Pith. sign in

REVIEW 4 major objections 5 minor 64 references

3DPillars: Pillar-based two-stage 3D object detection

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims a two-stage pillar-based 3D detector that processes voxel features as stacks of pseudo images can match slower voxel-based detectors while running in real time.

desk verdict Solid, well-ablated two-stage pillar detector; the headline claims oversell it and SVFM's exact computation needs to be pinned down before the efficiency story is credible. read the letter →

arxiv 2509.05780 v1 pith:DKZVI2FZ submitted 2025-09-06 cs.CV

Jongyoun Noh , Junghyup Lee , Hyekang Park , Bumsub Ham This is my paper · ORCID
classification cs.CV
keywords 3DobjectdetectionLiDARpointcloudspseudoimagerepresentationpillar-basedtwo-stagedetectorseparablevoxelfeaturemoduleautonomousdriving
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

The paper tries to establish that a pillar-style 3D detector, the fastest known family, does not have to give up accuracy to keep its speed. It claims the two reasons PointPillars lags behind state-of-the-art detectors are fixable: the pseudo-image representation collapses vertical structure, and 3D proposals cannot be refined in a second stage on a 2D feature map. The paper's answer is 3DPillars, a backbone that treats the 3D voxel feature volume as three stacks of pseudo images (bird's-eye, side, front) and runs cheap 2D convolutions on each, plus an RoI head that pools multi-scale features and a shared memory of scene contexts to refine proposals. On KITTI it reports large gains over PointPillars while running at 29.6 Hz, and it reports top cyclist and pedestrian results on the KITTI validation set. The significance, if true, is that real-time 3D perception for driving does not require choosing between speed and close-to-voxel accuracy.

What carries the argument

Two named modules carry the argument. SVFM (separable voxel feature module) treats the 4D voxel feature volume as a stack of pseudo images and decomposes a 3D convolution of size k×k×k into three 2D convolutions, 1×k×k over the X-Y plane, k×1×k over Y-Z, and k×k×1 over X-Z, so view-specific features are learned with 2D filters; sequential ordering of the three views outperforms parallel ordering in the paper's ablations. S2CFM (sparse scene context feature module) concatenates initial VFE features with trilinearly interpolated multi-scale backbone features at non-empty voxel positions to build a sparse scene feature, pools sub-RoI features via the voxel RoI pooling scheme, and uses a key-val

What would settle it

Run the same two-stage detector on KITTI with the SVFM backbone replaced by standard sparse 3D convolutions at matched channel counts, keeping the S2CFM RoI head and training schedule fixed. If the 3D-convolution version matches or exceeds the reported mAP while the separable version's claimed memory and latency advantage fails to appear under a profiler, the central claim that 2D axis-decomposed convolutions preserve needed 3D structure at lower cost is not supported. A second, quicker check: disable the side- and front-view pseudo-image stacks, keeping only the BEV stack; if accuracy hardly

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a two-stage 3D object detector can be built entirely on pseudo image representations, something previously available only to voxel-based or point-based pipelines. The authors state this is the first two-stage framework exploiting pseudo images, and they attribute its success to two components: SVFM, which splits the sparse voxel feature volume along the X, Y, and Z axes and applies 2D convolutions (1×k×k, k×1×k, k×k×1) to each stack, preserving 3D structure without 3D convolutions; and S2CFM, which forms a sparse scene feature from the initial voxel features plus multi-scale backbone features, pools RoI features from it, and enriches them with

Load-bearing premise

The whole accuracy-versus-speed trade-off rests on the claim that splitting the 3D voxel feature grid into three stacks of pseudo images and applying 2D convolutions along each axis captures enough 3D structure to match 3D convolutions, yet the paper never specifies how empty voxels are treated, what exactly collapsing the height dimension does, or the actual FLOPs and memory compared with sparse 3D convolutions.

Editorial extensions

If this is right

  • Two-stage refinement becomes available to any pillar-based or pseudo-image detector, not just sparse-3D-CNN pipelines.
  • Small and distant objects, such as cyclists, pedestrians, and far vehicles, benefit most because multi-scale pooling and global context compensate for sparse point clouds.
  • Real-time operation is preserved: 29.6 Hz on KITTI with 8.1M parameters, and the added RoI head costs about 6 ms and 115 MB over the single-stage backbone.
  • The same architecture transfers across KITTI, Waymo Open, and nuScenes with competitive BEV-family results, indicating the gains are not dataset-specific.
  • Sequential rather than parallel processing of the three pseudo-image views yields better accuracy, so how the views interact matters as much as their existence.

