Pith. sign in

REVIEW 3 major objections 5 minor 65 references

PDM-SSD: Single-Stage Three-Dimensional Object Detector With Point Dilation

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read Point dilation gives 3D point-cloud detectors grid coverage at 68 FPS.

desk verdict The point dilation idea is real and the Car numbers support a modest gain, but the multi-class SOTA claim is unsupported because pedestrians are skipped and cyclist results trail SPSNet on moderate/hard. read the letter →

arxiv 2502.07822 v1 pith:6ZJ64VL5 submitted 2025-02-10 cs.CV cs.AI

classification cs.CVcs.AI
keywords autonomousdriving3Dobjectdetectiondeeplearningpointcloudprocessingdilationmechanismsingle-stagedetectorsphericalharmonicsLiDARclouds
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

Point-based 3D object detectors are fast and lightweight, but they can only learn from points that actually exist, so for sparse or partially occluded objects their receptive field stops growing no matter how large the query radius is. This paper proposes a Point Dilation Mechanism (PDM) that lifts point-wise features onto a 2D grid, dilates the occupied cells, and fills the newly covered empty space with features built from spherical-harmonic angular coefficients and Gaussian scale coefficients. The result is PDM-SSD, a single-stage detector whose hybrid head combines the original point features with the filled grid features and a predicted scene heatmap. The paper reports that on the KITTI benchmark this configuration outperforms prior point-based single-stage detectors on Car and Cyclist detection, runs at 68 FPS, and improves detection of sparse and incomplete objects; a training-only auxiliary variant keeps the base detector's 84 FPS speed while still gaining accuracy. If the paper is right, point-based detectors can get the robustness of grid-based perception without the deployment cost of 3D convolutions or sparse-convolution operators.

What carries the argument

The central object is the Point Dilation Mechanism (PDM), a neck module with two steps. Point dilation binarizes the projected occupancy of sampled point features and expands it with a $5 \times 5$ structuring element, so cells with no LiDAR points become part of the feature space. Feature filling then assigns those empty cells a backpropagatable feature: the dilation center's feature is split, one half weighted by spherical-harmonic angular coefficients (direction) and the other by a Gaussian scale coefficient (distance and learned variance), and the two halves are summed, while overlapping centers are combined by height compression. This gives the point-based backbone a continuous receptive field over the object's center region, exactly the space a sparse or occluded object tends to lack.

What would settle it

Run PDM-SSD on the KITTI pedestrian class with the same $176 \times 200$ grid and $5 \times 5$ structuring element: if pedestrian average precision does not improve over the base detector, the single-setting assumption behind the multi-class claim is falsified. The paper's own limitation section predicts this failure.

Watch

Extended reading notes

Core claim

The paper's central claim is that the discontinuous receptive field of point-based detectors can be fixed without abandoning the point-based pipeline. Point Dilation projects the final-stage point features onto a $176 \times 200$ occupancy grid, applies a $5 \times 5$ all-ones structuring element to expand occupancy, and then fills every newly covered cell with a learnable feature: an angle coefficient obtained by weighting 16 fourth-order spherical-harmonic coefficients by the direction from the dilation center, plus a scale coefficient from a bivariate Gaussian whose variance is regressed per center. Features from overlapping dilation centers are added together through height compression, giving a sparse grid feature with continuous spatial coverage. A hybrid head then regresses a heatmap from the grid features, adds the top-256 heatmap peaks to the vote point set, and fuses point and grid features through channel attention before box prediction. On the KITTI val set the paper reports Car AP3D R40 moderate of 83.31 versus 82.86 for the base detector, Cyclist easy AP3D R11 of 92.04 versus 87.28, and an inference speed of 68 FPS, and it attributes these gains to correcting vote-point center errors and raising low classification probabilities on sparse and incomplete objects.

Load-bearing premise

The central accuracy claim rests on the assumption that one fixed grid resolution and one fixed $5 \times 5$ dilation window work for all object classes; the paper itself notes this is why pedestrians are excluded from the experiments.

Editorial extensions

If this is right

  • PDM-SSD reports the best KITTI Car results among point-based single-stage detectors, so point-based detectors can close the accuracy gap to grid-based ones without changing backbone type.
  • The auxiliary-learning variant gains accuracy over the base detector at the same 84 FPS inference speed, meaning PDM can be added as a training-only module to existing point-based detectors.
  • The hybrid head's heatmap-supplemented vote points and fused grid features address two failure modes on sparse objects: vote points that drift from the object center and boxes whose classification confidence falls below threshold.
  • Cyclist results indicate the mechanism helps elongated or partial objects beyond cars, supporting the paper's broader multi-class claim.
  • With 3.3 million total parameters and a neck of 0.53 million parameters, the model keeps the deployment advantages of point-based detectors while adding grid-like coverage.

