Pith. sign in

REVIEW 4 major objections 4 minor 54 references

SHTOcc: Effective 3D Occupancy Prediction with Sparse Head and Tail Voxels

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

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

Pith's one-line read SHTOcc claims a sparse head-tail voxel construction improves 3D occupancy accuracy while cutting GPU memory and inference time.

desk verdict Useful plug-and-play idea buried under unsupported headline numbers; the accuracy claim doesn't survive contact with the paper's own tables. read the letter →

arxiv 2505.22461 v2 pith:XXSIK2H6 submitted 2025-05-28 cs.CV

Qiucheng Yu , Yuan Xie , Xin Tan This is my paper
classification cs.CV
keywords 3Doccupancypredictionlong-taildistributionsparsevoxelselectionattention-guideddecoupledlearninglabelsmoothingsemanticscenecompletionautonomousdriving
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

SHTOcc argues that state-of-the-art 3D occupancy predictors waste computation on unimportant voxels and are systematically biased toward frequent classes. The paper proposes a plug-in module that keeps only two sparse sets of voxels — head voxels chosen by the model's own attention and tail voxels chosen by predicted tail-class probability — and a two-phase decoding scheme that retrains the classifier with label smoothing. The authors report that the module improves mIoU on every baseline they test, including SparseOcc, Symphonize, COTR, and OccFormer, while cutting training memory and inference latency; the largest stated savings are 42.2% memory and 58.6% inference time. The central claim is that sparsity should follow model-judged importance and class rarity, not merely voxel emptiness.

What carries the argument

The load-bearing object is the Sparse Head-Tail-Voxel Construction module, which turns a dense encoded voxel volume into two sparse sets of equal size. Head voxels come from a top-k selection over normalized attention weights, taking the argmax over queries for each key voxel and aggregating across attention heads; tail voxels come from a robust top-k selection over coarse tail-class probabilities produced by a single-layer MLP, with a rule that re-samples the second-best tail class when needed. Only these selected voxels pass through the 3D decoder layers, which is where the memory and speed gains come from. The accuracy gains come from the tail-voxel loss, which forces supervision on rare classes, and from the Decouple Decoder, a two-phase schedule that separates representation learning from classifier retraining and adds label smoothing to reduce head-class overconfidence.

What would settle it

Retrain a baseline with SHTOcc but keep the same tail-voxel selection and losses while drawing the head voxels randomly at the same budget; if mIoU does not drop, the attention-based head selection is not carrying the accuracy gain.

Watch

Extended reading notes

Core claim

The paper's central discovery is that 3D occupancy voxels have two exploitable patterns: a severe inter-class long tail and a highly redundant geometric layout, and both can be handled by the same sparse-selection mechanism. SHTOcc constructs a sparse working set by selecting top-k head voxels from averaged multi-head attention scores in the 3D decoder — the model's own focus — and an equal number of tail voxels by sorting coarse tail-class probabilities from a single-layer MLP. These head and tail voxels go through the decoder's attention layers, are blended back into the full voxel volume, and are supervised by an extra tail-voxel cross-entropy loss against ground-truth tail voxels. A second training phase freezes the learned representation and retrains only the segmentation head with label smoothing, which the paper argues stops the model from being overconfident about head classes. The result, as the paper reports it, is higher mIoU on multiple baselines at lower memory and latency cost.

Load-bearing premise

The method assumes that attention weights in the 3D decoder and coarse tail-class probabilities from a single MLP are reliable proxies for which voxels matter, so that deleting all other voxels does not discard information the model needs.

Editorial extensions

If this is right

  • SHTOcc is designed as a plug-and-play module and is reported to improve mIoU when inserted into SparseOcc, Symphonize, COTR, and OccFormer on four benchmarks.
  • Because only a small selected set of voxels is refined, the largest reported savings are 42.2% lower training memory and 58.6% faster inference compared with the relevant baselines.
  • The equal-size head/tail construction acts as a geometry-preserving voxel resampling scheme, giving 3D occupancy a long-tail remedy that avoids the spatial disorder of naive 2D-style resampling.
  • The decoupled decoder with label smoothing contributes a further accuracy gain on top of the sparse construction, so the two mechanisms are complementary rather than redundant.