Reading between the lines

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

  • The axis-decomposed 2D convolution idea is a general recipe: any 3D feature volume on a dense grid could be processed as three view stacks, which may transfer to other perception tasks such as 3D segmentation or occupancy prediction at lower cost than 3D convolutions.
  • Because the paper leaves open how empty voxels are handled in SVFM, the reported efficiency depends on whether the volume is processed densely; a sparse-aware implementation of the same decomposition could be a natural next step and might close the remaining gap to fully sparse voxel transformers.
  • The paper's arithmetic for parameter reduction (45% for k=3) does not match 3·k² versus k³, so the actual savings come from channel and depth structure rather than the kernel decomposition alone; a fair comparison would report FLOPs and memory, not just parameter counts.
  • The shared key-value memory of scene prototypes suggests a learned prior over driving environments; one testable extension is to update or condition the memory on the current scene's layout, such as map or lane information, to reduce context ambiguity like the reported garden-car false positive.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 3DPillars, a two-stage LiDAR-based 3D object detector built on pseudo image representations. Its first stage (SVFM) treats the sparse voxel feature volume as stacks of pseudo images and applies 2D convolutions along the X, Y, and Z axes instead of 3D convolutions, in order to preserve 3D structure while retaining efficiency. The second stage (S2CFM) aggregates multi-scale voxel features into a sparse scene feature and refines RoI features with a key-value memory that provides global context. Experiments are reported on KITTI, Waymo Open, and nuScenes, with a runtime of 29.6 Hz on KITTI and substantial gains over PointPillars. The paper's central claim is that this is the first two-stage pseudo-image-based 3D detector that narrows the accuracy gap to voxel-based state-of-the-art methods while retaining real-time speed.

Significance. If the method and measurements hold up, the paper would offer a useful speed/accuracy trade-off for pillar-based 3D detection: a two-stage detector that avoids 3D convolutions and still reaches real-time on KITTI. The paper has genuine strengths: it evaluates on three datasets (KITTI, Waymo, nuScenes), provides component-wise ablations (Tables 6–9), reports a runtime breakdown (Table 11), and the design of the RoI head with a memory module is a plausible contribution. However, the central efficiency and novelty claims contain errors and ambiguities that must be resolved before the contribution can be assessed: the SVFM description is underspecified, the stated parameter reduction is arithmetically false, and the 'first' and 'best at real-time' claims are contradicted by the paper's own tables.

