Pith. sign in

REVIEW 4 major objections 5 minor 33 references

Co-Win: Joint Object Detection and Instance Segmentation in LiDAR Point Clouds via Collaborative Window Processing

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Co-Win claims that predicting LiDAR objects as complete footprint masks, using a windowed linear-attention backbone, outperforms prior box-based and mask-based bird's-eye-view methods.

desk verdict No experiments, and the one quantitative argument contradicts itself—this is a placeholder submission, not a paper. read the letter →

arxiv 2507.19691 v1 pith:E46FQKFJ submitted 2025-07-25 cs.CV cs.RO

Haichuan Li , Tomi Westerlund This is my paper · ORCID
classification cs.CVcs.RO
keywords LiDARpointcloudsbird's-eyeviewperceptioninstancesegmentation3Dobjectdetectionlinearattentionmask-baseddecoderautonomousdrivingBEV
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper is trying to establish that bird's-eye-view perception for autonomous driving should predict each vehicle as a complete footprint mask rather than as a bounding box, and that this mask-based formulation can be made efficient with a windowed linear-attention backbone. The proposed Co-Win pipeline has three components: an Axis-Fusion Network (AFN) that encodes raw point clouds from three orthogonal projections, a Sub-window Parallel Computing Network (SPCN) whose linear attention reduces complexity from $O(N^2)$ to $O(N)$, and a mask-based decoder that outputs class labels and instance masks from learned object queries. The authors claim that this combination yields significant performance gains over prior state-of-the-art methods on KITTI, Waymo Open Dataset, and SemanticKITTI, with qualitatively better footprint completion than prior BEV mask work. If the central claim is right, LiDAR perception can deliver planner-ready object shapes at lower compute than global-attention alternatives.

What carries the argument

The load-bearing mechanism is the Sub-window Parallel Computing Network (SPCN), a windowed transformer backbone whose linear attention is written as $\phi(Q)(\phi(K)^T V)$ with kernel feature map $\phi(x) = \text{ELU}(x) + 1$. This reordering lets the network compute the key-value product first, reducing complexity from $O(N^2 d)$ to $O(N d^2)$, and partitioning the BEV map into non-overlapping sub-windows reduces memory relative to global attention; the paper estimates a 400x complexity reduction for a 200x200 map with window size 10. The mask-based decoder completes the argument by taking features from four SPCN stages and producing instance masks through mask embeddings dotted with pixel features, while the AFN supplies multi-plane geometric features and global tokens that are injected into both the backbone and decoder. What this machinery does is convert a normally box-regression problem into a mask-completion problem without paying global-attention cost.

What would settle it

Train Co-Win with global softmax attention in place of SPCN linear attention under identical settings and compare mask IoU on heavily occluded vehicles in SemanticKITTI; if the linear-attention version drops materially, the paper's claim that linear attention preserves representational power is falsified, and a plot of per-instance mask IoU against occlusion fraction would also test the boundary-completion claim directly.

Watch

Extended reading notes

Core claim

The paper's central claim is that a single feedforward pass over a BEV point-cloud representation can jointly detect and segment instances by predicting binary footprint masks, and that Co-Win's specific architecture—AFN encoding, SPCN windowed linear attention, and query-based mask decoder—makes that mask prediction both more accurate and cheaper than previous box-based or mask-based BEV methods. The authors report that using the ELU+1 feature map in linear attention preserves representational power while lowering attention complexity, and that mask prediction with Hungarian matching over classification, binary cross-entropy, and Dice losses recovers object boundaries, including completing the unobserved portions of occluded vehicles. They evaluate on KITTI, Waymo Open Dataset, and SemanticKITTI and state that Co-Win achieves significant performance gains over existing state-of-the-art methods.

Load-bearing premise

The load-bearing premise is that the SPCN's windowed linear attention with the feature map $\phi(x) = \text{ELU}(x) + 1$ keeps enough spatial detail to predict precise footprint masks while cutting complexity from quadratic to linear; if it loses fine boundary or long-range information, the whole architecture underperforms regardless of the decoder.

