Pith. sign in

REVIEW 3 major objections 6 minor 64 references

Collaborative Perceiver: Elevating Vision-based 3D Object Detection via Local Density-Aware Spatial Occupancy

T0 review · 3 major / 6 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 The paper claims that training a camera-only BEV 3D detector jointly with local-density-aware 3D occupancy prediction improves both tasks, reaching 49.5% mAP and 59.2% NDS on the nuScenes test set and outperforming prior vision-based…

desk verdict Plausible method, but the central numbers don't line up: the same base config is 53.5/42.2 in one table and 52.6/41.1 in another, so the claimed gains are not yet reliable. read the letter →

arxiv 2507.21358 v4 pith:TTSTAVBY submitted 2025-07-28 cs.CV

Jicheng Yuan , Manh Nguyen Duc , Qian Liu , Manfred Hauswirth , Danh Le Phuoc This is my paper
classification cs.CV
keywords 3Dobjectdetectionbird's-eye-viewoccupancypredictionmulti-tasklearninglocaldensityvoxel-heightsamplingfeaturefusionnuScenes
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 argues that camera-only 3D object detection gets better when the detector is trained alongside a second task: predicting dense 3D occupancy with voxel-wise local point-density weights. The proposed framework, Collaborative Perceiver (CoP), builds on a BEVDet4D-style baseline and reports 49.5% mAP and 59.2% NDS on the nuScenes test set, gains of +2.1 mAP and +1.1 NDS over the cited SOGDet baseline. The central idea is that dense, density-aware occupancy ground truths generated from accumulated multi-frame LiDAR give the shared bird's-eye-view features fine-grained structural information that box-level supervision alone omits. If the claim holds, occupancy prediction becomes a practical offline supervision source for vision-based detection, and the same labels improve existing occupancy models by small but consistent mIoU margins.

What carries the argument

The load-bearing mechanism is the local-density-aware dense occupancy (LDO) ground-truth generation pipeline. It accumulates LiDAR from multiple frames, separates static and dynamic points, aligns static points with ego poses and dynamic points with ground-truth bounding-box indices, voxelizes the merged cloud, and computes a per-voxel density weight $W_d^{ki} = P_d^{ki} / \sum_{j} P_d^{kj}$ that reweights the occupancy loss. On the feature side, VHS voxel-pools lifted 3D features over height intervals of interest derived from occupied voxels, and CFF fuses the resulting local features $f_l$ with height-collapsed global features $f_g$ through an adaptive scalar $\alpha = \sigma(\mathrm{MLP}(\mathrm{pool}(\mathrm{conv}(f_l))) + \mathrm{MLP}(\mathrm{pool}(\mathrm{conv}(f_g))))$, producing unified BEV features $f_u = \alpha \cdot \mathrm{conv}(f_g) + (1-\alpha) \cdot \mathrm{conv}(f_l)$, which are then re-elevated to voxel space by a channel-to-height transform.

What would settle it

Train CoP with the same settings but deliberately perturb the LDO generation: randomly reassign bounding-box indices or add moderate Gaussian noise to ego poses when aligning multi-frame LiDAR, then measure mAP/NDS on nuScenes validation. If performance does not drop, the claim that accurate density-aware occupancy alignment drives the detection gains would be falsified; if it drops sharply, the alignment assumption is confirmed as load-bearing.

Watch

Extended reading notes

Core claim

CoP's central claim is that multi-task learning with occupancy prediction elevates vision-only BEV detection rather than merely adding a head. The paper attributes the gain to three components: local-density-aware dense occupancy (LDO) labels, which encode per-voxel point-density variations; voxel-height-guided sampling (VHS), which pools lifted features over height intervals of interest; and global-local collaborative feature fusion (CFF), which adaptively selects between pooled global and locally sampled features. On nuScenes test with ResNet-101, CoP reports 49.5% mAP and 59.2% NDS, and on validation with ResNet-50 it reports 42.2% mAP and 53.5% NDS. Ablations isolate the components: replacing sparse occupancy with dense occupancy adds +1.5 mAP, adding local-density weighting adds +0.4 mAP, adding VHS adds +1.4 mAP, and the full fusion reaches 41.1% mAP versus 37.3% for the sparse baseline. The LDO labels also raise semantic scene completion mIoU of two existing occupancy models by +0.3 and +0.4 points.

