-K nearest class prototypes in feature space, then trains with partial-label, negative, and not-true distillation losses. On the hardest 90% noise setting the method still reports roughly 35% geometric IoU and 8% mIoU, gains of up to 13.91 IoU and 2.57 mIoU over five adapted 2D baselines that largely lose structure and rare classes.

What carries the argument

DPR-Occ dual-source partial label: for each voxel the candidate set is the union of the Top-K EMA-teacher predictions and the Top-K feature-prototype similarities (Eq. 4), then optimized by partial-label learning, negative learning on the complement, and EMA-guided self-not-true distillation under a decaying K schedule.

What would settle it

On the 90% asymmetric-noise OccNL split, ablate the prototype branch or force a fixed tiny K so ground-truth coverage of sparse dynamic classes falls; if geometric IoU and rare-class mIoU then collapse to the level of the pure 2D baselines, the recovery claim fails.

Watch

Extended reading notes

Core claim

State-of-the-art 2D label-noise methods collapse in sparse 3D occupancy spaces, yet a dual-source partial-label construction that fuses temporal model memory with prototype affinity can preserve both geometric completeness and sparse semantic classes even under 90% occupancy-asymmetric corruption and real dynamic trailing noise.

Load-bearing premise

The dual-source candidate set, after warm-up, still contains the true class often enough for partial-label learning to recover it, even when 90% of labels are flipped and the class distribution is driven toward uniformity.

Editorial extensions

If this is right

  • OccNL becomes a required stress test for any new 3D occupancy network that claims robustness.
  • Training pipelines can deliberately tolerate cheaper, noisier multi-frame labels instead of demanding perfectly cleaned voxels.
  • Sparse dynamic classes (bicyclists, motorcyclists) remain learnable under extreme corruption when candidates are structurally constrained.
  • Safety-critical planners receive maps that keep road and free-space geometry even when annotation quality is catastrophic.
  • Hypothesis-space regularization, not stronger loss reweighting, is the more transferable principle from 2D to sparse 3D.

Reading between the lines

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

  • The same dual-source candidate idea should transfer to other sparse voxel tasks such as LiDAR semantic segmentation under map-projection noise.
  • If the warm-up window is too short for a new backbone, the method may need an automatic early-stop on teacher-student agreement rather than a fixed epoch count.
  • Closed-loop driving simulators could inject OccNL-style trailing noise to quantify how much map corruption a planner can absorb before collision rates rise.
  • Long-tailed open-vocabulary occupancy may benefit from the same prototype-affinity term once language embeddings replace fixed class prototypes.
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 / 5 minor

Summary. The paper introduces OccNL, a benchmark for 3D semantic occupancy prediction under occupancy-asymmetric category-flipping noise and real-world dynamic trailing artifacts on SemanticKITTI, and proposes DPR-Occ, a dual-source partial-label framework. After a warm-up stage that exploits memorization under noisy labels while maintaining an EMA teacher and class prototypes, the method constructs candidate sets by union of Top-K EMA predictions and Top-K feature-prototype similarities (Eq. 4), then optimizes partial-label learning, negative learning on the complementary set, and EMA-guided self-not-true distillation (Eqs. 6–11). Experiments adapt five 2D label-noise methods (AGCE, ANL, JAL, VBL, SNTD) to the voxel setting and report that DPR-Occ preserves geometric IoU and improves mIoU under 50–90% asymmetric noise and three trailing severities (Tables II–III), with ablations on warm-up length, dynamic-K, dual-source fusion, and loss terms (Tables IV–VII).