Reading between the lines

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

  • A natural extension the authors do not test is per-class dilation: using different structuring-element sizes or grid resolutions for pedestrians and cyclists would likely extend PDM to small objects, and the paper's own limitation section points to this.
  • Because the filled grid supplies features where no LiDAR points exist, PDM should be tested under artificial point dropout, simulating range attenuation or sensor degradation; if gains persist, the mechanism is genuinely about missing space rather than about extra parameters.
  • The same dilation-and-fill recipe could transfer to other point-level tasks, such as 3D semantic segmentation or point tracking, where receptive-field discontinuity also limits performance; the paper only evaluates detection.
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 proposes PDM-SSD, a single-stage point-based 3D object detector whose key novelty is a Point Dilation Mechanism (PDM) placed in the neck. PDM lifts sampled point features onto a 2D grid, dilates the occupancy pattern with a fixed 5x5 structuring element, and fills newly occupied cells with features weighted by predicted spherical-harmonic angle coefficients and Gaussian scale coefficients. A hybrid head then combines a vote-based point head with a heatmap-prediction branch and grid-feature fusion. Experiments on KITTI report Car test and val results, Cyclist val results, ablations of the spherical-harmonic degree and coefficient-fusion strategy, runtime measurements, and qualitative object-level analyses. The authors claim state-of-the-art multi-class detection among single-modal methods at 68 FPS and also propose an auxiliary-training variant that preserves the base detector's speed.

Significance. The paper targets a genuine weakness of point-based detectors--the discontinuous receptive field on sparse and incomplete objects--and the proposed mechanism is simple, lightweight, and well motivated. The Car results are credible and competitive: PDM-SSD achieves 80.87 R40 moderate 3D AP on the KITTI test set for Car, the best among the point-based single-stage methods compared, at 68 FPS with only 3.3M parameters. The auxiliary-training variant, which keeps the full 84 FPS inference speed while improving accuracy over IA-SSD, is a practically useful and clean contribution. The ablations are informative, and the paper identifies a concrete limitation of its fixed structuring element. However, the significance is substantially qualified by the gap between the paper's multi-class state-of-the-art claim and the evidence actually reported; the headline claim is neither fully testable nor fully consistent with the included tables.

major comments (3)
  1. [Abstract; Section 1; Section 4.2, Tables 1-3] The central claim that PDM-SSD achieves state-of-the-art multi-class detection among single-modal methods is not supported by the reported experiments. KITTI's standard multi-class evaluation includes Car, Pedestrian, and Cyclist, but test-set results are provided only for Car (Table 1) and validation results only for Car and Cyclist (Tables 2-3); no Pedestrian results appear anywhere. Section 5 (Limitations) states: 'We avoided the detection of pedestrians in the KITTI dataset because the small volume and limited impact of pedestrians make the 5x5 structural element unsuitable for detecting such targets.' Omitting one of the three classes makes the multi-class claim untestable in the paper, and the one reported non-Car class actually contradicts it: in Table 3, SPSNet outperforms PDM-SSD on Cyclist 3D AP R40 at moderate (75.51 vs 73.44) and hard (71.10 vs 68.67). The authors should report the missing Pedestrian results (for example with a class-adaptive structuring element) or resubmit with an abstract and contribution statement scoped to what the data support, such as state-of-the-art KITTI Car performance among point-based single-stage detectors.
  2. [Section 3.4; Section 5 (Limitations)] The fixed grid resolution (176x200), fixed 5x5 dilation element, and fixed spherical-harmonic degree are introduced as architecture constants in Section 3.4, and Section 5 acknowledges that these settings are unsuitable for Pedestrian targets. No ablation or class-wise analysis is provided to show that the mechanism transfers across object classes. Because the paper's headline is multi-class performance, the absence of any demonstration that these hyperparameters work beyond Car and partially Cyclist is a load-bearing gap; an ablation varying the structuring element and grid resolution per class, or a principled criterion for selecting them, is needed to support the claimed generality.
  3. [Section 4.2, analysis item 4; Table 3] The discussion of Cyclist results says PDM-SSD 'outperforms IA-SSD by a large margin (4.76%, 0.64%, 1.21%) with 11 recall points' and 'achieves the best performance in the easy level.' This is true as far as it goes, but it does not address the relevant state-of-the-art comparison: against SPSNet, the strongest point-based single-stage baseline, PDM-SSD trails on Cyclist 3D AP R40 at moderate and hard by 2.07 and 2.43 points, respectively. The paper should present the full R40 comparison against SPSNet in the analysis rather than selecting the easy-level R11 result, and should avoid describing the Cyclist results as best without qualifying the moderate/hard deficit.
minor comments (5)
  1. [Section 3.4, Eq. (12)] The summation bounds in Eq. (12) are malformed as printed; the expression should be sum over l from 0 to L and m from -l to l of c_l^m Y_l^m(theta, phi). In addition, the text says 'fourth order' while Section 4.1 and Table 5 refer to a scale of 3 with 16 coefficients; the order/degree terminology should be made consistent.
  2. [Section 3.4, Eqs. (6), (13)-(14)] The symbol G is used both for the grid projection function in Eq. (6) and for the Gaussian density in Eqs. (13)-(14). Using distinct symbols (for example, Pi and N) would remove avoidable ambiguity.
  3. [Section 4.1, Training details] The augmentation description states that the scene is randomly rotated 'within the range of [-4/pi, 4/pi]'; this should presumably be [-pi/4, pi/4]. Please correct the notation.
  4. [Section 3.5; Section 4.1] Section 3.5 refers to adding the top K heatmap peaks to the voting point set, while Section 4.1 specifies top 256; the notation should be unified and the choice of K should be stated as a hyperparameter.
  5. [Section 3.2, Eqs. (2)-(3)] Eq. (2) contains the typo 'Sride i' for stride, and the index range of the product in Eq. (3) should be made explicit to match standard receptive-field formulas.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the multi-class SOTA overclaim is an evidence gap, not a circular derivation.