Load-bearing premise

The load-bearing assumption is that the automatically generated dense occupancy labels are accurate enough to supervise shared features: dynamic objects are aligned by ground-truth bounding-box indices and static points by ego poses, and the paper provides no quantitative check of those labels before training the detector.

Editorial extensions

If this is right

  • If CoP is right, occupancy prediction can be used as an offline auxiliary supervision for camera-only 3D detection, so the LiDAR used to build labels is not needed at inference.
  • The same LDO labels transfer to other occupancy models, so dense occupancy training can be improved without changing model architecture or inference cost.
  • Height-aware local sampling should reduce detection errors on objects with different vertical extents, matching the reported drops in translation and velocity errors when local pooling is used.
  • Because the framework produces both boxes and voxel semantics from one BEV representation, a single model can serve detection and occupancy-based downstream tasks such as drivable-surface reasoning.
  • On a standard validation split with ResNet-50, CoP exceeds the cited prior method by +2.2 mAP and +1.9 NDS, suggesting the gains are not dependent on a large backbone.

Reading between the lines

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

  • A testable extension the paper does not run: corrupting the LDO labels by randomly reassigning dynamic-object tracklets or adding pose noise should degrade mAP proportionally to the paper's reliance on alignment; if it does not, the alignment part of the pipeline is not the source of gain.
  • The reported +0.3 and +0.4 mIoU gains on existing occupancy models are small; a natural class-wise analysis would show whether the density weighting mostly helps sparse objects such as pedestrians and traffic cones rather than large surfaces.
  • Since LDO is generated offline from annotations, the method's benefit likely depends on datasets with consistent instance IDs; on unlabeled or noisy-annotation data, pseudo-label-based versions would be a natural next step, but the paper does not test them.
  • The CFF fusion can be read as a learned routing between global and local features; a possible extension would make the routing per class or per height layer, which the paper leaves unexplored.
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 / 6 minor

Summary. The paper proposes Collaborative Perceiver (CoP), a multi-task learning framework for camera-based 3D object detection that jointly predicts 3D bounding boxes and a local-density-aware semantic occupancy map. The method consists of an offline LDO ground-truth generation pipeline that aggregates multi-frame LiDAR by ego-pose for static content and by instance index for dynamic objects, a voxel-height-guided sampling (VHS) module that extracts local features from voxel height intervals, and a global-local collaborative feature fusion (CFF) module that combines these local features with global BEV features. On nuScenes, the paper reports 49.5% mAP / 59.2% NDS on the test set and claims state-of-the-art gains over SOGDet, along with ablations on the validation set. However, the validation result for the base model is reported inconsistently: Table II lists 53.5% NDS / 42.2% mAP, while Table III and Supplementary B.1 list 52.6% NDS / 41.1% mAP for the same configuration.

Significance. The core idea is timely and potentially useful: rather than treating occupancy as a separate output, CoP uses an automatically generated, density-weighted occupancy target as auxiliary supervision to bias the shared BEV representation toward fine-grained geometry. The paper includes a clear ablation path, and the reported occupancy-task gains on TPVFormer and SurroundOcc (Table IV) suggest that the LDO labels may have value as a plug-in signal. The main significance hinges on the detection numbers, which are currently not stable across tables; the difference between the two reported validation scores (0.9 NDS / 1.1 mAP) is of the same order as the claimed advantages over prior work. If the discrepancies are resolved and the LDO label quality is demonstrated, the method would be a solid contribution. As written, the evidence does not yet support the headline claim.