Reading between the lines

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

  • A natural stress test would be to replace the attention-based head selection with random selection of the same size; if mIoU holds, the gains come from the tail loss and decoupled training rather than from the attention proxy.
  • The same importance-proxy idea could be tried with other ranking signals, such as gradient magnitude, prediction entropy, or distance to the decision boundary, to see whether the attention weights are the best choice for choosing head voxels.
  • The reported gains on rare classes suggest SHTOcc could serve as a data-selection rule for other voxel-level 3D tasks, such as panoptic occupancy or scene flow, where long-tailed semantics and geometric redundancy coexist.
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

4 major / 4 minor

Summary. The paper proposes SHTOcc, a plug-and-play module for 3D occupancy prediction that constructs sparse sets of 'head' voxels (selected by attention weights) and 'tail' voxels (selected by coarse predicted tail-class probabilities), applies an auxiliary tail-voxel loss, and then retrains only the segmentation head with label smoothing in a decoupled-learning phase. The method is integrated into several existing backbones and evaluated on SemanticKITTI semantic scene completion, nuScenes-Occupancy, Occ3D-nuScenes, and nuScenes LiDAR segmentation. The authors claim that SHTOcc reduces GPU memory by 42.2%, increases inference speed by 58.6%, and improves accuracy by about 7%.

Significance. If the reported gains were fully supported, SHTOcc would be a valuable plug-and-play contribution: it addresses a real problem (the long-tail distribution and geometric redundancy in 3D occupancy voxels), ships code, and reports experiments across multiple datasets and baselines. The idea of letting the model select important voxels via attention and of coupling sparse voxel construction with decoupled training is reasonable and worth exploring. However, the headline quantitative claims in the abstract and introduction are not supported by the table entries, and several efficiency numbers come from different configurations, so the significance as stated is currently overstated.

major comments (4)
  1. [Abstract and Introduction; Tables 1–4]
  2. [Section 4.2, Table 1]
  3. [Section 5, Table 5 vs. Table 1]
  4. [Section 3.3 and Table 5]
minor comments (4)
  1. [Title and abstract]
  2. [Section 4.1]
  3. [Section 3.3, Eq. (2)]
  4. [General]

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SHTOcc's claims are benchmark-driven and its selection mechanisms are not definitionally equivalent to their outputs.

full rationale

The paper's contribution is an architecture and training scheme evaluated against external benchmarks; no step in the derivation chain reduces to its own inputs. The tail-voxel selection uses the model's own coarse probabilities to choose which voxels to supervise, but the supervision is ground-truth labels via cross-entropy (Eq. 3), so the selected set is not the predicted output and no equation is equivalent to itself by construction. The head-voxel selection uses attention scores, again an internal signal, but the final mIoU numbers come from external datasets (SemanticKITTI, nuScenes-Occupancy, Occ3D, nuScenes). The only self-citation is COTR [11] as a baseline in Table 3; it is not used to justify the method's design or to forbid alternatives, so it is not load-bearing. The abstract's 'about 7%' accuracy claim is not supported by the tables, but that is a correctness or reporting issue, not circularity. Inconsistent memory numbers between Table 5 and Table 1 likewise concern experimental consistency, not circular derivation. No fitted parameter is renamed as a prediction, and no prior work by the same authors is invoked as an external theorem to make the method's choice forced.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

SHTOcc is an empirical architecture paper; its central claims rest on benchmark experiments. The paper does not report the values of its key hyperparameters (K, epsilon), and its selection mechanism relies on two unvalidated proxy assumptions: attention and coarse tail probabilities indicate voxel importance. No new physical entities are introduced.