Editorial extensions

If this is right

  • LiDAR object detection can be framed as mask prediction instead of box regression without sacrificing efficiency.
  • A windowed linear-attention backbone with the ELU+1 kernel is sufficient for joint detection and instance segmentation in BEV point clouds.
  • A single pass can both localize vehicles and complete their occluded footprint masks, giving downstream planners shape information that boxes cannot provide.
  • The reported complexity reduction supports real-time BEV perception on resource-constrained hardware.
  • The method transfers across KITTI, Waymo Open Dataset, and SemanticKITTI, suggesting the mask-based formulation generalizes across LiDAR benchmarks.

Reading between the lines

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

  • Because the decoder is query-based and mask-based rather than class-specific, the same architecture could be pointed at other BEV map elements—lane markings, curbs, pedestrians—by adding queries and mask annotations, turning Co-Win into a unified panoptic BEV predictor.
  • The paper claims a 400x complexity reduction for a 200x200 map with window size 10; a direct wall-clock comparison of SPCN against global softmax attention at the same feature resolution would tell whether that theoretical reduction survives in practice on embedded hardware.
  • The ELU+1 linear-attention kernel is a natural point of comparison: swapping in other positive kernels and measuring mask IoU would show whether the exact kernel or just linear attention per se is responsible for the reported quality.
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 / 5 minor

Summary. The paper proposes Co-Win, a bird's-eye-view (BEV) LiDAR perception framework for joint object detection and instance segmentation. The architecture has three components: an Axis-Fusion Network (AFN) that encodes raw point clouds into BEV representations, a Sub-window Parallel Computing Network (SPCN) backbone that applies window-based linear attention, and a mask-based query decoder with Hungarian matching. The authors claim significant performance gains over state-of-the-art methods on KITTI, Waymo Open Dataset, and SemanticKITTI, while also reducing computational complexity relative to global attention. The manuscript provides detailed architectural equations and a qualitative figure, but contains no quantitative experimental results; Section 4.3 and Section 4.4 are empty placeholders.

Significance. If the claimed performance were substantiated, Co-Win would be a meaningful contribution to BEV-based LiDAR perception, combining mask-based instance segmentation with windowed linear attention and reporting results on multiple major benchmarks. The system design is coherent and builds on established components such as deformable attention, transformer decoders, and Hungarian matching, and the paper is explicit about its intended computational benefits. However, because no experiments, metrics, ablations, or training details are presented, the central claim of state-of-the-art performance cannot be evaluated; the significance of the work is therefore unsubstantiated.