major comments (3)
  1. [Table II vs. Table III / Supplementary B.1] The same base configuration (ResNet-50, voxel-128, 256×704 input) is reported as 53.5% NDS / 42.2% mAP in Table II but as 52.6% NDS / 41.1% mAP in Table III and Supplementary B.1. The text around Table II states +1.9 NDS / +2.2 mAP over IA-BEV, whereas Supplementary B.1 states +1.1 NDS / +1.0 mAP for the same comparison. This discrepancy is comparable to the claimed test-set gains over SOGDet (Table I) and to the individual ablation increments in Table III, so the central performance claim is not reliably established. Please identify the cause (for example, test-time augmentation, voxel size, or checkpoint selection) and present a single consistent set of validation numbers for the base model.
  2. [Section III-B, Eq. (4)] The LDO ground truths are generated by merging multi-frame LiDAR with ground-truth box indices for dynamic objects and ego poses for static content. No quantitative check of the resulting label quality is reported before these labels are used to supervise the shared features through the loss in Eq. (3). Annotation noise or misalignment would propagate directly into the detection head. Please add a label-quality evaluation (for example, mIoU against the dense occupancy ground truth on a held-out subset, or an ablation that perturbs the box indices and ego poses) to show that the generated occupancy targets are accurate enough to serve as auxiliary supervision.
  3. [Table I and Section IV.B] The test-set results (49.5% mAP / 59.2% NDS) are presented without an official test-server report, and the abstract's code link conflicts with Supplementary A.2, which states 'we will release the code after publication'. Given the internal inconsistency in the validation numbers, the test-set comparison cannot be independently checked from the manuscript. Please provide the official evaluation logs for all compared methods or a reproducible code release with exact version/checkpoint identifiers, so the claimed state-of-the-art performance is verifiable.
minor comments (6)
  1. [Figure 2] The caption says 'ConsistentPerceiver (CoP)' and the figure contains elements such as 'GLT Module', 'Teacher Network / Student Network', 'EMA', and 'Pseudo-Labels' that are never described in the method; please update the figure to match the actual CoP architecture or remove the unrelated components.
  2. [Supplementary A.2] The implementation details contain unresolved placeholders '[?]' for PyTorch, MMDetection3D, ImageNet, and nuImg; please fill in the missing citations.
  3. [Section IV.B] The phrase 'marking an improvement of +1.1% NSD' should read 'NDS'.
  4. [Section IV.C and Table V] The text says uniformly distributed heights are used at intervals of 2 m, but the HoI list in Table VI includes multiple 1 m intervals; please clarify how the uniform baseline was constructed.
  5. [Table VII] The sentence 'the coarsest voxel sizes [64,64,5] result in lower mAP and NDS values compared to larger voxel sizes' should say 'compared to finer voxel sizes'.
  6. [Supplementary B.1] The sentence 'with 52.6%/41.1% mAP/NDS' mixes the order of mAP and NDS; Table VII reports 41.1 mAP / 52.6 NDS, so the text should be made consistent.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the detection claim is evaluated on the external nuScenes benchmark, and the LDO auxiliary supervision is a training signal derived from nuScenes labels rather than from the model's own predictions.

full rationale

The paper's derivation chain is not circular. The LDO occupancy ground truths are generated from offline multi-frame LiDAR data plus nuScenes ground-truth bounding boxes, instance IDs, and ego poses (Eq. 4 and Section III-B), and then used as auxiliary supervision in the joint loss (Eq. 3). The 3D detection outputs are not defined in terms of the LDO labels; the occupancy loss is a regularizer/auxiliary task, and detection accuracy is measured on the official nuScenes validation and test splits, which are external to the LDO generation procedure. The ablations in Table III compare models trained with and without sparse/dense/LDO occupancy under the same external detection metrics, so the reported gains are empirical and not forced by construction. The VHS and CFF modules are architectural components trained with the same detection supervision; they do not rename or re-fit the evaluation labels. The only self-citation is Reference [49] (VisionKG) as a data-hosting statement ('All the generated LDO data can be found in VisionKG [49]'); this is not load-bearing evidence for any performance claim and does not make the derivation circular. The inconsistency between Table II (53.5 NDS / 42.2 mAP) and Table III / Supplementary B.1 (52.6 NDS / 41.1 mAP) for the same base configuration is a reproducibility or reporting concern, not a circularity concern, because both sets of numbers are external benchmark evaluations rather than restatements of the method's inputs. Overall, no step in the claimed derivation reduces by construction to its own inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a hand-designed data generation pipeline and several hand-chosen hyperparameters (beta, HoIs, voxel size). No new physical entities are introduced. The LDO label is a new data representation, not an entity, and its quality is assumed rather than verified.