full rationale

PDM-SSD is an empirical detector paper: the PDM neck, heatmap head, and hybrid head are learned components trained with standard losses on KITTI, and no central derivation reduces to its own inputs by construction. The only self-reference is the authors' prior work SPSNet [40], cited in Related Work and in Section 3.6 ('Although SPSNet has shown that points closer to the center are not necessarily better...'). That citation is used for background and as a baseline comparison, not as a load-bearing premise; the detector is built on IA-SSD and evaluated against external KITTI test/val benchmarks with independently reported baselines, so the core accuracy claims are not forced by self-citation. The self-referential use of the model's own predicted heatmap to augment the voting point set is an architectural choice, and the ablation (PDM-SSD(A) vs PDM-SSD(J)) quantifies its contribution rather than disguising a fitted parameter as a prediction. The Section 5 limitation—'We avoided the detection of pedestrians in the KITTI dataset because the small volume and limited impact of pedestrians make the 5 × 5 structural element unsuitable for detecting such targets'—undermines the abstract's multi-class SOTA claim on evidential grounds, because the standard multi-class benchmark includes pedestrians, but this is an overclaim/support issue, not circularity. No equation is defined in terms of its target result, no fitted constant is renamed as a prediction, and no uniqueness theorem is imported from the authors' prior work. Score 1 reflects only the presence of a minor, non-load-bearing self-citation.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

No new physical or mathematical entities are postulated. The PDM is a neural network module. The free parameters listed are hand-selected or ablated design choices that the central accuracy claim depends on. The axioms include standard math and the paper's own domain assumptions about how to fill empty space.

free parameters (5)
  • Grid size (W x H) = 176 x 200
    Hand-chosen for the KITTI range [0,70.4] x [-40,40]; not ablated. Affects resolution and memory footprint.
  • Structuring element size = 5 x 5 all-ones
    Chosen to expand occupancy; the authors admit in Section 5 that it is unsuitable for pedestrians, making it class-dependent and critical to the multi-class claim.
  • Spherical harmonic degree = degree 3, 16 coefficients
    Ablated in Table 5; degree 4 gives similar performance, degree 3 is chosen for efficiency, so the choice is performance-fitted.
  • Top-K heatmap points added to vote set = 256
    Mentioned in training details; not ablated, hand-chosen.
  • Object-level augmentation copies = 15 cars
    Hand-chosen number of cars copied into the scene during training; no ablation.
assumptions (5)
  • standard math Spherical harmonics are a complete angular basis (Eq. 9).
    Used to define angle coefficients alpha in Eq. 12; standard result from mathematical physics.
  • standard math The Gaussian density function models scale dependence (Eq. 14).
    Used to define scale coefficients beta; standard probability density.
  • ad hoc to paper The angle between a new cell and the dilation center is a valid proxy for the feature relationship (Section 3.4, Angle Coefficient).
    The paper postulates that SH-weighted features of the center describe surrounding cells; no independent evidence is given.
  • ad hoc to paper The learned variance in the Gaussian captures the spatial extent of the target (Section 3.4, Scale Coefficient).
    The predicted sigma is assumed to encode scale; this is not derived or externally validated.
  • domain assumption Supervising the dilated grid features with GT center heatmaps transfers global object information back to point features (Section 3.5, Auxiliary Learning).
    The improvement of PDM-SSD(A) over IA-SSD is empirical evidence, but the transfer mechanism is asserted rather than shown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PDM-SSD: Single-Stage Three-Dimensional Object Detector With Point Dilation." pith.science (2026). https://pith.science/paper/6ZJ64VL5