major comments (4)
  1. [Abstract; §4.3] The central claim of 'significant performance gains over existing state-of-the-art methods' is unsupported. Section 4.3 contains only the sentence 'In ??, ??, and ??, the above evaluation metrics are compared with previous works,' with no tables, metrics, error bars, or comparison results, and Section 4.4 similarly refers to '?? and ??' for ablation results. The paper therefore provides no experimental evidence for its primary contribution, and the claim cannot be verified from the manuscript.
  2. [§3.2] The computational complexity analysis is internally inconsistent. After introducing linear attention with the kernel feature map φ(x)=ELU(x)+1 and claiming O(N d²) complexity, the 'Computational Complexity Analysis' paragraph gives Ω(SPCN)=K·O(M⁴)=O(HW·M²). The latter is the cost of standard softmax attention inside M×M windows, not of the linear-attention formulation; for linear attention, the per-window cost is O(M² d²), and summing over K=HW/M² windows yields O(HW d²). Consequently, the claimed 400× reduction over global attention is not a consequence of the stated linear-attention mechanism, and the theoretical efficiency justification needs to be corrected.
  3. [§4.5] The qualitative comparison in Figure 3 does not support the claim of 'significant improvement': the figure caption asserts this improvement without quantitative backing, and the accompanying text notes objects that 'didn't be recognized by previous method' based on a single visual example. No detection or segmentation metrics are reported, so this result is anecdotal and cannot substantiate the abstract's performance claim.
  4. [§3.1; §3.3] Several components of the proposed architecture are specified at a level that prevents reproducibility. For example, the statistical fusion function S, the feature similarity measure sim(·,·), the token generator T for GGIT, and the exact structure of the 'Sub-window Block' are not defined unambiguously; additionally, the number of layers L, block depths D_l, channel counts C, and loss weights λ are never given, nor is any training schedule or data preprocessing detail provided. As a methods paper, the absence of these specifications is a load-bearing deficiency even beyond the missing experiments.
minor comments (5)
  1. [Throughout] There are multiple unresolved placeholders, including 'Figure. ??' in §3.1, 'Fig. ??' in §3.3, and '??' in §4.3/§4.4; these should be replaced with actual references before submission.
  2. [§4.5] The caption of Figure 3 contains grammatical errors and an unsupported evaluative statement; for example, 'ours shows a significant improvement (e.g. arrows in (b) (c) (d)) didn't be recognized by previous method' is not a complete sentence and should be rewritten.
  3. [§2] In the related work, the description of PointRCNN++ as 'enhanced accuracy through distance bin-based encoding' is not reflected in the cited reference [17]; consider clarifying the contribution or revisiting the citation.
  4. [§3.2; §3.3] The notation for sub-window size M conflicts with the use of M in the multi-scale deformable attention formula in §3.3, where M denotes the number of attention heads; consider renaming one of them to avoid ambiguity.
  5. [§1] The phrase 'multi-modality inherent in environmental understanding' is vague; the method itself is LiDAR-only, so the relevance of 'multi-modality' is not clear.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper's performance claims are unsupported by missing experiments, but no equation or fitted value is invoked as its own output.

full rationale

The claimed derivation chain is purely architectural; no parameter is fitted to a subset of data and then presented as a prediction of that same subset. The AFN, SPCN, and mask decoder are defined by ordinary learned modules trained with classification, mask BCE, and Dice losses, and the reported state-of-the-art gains are asserted without quantitative tables: Section 4.3 reads 'In ??, ??, and ??, the above evaluation metrics are compared with previous works' and Section 4.4 refers to '?? and ?? demonstrate the impact of each component.' Missing experimental evidence is a soundness and completeness defect, not circularity. The SPCN complexity analysis in Section 3.2 is internally inconsistent: it introduces linear attention with phi(x)=ELU(x)+1 and claims O(N) complexity, but then computes the per-window cost as K times O(M^4)=O(HW*M^2), which is the cost of standard softmax attention inside M by M windows rather than the stated linear attention. This invalidates the 400x efficiency justification, but it is a non-sequitur, not a self-referential reduction: the O(N) claim does not assume the conclusion it is supposed to establish. There are also no load-bearing self-citations: MaskBEV [7] and [30] are external prior works, and no uniqueness theorem or prior result by the present authors is imported to force the architecture. Therefore no circular step that meets the quoted-reduction bar can be identified.

Assumptions & free parameters 7 free parameters · 5 assumptions · 3 invented entities

The architecture introduces many hand-chosen hyperparameters and unstated implementation details; no values are given. The behavior relies on standard ML training assumptions plus a few paper-specific design choices such as trigonometric feature modulation and learned GGIT tokens. The datasets and their ground-truth masks are assumed to be correct.