free parameters (4)
  • beta (task balance temperature) = 0.9
    Equation (3) weights the occupancy loss; value from Table VI. Chosen by hand, not derived.
  • Height intervals of interest (HoIs) = [−3,−2], [−2,−1], [−1,0], [0,2], [−5,3], [−4,2], [−6,−4], [−2,1] m
    Design choice for VHS sampling, based on training-set voxel height distributions (Fig. 7) and ablated on validation in Table VIII; affects which local features are extracted.
  • Voxel grid resolution (occupancy) = [128,128,10] voxels (0.8 m cells) for base
    Chosen after ablating [64,64,5] and [256,256,16] on validation (Table VII); affects granularity of LDO supervision and compute.
  • Depth discretization and spatial range = 60 depth bins; range [−51.2,51.2] x [−51.2,51.2] x [−3,5] m
    Standard BEVDet-style configuration from Table VI; hand-set and not justified beyond prior work.
assumptions (4)
  • domain assumption nuScenes ground-truth 3D boxes, instance IDs, ego poses, and segmentation labels are accurate enough to support offline LiDAR aggregation
    Section III-B uses these annotations to separate and merge static/dynamic points; any errors propagate into the LDO labels and then into detection features.
  • domain assumption Multi-frame world-coordinate aggregation with per-object box indices produces physically meaningful dense occupancy labels
    Eq. (4) assumes concatenating points object-by-object and transforming back to LiDAR coordinates yields correct geometry; no label-quality validation is reported.
  • domain assumption Voxel height distributions and HoI choices computed from the training set transfer to validation and test scenes
    Section III-C and Fig. 7 set fixed height intervals from nuScenes training data; if the test distribution shifts, the sampling priors become stale.
  • domain assumption The occupancy auxiliary task improves detection through shared structural features rather than through added capacity, training time, or richer supervision alone
    Section III-A and Eq. (3) assume multi-task collaboration is the cause of the gains; no control for extra parameters or training cost is presented.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Collaborative Perceiver: Elevating Vision-based 3D Object Detection via Local Density-Aware Spatial Occupancy." pith.science (2026). https://pith.science/paper/TTSTAVBY

@misc{pith2026250721358,
  author       = {Pith},
  title        = {Pith review of: Collaborative Perceiver: Elevating Vision-based 3D Object Detection via Local Density-Aware Spatial Occupancy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TTSTAVBY}},
  note         = {Machine review of arXiv:2507.21358}
}
read the original abstract

Vision-based bird's-eye-view (BEV) 3D object detection has advanced significantly in autonomous driving by offering cost-effectiveness and rich contextual information. However, existing methods often construct BEV representations by collapsing extracted object features, neglecting intrinsic environmental contexts, such as roads and pavements. This hinders detectors from comprehensively perceiving the characteristics of the physical world. To alleviate this, we introduce a multi-task learning framework, Collaborative Perceiver (CoP), that leverages spatial occupancy as auxiliary information to mine consistent structural and conceptual similarities shared between 3D object detection and occupancy prediction tasks, bridging gaps in spatial representations and feature refinement. To this end, we first propose a pipeline to generate dense occupancy ground truths incorporating local density information (LDO) for reconstructing detailed environmental information. Next, we employ a voxel-height-guided sampling (VHS) strategy to distill fine-grained local features according to distinct object properties. Furthermore, we develop a global-local collaborative feature fusion (CFF) module that seamlessly integrates complementary knowledge between both tasks, thus composing more robust BEV representations. Extensive experiments on the nuScenes benchmark demonstrate that CoP outperforms existing vision-based frameworks, achieving 49.5\% mAP and 59.2\% NDS on the test set. Code and supplementary materials are available at this link https://github.com/jichengyuan/Collaborative-Perceiver.

Figures

Figures reproduced from arXiv: 2507.21358 by the authors.

Figure 1
Figure 1. Overview of the proposed Collaborative Perceiver (CoP), which leverages complementary and consistent knowledge from 3D occupancy and 3D object detection to achieve comprehensive scene understanding. an occupied probability to each voxel in the 3D space, de￾livering a geometrically consistent and precise description of dynamic scenes. However, many existing methods [10], [11] inherently assume a homogeneous point den… view at source ↗
Figure 2
Figure 2. Overview of the proposed Consistent Perceiver (CoP) and 3D occupancy ground truth (GT) generation pipeline. First, multi-view image features are extracted and transformed into the 3D space following the LSS [8]. Then, global pooling and local sampling are employed to extract coarse global and fine-grained local features guided by the voxel-height distributions. Next, a global-local collaborative feature fusion block… view at source ↗
Figure 3
Figure 3. Illustration of different 3D Occupancy ground truth [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: The proposed two-stage feature selection and fusion [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results of the 3D detection and occupancy prediction on the validation set of nuScenes [14]. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Class distribution of the nuScenes [14] dataset used [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Height distribution of the generated occupied voxels [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Qualitative results of CoP (large) for the nuScenes [14] dataset are displayed from left to right, showcasing 3D [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Class-wise Evaluation of CoP (large) on the test set of nuScenes [14]. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