Significance. If the results hold, the work is a useful first systematic study of voxel-level label noise for semantic occupancy, a practically important and previously underexplored failure mode for safety-critical robotics. OccNL supplies controllable synthetic and real-world-style noise protocols plus a refined clean evaluation set (Eq. 1), and the public code/benchmark commitment is a concrete contribution. The empirical demonstration that several strong 2D robust losses collapse in sparse 3D voxel spaces (Table II, 90% row) is informative for the community. DPR-Occ’s gains in geometric integrity under extreme corruption are the most actionable finding; the dual-source partial-label construction is a reasonable, if not fully validated, design for sparse 3D settings.

major comments (3)
  1. The central 90% claim (abstract; Table II) rests on an unmeasured coverage assumption for the dual-source candidate set PL_v (Eq. 4). The paper never reports hit-rate of PL_v versus true labels Y, nor how often the true class is excluded when both EMA rank and prototype rank drop. At η=90% the label distribution is driven toward uniformity (Fig. 2), which undermines the long-tail structure that prototypes and early EMA memory rely on. Ablations (Tables VI–VII) only show complementary gains at 70%; they do not establish that coverage remains high enough at 90% for PLL/NL/SNTD to recover sparse classes. Without this diagnostic, the claim that the method “prevents geometric and semantic collapse” is only partially supported: geometry is largely preserved, but many minority semantics remain near zero IoU even for DPR-Occ.
  2. Table II, 90% row: the headline “up to 2.57% mIoU and 13.91% IoU” gains are real relative to the adapted baselines, yet absolute mIoU is only 8.23 and several dynamic/rare classes (bicycle, motorcycle, person, bicyclist, motorcyclist, other-ground, traffic-sign) stay at or near 0. The abstract and conclusion language of preserving “sparse dynamic semantics” and providing a “reliable foundation for safety-critical” perception overstates what the per-class numbers show. The manuscript should either qualify these claims or provide additional evidence (e.g., coverage curves, class-wise recovery analysis) that minority semantics are meaningfully retained rather than merely less collapsed than the baselines.
  3. All quantitative results appear to be single-run point estimates with no error bars, seeds, or multi-run statistics (Tables I–VII). Given the free parameters (Ew, Kstart/Kend/γ, τs, EMA momentum) and the known sensitivity of noisy-label methods to initialization and schedule, the reported margins—especially the 2.57 mIoU gap at 90%—cannot be assessed for stability. At least multi-seed means and standard deviations on the primary 70% and 90% settings are needed for the load-bearing comparisons.
minor comments (5)
  1. The empty-voxel flip rate is fixed at 10^{-3}η without sensitivity analysis; a short ablation would clarify whether geometric preservation is an artifact of this design choice.
  2. Fig. 1 and Fig. 4 captions are dense; a clearer legend for noise type (asymmetric vs. trailing) and a consistent color key for classes would improve readability.
  3. Notation for the candidate size switches between K, Ke, Kstart, Kend; a single consistent symbol with a short definition list would help.
  4. Related work on uncertainty-aware occupancy and pseudo-label filtering is cited, but a brief discussion of how OccNL differs from confidence-based filtering (rather than only from 2D LNL) would strengthen positioning.
  5. Implementation details state consistency with ProtoOcc; explicitly listing any ProtoOcc-specific losses that remain in L_base would aid reproducibility before code release.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: empirical method and benchmark gains rest on external baselines, controlled synthetic noise, and a clean refined eval set, not on quantities defined to equal the reported metrics.

full rationale

The paper's central claims are experimental (Tables II–III): DPR-Occ yields higher IoU/mIoU than five adapted external 2D noisy-label losses (AGCE, ANL, JAL, VBL, SNTD) under occupancy-asymmetric and trailing noise on SemanticKITTI. The dual-source candidate set (Eq. 4), dynamic-K schedule (Eq. 5), PLL/NL/SNTD losses (Eqs. 6–9), and total objective (Eq. 11) are constructive training devices; none is algebraically forced to equal the evaluation metrics. Warm-up exploits the known memorization effect (cited externally), prototypes follow the ProtoOcc backbone (different authors), and EMA is a standard momentum average. Evaluation uses a refined clean set (Eq. 1, following SCP-Net) distinct from the noisy training labels ˜Y. No parameter is fitted to a subset and then re-reported as a prediction of a closely related quantity; no uniqueness theorem or ansatz is imported from overlapping authors to forbid alternatives; no known empirical pattern is merely renamed. Minor citation of ProtoOcc for the base architecture and loss does not load-bear the robustness gains. The work is therefore self-contained against external benchmarks and free of the enumerated circularity patterns.