major comments (4)
  1. [§3.1.2, Fig. 3] The SVFM specification is ambiguous and load-bearing. The text says SVFM 'applies 2D convolutions to each slide' but then says it 'reshapes the 4D features ... by collapsing the height dimension' and applies 2D convolutions along X-Y. These are different operations: per-slice convolutions scale FLOPs with the number of slices, while collapsing height changes the filter's receptive field and contradicts the stated 1×k×k kernel shape. The paper also never states how empty voxels are handled, whether the convolutions are sparse, dense, or submanifold, or what the actual FLOPs/memory are. Without this specification, the reported 8.9 ms backbone time and the claim that SVFM avoids 3D convolutions cannot be verified.
  2. [§3.1.2, parameter count] The claimed '45% parameter reduction compared to the 3D case when k = 3' is arithmetically incorrect. Three 2D kernels with sizes 1×k×k, k×1×k, and k×k×1 contain 3·k^2 parameters; a single 3D kernel k×k×k contains k^3. For k=3 both are 27, so there is zero parameter reduction. A reduction only appears for k>3. This quantitative pillar of the efficiency story needs to be corrected or removed.
  3. [Abstract, §1, Table 1] The claim of being the 'first two-stage 3D detection framework exploiting pseudo image representations' is contradicted by Table 1, which lists LiDAR R-CNN as a BEV-based two-stage method (35.6 Hz). In the same table, the statement that 'ours achieves the best mAP across all classes at real-time speed' is falsified by Voxel R-CNN (25.2 Hz) with higher Car Easy/Moderate mAP, and by PVT-SSD (20.4 Hz) with higher Car Easy/Moderate/Hard mAP, both above the 20 Hz real-time threshold used in the paper. These claims should be narrowed or qualified.
  4. [§4.2.2, Table 3] On Waymo Open, the gap to recent fully sparse voxel methods is substantial (e.g., vehicle L2 mAP 70.42 vs 73.10 for ScatterFormer and 72.70 for SAFDNet; pedestrian L2 mAP 73.12 vs 77.00 and 77.30). The text acknowledges this, but the abstract's 'narrowing the performance gaps ... while retaining its efficiency' overstates the case without also reporting the corresponding efficiency comparison for those methods. The accuracy/efficiency trade-off should be presented more carefully.
minor comments (5)
  1. [§3.3, Eq. (10)] The confidence loss is denoted \mathcal{L}_{cfd} but the text says 'confidence (cfd)'; likely a typo for \mathcal{L}_{conf} or similar.
  2. [§4.3.6, Table 10] The paper says SVFM 'reduces the number of parameters significantly by replacing the expensive 3D operations with 2D convolutions,' but no comparison of SVFM's parameter count against a 3D-convolution baseline is actually reported. Table 10 compares whole detectors, not the SVFM vs a 3D-convolution backbone under matched design.
  3. [§4.2.2] Typos: 'perfomance' and 'PoinPillars' should be corrected.
  4. [Fig. 4] The four SVFM variants are described in the caption and text, but it would help to specify the exact tensor reshape/grouping used for each variant, especially for the sequential-parallel and parallel-sequential cases, to make the ablation reproducible.
  5. [§4.1.4] Key and value memory sizes are tuned on the validation split (K=10,V=50 on KITTI; K=20,V=200 on Waymo). This should be stated as validation-tuned hyperparameters, which the paper does, but the sensitivity analysis in Fig. 6 is only for V and only on KITTI. Reporting sensitivity to K would strengthen the memory module's robustness claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims rest on held-out benchmark measurements, not on self-referential derivation.

full rationale

The paper (arXiv:2509.05780) is an empirical architecture paper. The headline claims—two-stage performance gains over PointPillars, real-time runtime, competitiveness on KITTI/Waymo/nuScenes—are supported by benchmark results on test/validation splits compared against external baselines. No equation derives a prediction from the same data used to fit it. Hyperparameters such as memory sizes K and V are explicitly tuned on the KITTI validation split (Sec. 4.1.4) and reported as such; this is standard model selection, not a fitted input disguised as a prediction. The SVFM and S2CFM modules are ablated in Sec. 4.3.1–4.3.5; those ablations compare measured mAP under different architectural choices, so there is no reduction-by-construction. Self-citations appear only as related work and baselines (e.g., HVPR in Tables 1 and 2), not as load-bearing justification. No uniqueness theorem or prior result by the same authors is invoked to force a choice. The limitations section (Sec. 5) candidly documents failure cases on distant sparse objects. Two non-circular concerns are worth noting outside the circularity score: (1) the claimed '45% parameter reduction' for k=3 (Sec. 3.1.2) is arithmetically inconsistent with the three stated 2D kernel sizes, which together contain 3*3*3 = 27 parameters, equal to the 3D kernel; (2) how SVFM handles empty voxels and 'collapsing the height dimension' is underspecified. These are correctness/reproducibility issues, not circular reasoning.

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

The central claim depends on empirical assumptions about 2D-conv approximations of 3D structure, shared scene contexts, and fair benchmark comparisons. Free parameters (memory sizes, loss weight, SVFM variant) are tuned on validation splits or adopted from prior work; this is standard but slightly weakens the generalization claim. No invented physical entities are introduced, only new network modules.