free parameters (2)
  • Top-k voxel count K (head and tail selection size)
    The sparsity and balance of the selected voxel set are controlled by K, but no value is reported in the paper. It directly affects the efficiency-accuracy tradeoff.
  • Label smoothing factor epsilon_c
    Used in Eq. (5) in the decoupled decoder. The paper says a small factor works best but does not give the value or a sensitivity study.
assumptions (3)
  • domain assumption Attention weights in the decoder are a valid proxy for voxel importance.
    Head voxel selection in Sec 3.3 keeps top-k voxels by attention; no evidence is given that top-attention voxels are the most informative for occupancy labels.
  • domain assumption Coarse tail-class probabilities from a single-layer MLP reliably identify tail voxels.
    Tail voxel selection in Sec 3.3 sorts voxels by predicted tail probability and refines them with tail loss; this assumes the coarse predictions are accurate enough for mining.
  • domain assumption The observed class-frequency imbalance is the cause of the occupancy model's poor tail-class performance.
    The long-tail motivation in Sec 1 is asserted from dataset statistics, but no correlation or causal analysis links class frequency to the mIoU gaps.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SHTOcc: Effective 3D Occupancy Prediction with Sparse Head and Tail Voxels." pith.science (2026). https://pith.science/paper/XXSIK2H6

@misc{pith2026250522461,
  author       = {Pith},
  title        = {Pith review of: SHTOcc: Effective 3D Occupancy Prediction with Sparse Head and Tail Voxels},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XXSIK2H6}},
  note         = {Machine review of arXiv:2505.22461}
}
read the original abstract

3D occupancy prediction has attracted much attention in the field of autonomous driving due to its powerful geometric perception and object recognition capabilities. However, existing methods have not explored the most essential distribution patterns of voxels, resulting in unsatisfactory results. This paper first explores the inter-class distribution and geometric distribution of voxels, thereby solving the long-tail problem caused by the inter-class distribution and the poor performance caused by the geometric distribution. Specifically, this paper proposes SHTOcc (Sparse Head-Tail Occupancy), which uses sparse head-tail voxel construction to accurately identify and balance key voxels in the head and tail classes, while using decoupled learning to reduce the model's bias towards the dominant (head) category and enhance the focus on the tail class. Experiments show that significant improvements have been made on multiple baselines: SHTOcc reduces GPU memory usage by 42.2%, increases inference speed by 58.6%, and improves accuracy by about 7%, verifying its effectiveness and efficiency. The code is available at https://github.com/ge95net/SHTOcc

Figures

Figures reproduced from arXiv: 2505.22461 by the authors.

Figure 1
Figure 1. Comparisons of the SHTOcc of various 3D semantic scene completion methods on the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of SHTOcc . The images featured are initially extracted by the image backbone [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Visualization of SparseOcc [22] and SHTOcc. The figure demonstrates that SHTOcc can [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 27 canonical work pages

  1. [1]

    Se- mantickitti: A dataset for semantic scene understanding of lidar sequences,

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stachniss, and J. Gall, “Se- mantickitti: A dataset for semantic scene understanding of lidar sequences,” inProceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 9297–9307

    work page 2019

  2. [2]

    Openoccupancy: A large scale benchmark for surrounding semantic occupancy perception,

    X. Wang, Z. Zhu, W. Xu, Y . Zhang, Y . Wei, X. Chi, Y . Ye, D. Du, J. Lu, and X. Wang, “Openoccupancy: A large scale benchmark for surrounding semantic occupancy perception,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 17 850–17 859

    2023

  3. [3]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 11 621–11 631

    2020

  4. [4]

    Inversematrixvt3d: An efficient projection matrix-based approach for 3d occupancy prediction,

    Z. Ming, J. S. Berrio, M. Shan, and S. Worrall, “Inversematrixvt3d: An efficient projection matrix-based approach for 3d occupancy prediction,” in2024 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 9565–9572

    work page 2024

  5. [5]

    Surroundocc: Multi-camera 3d occupancy prediction for autonomous driving,

    Y . Wei, L. Zhao, W. Zheng, Z. Zhu, J. Zhou, and J. Lu, “Surroundocc: Multi-camera 3d occupancy prediction for autonomous driving,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 21 729–21 740

    work page 2023

  6. [6]

    Occformer: Dual-path transformer for vision-based 3d semantic occupancy prediction,