Assumptions & free parameters 5 free parameters · 4 assumptions · 2 invented entities

The central empirical claim rests on standard deep-learning practice (memorization effect, EMA, prototypes), on the authors’ synthetic noise model and trailing construction, and on a handful of hand-chosen schedule parameters. No new physical entities are postulated; the free parameters are training hyper-parameters whose values are reported and ablated.

free parameters (5)
  • warm-up epochs Ew = 12
    Chosen by ablation (Table IV); set to 12. Controls when robust losses activate and how much clean signal is memorized.
  • dynamic-K schedule (Kstart, Kend, γ) = 9, 2, 2
    Linear decay of candidate-set size (Eq. 5); values 9→2 with step 2 selected by ablation (Table V).
  • SNTD temperature τs = 3.0
    Temperature in the not-true distillation KL (Eq. 8–9); fixed at 3.0 without extensive sensitivity reported.
  • EMA momentum dt = 0.999
    Teacher update rate (Eq. 2); set to 0.999 following common practice.
  • empty-voxel flip rate 10^{-3} η = 0.001 * η
    Asymmetric noise model deliberately flips empty voxels far less often than occupied ones to avoid total structure collapse; the constant 10^{-3} is a design choice of the benchmark.
assumptions (4)
  • domain assumption Deep networks first memorize clean patterns before fitting noise (memorization effect).
    Invoked to justify the warm-up stage (Sec. IV-B, citation to Arpit et al.).
  • ad hoc to paper EMA teacher predictions and feature-prototype cosine similarities supply complementary evidence that together cover the true label with high probability.
    Core construction of the dual-source candidate set (Eq. 3–4, Sec. IV-C); not independently proven for 90% noise.
  • domain assumption Refining dynamic voxels by current-frame instance masks yields an unbiased evaluation set.
    Ground-truth refinement (Eq. 1, Sec. III-B) following SCPNet; assumes masks correctly identify true object extent.
  • ad hoc to paper Occupancy-asymmetric flipping (occupied→any other class at rate η; empty→occupied at 10^{-3}η) and multi-frame trailing are representative of real annotation artifacts.
    Defines the OccNL noise protocols (Sec. III-C); the extreme 90% rate is explicitly a stress test beyond typical sensor error.
invented entities (2)
  • OccNL benchmark (occupancy-asymmetric + dynamic trailing noise protocols)
    purpose: Provide the first controllable testbed for 3D occupancy under label noise.
    New evaluation construct; independent evidence will exist once the promised public release is used by others.
  • DPR-Occ dual-source partial-label candidate set
    purpose: Construct reliable supervision by union of EMA Top-K and prototype Top-K.
    Methodological construct internal to the paper; no external measurement of coverage rate is supplied beyond final task metrics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Can we Trust Unreliable Voxels? Exploring 3D Semantic Occupancy Prediction under Label Noise." pith.science (2026). https://pith.science/paper/C6IU2A2F

@misc{pith2026260306279,
  author       = {Pith},
  title        = {Pith review of: Can we Trust Unreliable Voxels? Exploring 3D Semantic Occupancy Prediction under Label Noise},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C6IU2A2F}},
  note         = {Machine review of arXiv:2603.06279}
}
read the original abstract

3D semantic occupancy prediction is a cornerstone of robotic perception, yet real-world voxel annotations are inherently corrupted by structural artifacts and dynamic trailing effects. This raises a critical but underexplored question: can autonomous systems safely rely on such unreliable occupancy supervision? To systematically investigate this issue, we establish OccNL, the first benchmark dedicated to 3D occupancy under occupancy-asymmetric and dynamic trailing noise. Our analysis reveals a fundamental domain gap: state-of-the-art 2D label noise learning strategies collapse catastrophically in sparse 3D voxel spaces, exposing a critical vulnerability in existing paradigms. To address this challenge, we propose DPR-Occ, a principled label-noise-robust framework that constructs reliable supervision through dual-source partial label reasoning. By synergizing temporal model memory with representation-level structural affinity, DPR-Occ dynamically expands and prunes candidate label sets to preserve true semantics while suppressing noise propagation. Extensive experiments on SemanticKITTI demonstrate that DPR-Occ prevents geometric and semantic collapse under extreme corruption. Notably, even at 90% label noise, our method achieves significant performance gains (up to 2.57% mIoU and 13.91% IoU) over existing label noise learning baselines adapted to the 3D occupancy prediction task. By bridging label noise learning and 3D perception, OccNL and DPR-Occ provide a reliable foundation for safety-critical robotic perception in dynamic environments. The benchmark and source code will be made publicly available at https://github.com/mylwx/OccNL.