free parameters (5)
  • Key memory size K = 10 (KITTI), 20 (Waymo)
    Set by grid search on KITTI validation mAP (Sec 4.1.4); adjusted for Waymo based on training set size.
  • Value memory size V = 50 (KITTI), 200 (Waymo)
    Grid-searched on KITTI validation (Sec 4.1.4, Fig. 6); adjusted for Waymo.
  • Memory loss weight lambda_mem = 0.5
    Chosen by hand (Sec 4.1.3); no sensitivity analysis reported beyond memory size.
  • SVFM variant (sequential) = sequential
    Selected by validation ablation among four variants (Table 7); the paper notes this choice is empirical.
  • Voxel sizes = KITTI (0.16, 0.16, 0.25); Waymo (0.16, 0.16, 0.3); nuScenes (0.2, 0.2, 0.4)
    Adopted from prior work (PointPillars, PillarNet) rather than fitted, but they define the representation and are hand-set hyperparameters of the central system.
assumptions (4)
  • domain assumption A sparse 3D voxel feature volume can be processed as a stack of dense pseudo images with axis-decomposed 2D convolutions without explicit sparsity handling
    Core of SVFM, used throughout Sec 3.1.2; if false, the efficiency or accuracy claim of the backbone fails.
  • domain assumption Outdoor scenes share prototypical global contexts that can be stored in a key-value memory shared across scenes
    Motivates the memory module in Sec 3.2.2; supported only by in-paper ablations, not by external evidence.
  • domain assumption Voxel RoI pooling (Deng et al. 2021) transfers to features produced by a 2D-conv backbone
    Used in Sec 3.2.1 without modification; relies on the multi-scale sparse scene feature having a voxel-like structure.
  • standard math Trilinear interpolation, cross-attention, and smooth-L1 losses behave as standard
    Standard building blocks invoked in Secs 3.2.1-3.3; no proof needed for the central claim.
invented entities (2)
  • SVFM (separable voxel feature module)
    purpose: Extract voxel-based 3D features using 2D convolutions applied along X, Y, Z axes of a pseudo-image stack, replacing 3D convolutions
    No external evidence outside this paper; its benefit is shown only through in-paper ablations on KITTI and Waymo.
  • S2CFM with key-value memory (sparse scene context feature module)
    purpose: Aggregate multi-scale sparse scene features, pool RoI features, and refine them with global context prototypes
    Validated by ablation tables within the paper; the memory concept is borrowed from prior work, but the specific module is new and lacks independent validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 3DPillars: Pillar-based two-stage 3D object detection." pith.science (2026). https://pith.science/paper/DKZVI2FZ

@misc{pith2026250905780,
  author       = {Pith},
  title        = {Pith review of: 3DPillars: Pillar-based two-stage 3D object detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DKZVI2FZ}},
  note         = {Machine review of arXiv:2509.05780}
}
read the original abstract

PointPillars is the fastest 3D object detector that exploits pseudo image representations to encode features for 3D objects in a scene. Albeit efficient, PointPillars is typically outperformed by state-of-the-art 3D detection methods due to the following limitations: 1) The pseudo image representations fail to preserve precise 3D structures, and 2) they make it difficult to adopt a two-stage detection pipeline using 3D object proposals that typically shows better performance than a single-stage approach. We introduce in this paper the first two-stage 3D detection framework exploiting pseudo image representations, narrowing the performance gaps between PointPillars and state-of-the-art methods, while retaining its efficiency. Our framework consists of two novel components that overcome the aforementioned limitations of PointPillars: First, we introduce a new CNN architecture, dubbed 3DPillars, that enables learning 3D voxel-based features from the pseudo image representation efficiently using 2D convolutions. The basic idea behind 3DPillars is that 3D features from voxels can be viewed as a stack of pseudo images. To implement this idea, we propose a separable voxel feature module that extracts voxel-based features without using 3D convolutions. Second, we introduce an RoI head with a sparse scene context feature module that aggregates multi-scale features from 3DPillars to obtain a sparse scene feature. This enables adopting a two-stage pipeline effectively, and fully leveraging contextual information of a scene to refine 3D object proposals. Experimental results on the KITTI and Waymo Open datasets demonstrate the effectiveness and efficiency of our approach, achieving a good compromise in terms of speed and accuracy.

Figures

Figures reproduced from arXiv: 2509.05780 by the authors.