@misc{pith2026250207822,
  author       = {Pith},
  title        = {Pith review of: PDM-SSD: Single-Stage Three-Dimensional Object Detector With Point Dilation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6ZJ64VL5}},
  note         = {Machine review of arXiv:2502.07822}
}
read the original abstract

Current Point-based detectors can only learn from the provided points, with limited receptive fields and insufficient global learning capabilities for such targets. In this paper, we present a novel Point Dilation Mechanism for single-stage 3D detection (PDM-SSD) that takes advantage of these two representations. Specifically, we first use a PointNet-style 3D backbone for efficient feature encoding. Then, a neck with Point Dilation Mechanism (PDM) is used to expand the feature space, which involves two key steps: point dilation and feature filling. The former expands points to a certain size grid centered around the sampled points in Euclidean space. The latter fills the unoccupied grid with feature for backpropagation using spherical harmonic coefficients and Gaussian density function in terms of direction and scale. Next, we associate multiple dilation centers and fuse coefficients to obtain sparse grid features through height compression. Finally, we design a hybrid detection head for joint learning, where on one hand, the scene heatmap is predicted to complement the voting point set for improved detection accuracy, and on the other hand, the target probability of detected boxes are calibrated through feature fusion. On the challenging Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) dataset, PDM-SSD achieves state-of-the-art results for multi-class detection among single-modal methods with an inference speed of 68 frames. We also demonstrate the advantages of PDM-SSD in detecting sparse and incomplete objects through numerous object-level instances. Additionally, PDM can serve as an auxiliary network to establish a connection between sampling points and object centers, thereby improving the accuracy of the model without sacrificing inference speed. Our code will be available at https://github.com/AlanLiangC/PDM-SSD.git.

Figures

Figures reproduced from arXiv: 2502.07822 by the authors.

Figure 1
Figure 1. (a) The basic structure of the Grid-based 3D detector. P/VFE means Pillar/Voxel Feature Encoder. This [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall workflow of PDM-SSD. In the joint training phase, the input LiDAR point clouds are first passed [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Visualization on some very sparse and extremely incomplete targets on the KITTI dataset. For grid-based [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Point dilation operation. The point cloud is first projected onto a 2D binary occupancy grid and then dilated [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Feature Filling operation. We propose a feature filling method based on spatial separation coefficient. We use [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: (a) Coefficients fusion. In order to maintain the nonlinearity of different cell features, we first decompose the [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Joint Learning. In the case of sparsity and extremely incomplete targets, on one hand, the vote points generated [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: The detection results of PDM-SSD on a subset of KITTI [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Issue1. The error in the voting point position regressed from the sampling points increases the difficulty of [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Issue2. The detection box regressed from the features learned from context has a probability lower than the [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 57 canonical work pages

  1. [1]

    Are we ready for autonomous driving? the kitti vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In 2012 IEEE conference on computer vision and pattern recognition, pages 3354–3361. IEEE, 2012

  2. [2]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aurelien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, et al. Scalability in perception for autonomous driving: Waymo open dataset. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2446–2454, 2020

  3. [3]

    Rangedet: In defense of range view for lidar-based 3d object detection

    Lue Fan, Xuan Xiong, Feng Wang, Naiyan Wang, and Zhaoxiang Zhang. Rangedet: In defense of range view for lidar-based 3d object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 2918–2927, 2021

  4. [4]

    Complex-YOLO: Real-time 3d object detection on point clouds

    Martin Simon, Stefan Milz, Karl Amende, and Horst-Michael Gross. Complex-YOLO: Real-time 3d object detection on point clouds

  5. [5]

    Multi-view 3d object detection network for autonomous driving

  6. [6]

    Birdnet: a 3d object detection framework from lidar information

    Jorge Beltrán, Carlos Guindel, Francisco Miguel Moreno, Daniel Cruzado, Fernando Garcia, and Arturo De La Es- calera. Birdnet: a 3d object detection framework from lidar information. In 2018 21st International Conference on Intelligent Transportation Systems (ITSC), pages 3517–3523. IEEE, 2018