Figures

Figures reproduced from arXiv: 2603.06279 by the authors.

Figure 1
Figure 1. Comparison of our proposed DPR-Occ with state-of-the-art label-noise learning methods on our OccNL benchmark. The upper part presents examples of semantic occupancy predictions under noisy supervision, including occupancy-asymmetric and real-world dynamic trailing noise. The lower part shows that existing robust learning strategies struggle to alleviate the adverse effects of voxel-level noise, whereas our proposed … view at source ↗
Figure 2
Figure 2. Semantic distribution evolution under voxel-level category￾flipping noise. Due to orders-of-magnitude differences in voxel counts, we employ a logarithmic scale for visualization (empty voxels omitted as they remain constant at 10−3 η flip rate). Increased noise drives the distribution toward uniformity: dominant classes (e.g., vegetation, road) are suppressed, while rare classes (e.g., motorcyclist, person) are art… view at source ↗
Figure 3
Figure 3. The overall framework of our proposed DPR-Occ. Warm-up Stage: The model captures clean patterns via standard training on noisy labels Y˜ . Robust Stage: Guided by dynamic-K scheduling, we construct dual-source partial label sets by fusing Top-K predictions from the EMA teacher and feature-prototype similarities. The network is then optimized using Partial Label Learning (PLL) and Negative Learning (NL), with EMA-gui… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative results under 90% asymmetric noise on OccNL benchmark. Compared to collapsing baselines, DPR-Occ preserves structural integrity and reliable semantics. The final column shows a failure case (yellow box), where DPR-Occ still reconstructs basic road and veget…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 1 linked inside Pith

  1. [1]

    V oxDet: Rethinking 3D semantic occupancy prediction as dense object detection,

    W. Li, Z. Yu, and A. Alahi, “V oxDet: Rethinking 3D semantic occupancy prediction as dense object detection,” inProc. NeurIPS, 2025

    2025

  2. [2]

    UniOcc: A unified benchmark for occupancy fore- casting and prediction in autonomous driving,

    Y . Wanget al., “UniOcc: A unified benchmark for occupancy fore- casting and prediction in autonomous driving,” inProc. ICCV, 2025, pp. 25 560–25 570

    2025

  3. [3]

    SfmOcc: Vision-based 3D semantic occupancy prediction in urban environments,

    R. Marcuzziet al., “SfmOcc: Vision-based 3D semantic occupancy prediction in urban environments,”IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 5074–5081, 2025

    2025

  4. [4]

    VPOcc: Exploiting vanishing point for 3D semantic occupancy prediction,

    J. Kim, J. Lee, U. Shin, J. Oh, and K. Joo, “VPOcc: Exploiting vanishing point for 3D semantic occupancy prediction,” inProc. IROS, 2025, pp. 4307–4314

    2025

  5. [5]

    Quadric- Former: Scene as superquadrics for 3D semantic occupancy predic- tion,

    S. Zuo, W. Zheng, X. Han, L. Yang, Y . Pan, and J. Lu, “Quadric- Former: Scene as superquadrics for 3D semantic occupancy predic- tion,” inProc. NeurIPS, 2025

    2025

  6. [6]

    Online diffusion-based 3D occupancy prediction at the frontier with proba- bilistic map reconciliation,

    A. Reed, L. Achey, B. Crowe, B. Hayes, and C. Heckman, “Online diffusion-based 3D occupancy prediction at the frontier with proba- bilistic map reconciliation,” inProc. ICRA, 2025, pp. 2846–2852