Figure 1
Figure 1. Limitations of PointPillars (Lang et al., 2019). A pseudo image representation in PointPillars allows to exploit 2D convolutions for efficient 3D object detection. Left: The pseudo image representation however collapses point clouds in vertical pillars, and thus PointPillars does not preserve precise 3D structures of a scene. Right: 3D object proposals cannot be projected onto BEV features directly, suggesting that … view at source ↗
Figure 2
Figure 2. An overview of our approach. Our two-stage framework mainly consists of 3DPillars and an RoI head with S2CFM. 3DPillars extracts multi-scale features from 3D voxels, without using 3D convolutions. To this end, SVFM considers the features from VFE as a stack of pseudo images, and extracts view-specific features by applying a 2D convolution to each slice of the stack along X-, Y- and Z-axes ( [PITH_FULL_IMAGE:figures… view at source ↗
Figure 4
Figure 4. Illustration of different versions of SVFM. The sequential SVFM in (a) learns view-specific features con￾secutively, while the parallel one in (b) aggregates the features simultaneously. The sequential-parallel and parallel￾sequential SVFMs in (c) and (d), respectively, consist of two parts that extract BEV-specific and height-related features. The former extracts the features sequentially, while the latter performs… view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Illustration of S2CFM. S2CFM aggregates multi-scale features and a stack of pseudo images into a sparse scene feature. It then pools RoI features from the sparse scene feature and combines each RoI feature with global context features to obtain a context-aware RoI repr…
Figure 6
Figure 6. Figure 6: Quantitative comparison according to the size of a value memory (𝑉 ). We compute mAP(%) with 40 recall positions on the validation split of KITTI (Geiger et al., 2012). 4.3.4. Global context feature We show in [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Qualitative results on the validation set of KITTI (Geiger et al., 2012). We show our predictions and ground-truth bounding boxes as red and green boxes, respectively. In particular, our model with all components localizes small objects well consisting of sparse point …
Figure 8
Figure 8. Figure 8: Qualitative comparison with the state of the art on the validation set of KITTI (Geiger et al., 2012). We show ground-truth bounding boxes as green boxes, and predictions as red, black, and yellow boxes for car, pedestrian, and cyclist, respectively. We can clearly see…
Figure 9
Figure 9. Figure 9: Qualitative results on the validation set of the Waymo Open Dataset (Sun et al., 2020). We show ground-truth bounding boxes as green boxes, while our predictions for the vehicle class are represented with red boxes. We visualize predictions of our model for the cyclist…
Figure 10
Figure 10. Figure 10: Qualitative results on the validation set of KITTI (Geiger et al., 2012). We show ground-truth bounding boxes as green boxes, while our predictions for the vehicle class are represented with red boxes. We visualize our predictions for the cyclist and pedestrian classe…
Figure 11
Figure 11. Figure 11: Qualitative results on the validation set of KITTI (Geiger et al., 2012). We visualize (a) the RGB image, (b) the input point cloud with ground-truth bounding boxes, and (c) the scene feature activations with predicted object boxes. Ground-truth boxes are shown in gre…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 64 canonical work pages

  1. [1]

    H., V ora, S., Liong, V

    Caesar, H., Bankiti, V ., Lang, A. H., V ora, S., Liong, V . E., Xu, Q., Krishnan, A., Pan, Y ., Baldan, G., & Beijbom, O. (2020). nuScenes: A multimodal dataset for autonomous driving. In CVPR

    work page 2020

  2. [2]

    Chen, Q., Sun, L., Wang, Z., Jia, K., & Yuille, A. (2020). Object as hotspots: An anchor-free 3D object detection approach via firing of hotspots. In ECCV

    work page 2020

  3. [3]

    G., Ma, H., Fidler, S., & Urtasun, R

    Chen, X., Kundu, K., Zhu, Y ., Berneshawi, A. G., Ma, H., Fidler, S., & Urtasun, R. (2015). 3D object proposals for accurate object class detection. In NeurIPS

    work page 2015

  4. [4]

    Chen, Y ., Li, Y ., Zhang, X., Sun, J., & Jia, J. (2022). Focal sparse convolutional networks for 3D object detection. In CVPR

    work page 2022

  5. [5]

    Chen, Y ., Liu, S., Shen, X., & Jia, J. (2019). Fast point R-CNN. InICCV

    work page 2019

  6. [6]