  7. [7]

    Rt3d: Real-time 3-d vehicle detection in lidar point cloud for autonomous driving

    Yiming Zeng, Yu Hu, Shice Liu, Jing Ye, Yinhe Han, Xiaowei Li, and Ninghui Sun. Rt3d: Real-time 3-d vehicle detection in lidar point cloud for autonomous driving. IEEE Robotics and Automation Letters, 3(4):3434–3440, 2018

  8. [8]

    Yolo3d: End-to-end real-time 3d oriented object bounding box detection from lidar point cloud

    Waleed Ali, Sherif Abdelkarim, Mahmoud Zidan, Mohamed Zahran, and Ahmad El Sallab. Yolo3d: End-to-end real-time 3d oriented object bounding box detection from lidar point cloud. In Proceedings of the European conference on computer vision (ECCV) workshops, pages 0–0, 2018

Show all 65 references
  1. [9]

    Birdnet+: End-to-end 3d object detection in lidar bird’s eye view

    Alejandro Barrera, Carlos Guindel, Jorge Beltrán, and Fernando García. Birdnet+: End-to-end 3d object detection in lidar bird’s eye view. In 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC), pages 1–6. IEEE, 2020

  2. [10]

    Bcaf-3d: Bilateral content awareness fusion for cross-modal 3d object detection

    Mu Chen, Pengfei Liu, and Huaici Zhao. Bcaf-3d: Bilateral content awareness fusion for cross-modal 3d object detection. Knowledge-Based Systems, 279:110952, 2023

  3. [11]

    Se-ssd: Self-ensembling single-stage object detector from point cloud

    Wu Zheng, Weiliang Tang, Li Jiang, and Chi-Wing Fu. Se-ssd: Self-ensembling single-stage object detector from point cloud. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14494–14503, 2021

  4. [12]

    Not all points are equal: Learning highly efficient point-based detectors for 3d lidar point clouds

    Yifan Zhang, Qingyong Hu, Guoquan Xu, Yanxin Ma, Jianwei Wan, and Yulan Guo. Not all points are equal: Learning highly efficient point-based detectors for 3d lidar point clouds. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18953–1...

  5. [13]

    Pointrcnn: 3D object proposal generation and detection from point cloud

    Shaoshuai Shi, Xiaogang Wang, and Hongsheng Li. Pointrcnn: 3D object proposal generation and detection from point cloud. In CVPR, pages 770–779, 2019

  6. [14]

    Point-gnn: Graph neural network for 3D object detection in a point cloud

    Weijing Shi and Raj Rajkumar. Point-gnn: Graph neural network for 3D object detection in a point cloud. In CVPR, pages 1711–1719, 2020

  7. [15]

    3dssd: Point-based 3D single stage object detector

    Zetong Yang, Yanan Sun, Shu Liu, and Jiaya Jia. 3dssd: Point-based 3D single stage object detector. In CVPR, pages 11040–11048, 2020

  8. [16]

    Learning semantic segmentation of large-scale point clouds with random sampling

    Qingyong Hu, Bo Yang, Linhai Xie, Stefano Rosa, Yulan Guo, Zhihua Wang, Niki Trigoni, and Andrew Markham. Learning semantic segmentation of large-scale point clouds with random sampling. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

  9. [17]

    Sqn: Weakly-supervised semantic segmentation of large-scale 3d point clouds with 1000x fewer labels

    Qingyong Hu, Bo Yang, Guangchi Fang, Yulan Guo, Ales Leonardis, Niki Trigoni, and Andrew Markham. Sqn: Weakly-supervised semantic segmentation of large-scale 3d point clouds with 1000x fewer labels. arXiv preprint arXiv:2104.04891, 2021

  10. [18]

    Sensaturban: Learning semantics from urban-scale photogrammetric point clouds

    Qingyong Hu, Bo Yang, Sheikh Khalid, Wen Xiao, Niki Trigoni, and Andrew Markham. Sensaturban: Learning semantics from urban-scale photogrammetric point clouds. International Journal of Computer Vision, pages 1–28, 2022

  11. [19]

    Spatial-temporal transformer for 3d point cloud sequences

    Yimin Wei, Hao Liu, Tingting Xie, Qiuhong Ke, and Yulan Guo. Spatial-temporal transformer for 3d point cloud sequences. In W ACV, pages 1171–1180, 2022

  12. [20]