free parameters (7)
  • Voxel size (vx, vy, vz) in AFN
    Determines sparsity and resolution of the BEV representation; no value is reported.
  • Sub-window size M in SPCN
    Controls the attention locality and complexity; the example H=W=200, M=10 is illustrative, but the actual value used in experiments is not stated.
  • Number of SPCN stages and block depths (L, Dl, channels C)
    Defines the feature hierarchy; no configuration is provided.
  • Height encoding frequency bands L=6 = 6
    Chosen by hand for multi-frequency positional encoding in AFN.
  • Loss weights lambda_cls, lambda_mask, lambda_dice
    Trade-off between classification, mask BCE, and Dice loss; values not given.
  • Inference confidence threshold tau and NMS IoU threshold
    Post-processing hyperparameters; no values reported.
  • Learnable angular scaling factor alpha in multi-perspective fusion
    Learned parameter modulating trigonometric features; specific initial value and range not given.
assumptions (5)
  • domain assumption BEV projection from sparse LiDAR points to a dense HxWxC grid preserves sufficient geometric information for mask-level instance prediction.
    Invoked in Section 3.1 'BEV Representation'; the entire downstream pipeline operates on the projected grid. If projection loses shape detail (e.g., for thin or occluded objects), mask prediction cannot recover it.
  • domain assumption Ground-truth masks in KITTI, Waymo, and SemanticKITTI are complete and correct footprints for all vehicles with at least one visible LiDAR point.
    Used in Section 4.3 where the paper states all visible vehicle instances are evaluated; this assumes the annotations are reliable for heavily occluded vehicles.
  • ad hoc to paper The ELU+1 kernel feature map yields a valid linear attention approximation without significant information loss.
    Section 3.2 states 'This formulation ensures positive values and enables reordering matrix multiplications'; no experiments substantiate the accuracy impact of this approximation.
  • ad hoc to paper The specific trigonometric modulation (1 + mean(sin, cos)) and exponential height weighting improve directional awareness.
    Introduced in Section 3.1 'Multi-Perspective Feature Extraction' and 'Advanced Height Encoding'; the functional forms are arbitrary and untested.
  • standard math Standard backpropagation and stochastic optimization can train the full query-based mask decoder with Hungarian matching.
    Implicit in Sections 3.3 and 3.4; this is a routine assumption for transformer-based detectors.
invented entities (3)
  • Global Geographic Information Tokens (GGIT)
    purpose: Scene-level context tokens that are injected into both the SPCN backbone and the decoder to provide global awareness.
    Learned embeddings with no external validation; the claimed benefit is only qualitatively described. No ablation numbers isolate their contribution.
  • Geometric consistency factor C in GAF
    purpose: A hand-designed multiplicative factor that enforces agreement between the three projection-plane features.
    Defined as the average of feature similarities in Section 3.1; no analysis shows it improves performance.
  • Sub-window Block (SWB)
    purpose: Basic processing unit performing windowed multi-head self-attention and feed-forward layers.
    A re-configuration of standard transformer blocks; no independent evidence beyond the paper's missing experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Co-Win: Joint Object Detection and Instance Segmentation in LiDAR Point Clouds via Collaborative Window Processing." pith.science (2026). https://pith.science/paper/E46FQKFJ

@misc{pith2026250719691,
  author       = {Pith},
  title        = {Pith review of: Co-Win: Joint Object Detection and Instance Segmentation in LiDAR Point Clouds via Collaborative Window Processing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E46FQKFJ}},
  note         = {Machine review of arXiv:2507.19691}
}
read the original abstract

Accurate perception and scene understanding in complex urban environments is a critical challenge for ensuring safe and efficient autonomous navigation. In this paper, we present Co-Win, a novel bird's eye view (BEV) perception framework that integrates point cloud encoding with efficient parallel window-based feature extraction to address the multi-modality inherent in environmental understanding. Our method employs a hierarchical architecture comprising a specialized encoder, a window-based backbone, and a query-based decoder head to effectively capture diverse spatial features and object relationships. Unlike prior approaches that treat perception as a simple regression task, our framework incorporates a variational approach with mask-based instance segmentation, enabling fine-grained scene decomposition and understanding. The Co-Win architecture processes point cloud data through progressive feature extraction stages, ensuring that predicted masks are both data-consistent and contextually relevant. Furthermore, our method produces interpretable and diverse instance predictions, enabling enhanced downstream decision-making and planning in autonomous driving systems.