    Deng, J., Shi, S., Li, P., Zhou, W., Zhang, Y ., & Li, H. (2021). V oxel R-CNN: Towards high performance voxel-based 3D object detection. In AAAI

    work page 2021

  7. [7]

    Du, L., Ye, X., Tan, X., Feng, J., Xu, Z., Ding, E., & Wen, S. (2020). Associate-3Ddet: Perceptual-to-conceptual association for 3D point cloud object detection. In CVPR

    work page 2020

  8. [8]

    Fan, L., Pang, Z., Zhang, T., Wang, Y .-X., Zhao, H., Wang, F., Wang, N., & Zhang, Z. (2022). Embracing single stride 3D object detector with sparse transformer. In CVPR

    work page 2022

Show all 64 references
  1. [9]

    Geiger, A., Lenz, P., & Urtasun, R. (2012). Are we ready for autonomous driving? the KITTI vision benchmark suite. In CVPR

    2012

  2. [10]

    R., Venkatesh, S., & Hengel, A

    Gong, D., Liu, L., Le, V ., Saha, B., Mansour, M. R., Venkatesh, S., & Hengel, A. v. d. (2019). Memorizing normality to detect anomaly: Memory- augmented deep autoencoder for unsupervised anomaly detection. In ICCV

    2019

  3. [11]

    Guang, J., Hu, Z., Wu, S., Zhang, Q., & Liu, J. (2024). Rpea: A residual path network with efficient attention for 3d pedestrian detection from lidar point clouds. Expert Systems with Applications, 249, 123497

    2024

  4. [12]

    Guo, C., Fan, B., Zhang, Q., Xiang, S., & Pan, C. (2020). AugFPN: Improving multi-scale feature learning for object detection. In CVPR

    2020

  5. [13]

    Guo, C., Pleiss, G., Sun, Y ., & Weinberger, K. Q. (2017). On calibration of modern neural networks

    2017

  6. [14]

    He, C., Li, R., Li, S., & Zhang, L. (2022). V oxel set transformer: A set-to-set approach to 3D object detection from point clouds. In CVPR

    2022

  7. [15]

    He, C., Li, R., Zhang, G., & Zhang, L. (2024). Scatterformer: Efficient voxel transformer with scattered linear attention. In ECCV: Springer

    2024

  8. [16]

    He, C., Zeng, H., Huang, J., Hua, X.-S., & Zhang, L. (2020). Structure aware single-stage 3D object detection from point cloud. In CVPR

    2020

  9. [17]

    Hu, J. S. K., Kuai, T., & Waslander, S. L. (2022). Point density-aware voxels for lidar 3D object detection. In CVPR

    2022

  10. [18]

    & Szegedy, C

    Ioffe, S. & Szegedy, C. (2015). Batch normalization: Accelerating deep network training by reducing internal covariate shift. In ICML

    2015

  11. [19]

    Konushin, A. (2025). Unidet3d: Multi-dataset indoor 3d object detection. In AAAI

    2025

  12. [20]

    Kong, T., Sun, F., Yao, A., Liu, H., Lu, M., & Chen, Y . (2017). Ron: Reverse connection with objectness prior networks for object detection. In CVPR

    2017

  13. [21]

    Koo, I., Lee, I., Kim, S.-H., Kim, H.-S., Jeon, W.-j., & Kim, C. (2023). PG-RCNN: Semantic surface point generation for 3d object detection. In ICCV

    2023

  14. [22]

    H., V ora, S., Caesar, H., Zhou, L., Yang, J., & Beijbom, O

    Lang, A. H., V ora, S., Caesar, H., Zhou, L., Yang, J., & Beijbom, O. (2019). PointPillars: Fast encoders for object detection from point clouds. In CVPR

    2019

  15. [23]

    Li, J., Luo, C., & Yang, X. (2023). PillarNeXt: Rethinking network designs for 3d object detection in lidar point clouds. In CVPR

    2023

  16. [24]

    Li, X., Wang, C., & Zeng, Z. (2024). Ws-ssd: Achieving faster 3d object detection for autonomous driving via weighted point cloud sampling. Expert Systems with Applications, 249, 123805

    2024

  17. [25]