    2025

  7. [7]

    H3O: Hyper-efficient 3D occupancy prediction with heterogeneous supervision,

    Y . Shi, H. Cai, A. Ansari, and F. Porikli, “H3O: Hyper-efficient 3D occupancy prediction with heterogeneous supervision,” inProc. ICRA, 2025, pp. 4869–4876

    2025

  8. [8]

    α- OCC: Uncertainty-aware camera-based 3D semantic scene comple- tion,

    S. Su, N. Chen, C. Lin, F. Juefei-Xu, C. Feng, and F. Miao, “α- OCC: Uncertainty-aware camera-based 3D semantic scene comple- tion,”arXiv preprint arXiv:2406.11021, 2024

    arXiv 2024

Show all 34 references
  1. [9]

    Semantic causality-aware vision-based 3D occupancy prediction,

    D. Chenet al., “Semantic causality-aware vision-based 3D occupancy prediction,” inProc. ICCV, 2025, pp. 24 878–24 888

    2025

  2. [10]

    ProtoOcc: Ac- curate, efficient 3D occupancy prediction using dual branch encoder- prototype query decoder,

    J. Kim, C. Kang, D. Lee, S. Choi, and J. W. Choi, “ProtoOcc: Ac- curate, efficient 3D occupancy prediction using dual branch encoder- prototype query decoder,” inProc. AAAI, 2025, pp. 4284–4292

    2025

  3. [11]

    AutoOcc: Automatic open-ended semantic occupancy annotation via vision-language guided gaussian splatting,

    X. Zhou, J. Wang, Y . Wang, Y . Wei, N. Dong, and M.-H. Yang, “AutoOcc: Automatic open-ended semantic occupancy annotation via vision-language guided gaussian splatting,” inProc. ICCV, 2025, pp. 3367–3377

    2025

  4. [12]

    SemanticKITTI: A dataset for semantic scene understanding of LiDAR sequences,

    J. Behleyet al., “SemanticKITTI: A dataset for semantic scene understanding of LiDAR sequences,” inProc. ICCV, 2019, pp. 9296– 9306

    2019

  5. [13]

    Asymmetric loss functions for noise-tolerant learning: Theory and applications,

    X. Zhou, X. Liu, D. Zhai, J. Jiang, and X. Ji, “Asymmetric loss functions for noise-tolerant learning: Theory and applications,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 7, pp. 8094–8109, 2023

    2023

  6. [14]

    Variation-bounded loss for noise-tolerant learning,

    J. Wanget al., “Variation-bounded loss for noise-tolerant learning,” in Proc. AAAI, 2026

    2026

  7. [15]

    Active negative loss: A robust framework for learning with noisy labels,

    X. Ye, Y . Wu, Y . Wang, X. Li, W. Zhang, and Y . Chen, “Active negative loss: A robust framework for learning with noisy labels,” arXiv preprint arXiv:2412.02373, 2024

    2024

  8. [16]

    Joint asymmetric loss for learning with noisy labels,

    J. Wanget al., “Joint asymmetric loss for learning with noisy labels,” inProc. ICCV, 2025, pp. 1947–1956

    2025

  9. [17]

    Continuous review and timely correction: Enhancing the resistance to noisy labels via self-not-true and class-wise distillation,

    L. Lan, J. Wang, X. Wu, B. Han, and X. Liu, “Continuous review and timely correction: Enhancing the resistance to noisy labels via self-not-true and class-wise distillation,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    2025

  10. [18]

    OCCUQ: Exploring efficient uncertainty quantification for 3D occupancy prediction,

    S. Heidrich, T. Beemelmanns, A. Nekrasov, B. Leibe, and L. Eck- stein, “OCCUQ: Exploring efficient uncertainty quantification for 3D occupancy prediction,” inProc. ICRA, 2025, pp. 1–8

    2025

  11. [19]

    Reliable and calibrated semantic occupancy prediction by hybrid uncertainty learning,

    S. Wanget al., “Reliable and calibrated semantic occupancy prediction by hybrid uncertainty learning,” inProc. IJCAI, 2025, pp. 1973–1981

