pith. machine review for the scientific record. sign in

arxiv: 2605.01924 · v1 · submitted 2026-05-03 · 💻 cs.CV

Recognition: unknown

SimPB++: Simultaneously Detecting 2D and 3D Objects from Multiple Cameras

Yingqi Tang , Zhaotie Meng , Erkang Cheng , Haibin Ling
Authors on Pith no claims yet

Pith reviewed 2026-05-09 17:19 UTC · model grok-4.3

classification 💻 cs.CV
keywords multi-camera detection2D and 3D joint detectionBEV perceptionhybrid decoderend-to-end modelmixed supervisionnuScenesArgoverse2
0
0 comments X

The pith

SimPB++ unifies 2D perspective detection and 3D BEV detection into one end-to-end multi-camera model with interactive decoders.

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

The paper seeks to prove that 2D object detection in camera views and 3D detection in bird's-eye view can be handled together in a single model instead of separate or sequential stages. It builds a hybrid decoder where the 2D and 3D parts exchange information through dynamic query mechanisms to create a cycle of mutual refinement. This would matter for autonomous driving because simultaneous perception could raise accuracy on both tasks while cutting the amount of costly 3D labels needed. The design also adds query-group attention for 2D, crop-and-scale handling for distant objects, and support for mixed 2D-only and full supervision, with results reported as state-of-the-art on nuScenes and strong long-range performance on Argoverse2.

Core claim

SimPB++ unifies both tasks into an end-to-end model with a hybrid decoder architecture, coupling multi-view 2D and 3D decoders interactively via Dynamic Query Allocation and Adaptive Query Aggregation, forming a cyclic 3D-2D-3D refinement. For multi-view 2D detection, Query-group Attention is used for intra-group communication along with a Crop-and-Scale strategy for long-range perception and a Propagating Denoising strategy with an auxiliary RoI detector. The model supports mixed supervision with 2D-only and fully annotated data.

What carries the argument

Hybrid decoder with Dynamic Query Allocation that adaptively assigns 2D queries to 3D candidates and Adaptive Query Aggregation that refines 3D representations using multi-view 2D features to enable cyclic refinement.

If this is right

  • End-to-end joint training improves accuracy on both 2D perspective and 3D BEV detection compared with two-stage pipelines.
  • The model reaches state-of-the-art performance on the nuScenes benchmark for both detection tasks.
  • Mixed supervision allows training with cheaper 2D-only labels while still delivering strong 3D results.
  • Long-range 3D detection extends reliably to 150 meters on Argoverse2.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The query-based interaction pattern could be tested on other coupled tasks such as joint depth estimation and semantic segmentation.
  • Lower dependence on full 3D labels might speed up deployment in new geographic regions where only 2D annotations are available.
  • The cyclic refinement loop suggests a route to improve robustness under partial occlusions or sensor dropout by letting 2D and 3D streams correct each other.

Load-bearing premise

The Dynamic Query Allocation and Adaptive Query Aggregation modules will create effective deep interactions between the 2D and 3D decoders that produce measurable gains over two-stage pipelines without adding instability.

What would settle it

An ablation experiment on nuScenes in which removing or replacing the Dynamic Query Allocation and Adaptive Query Aggregation modules yields no accuracy gain or a drop in either 2D or 3D detection metrics would falsify the value of the interactive coupling.

Figures

Figures reproduced from arXiv: 2605.01924 by Erkang Cheng, Haibin Ling, Yingqi Tang, Zhaotie Meng.