    Li, Y ., Pang, Y ., Shen, J., Cao, J., & Shao, L. (2020). NETNet: Neighbor erasing and transferring network for better single shot object detection. In CVPR

    2020

  18. [26]

    Li, Z., Wang, F., & Wang, N. (2021). LiDAR R-CNN: An efficient and universal 3D object detector. In CVPR

    2021

  19. [27]

    Lin, T.-Y ., Goyal, P., Girshick, R., He, K., & Dollár, P. (2017). Focal loss for dense object detection. In ICCV

    2017

  20. [28]

    Liu, Z., Zhao, X., Huang, T., Hu, R., Zhou, Y ., & Bai, X. (2020). TANet: Robust 3D object detection from point clouds with triple attention. In AAAI

    2020

  21. [29]

    Zhan, X. (2021). PVGNet: A bottom-up one-stage 3D object detector with integrated multi-level features. In CVPR

    2021

  22. [30]

    Miller, A., Fisch, A., Dodge, J., Karimi, A.-H., Bordes, A., & Weston, J. (2016). Key-value memory networks for directly reading documents. In EMNLP

    2016

  23. [31]

    Noh, J., Lee, S., & Ham, B. (2021). HVPR: Hybrid voxel-point representation for single-stage 3D object detection. In CVPR

    2021

  24. [32]

    E., & Huang, G

    Pan, X., Xia, Z., Song, S., Li, L. E., & Huang, G. (2021). 3d object detection with pointformer. In CVPR. Noh et al.: Preprint submitted to Elsevier Page 18 of 19

    2021

  25. [33]

    R., Liu, W., Wu, C., Su, H., & Guibas, L

    Qi, C. R., Liu, W., Wu, C., Su, H., & Guibas, L. J. (2018). Frustum PointNets for 3D object detection from RGB-D data. In CVPR

    2018

  26. [34]

    Qin, Z., Li, Z., Zhang, Z., Bao, Y ., Yu, G., Peng, Y ., & Sun, J. (2019). ThunderNet: Towards real-time generic object detection on mobile devices. In ICCV

    2019

  27. [35]

    Sheng, H., Cai, S., Liu, Y ., Deng, B., Huang, J., Hua, X.-S., & Zhao, M.-J. (2021). Improving 3D object detection with channel-wise transformer. In ICCV

    2021

  28. [36]

    Sheng, H., Cai, S., Zhao, N., Deng, B., Huang, J., Hua, X.-S., Zhao, M.-J., & Lee, G. H. (2022). Rethinking iou-based optimization for single-stage 3d object detection. In ECCV

    2022

  29. [37]

    Shi, G., Li, R., & Ma, C. (2022). Pillarnet: Real-time and high-performance pillar-based 3D object detection. In ECCV: Springer

    2022

  30. [38]

    Shi, S., Jiang, L., Deng, J., Wang, Z., Guo, C., Shi, J., Wang, X., & Li, H. (2023). PV-RCNN++: Point-voxel feature set abstraction with local vector representation for 3D object detection. IJCV, 131(2), 531–551

    2023

  31. [39]

    Shi, S., Wang, X., & Li, H. (2019). PointRCNN: 3D object proposal generation and detection from point cloud. In CVPR

    2019

  32. [40]

    & Rajkumar, R

    Shi, W. & Rajkumar, R. (2020). Point-GNN: Graph neural network for 3D object detection in a point cloud. In CVPR

    2020

  33. [41]

    Simon, M., Milzy, S., Amendey, K., & Gross, H.-M. (2018). Complex- YOLO: An euler-region-proposal for real-time 3D object detection on point clouds. In ECCVW

    2018

  34. [42]

    Smith, L. N. & Topin, N. (2019). Super-convergence: Very fast training of neural networks using large learning rates. In Artificial intelligence and machine learning for multi-domain operations applications

    2019

  35. [43]

    Sukhbaatar, S., Weston, J., Fergus, R., et al. (2015). End-to-end memory networks. In NeurIPS

    2015

  36. [44]

    Sun, P., Kretzschmar, H., Dotiwalla, X., Chouard, A., Patnaik, V ., Tsui, P., Guo, J., Zhou, Y ., Chai, Y ., Caine, B., et al. (2020). Scalability in perception for autonomous driving: Waymo open dataset. In CVPR