    Center-based 3d object detection and tracking

    Tianwei Yin, Xingyi Zhou, and Philipp Krahenbuhl. Center-based 3d object detection and tracking. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11784–11793, 2021

  13. [21]

    V oxelNet: End-to-end learning for point cloud based 3d object detection

    Yin Zhou and Oncel Tuzel. V oxelNet: End-to-end learning for point cloud based 3d object detection

  14. [22]

    V oxelnext: Fully sparse voxelnet for 3d object detection and tracking

    Yukang Chen, Jianhui Liu, Xiangyu Zhang, Xiaojuan Qi, and Jiaya Jia. V oxelnext: Fully sparse voxelnet for 3d object detection and tracking. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21674–21683, 2023

  15. [23]

    Distillbev: Boosting multi-camera 3d object detection with cross-modal knowledge distillation

    Zeyu Wang, Dingwen Li, Chenxu Luo, Cihang Xie, and Xiaodong Yang. Distillbev: Boosting multi-camera 3d object detection with cross-modal knowledge distillation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8637–8646, 2023

  16. [24]

    LidarMulti- Net: Towards a unified multi-task network for LiDAR perception

    Dongqiangzi Ye, Zixiang Zhou, Weijia Chen, Yufei Xie, Yu Wang, Panqu Wang, and Hassan Foroosh. LidarMulti- Net: Towards a unified multi-task network for LiDAR perception

  17. [25]

    Lang, Sourabh V ora, Holger Caesar, Lubing Zhou, Jiong Yang, and Oscar Beijbom

    Alex H. Lang, Sourabh V ora, Holger Caesar, Lubing Zhou, Jiong Yang, and Oscar Beijbom. PointPillars: Fast encoders for object detection from point clouds

  18. [26]

    PillarNet: Real-time and high-performance pillar-based 3d object detection

    Guangsheng Shi, Ruifeng Li, and Chao Ma. PillarNet: Real-time and high-performance pillar-based 3d object detection

  19. [27]

    FastPillars: A deployment-friendly pillar-based 3d detector

    Sifan Zhou, Zhi Tian, Xiangxiang Chu, Xinyu Zhang, Bo Zhang, Xiaobo Lu, Chengjian Feng, Zequn Jie, Patrick Yin Chiang, and Lin Ma. FastPillars: A deployment-friendly pillar-based 3d detector

  20. [28]

    Driveadapter: Breaking the coupling barrier of perception and planning in end-to-end autonomous driving

    Xiaosong Jia, Yulu Gao, Li Chen, Junchi Yan, Patrick Langechuan Liu, and Hongyang Li. Driveadapter: Breaking the coupling barrier of perception and planning in end-to-end autonomous driving. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 7953...

  21. [29]

    Planning-oriented autonomous driving

    Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, Wenhai Wang, et al. Planning-oriented autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17853–17862, 2023

  22. [30]

    Delving into the devils of bird’s-eye-view perception: A review, evaluation and recipe

    Hongyang Li, Chonghao Sima, Jifeng Dai, Wenhai Wang, Lewei Lu, Huijie Wang, Jia Zeng, Zhiqi Li, Jiazhi Yang, Hanming Deng, et al. Delving into the devils of bird’s-eye-view perception: A review, evaluation and recipe. IEEE Transactions on Pattern Analysis and Machine Intellige...

  23. [31]

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

    Zhiqi Li, Wenhai Wang, Hongyang Li, Enze Xie, Chonghao Sima, Tong Lu, Yu Qiao, and Jifeng Dai. Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers. In European conference on computer vision, pages 1–18. Springer, 2022

  24. [32]

    Lasernet: An efficient probabilistic 3d object detector for autonomous driving

    Gregory P Meyer, Ankit Laddha, Eric Kee, Carlos Vallespi-Gonzalez, and Carl K Wellington. Lasernet: An efficient probabilistic 3d object detector for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12677–12686, 2019

  25. [33]

    Deep layer aggregation

    Fisher Yu, Dequan Wang, Evan Shelhamer, and Trevor Darrell. Deep layer aggregation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2403–2412, 2018. 20 Running Title for Header

  26. [34]

    Laserflow: Efficient and probabilistic object detection and motion forecasting

    Gregory P Meyer, Jake Charland, Shreyash Pandey, Ankit Laddha, Shivam Gautam, Carlos Vallespi-Gonzalez, and Carl K Wellington. Laserflow: Efficient and probabilistic object detection and motion forecasting. IEEE Robotics and Automation Letters, 6(2):526–533, 2020

  27. [35]

    Rangercnn: Towards fast and accurate 3d object detection with range image representation