Figure 1
Figure 1. Figure 1: Comparisons of different multi-view object detection view at source ↗
Figure 2
Figure 2. Figure 2: Overview of SimPB++, a unified multi-view 2D and 3D object detection framework. An image backbone and an view at source ↗
Figure 3
Figure 3. Figure 3: Dynamic Query Allocation. It establishes the query association between the BEV and the perspective view by view at source ↗
Figure 4
Figure 4. Figure 4: Query-Group Attention. We enforce interaction among 2D queries only within the same camera group. DA represents view at source ↗
Figure 5
Figure 5. Figure 5: Adaptive Query Aggregation. The indicator vector represents whether a 2D query is truncated or not. view at source ↗
Figure 6
Figure 6. Figure 6: The Crop-and-Scale strategy generates a high view at source ↗
Figure 7
Figure 7. Figure 7: The Propagating Denoising in multi-view 2D decoder view at source ↗
Figure 8
Figure 8. Figure 8: Performance analysis of Partial Labeling Training (PLT) with additional 2D-only labeled data for training. view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of long-range detection on Argoverse2. view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of association strategies. (a) MV2D (2D-to-3D association): view at source ↗
Figure 11
Figure 11. Figure 11: Visualization results of StreamPETR and SimPB++. view at source ↗
Figure 12
Figure 12. Figure 12: Detection results on the nuScenes validation dataset. view at source ↗
Figure 13
Figure 13. Figure 13: Detection results on the Argoverse2 validation dataset. view at source ↗
read the original abstract

Simultaneous perception of 2D objects in perspective view and 3D objects in Bird's Eye View (BEV) is challenging for multi-camera autonomous driving. Existing two-stage pipelines use 2D results only as a one-time cue for 3D detection. We propose SimPB++, which simultaneously detects 2D objects in perspective and 3D objects in BEV from multiple cameras. It unifies both tasks into an end-to-end model with a hybrid decoder architecture, coupling multi-view 2D and 3D decoders interactively. Two novel modules enable deep interaction: Dynamic Query Allocation adaptively assigns 2D queries to 3D candidates, and Adaptive Query Aggregation refines 3D representations using multi-view 2D features, forming a cyclic 3D-2D-3D refinement. For multi-view 2D detection, we use Query-group Attention for intra-group communication. We also design a Crop-and-Scale strategy for long-range perception and a Propagating Denoising strategy with an auxiliary RoI detector. SimPB++ supports mixed supervision with 2D-only and fully annotated data, reducing reliance on expensive 3D labels. Experiments show state-of-the-art performance on nuScenes for both tasks and strong long-range detection (up to 150m) on Argoverse2.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces SimPB++, an end-to-end model for simultaneous multi-view 2D perspective and 3D BEV object detection. It employs a hybrid decoder architecture that interactively couples 2D and 3D decoders using two novel modules—Dynamic Query Allocation for adaptive assignment of 2D queries to 3D candidates and Adaptive Query Aggregation for refining 3D representations with multi-view 2D features—creating a cyclic 3D-2D-3D refinement process. Additional components include Query-group Attention for 2D intra-group communication, a Crop-and-Scale strategy for long-range perception, and a Propagating Denoising strategy with an auxiliary RoI detector. The model supports mixed supervision using 2D-only and fully annotated data, and the authors claim state-of-the-art performance on nuScenes for both tasks along with strong long-range detection results on Argoverse2.

Significance. If the performance claims and the effectiveness of the interactive modules hold, this work could meaningfully advance unified 2D-3D perception pipelines in autonomous driving by moving beyond one-way cueing in two-stage systems and by enabling reduced reliance on costly 3D labels via mixed supervision. The hybrid decoder approach addresses a practical limitation in current multi-camera detectors.

major comments (2)
  1. [§4.3] §4.3 (Ablation Studies): The central claim that Dynamic Query Allocation and Adaptive Query Aggregation enable deep cyclic interaction and measurable gains over non-interactive hybrids is not load-bearingly demonstrated. No ablation removes only these two modules while retaining the hybrid decoder, Query-group Attention, Crop-and-Scale, and Propagating Denoising; without this control, it remains unclear whether the reported improvements on nuScenes NDS and mAP are carried by the claimed bidirectional coupling or by the overall architecture.
  2. [§4.4] §4.4 (Mixed Supervision Experiments): The results using 2D-only labels show gains, but the paper does not include a control that applies the auxiliary RoI detector without the 3D-2D-3D cycle. This leaves open whether the unification benefit, rather than the auxiliary detector alone, enables effective use of partial annotations, weakening the argument that the cyclic refinement is key to reducing 3D label dependency.
minor comments (2)
  1. [Abstract] Abstract: The claim of state-of-the-art performance on nuScenes is stated without any numerical values, baseline names, or dataset splits, which reduces immediate readability even though the full experiments section supplies these details.
  2. [Figure 2] Figure 2 (Architecture Diagram): The data-flow arrows between the 2D and 3D decoder branches could be labeled more explicitly to highlight the exact points at which Dynamic Query Allocation and Adaptive Query Aggregation operate, aiding readers in tracing the cyclic refinement.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address the major comments point-by-point below and will revise the manuscript accordingly to strengthen the empirical support for our claims.