    Y . Zhang, Z. Zhu, and D. Du, “Occformer: Dual-path transformer for vision-based 3d semantic occupancy prediction,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 9433–9443

    2023

  7. [7]

    Fastocc: Accelerating 3d occupancy prediction by fusing the 2d bird’s-eye view and perspective view,

    J. Hou, X. Li, W. Guan, G. Zhang, D. Feng, Y . Du, X. Xue, and J. Pu, “Fastocc: Accelerating 3d occupancy prediction by fusing the 2d bird’s-eye view and perspective view,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 16 425–16 431

    work page 2024

  8. [8]

    Fast Occupancy Network

    M. Lu, Y . Huang, J. Liu, X. Huang, D. Li, J. Peng, L. Tian, and E. Barsoum, “Fast occupancy network,”arXiv preprint arXiv:2412.07163, 2024

    work page Pith review arXiv 2024

Show all 54 references
  1. [9]

    Mambaocc: Visual state space model for bev-based occupancy prediction with local adaptive reordering,

    Y . Tian, S. Bai, Z. Luo, Y . Wang, Y . Lv, and F.-Y . Wang, “Mambaocc: Visual state space model for bev-based occupancy prediction with local adaptive reordering,”arXiv preprint arXiv:2408.11464, 2024

    2024 arXiv

  2. [10]

    Flashocc: Fast and memory-efficient occupancy prediction via channel-to-height plugin,

    Z. Yu, C. Shu, J. Deng, K. Lu, Z. Liu, J. Yu, D. Yang, H. Li, and Y . Chen, “Flashocc: Fast and memory-efficient occupancy prediction via channel-to-height plugin,”arXiv preprint arXiv:2311.12058, 2023

    2023 arXiv

  3. [11]

    Cotr: Compact occupancy transformer for vision-based 3d occupancy prediction,

    Q. Ma, X. Tan, Y . Qu, L. Ma, Z. Zhang, and Y . Xie, “Cotr: Compact occupancy transformer for vision-based 3d occupancy prediction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 936–19 945

    2024

  4. [12]

    Rethinking temporal fusion with a unified gradient descent view for 3d semantic occupancy prediction,

    D. Chen, H. Zheng, J. Fang, X. Dong, X. Li, W. Liao, T. He, P. Peng, and J. Shen, “Rethinking temporal fusion with a unified gradient descent view for 3d semantic occupancy prediction,” arXiv preprint arXiv:2504.12959, 2025

    2025 arXiv

  5. [13]

    Stcocc: Sparse spatial-temporal cascade renovation for 3d occupancy and scene flow prediction,

    Z. Liao, P. Wei, S. Chen, H. Wang, and Z. Ren, “Stcocc: Sparse spatial-temporal cascade renovation for 3d occupancy and scene flow prediction,”arXiv preprint arXiv:2504.19749, 2025

    2025 arXiv

  6. [14]

    Tri-perspective view for vision-based 3d semantic occupancy prediction,

    Y . Huang, W. Zheng, Y . Zhang, J. Zhou, and J. Lu, “Tri-perspective view for vision-based 3d semantic occupancy prediction,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 9223–9232

    2023

  7. [15]

    Lightweight spatial embedding for vision-based 3d occupancy prediction,

    J. Zhang, Y . Zhang, Q. Liu, and Y . Wang, “Lightweight spatial embedding for vision-based 3d occupancy prediction,”arXiv preprint arXiv:2412.05976, 2024

    2024 arXiv

  8. [16]

    Et-former: Efficient triplane deformable attention for 3d semantic scene completion from monocular camera,

    J. Liang, H. Yin, X. Qi, J. J. Park, M. Sun, R. Madhivanan, and D. Manocha, “Et-former: Efficient triplane deformable attention for 3d semantic scene completion from monocular camera,”arXiv preprint arXiv:2410.11019, 2024

    2024 arXiv

  9. [17]

    Gaussianformer: Scene as gaussians for vision-based 3d semantic occupancy prediction,