64 extracted references · 39 canonical work pages

  1. [1]

    Aedet: Azimuth- invariant multi-view 3d object detection,

    C. Feng, Z. Jie, Y . Zhong, X. Chu, and L. Ma, “Aedet: Azimuth- invariant multi-view 3d object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 21 580–21 588

    work page 2023

  2. [2]

    3d object detection for autonomous driving: A comprehensive survey,

    J. Mao, S. Shi, X. Wang, and H. Li, “3d object detection for autonomous driving: A comprehensive survey,”International Journal of Computer Vision, vol. 131, no. 8, pp. 1909–1963, 2023

    work page 1909

  3. [3]

    Bevdet: High- performance multi-camera 3d object detection in bird-eye-view,

    J. Huang, G. Huang, Z. Zhu, Y . Ye, and D. Du, “Bevdet: High- performance multi-camera 3d object detection in bird-eye-view,”arXiv preprint arXiv:2112.11790, 2021

    arXiv 2021

  4. [4]

    Bevdepth: Acquisition of reliable depth for multi-view 3d object detection,

    Y . Li, Z. Ge, G. Yu, J. Yang, Z. Wang, Y . Shi, J. Sun, and Z. Li, “Bevdepth: Acquisition of reliable depth for multi-view 3d object detection,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 2, 2023, pp. 1477–1485

    work page 2023

  5. [5]

    Bevdet4d: Exploit temporal cues in multi- camera 3d object detection,

    J. Huang and G. Huang, “Bevdet4d: Exploit temporal cues in multi- camera 3d object detection,”arXiv preprint arXiv:2203.17054, 2022

    arXiv 2022

  6. [6]

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

    Z. Li, W. Wang, H. Li, E. Xie, C. Sima, T. Lu, Y . Qiao, and J. Dai, “Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers,” inEuropean conference on computer vision. Springer, 2022, pp. 1–18

    2022

  7. [7]

    Petrv2: A unified framework for 3d perception from multi-camera images,

    Y . Liu, J. Yan, F. Jia, S. Li, A. Gao, T. Wang, and X. Zhang, “Petrv2: A unified framework for 3d perception from multi-camera images,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 3262–3272

    work page 2023

  8. [8]

    Lift, splat, shoot: Encoding images from arbitrary camera rigs by implicitly unprojecting to 3d,

    J. Philion and S. Fidler, “Lift, splat, shoot: Encoding images from arbitrary camera rigs by implicitly unprojecting to 3d,” inComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XIV 16. Springer, 2020, pp. 194–210

    2020

Show all 64 references
  1. [9]

    Did we test all scenarios for automated and autonomous driving systems?

    F. Hauer, T. Schmidt, B. Holzm ¨uller, and A. Pretschner, “Did we test all scenarios for automated and autonomous driving systems?” in2019 IEEE Intelligent Transportation Systems Conference (ITSC). IEEE, 2019, pp. 2950–2955

    2019

  2. [10]

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

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

    2023

  3. [11]

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

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

    2023 arXiv

  4. [12]

    Bevfusion: Multi-task multi-sensor fusion with unified bird’s- eye view representation,

    Z. Liu, H. Tang, A. Amini, X. Yang, H. Mao, D. L. Rus, and S. Han, “Bevfusion: Multi-task multi-sensor fusion with unified bird’s- eye view representation,” in2023 IEEE international conference on robotics and automation (ICRA). IEEE, 2023, pp. 2774–2781

    2023

  5. [13]

    Anyloc: Towards universal visual place recognition,

    N. Keetha, A. Mishra, J. Karhade, K. M. Jatavallabhula, S. Scherer, M. Krishna, and S. Garg, “Anyloc: Towards universal visual place recognition,”IEEE Robotics and Automation Letters, 2023

    2023

  6. [14]

    nuscenes: A multimodal dataset for autonomous driving,

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

    2020

  7. [15]

    Performance and challenges of 3d object detection methods in complex scenes for autonomous driving,

    K. Wang, T. Zhou, X. Li, and F. Ren, “Performance and challenges of 3d object detection methods in complex scenes for autonomous driving,”IEEE Transactions on Intelligent Vehicles, vol. 8, no. 2, pp. 1699–1716, 2022

    2022

  8. [16]

    Objects are different: Flexible monocu- lar 3d object detection,

    Y . Zhang, J. Lu, and J. Zhou, “Objects are different: Flexible monocu- lar 3d object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 3289–3298

    2021

  9. [17]

    Is pseudo- lidar needed for monocular 3d object detection?