read point-by-point responses

  1. Referee: [§4.3] §4.3 (Ablation Studies): The central claim that Dynamic Query Allocation and Adaptive Query Aggregation enable deep cyclic interaction and measurable gains over non-interactive hybrids is not load-bearingly demonstrated. No ablation removes only these two modules while retaining the hybrid decoder, Query-group Attention, Crop-and-Scale, and Propagating Denoising; without this control, it remains unclear whether the reported improvements on nuScenes NDS and mAP are carried by the claimed bidirectional coupling or by the overall architecture.

    Authors: We agree that a more targeted control is needed to isolate the contribution of Dynamic Query Allocation and Adaptive Query Aggregation. Our existing §4.3 ablations compare the full model to variants lacking the interactive modules, but they do not hold the remainder of the hybrid decoder and auxiliary components fixed. We will add a new ablation that disables only these two modules while retaining Query-group Attention, Crop-and-Scale, and Propagating Denoising. This will directly test whether the bidirectional 3D-2D-3D coupling accounts for the observed gains on nuScenes. revision: yes

  2. Referee: [§4.4] §4.4 (Mixed Supervision Experiments): The results using 2D-only labels show gains, but the paper does not include a control that applies the auxiliary RoI detector without the 3D-2D-3D cycle. This leaves open whether the unification benefit, rather than the auxiliary detector alone, enables effective use of partial annotations, weakening the argument that the cyclic refinement is key to reducing 3D label dependency.

    Authors: We acknowledge the validity of this point. The mixed-supervision results in §4.4 demonstrate that our framework can leverage 2D-only data, yet they do not separate the effect of the auxiliary RoI detector from the cyclic refinement. We will add a control experiment that applies the auxiliary RoI detector without Dynamic Query Allocation and Adaptive Query Aggregation (i.e., without the 3D-2D-3D cycle). This will clarify whether the interactive unification is necessary for the observed benefits under partial annotation. revision: yes

Circularity Check

0 steps flagged

No circularity: architectural description with empirical validation

full rationale

The paper proposes a hybrid decoder architecture for simultaneous 2D/3D detection, introducing modules like Dynamic Query Allocation and Adaptive Query Aggregation to enable interaction. No mathematical derivations, equations, fitted parameters renamed as predictions, or self-referential definitions appear in the abstract or description. Claims rest on end-to-end training and benchmark results rather than any reduction to inputs by construction. No load-bearing self-citations or uniqueness theorems are invoked in the provided text. This is a standard empirical architecture paper whose central claims are externally falsifiable via ablation and comparison experiments.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 5 invented entities

The central claim rests on the unproven effectiveness of the newly introduced architectural modules; no external benchmarks or prior derivations are invoked in the abstract to support them.

invented entities (5)
  • Dynamic Query Allocation no independent evidence
    purpose: adaptively assigns 2D queries to 3D candidates
    Core new module enabling 2D-to-3D interaction
  • Adaptive Query Aggregation no independent evidence
    purpose: refines 3D representations using multi-view 2D features
    Core new module enabling cyclic 3D-2D-3D refinement
  • Query-group Attention no independent evidence
    purpose: enables intra-group communication for multi-view 2D detection
    New attention mechanism for 2D branch
  • Crop-and-Scale strategy no independent evidence
    purpose: supports long-range perception
    New preprocessing strategy
  • Propagating Denoising strategy no independent evidence
    purpose: improves training with auxiliary RoI detector
    New denoising training technique

pith-pipeline@v0.9.0 · 5550 in / 1504 out tokens · 34041 ms · 2026-05-09T17:19:00.151294+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

60 extracted references · 10 canonical work pages

  1. [1]