    Y . Huang, W. Zheng, Y . Zhang, J. Zhou, and J. Lu, “Gaussianformer: Scene as gaussians for vision-based 3d semantic occupancy prediction,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 376–393. 10

    2024

  10. [18]

    Gausstr: Foundation model-aligned gaussian transformer for self-supervised 3d spatial understanding,

    H. Jiang, L. Liu, T. Cheng, X. Wang, T. Lin, Z. Su, W. Liu, and X. Wang, “Gausstr: Foundation model-aligned gaussian transformer for self-supervised 3d spatial understanding,”arXiv preprint arXiv:2412.13193, 2024

    2024 arXiv

  11. [19]

    Probabilistic gaussian superposition for efficient 3d occupancy prediction,

    Y . Huang, A. Thammatadatrakoon, W. Zheng, Y . Zhang, D. Du, and J. Lu, “Probabilistic gaussian superposition for efficient 3d occupancy prediction,”arXiv preprint arXiv:2412.04384, 2024

    2024 arXiv

  12. [20]

    Opus: occupancy prediction using a sparse set,

    J. Wang, Z. Liu, Q. Meng, L. Yan, K. Wang, J. Yang, W. Liu, Q. Hou, and M.-M. Cheng, “Opus: occupancy prediction using a sparse set,”arXiv preprint arXiv:2409.09350, 2024

    2024 arXiv

  13. [21]

    Fully sparse 3d occupancy prediction,

    H. Liu, Y . Chen, H. Wang, Z. Yang, T. Li, J. Zeng, L. Chen, H. Li, and L. Wang, “Fully sparse 3d occupancy prediction,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 54–71

    2024

  14. [22]

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

    P. Tang, Z. Wang, G. Wang, J. Zheng, X. Ren, B. Feng, and C. Ma, “Sparseocc: Rethinking sparse latent representation for vision-based semantic occupancy prediction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 15 035– 15 044

    2024

  15. [23]

    3d occupancy prediction with low-resolution queries via prototype-aware view transformation,

    G. Oh, S. Kim, H. Ko, H.-g. Chi, J. Kim, D. Lee, D. Ji, S. Choi, S. Jang, and S. Kim, “3d occupancy prediction with low-resolution queries via prototype-aware view transformation,” arXiv preprint arXiv:2503.15185, 2025

    2025 arXiv

  16. [24]

    Long-tailed recognition via weight balancing,

    S. Alshammari, Y .-X. Wang, D. Ramanan, and S. Kong, “Long-tailed recognition via weight balancing,” inProceedings of the IEEE/CVF conference on computer vision and pattern recog- nition, 2022, pp. 6897–6907

    2022

  17. [25]

    The devil is in classification: A simple framework for long-tail instance segmentation,

    T. Wang, Y . Li, B. Kang, J. Li, J. Liew, S. Tang, S. Hoi, and J. Feng, “The devil is in classification: A simple framework for long-tail instance segmentation,” inComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XIV 16. S...

    2020

  18. [26]

    Decoupling representation and classifier for long-tailed recognition,

    B. Kang, S. Xie, M. Rohrbach, Z. Yan, A. Gordo, J. Feng, and Y . Kalantidis, “Decoupling representation and classifier for long-tailed recognition,”arXiv preprint arXiv:1910.09217, 2019

    1910 arXiv

  19. [27]

    Occ3d: A large-scale 3d occupancy prediction benchmark for autonomous driving,

    X. Tian, T. Jiang, L. Yun, Y . Mao, H. Yang, Y . Wang, Y . Wang, and H. Zhao, “Occ3d: A large-scale 3d occupancy prediction benchmark for autonomous driving,”Advances in Neural Information Processing Systems, vol. 36, pp. 64 318–64 330, 2023

    2023

  20. [28]

    Occupancy networks: Learning 3d reconstruction in function space,

    L. Mescheder, M. Oechsle, M. Niemeyer, S. Nowozin, and A. Geiger, “Occupancy networks: Learning 3d reconstruction in function space,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 4460–4470

    2019

  21. [29]