    2025

  12. [20]

    Particle-based instance-aware semantic occupancy mapping in dynamic environ- ments,

    G. Chen, Z. Wang, W. Dong, and J. Alonso-Mora, “Particle-based instance-aware semantic occupancy mapping in dynamic environ- ments,”IEEE Transactions on Robotics, vol. 41, pp. 1155–1171, 2025

    2025

  13. [21]

    HD-CCSOM: Hierarchical and dense collaborative continuous semantic occupancy mapping through label diffusion,

    Y . Deng, M. Wang, Y . Yang, and Y . Yue, “HD-CCSOM: Hierarchical and dense collaborative continuous semantic occupancy mapping through label diffusion,” inProc. IROS, 2022, pp. 2417–2422

    2022

  14. [22]

    Sparse annotation, dense supervision: Unleashing self-training power for occupancy prediction with 2D labels,

    Z. Liuet al., “Sparse annotation, dense supervision: Unleashing self-training power for occupancy prediction with 2D labels,”IEEE Robotics and Automation Letters, vol. 11, no. 1, pp. 418–425, 2026

    2026

  15. [23]

    Enhancing generalizability via utilization of unlabeled data for occupancy perception,

    R. Liet al., “Enhancing generalizability via utilization of unlabeled data for occupancy perception,” inProc. AAAI, 2025, pp. 4896–4904

    2025

  16. [24]

    Language driven occupancy prediction,

    Z. Yuet al., “Language driven occupancy prediction,” inProc. ICCV, 2025, pp. 7548–7558

    2025

  17. [25]

    nuCraft: Crafting high resolution 3D semantic occupancy for unified 3D scene understanding,

    B. Zhu, Z. Wang, and H. Li, “nuCraft: Crafting high resolution 3D semantic occupancy for unified 3D scene understanding,” inProc. ECCV, 2024, pp. 125–141

    2024

  18. [26]

    Knowledge distillation meets label noise learning: Ambiguity-guided mutual label refinery,

    R. Jiang, Y . Yan, J.-H. Xue, S. Chen, N. Wang, and H. Wang, “Knowledge distillation meets label noise learning: Ambiguity-guided mutual label refinery,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 1, pp. 939–952, 2025

    2025

  19. [27]

    Robust noisy label learning via two-stream sample distillation,

    S. Bai, S. Zhou, Z. Qin, L. Wang, and N. Zheng, “Robust noisy label learning via two-stream sample distillation,”IEEE Transactions on Multimedia, vol. 27, pp. 9072–9084, 2025

    2025

  20. [28]

    Understanding self-distillation in the pres- ence of label noise,

    R. Das and S. Sanghavi, “Understanding self-distillation in the pres- ence of label noise,” inProc. ICML, 2023, pp. 7102–7140

    2023

  21. [29]

    Federated learning with extremely noisy clients via negative distillation,

    Y . Luet al., “Federated learning with extremely noisy clients via negative distillation,” inProc. AAAI, 2024, pp. 14 184–14 192

    2024

  22. [30]

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results,

    A. Tarvainen and H. Valpola, “Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results,” inProc. NeurIPS, 2017, pp. 1195–1204

    2017

  23. [31]

    SparseOcc: Rethinking sparse latent representation for vision-based semantic occupancy prediction,

    P. Tanget al., “SparseOcc: Rethinking sparse latent representation for vision-based semantic occupancy prediction,” inProc. CVPR, 2024, pp. 15 035–15 044

    2024

  24. [32]

    SCPNet: Semantic scene completion on point cloud,

    Z. Xiaet al., “SCPNet: Semantic scene completion on point cloud,” inProc. CVPR, 2023, pp. 17 642–17 651

    2023

  25. [33]

    A closer look at memorization in deep networks,

    D. Arpitet al., “A closer look at memorization in deep networks,” in Proc. ICML, 2017, pp. 233–242

    2017

  26. [34]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProc. CVPR, 2016, pp. 770–778

    2016

Pith tools

Reviewed July 15, 2026 · model on record in the stance chip above.