    Zhidong Liang, Ming Zhang, Zehan Zhang, Xian Zhao, and Shiliang Pu. Rangercnn: Towards fast and accurate 3d object detection with range image representation. arXiv preprint arXiv:2009.00206, 2020

  28. [36]

    Qi, Or Litany, Kaiming He, and Leonidas J

    Charles R. Qi, Or Litany, Kaiming He, and Leonidas J. Guibas. Deep hough voting for 3d object detection in point clouds

  29. [37]

    3dssd: Point-based 3d single stage object detector

    Zetong Yang, Yanan Sun, Shu Liu, and Jiaya Jia. 3dssd: Point-based 3d single stage object detector

  30. [38]

    Pointnet: Deep learning on point sets for 3d clas- sification and segmentation

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d clas- sification and segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 652–660, 2017

  31. [39]

    Dbq-ssd: Dynamic ball query for efficient 3d object detection

    Jinrong Yang, Lin Song, Songtao Liu, Weixin Mao, Zeming Li, Xiaoping Li, Hongbin Sun, Jian Sun, and Nanning Zheng. Dbq-ssd: Dynamic ball query for efficient 3d object detection. In The Eleventh International Conference on Learning Representations, 2022

  32. [40]

    Spsnet: Boosting 3d point-based object detectors with stable point sampling

    Ao Liang, Hao Zhang, Haiyang Hua, Whenyu Chen, and Huaici Zhao. Spsnet: Boosting 3d point-based object detectors with stable point sampling. Engineering Applications of Artificial Intelligence, 126:106807, 2023

  33. [41]

    Understanding the robustness of 3d object detection with bird’s-eye-view representations in autonomous driving

    Zijian Zhu, Yichi Zhang, Hai Chen, Yinpeng Dong, Shu Zhao, Wenbo Ding, Jiachen Zhong, and Shibao Zheng. Understanding the robustness of 3d object detection with bird’s-eye-view representations in autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision ...

  34. [42]

    Point-voxel cnn for efficient 3d deep learning

    Zhijian Liu, Haotian Tang, Yujun Lin, and Song Han. Point-voxel cnn for efficient 3d deep learning. Advances in Neural Information Processing Systems, 32, 2019

  35. [43]

    Searching efficient 3d architectures with sparse point-voxel convolution

    Haotian Tang, Zhijian Liu, Shengyu Zhao, Yujun Lin, Ji Lin, Hanrui Wang, and Song Han. Searching efficient 3d architectures with sparse point-voxel convolution. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXVIII, p...

  36. [44]

    Pv-rcnn++: Point-voxel feature set abstraction with local vector representation for 3d object detection

    Shaoshuai Shi, Li Jiang, Jiajun Deng, Zhe Wang, Chaoxu Guo, Jianping Shi, Xiaogang Wang, and Hongsheng Li. Pv-rcnn++: Point-voxel feature set abstraction with local vector representation for 3d object detection. International Journal of Computer Vision, 131(2):531–551, 2023

  37. [45]

    Second: Sparsely embedded convolutional detection.Sensors, 18(10):3337, 2018

    Yan Yan, Yuxing Mao, and Bo Li. Second: Sparsely embedded convolutional detection.Sensors, 18(10):3337, 2018

  38. [46]

    Badet: Boundary-aware 3d object detection from point clouds

    Rui Qian, Xin Lai, and Xirong Li. Badet: Boundary-aware 3d object detection from point clouds. Pattern Recognition, 125:108524, 2022

  39. [47]

    Morphological dilation image coding with context weights prediction

    Jiaji Wu, Anand Paul, Yan Xing, Yong Fang, Jechang Jeong, Licheng Jiao, and Guangming Shi. Morphological dilation image coding with context weights prediction. Signal Processing: Image Communication, 25(10):717– 728, 2010

  40. [48]

    Iou loss for 2d/3d object detection

    Dingfu Zhou, Jin Fang, Xibin Song, Chenye Guan, Junbo Yin, Yuchao Dai, and Ruigang Yang. Iou loss for 2d/3d object detection. In 2019 international conference on 3D vision (3DV), pages 85–94. IEEE, 2019

  41. [49]

    Tanet: Robust 3d object detection from point clouds with triple attention

    Zhe Liu, Xin Zhao, Tengteng Huang, Ruolan Hu, Yu Zhou, and Xiang Bai. Tanet: Robust 3d object detection from point clouds with triple attention. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 11677–11684, 2020

  42. [50]

    From points to parts: 3d object detection from point cloud with part-aware and part-aggregation network