    D. Park, R. Ambrus, V . Guizilini, J. Li, and A. Gaidon, “Is pseudo- lidar needed for monocular 3d object detection?” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 3142–3152

    2021

  10. [18]

    M3d-rpn: Monocular 3d region proposal network for object detection,

    G. Brazil and X. Liu, “M3d-rpn: Monocular 3d region proposal network for object detection,” inProceedings of the IEEE/CVF in- ternational conference on computer vision, 2019, pp. 9287–9296

    2019

  11. [19]

    Delving into localization errors for monocular 3d object detection,

    X. Ma, Y . Zhang, D. Xu, D. Zhou, S. Yi, H. Li, and W. Ouyang, “Delving into localization errors for monocular 3d object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 4721–4730

    2021

  12. [20]

    Monocular 3d object detection for autonomous driving,

    X. Chen, K. Kundu, Z. Zhang, H. Ma, S. Fidler, and R. Urtasun, “Monocular 3d object detection for autonomous driving,” inPro- ceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2147–2156

    2016

  13. [21]

    Fcos: A simple and strong anchor-free object detector,

    Z. Tian, C. Shen, H. Chen, and T. He, “Fcos: A simple and strong anchor-free object detector,”IEEE transactions on pattern analysis and machine intelligence, vol. 44, no. 4, pp. 1922–1933, 2020

    1922

  14. [22]

    Center- net: Keypoint triplets for object detection,

    K. Duan, S. Bai, L. Xie, H. Qi, Q. Huang, and Q. Tian, “Center- net: Keypoint triplets for object detection,” inProceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 6569–6578

    2019

  15. [23]

    Fcos3d: Fully convolutional one-stage monocular 3d object detection,

    T. Wang, X. Zhu, J. Pang, and D. Lin, “Fcos3d: Fully convolutional one-stage monocular 3d object detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 913–922

    2021

  16. [24]

    Bev- san: Accurate bev 3d object detection via slice attention networks,

    X. Chi, J. Liu, M. Lu, R. Zhang, Z. Wang, Y . Guo, and S. Zhang, “Bev- san: Accurate bev 3d object detection via slice attention networks,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 17 461–17 470

    2023

  17. [25]

    M2bev: Multi-camera joint 3d detection and segmentation with unified birds-eye view representation. arxiv 2022,

    E. Xie, Z. Yu, D. Zhou, J. Philion, A. Anandkumar, S. Fidler, P. Luo, and J. Alvarez, “M2bev: Multi-camera joint 3d detection and segmentation with unified birds-eye view representation. arxiv 2022,” arXiv preprint arXiv:2204.05088

    2022 arXiv

  18. [26]

    Detr3d: 3d object detection from multi-view images via 3d-to-2d queries,

    Y . Wang, V . C. Guizilini, T. Zhang, Y . Wang, H. Zhao, and J. Solomon, “Detr3d: 3d object detection from multi-view images via 3d-to-2d queries,” inConference on Robot Learning. PMLR, 2022, pp. 180– 191

    2022

  19. [27]

    Petr: Position embedding transformation for multi-view 3d object detection,

    Y . Liu, T. Wang, X. Zhang, and J. Sun, “Petr: Position embedding transformation for multi-view 3d object detection,” inEuropean Con- ference on Computer Vision. Springer, 2022, pp. 531–548

    2022

  20. [28]

    Sparsebev: High- performance sparse 3d object detection from multi-camera videos,

    H. Liu, Y . Teng, T. Lu, H. Wang, and L. Wang, “Sparsebev: High- performance sparse 3d object detection from multi-camera videos,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 18 580–18 590

    2023

  21. [29]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European conference on computer vision. Springer, 2020, pp. 213– 229

    2020

  22. [30]

    Virtual occupancy grid map for submap-based pose graph slam and planning in 3d environments,