    Convolutional occupancy networks,

    S. Peng, M. Niemeyer, L. Mescheder, M. Pollefeys, and A. Geiger, “Convolutional occupancy networks,” inComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part III 16. Springer, 2020, pp. 523–540

    2020

  22. [30]

    Semantic scene completion from a single depth image,

    S. Song, F. Yu, A. Zeng, A. X. Chang, M. Savva, and T. Funkhouser, “Semantic scene completion from a single depth image,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1746–1754

    2017

  23. [31]

    Rgbd based dimensional decomposition residual network for 3d semantic scene completion,

    J. Li, Y . Liu, D. Gong, Q. Shi, X. Yuan, C. Zhao, and I. Reid, “Rgbd based dimensional decomposition residual network for 3d semantic scene completion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 7693–7702

    2019

  24. [32]

    Cascaded context pyramid for full-resolution 3d semantic scene completion,

    P. Zhang, W. Liu, Y . Lei, H. Lu, and X. Yang, “Cascaded context pyramid for full-resolution 3d semantic scene completion,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 7801–7810

    2019

  25. [33]

    View-volume network for semantic scene completion from a single depth image,

    Y .-X. Guo and X. Tong, “View-volume network for semantic scene completion from a single depth image,”arXiv preprint arXiv:1806.05361, 2018

    2018 arXiv

  26. [34]

    Anisotropic convolutional networks for 3d semantic scene completion,

    J. Li, K. Han, P. Wang, Y . Liu, and X. Yuan, “Anisotropic convolutional networks for 3d semantic scene completion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 3351–3359

    2020

  27. [35]

    See and think: Disentangling semantic scene completion,

    S. Liu, Y . Hu, Y . Zeng, Q. Tang, B. Jin, Y . Han, and X. Li, “See and think: Disentangling semantic scene completion,”Advances in Neural Information Processing Systems, vol. 31, 2018. 11

    2018

  28. [36]

    Monoscene: Monocular 3d semantic scene completion,

    A.-Q. Cao and R. De Charette, “Monoscene: Monocular 3d semantic scene completion,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 3991–4001

    2022

  29. [37]

    V oxformer: Sparse voxel transformer for camera-based 3d semantic scene completion,

    Y . Li, Z. Yu, C. Choy, C. Xiao, J. M. Alvarez, S. Fidler, C. Feng, and A. Anandkumar, “V oxformer: Sparse voxel transformer for camera-based 3d semantic scene completion,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 9087–9098

    2023

  30. [38]

    Occdepth: A depth-aware method for 3d semantic scene completion,

    R. Miao, W. Liu, M. Chen, Z. Gong, W. Xu, C. Hu, and S. Zhou, “Occdepth: A depth-aware method for 3d semantic scene completion,”arXiv preprint arXiv:2302.13540, 2023

    2023 arXiv

  31. [39]

    Ndc-scene: Boost monocular 3d semantic scene completion in normalized device coordinates space,

    J. Yao, C. Li, K. Sun, Y . Cai, H. Li, W. Ouyang, and H. Li, “Ndc-scene: Boost monocular 3d semantic scene completion in normalized device coordinates space,” in2023 IEEE/CVF International Conference on Computer Vision (ICCV). IEEE Computer Society, 2023, pp. 9421–9431

    2023

  32. [40]

    Scene as occupancy,

    W. Tong, C. Sima, T. Wang, L. Chen, S. Wu, H. Deng, Y . Gu, L. Lu, P. Luo, D. Linet al., “Scene as occupancy,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 8406–8415

    2023

  33. [41]

    Rethinking classifier re-training in long-tailed recognition: Label over-smooth can balance,

    S. Sun, H. Lu, J. Li, Y . Xie, T. Li, X. Yang, L. Zhang, and J. Yan, “Rethinking classifier re-training in long-tailed recognition: Label over-smooth can balance,” inThe Thirteenth International Conference on Learning Representations

  34. [42]

    Not all voxels are equal: Hardness- aware semantic scene completion with self-distillation,