    Shaoshuai Shi, Zhe Wang, Jianping Shi, Xiaogang Wang, and Hongsheng Li. From points to parts: 3d object detection from point cloud with part-aware and part-aggregation network. IEEE transactions on pattern analysis and machine intelligence, 43(8):2647–2664, 2020

  43. [51]

    Cia-ssd: Confident iou-aware single-stage object detector from point cloud

    Wu Zheng, Weiliang Tang, Sijin Chen, Li Jiang, and Chi-Wing Fu. Cia-ssd: Confident iou-aware single-stage object detector from point cloud. In Proceedings of the AAAI conference on artificial intelligence, volume 35, pages 3555–3562, 2021

  44. [52]

    Structure aware single-stage 3d object detection from point cloud

    Chenhang He, Hui Zeng, Jianqiang Huang, Xian-Sheng Hua, and Lei Zhang. Structure aware single-stage 3d object detection from point cloud. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11873–11882, 2020

  45. [53]

    Associate-3ddet: Perceptual-to-conceptual association for 3d point cloud object detection

    Liang Du, Xiaoqing Ye, Xiao Tan, Jianfeng Feng, Zhenbo Xu, Errui Ding, and Shilei Wen. Associate-3ddet: Perceptual-to-conceptual association for 3d point cloud object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13329–1...

  46. [54]

    Svga-net: Sparse voxel-graph attention net- work for 3d object detection from point clouds

    Qingdong He, Zhengning Wang, Hao Zeng, Yi Zeng, and Yijun Liu. Svga-net: Sparse voxel-graph attention net- work for 3d object detection from point clouds. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 870–878, 2022

  47. [55]

    Fast point r-cnn

    Yilun Chen, Shu Liu, Xiaoyong Shen, and Jiaya Jia. Fast point r-cnn. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9775–9784, 2019

  48. [56]

    Std: Sparse-to-dense 3d object detector for point cloud

    Zetong Yang, Yanan Sun, Shu Liu, Xiaoyong Shen, and Jiaya Jia. Std: Sparse-to-dense 3d object detector for point cloud. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1951–1960, 2019

  49. [57]

    Pv-rcnn: Point-voxel feature set abstraction for 3d object detection

    Shaoshuai Shi, Chaoxu Guo, Li Jiang, Zhe Wang, Jianping Shi, Xiaogang Wang, and Hongsheng Li. Pv-rcnn: Point-voxel feature set abstraction for 3d object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10529–10538, 2020

  50. [58]

    A unified query-based paradigm for point cloud understanding

    Zetong Yang, Li Jiang, Yanan Sun, Bernt Schiele, and Jiaya Jia. A unified query-based paradigm for point cloud understanding. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8541–8551, 2022

  51. [59]

    Vic-net: V oxelization information compensation network for point cloud 3d object detection

    Tianyuan Jiang, Nan Song, Huanyu Liu, Ruihao Yin, Ye Gong, and Jian Yao. Vic-net: V oxelization information compensation network for point cloud 3d object detection. In 2021 IEEE International Conference on Robotics and Automation (ICRA), pages 13408–13414. IEEE, 2021

  52. [60]

    Hvpr: Hybrid voxel-point representation for single-stage 3d object detection

    Jongyoun Noh, Sanghoon Lee, and Bumsub Ham. Hvpr: Hybrid voxel-point representation for single-stage 3d object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14605–14614, 2021

  53. [61]

    3d iou-net: Iou guided 3d object detector for point clouds

    Jiale Li, Shujie Luo, Ziqi Zhu, Hang Dai, Andrey S Krylov, Yong Ding, and Ling Shao. 3d iou-net: Iou guided 3d object detector for point clouds. arXiv preprint arXiv:2004.04962, 2020

  54. [62]

    Openpcdet: An open-source toolbox for 3d object detection from point clouds

    OpenPCDet Development Team. Openpcdet: An open-source toolbox for 3d object detection from point clouds. https://github.com/open-mmlab/OpenPCDet, 2020

  55. [63]

    Novel logic mining incorporating log linear approach

    Siti Zulaikha Mohd Jamaludin, Nurul Atiqah Romli, Mohd Shareduwan Mohd Kasihmuddin, Aslina Baharum, Mohd Asyraf Mansor, and Muhammad Fadhil Marsani. Novel logic mining incorporating log linear approach. Journal of King Saud University-Computer and Information Sciences, 34(10):...

  56. [64]

    Plenoxels: Radiance fields without neural networks

    Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5501–5510, 2022

  57. [65]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics (ToG), 42(4):1–14, 2023. 22

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.