    2020

  37. [45]

    Sun, P., Tan, M., Wang, W., Liu, C., Xia, F., Leng, Z., & Anguelov, D. (2022). SWFormer: Sparse window transformer for 3D object detection in point clouds. In ECCV: Springer

    2022

  38. [46]

    Wang, C., Yang, W., Liu, X., & Zhang, T. (2025). State space model meets transformer: A new paradigm for 3d object detection. In ICLR

    2025

  39. [47]

    Wang, T., Hu, X., Liu, Z., & Fu, C.-W. (2022). Sparse2Dense: Learning to densify 3D features to boost 3D object detection. In NeurIPS

    2022

  40. [48]

    Wang, W., Lu, Y ., Zheng, G., Zhan, S., Ye, X., Tan, Z., Wang, J., Wang, G., & Li, X. (2024). Bevspread: Spread voxel pooling for bird’s-eye-view representation in vision-based roadside 3d object detection. In CVPR

    2024

  41. [49]

    A., Pantofaru, C., Funkhouser, T., & Solomon, J

    Wang, Y ., Fathi, A., Kundu, A., Ross, D. A., Pantofaru, C., Funkhouser, T., & Solomon, J. (2020). Pillar-based object detection for autonomous driving. In ECCV

    2020

  42. [50]

    & Jia, K

    Wang, Z. & Jia, K. (2019). Frustum ConvNet: Sliding frustums to aggregate local point-wise features for amodal 3D object detection. In IROS

    2019

  43. [51]

    Weston, J., Chopra, S., & Bordes, A. (2015). Memory networks. In ICLR

    2015

  44. [52]

    Wu, X., Shi, G., Zhao, Z., Li, M., Gao, X., & Yan, X. (2024). Dccn: A dual-cross contrastive neural network for 3d point cloud representation learning. Expert Systems with Applications, 249, 123564

    2024

  45. [53]

    Yan, Y ., Mao, Y ., & Li, B. (2018). SECOND: Sparsely embedded convolutional detection. Sensors, 18(10)

    2018

  46. [54]

    Yang, B., Luo, W., & Urtasun, R. (2018). PIXOR: Real-time 3D object detection from point clouds. In CVPR

    2018

  47. [55]

    Yang, H., Wang, W., Chen, M., Lin, B., He, T., Chen, H., He, X., & Ouyang, W. (2023). PVT-SSD: Single-stage 3d object detector with point-voxel transformer. In CVPR

    2023

  48. [56]

    Yang, Z., Sun, Y ., Liu, S., & Jia, J. (2020). 3DSSD: Point-based 3D single stage object detector. In CVPR

    2020

  49. [57]

    Yang, Z., Sun, Y ., Liu, S., Shen, X., & Jia, J. (2019). STD: Sparse-to-dense 3D object detector for point cloud. In ICCV

    2019

  50. [58]

    Ye, M., Xu, S., & Cao, T. (2020). HVNet: Hybrid voxel network for lidar based 3D object detection. In CVPR

    2020

  51. [59]

    Yin, T., Zhou, X., & Krahenbuhl, P. (2021). Center-based 3D object detection and tracking. In CVPR

    2021

  52. [60]

    Zhang, G., Chen, J., Gao, G., Li, J., Liu, S., & Hu, X. (2024). SAFDNet: A simple and effective network for fully sparse 3d object detection. In CVPR

    2024

  53. [61]

    Zhang, H., Zhang, H., Wang, C., & Xie, J. (2019). Co-occurrent features in semantic segmentation. In CVPR

    2019

  54. [62]

    Zhao, H., Shi, J., Qi, X., Wang, X., & Jia, J. (2017). Pyramid scene parsing network. In CVPR

    2017

  55. [63]

    & Tuzel, O

    Zhou, Y . & Tuzel, O. (2018). V oxelNet: End-to-end learning for point cloud based 3D object detection. In CVPR

    2018

  56. [64]

    Zhou, Z., Zhao, X., Wang, Y ., Wang, P., & Foroosh, H. (2022). Centerformer: Center-based transformer for 3D object detection. In ECCV: Springer. Noh et al.: Preprint submitted to Elsevier Page 19 of 19

    2022

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.