    S. Wang, J. Yu, W. Li, W. Liu, X. Liu, J. Chen, and J. Zhu, “Not all voxels are equal: Hardness- aware semantic scene completion with self-distillation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 14 792–14 801

    2024

  35. [43]

    Symphonize 3d semantic scene completion with contextual instance queries,

    H. Jiang, T. Cheng, N. Gao, H. Zhang, T. Lin, W. Liu, and X. Wang, “Symphonize 3d semantic scene completion with contextual instance queries,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 258–20 267

    2024

  36. [44]

    Lmscnet: Lightweight multiscale 3d semantic completion,

    L. Roldao, R. De Charette, and A. Verroust-Blondet, “Lmscnet: Lightweight multiscale 3d semantic completion,” in2020 International Conference on 3D Vision (3DV). IEEE, 2020, pp. 111–119

    2020

  37. [45]

    Sparse single sweep lidar point cloud segmentation via learning contextual shape priors from scene completion,

    X. Yan, J. Gao, J. Li, R. Zhang, Z. Li, R. Huang, and S. Cui, “Sparse single sweep lidar point cloud segmentation via learning contextual shape priors from scene completion,” inProceedings of the AAAI conference on artificial intelligence, vol. 35, no. 4, 2021, pp. 3101–3109

    2021

  38. [46]

    Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers,

    Z. Li, W. Wang, H. Li, E. Xie, C. Sima, T. Lu, Y . Qiao, and J. Dai, “Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers,” arXiv preprint arXiv:2203.17270, 2022

    2022 arXiv

  39. [47]

    Fb-bev: Bev representation from forward-backward view transformations,

    Z. Li, Z. Yu, W. Wang, A. Anandkumar, T. Lu, and J. M. Alvarez, “Fb-bev: Bev representation from forward-backward view transformations,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 6919–6928

    2023

  40. [48]

    Panoocc: Unified occupancy representation for camera-based 3d panoptic segmentation,

    Y . Wang, Y . Chen, X. Liao, L. Fan, and Z. Zhang, “Panoocc: Unified occupancy representation for camera-based 3d panoptic segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 17 158–17 168

    2024

  41. [49]

    Protoocc: Accurate, efficient 3d occupancy prediction using dual branch encoder-prototype query decoder,

    J. Kim, C. Kang, D. Lee, S. Choi, and J. W. Choi, “Protoocc: Accurate, efficient 3d occupancy prediction using dual branch encoder-prototype query decoder,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 4, 2025, pp. 4284–4292

    2025

  42. [50]

    Rangenet++: Fast and accurate lidar semantic segmentation,

    A. Milioto, I. Vizzo, J. Behley, and C. Stachniss, “Rangenet++: Fast and accurate lidar semantic segmentation,” in2019 IEEE/RSJ international conference on intelligent robots and systems (IROS). IEEE, 2019, pp. 4213–4220

    2019

  43. [51]

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

    Y . Zhang, Z. Zhou, P. David, X. Yue, Z. Xi, B. Gong, and H. Foroosh, “Polarnet: An improved grid representation for online lidar point clouds semantic segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 9601–9610

    2020

  44. [52]

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

    T. Cortinhal, G. Tzelepis, and E. Erdal Aksoy, “Salsanext: Fast, uncertainty-aware semantic segmentation of lidar point clouds,” inAdvances in Visual Computing: 15th International Symposium, ISVC 2020, San Diego, CA, USA, October 5–7, 2020, Proceedings, Part II 15. Springer, 2...

    2020

  45. [53]

    Cylindrical and asymmetrical 3d convolution networks for lidar segmentation,

    X. Zhu, H. Zhou, T. Wang, F. Hong, Y . Ma, W. Li, H. Li, and D. Lin, “Cylindrical and asymmetrical 3d convolution networks for lidar segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 9939–9948

    2021

  46. [54]

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

    J. Xu, R. Zhang, J. Dou, Y . Zhu, J. Sun, and S. Pu, “Rpvnet: A deep and efficient range-point- voxel fusion network for lidar point cloud segmentation,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 16 024–16 033. 13

    2021

Pith tools

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