    B.-J. Ho, P. Sodhi, P. Teixeira, M. Hsiao, T. Kusnur, and M. Kaess, “Virtual occupancy grid map for submap-based pose graph slam and planning in 3d environments,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018, pp. 2175–2182

    2018

  23. [31]

    3d path planning and stereo-based obstacle avoidance for rotorcraft uavs,

    S. Hrabar, “3d path planning and stereo-based obstacle avoidance for rotorcraft uavs,” in2008 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2008, pp. 807–814

    2008

  24. [32]

    Using occupancy grids for mobile robot perception and navigation,

    A. Elfes, “Using occupancy grids for mobile robot perception and navigation,”Computer, vol. 22, no. 6, pp. 46–57, 1989

    1989

  25. [33]

    Occupancy anticipation for efficient exploration and navigation,

    S. K. Ramakrishnan, Z. Al-Halah, and K. Grauman, “Occupancy anticipation for efficient exploration and navigation,” inComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part V 16. Springer, 2020, pp. 400–418

    2020

  26. [34]

    Probabilistic and geometric depth: Detecting objects in perspective,

    T. Wang, Z. Xinge, J. Pang, and D. Lin, “Probabilistic and geometric depth: Detecting objects in perspective,” inConference on Robot Learning. PMLR, 2022, pp. 1475–1485

    2022

  27. [35]

    High resolution maps from wide an- gle sonar,

    H. Moravec and A. Elfes, “High resolution maps from wide an- gle sonar,” inProceedings. 1985 IEEE international conference on robotics and automation, vol. 2. IEEE, 1985, pp. 116–121

    1985

  28. [36]

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

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

    2023

  29. [37]

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

    X. Tian, T. Jiang, L. Yun, Y . Mao, H. Yang, Y . Wang, Y . Wang, and H. Zhao, “Occ3d: A large-scale 3d occupancy prediction benchmark for autonomous driving,”Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  30. [38]

    Occnerf: Self-supervised multi-camera occupancy prediction with neural radiance fields,

    C. Zhang, J. Yan, Y . Wei, J. Li, L. Liu, Y . Tang, Y . Duan, and J. Lu, “Occnerf: Self-supervised multi-camera occupancy prediction with neural radiance fields,”arXiv preprint arXiv:2312.09243, 2023

    2023 arXiv

  31. [39]

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

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

    2024

  32. [40]

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

    Z. Yang, Y . Sun, S. Liu, and J. Jia, “3dssd: Point-based 3d single stage object detector,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 11 040–11 048

    2020

  33. [41]

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

    Z. Yang, Y . Sun, S. Liu, X. Shen, and J. Jia, “Std: Sparse-to-dense 3d object detector for point cloud,” inProceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 1951–1960

    2019

  34. [42]

    A survey on 3d object detection methods for autonomous driving applications,

    E. Arnold, O. Y . Al-Jarrah, M. Dianati, S. Fallah, D. Oxtoby, and A. Mouzakitis, “A survey on 3d object detection methods for autonomous driving applications,”IEEE Transactions on Intelligent Transportation Systems, vol. 20, no. 10, pp. 3782–3795, 2019

    2019

  35. [43]

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

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

    2023

  36. [44]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    2016

  37. [45]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 2117–2125

    2017

  38. [46]

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

    J. Hou, X. Li, W. Guan, G. Zhang, D. Feng, Y . Du, X. Xue, and J. Pu, “Fastocc: Accelerating 3d occupancy prediction by fus- ing the 2d bird’s-eye view and perspective view,”arXiv preprint arXiv:2403.02710, 2024

    2024 arXiv

  39. [47]

    Pointocc: Cylindrical tri-perspective view for point-based 3d semantic occupancy predic- tion,

    S. Zuo, W. Zheng, Y . Huang, J. Zhou, and J. Lu, “Pointocc: Cylindrical tri-perspective view for point-based 3d semantic occupancy predic- tion,”arXiv preprint arXiv:2308.16896, 2023

    2023 arXiv

  40. [48]

    Density- preserving deep point cloud compression,

    Y . He, X. Ren, D. Tang, Y . Zhang, X. Xue, and Y . Fu, “Density- preserving deep point cloud compression,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 2333–2342

    2022

  41. [49]

    Visionkg: Unleashing the power of visual datasets via knowledge graph,

    J. Yuan, A. Le-Tuan, M. Nguyen-Duc, T.-K. Tran, M. Hauswirth, and D. Le-Phuoc, “Visionkg: Unleashing the power of visual datasets via knowledge graph,” inEuropean Semantic Web Conference. Springer, 2024, pp. 75–93