Figures

Figures reproduced from arXiv: 2507.19691 by the authors.

Figure 1
Figure 1. The data transformation process of the Co-Win algorithm is illustrated. Subfigures (a) and [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An example of object mask in BEV. The subfigure (a) shows the ground truth and point [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Qualitative test of predictions. Because the point clouds range on x-axis is [0,80], y-axis [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 31 canonical work pages

  1. [1]

    Behley, M

    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. Proceedings of the IEEE/CVF International Conference on Computer Vision , 2019

    work page 2019

  2. [2]

    X. Chen, T. Läbe, A. Milioto, T. Röhling, O. Vysotska, A. Haag, J. Behley, and C. Stachniss. Polarstream: Streaming lidar object detection and segmentation with polar pillars. In IEEE International Conference on Robotics and Automation (ICRA) , pages 14525–14531, 2021

    work page 2021

  3. [3]

    X. Chen, H. Ma, J. Wan, B. Li, and T. Xia. Multi-view 3d object detection network for autonomous driving. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 1907–1915, 2017

    work page 1907

  4. [4]

    Cheng, A

    B. Cheng, A. Schwing, and A. Kirillov. Mask2former for video instance segmentation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 1234–1243, 2022

    work page 2022

  5. [5]

    S. Choi, J. Kim, H. Shin, and J. W. Choi. Mask2Map: V ectorized HD Map Construction Using Bird’s Eye View Segmentation Masks. Springer Nature Switzerland, Dec. 2024

    work page 2024

  6. [6]

    Geiger, P

    A. Geiger, P. Lenz, and R. Urtasun. Vision meets robotics: The kitti dataset. International Journal of Robotics Research, 32(11):1231–1237, 2013

    work page 2013

  7. [7]

    Guimont-Martin, J.-M

    W. Guimont-Martin, J.-M. Fortin, F. Pomerleau, and P. Giguère. Maskbev: Joint object detection and footprint completion for bird’s-eye view 3d point clouds. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2023

    work page 2023

  8. [8]

    K. He, G. Gkioxari, P. Dollár, and R. Girshick. Mask r-cnn. In IEEE International Conference on Computer Vision (ICCV), pages 2961–2969, 2017

    work page 2017

Show all 33 references
  1. [9]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2015

    2015

  2. [10]

    F. Hong, H. Zhou, X. Zhu, H. Li, and Z. Liu. Lidar-based panoptic segmentation via dynamic shifting network. In IEEE/CVF International Conference on Computer Vision (ICCV) , pages 13046–13055, 2021

    2021

  3. [11]

    Huang, G

    J. Huang, G. Huang, Z. Zhu, and D. Du. Bevdet: High-performance multi-camera 3d object detection in bird-eye-view. In IEEE/CVF International Conference on Computer Vision (ICCV) , 2021

    2021

  4. [12]

    J. Jin, W. Liu, Z. Ning, Q. Zhao, S. Cheng, and J. Hu. 3d object detection for autonomous driving: A survey. In Chinese Control and Decision Conference (CCDC) , 2024

    2024

  5. [13]

    J. Ku, M. Mozifian, J. Lee, A. Harakeh, and S. L. Waslander. Joint 3d proposal generation and object detection from view aggregation. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 1–8, 2018

    2018

  6. [14]

    A. H. Lang, S. V ora, H. Caesar, L. Zhou, J. Yang, and O. Beijbom. Pointpillars: Fast encoders for object detection from point clouds. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019

    2019

  7. [15]

    Y . Li, L. Fan, Y . Liu, Z. Huang, Y . Chen, N. Wang, and Z. Zhang. Fully sparse fusion for 3d object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence , 46(11), 2024

    2024

  8. [16]

    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. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(8):9063–9079, 2022

    2022

  9. [17]

    Liu and Z

    D. Liu and Z. Wang. Pointrcnn++: Towards more accurate two-stage 3d object detection from point cloud. China Automation Congress (CAC), 2023. 10

    2023

  10. [18]

    Z. Liu, H. Tang, A. Amini, X. Yang, H. Mao, D. Rus, and S. Han. Bevfusion: Multi-task multi-sensor fusion with unified bird’s-eye view representation. IEEE Robotics and Automation Letters, 8(3):1522–1529, 2023

    2023

  11. [19]

    Marcuzzi, L

    R. Marcuzzi, L. Nunes, L. Wiesmann, J. Behley, and C. Stachniss. Mask-based panoptic lidar segmentation for autonomous driving. IEEE Robotics and Automation Letters , 8, 2023

    2023

  12. [20]

    Narita, T

    G. Narita, T. Seno, T. Ishikawa, and Y . Kaji. Panopticfusion: Online volumetric semantic mapping at the level of stuff and things. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 4205–4212, 2019

    2019

  13. [21]

    S. Shao, Z. Pei, W. Chen, Q. Liu, H. Yue, and Z. Li. Sparse pseudo-lidar depth assisted monocular depth estimation. IEEE Transactions on Intelligent V ehicles, 9, 2024

    2024

  14. [22]

    S. Shi, X. Wang, and H. Li. Pointrcnn: 3d object proposal generation and detection from point cloud. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 770–779, 2019

    2019

  15. [23]

    P. Sun, H. Kretzschmar, X. Dotiwalla, C. Chouard, V . Patnaik, P. Tsui, J. Guo, Y . Zhou, Y . Chai, B. Caine, et al. Scalability in perception for autonomous driving: Waymo open dataset. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020

    2020

  16. [24]

    Wang, W.-L

    Y . Wang, W.-L. Chao, D. Garg, B. Hariharan, M. E. Campbell, and K. Q. Weinberger. Pseudo- lidar from visual depth estimation: Bridging the gap in 3d object detection for autonomous driving. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2018

    2018

  17. [25]

    Y . Weng, M. Han, H. He, M. Li, L. Yao, X. Chang, and B. Zhuang. Mask propagation for efficient video semantic segmentation. ArXiv, abs/2310.18954, 2023

    2023 arXiv

  18. [26]

    W. Yan, S. Liu, C. Tang, and W. Zhou. Pisfanet: Pillar scale-aware feature aggregation network for real-time 3d pedestrian detection. IEEE Signal Processing Letters, 31:2000–2004, 2024

    2000

  19. [27]

    Y . Yan, Y . Mao, and B. Li. Second: Sparsely embedded convolutional detection. InSensors, volume 18, page 3337, 2018

    2018

  20. [28]

    Z. Yang, Y . Sun, S. Liu, X. Shen, and J. Jia. 3dssd: Point-based 3d single stage object detector. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 11040–11048, 2020

    2020

  21. [29]

    T. Yin, X. Zhou, and P. Krahenbuhl. Center-based 3d object detection and tracking. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 11784–11793, 2021

    2021

  22. [30]

    X. Zhao, X. Zhang, D. Yang, M. Sun, M. Li, S. Wang, and L. Zhang. Maskbev: Towards a unified framework for bev detection and map segmentation. ArXiv, abs/2408.09122, 2024

    2024 arXiv

  23. [31]

    Y . Zhou, P. Sun, Y . Zhang, D. Anguelov, J. Gao, T. Y . Ouyang, J. Guo, J. Ngiam, and V . Vasude- van. End-to-end multi-view fusion for 3d object detection in lidar point clouds. In Conference on Robot Learning, 2019

    2019

  24. [32]

    Zhou and O

    Y . Zhou and O. Tuzel. V oxelnet: End-to-end learning for point cloud based 3d object detection. IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2017

    2017

  25. [33]

    Z. Zong, D. Jiang, G. Song, Z. Xue, J. Su, H. Li, and Y . Liu. Temporal enhanced training of multi-view 3d object detector via historical object prediction. IEEE/CVF International Conference on Computer Vision (ICCV), 2023. 11

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.