    2024

  42. [50]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141

    2018

  43. [51]

    Density-based clustering for 3d object detection in point clouds,

    S. M. Ahmed and C. M. Chew, “Density-based clustering for 3d object detection in point clouds,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 10 608–10 617

    2020

  44. [52]

    Infofocus: 3d object detec- tion for autonomous driving with dynamic information modeling,

    J. Wang, S. Lan, M. Gao, and L. S. Davis, “Infofocus: 3d object detec- tion for autonomous driving with dynamic information modeling,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part X 16. Springer, 2020, pp. 405–420

    2020

  45. [53]

    Pointpillars: Fast encoders for object detection from point clouds,

    A. H. Lang, S. V ora, H. Caesar, L. Zhou, J. Yang, and O. Beijbom, “Pointpillars: Fast encoders for object detection from point clouds,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 12 697–12 705

    2019

  46. [54]

    Centerfusion: Center-based radar and camera fusion for 3d object detection,

    R. Nabati and H. Qi, “Centerfusion: Center-based radar and camera fusion for 3d object detection,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2021, pp. 1527–1536

    2021

  47. [55]

    Mvfusion: Multi-view 3d object detection with semantic-aligned radar and camera fusion,

    Z. Wu, G. Chen, Y . Gan, L. Wang, and J. Pu, “Mvfusion: Multi-view 3d object detection with semantic-aligned radar and camera fusion,” in2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2023, pp. 2766–2773

    2023

  48. [56]

    Learning ego 3d representation as ray tracing,

    J. Lu, Z. Zhou, X. Zhu, H. Xu, and L. Zhang, “Learning ego 3d representation as ray tracing,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 129–144

    2022

  49. [57]

    Cross modal transformer: Towards fast and robust 3d object detection,

    J. Yan, Y . Liu, J. Sun, F. Jia, S. Li, T. Wang, and X. Zhang, “Cross modal transformer: Towards fast and robust 3d object detection,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 18 268–18 278

    2023

  50. [58]

    X3kd: Knowledge distillation across modalities, tasks and stages for multi-camera 3d object detection,

    M. Klingner, S. Borse, V . R. Kumar, B. Rezaei, V . Narayanan, S. Yogamani, and F. Porikli, “X3kd: Knowledge distillation across modalities, tasks and stages for multi-camera 3d object detection,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni...

    2023

  51. [59]

    Sogdet: Semantic-occupancy guided multi-view 3d object detection,

    Q. Zhou, J. Cao, H. Leng, Y . Yin, Y . Kun, and R. Zimmermann, “Sogdet: Semantic-occupancy guided multi-view 3d object detection,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 7, 2024, pp. 7668–7676

    2024

  52. [60]

    Cobev: Elevating roadside 3d object detection with depth and height complementarity,

    H. Shi, C. Pang, J. Zhang, K. Yang, Y . Wu, H. Ni, Y . Lin, R. Stiefel- hagen, and K. Wang, “Cobev: Elevating roadside 3d object detection with depth and height complementarity,”IEEE Transactions on Image Processing, 2024

    2024

  53. [61]

    Dualbev: Cnn is all you need in view transformation,

    P. Li, W. Shen, Q. Huang, and D. Cui, “Dualbev: Cnn is all you need in view transformation,”arXiv preprint arXiv:2403.05402, 2024

    2024 arXiv

  54. [62]

    Decoupled weight decay regularization,

    I. Loshchilov, “Decoupled weight decay regularization,”arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  55. [63]

    Class-balanced grouping and sampling for point cloud 3d object detection,

    B. Zhu, Z. Jiang, X. Zhou, Z. Li, and G. Yu, “Class-balanced grouping and sampling for point cloud 3d object detection,”arXiv preprint arXiv:1908.09492, 2019

    1908 arXiv

  56. [64]

    Instance-aware multi-camera 3d object detection with structural pri- ors mining and self-boosting learning,

    Y . Jiao, Z. Jie, S. Chen, L. Cheng, J. Chen, L. Ma, and Y .-G. Jiang, “Instance-aware multi-camera 3d object detection with structural pri- ors mining and self-boosting learning,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 3, 2024, pp. 2598–...

    2024

Pith tools

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