    Simpb: A single model for 2d and 3d object detection from multiple cameras,

    Y. Tang, Z. Meng, and E. Cheng, “Simpb: A single model for 2d and 3d object detection from multiple cameras,” inEuropean Conference on Computer Vision, 2024

    2024

  2. [2]

    Faster r-cnn: Towards real-time object detection with region proposal networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,”IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1137– 1149, 2017

    2017

  3. [3]

    You only look once: Unified, real-time object detection,

    J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-time object detection,” inIEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 779–788

    2016

  4. [4]

    Fcos: Fully convolutional one-stage object detection,

    Z. Tian, C. Shen, H. Chen, and T. He, “Fcos: Fully convolutional one-stage object detection,” inIEEE International Conference on Computer Vision, 2019, pp. 9627–9636

    2019

  5. [5]

    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,” inIEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 2147–2156

    2016

  6. [6]

    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,” inIEEE International Conference on Computer Vision, 2019, pp. 9287–9296

    2019

  7. [7]

    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,” inIEEE International Conference on Computer Vision, 2021, pp. 913–922

    2021

  8. [8]

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

    Y. Wang, V . Guizilini, T. Zhang, Y. Wang, H. Zhao, and J. Solomon, “Detr3d: 3d object detection from multi-view images via 3d-to-2d queries,” 2021, pp. 180–191

    2021

  9. [9]

    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 Con- ference on Computer Vision, 2022, pp. 1–18

    2022

  10. [10]

    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,” inEuropean Conference on Computer Vision, 2020, pp. 194–210

    2020

  11. [11]

    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,” inAAAI Conference on Artificial Intelligence, 2023, pp. 1477–1485

    2023

  12. [12]

    Bevnext: Reviving dense bev frameworks for 3d object detection,

    Z. Li, S. Lan, J. M. Alvarez, and Z. Wu, “Bevnext: Reviving dense bev frameworks for 3d object detection,” inIEEE Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 113–20 123

    2024

  13. [13]

    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 Conference on Computer Vision, 2022, pp. 531–548

    2022

  14. [14]

    arXiv preprint arXiv:2211.10581 (2022)

    X. Lin, T. Lin, Z. Pei, L. Huang, and Z. Su, “Sparse4d: Multi-view 3d object detection with sparse spatial-temporal fusion,”arXiv preprint arXiv:2211.10581, 2022

    work page arXiv 2022

  15. [15]

    Exploring object- centric temporal modeling for efficient multi-view 3d object detec- tion,

    S. Wang, Y. Liu, T. Wang, Y. Li, and X. Zhang, “Exploring object- centric temporal modeling for efficient multi-view 3d object detec- tion,” inIEEE International Conference on Computer Vision, 2023, pp. 3621–3631

    2023

  16. [16]

    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,” inIEEE International Conference on Computer Vision, 2023, pp. 18 580–18 590

    2023

  17. [17]

    End-to-end object detection with transformers,

    C. Nicolas, M. Francisco, S. Gabriel, U. Nicolas, K. Alexander, and Z. Sergey, “End-to-end object detection with transformers,” inEuropean Conference on Computer Vision, 2020, pp. 213—-229

    2020

  18. [18]

    Focal-petr: Embracing foreground for efficient multi-camera 3d object detection,

    S. Wang, X. Jiang, and Y. Li, “Focal-petr: Embracing foreground for efficient multi-camera 3d object detection,”IEEE Transactions on Intelligent Vehicles, 2023

    2023

  19. [19]

    Object as query: Equipping any 2d object detector with 3d detection ability,

    Z. Wang, Z. Huang, J. Fu, N. Wang, and S. Liu, “Object as query: Equipping any 2d object detector with 3d detection ability,” in IEEE International Conference on Computer Vision, 2023, pp. 3791– 3800

    2023

  20. [20]

    Far3d: Expanding the horizon for surround-view 3d object detection,

    X. Jiang, S. Li, Y. Liu, S. Wang, F. Jia, T. Wang, L. Han, and X. Zhang, “Far3d: Expanding the horizon for surround-view 3d object detection,” inAAAI Conference on Artificial Intelligence, 2023

    2023

  21. [21]

    arXiv preprint arXiv:2311.11722 (2023)

    X. Lin, Z. Pei, T. Lin, L. Huang, and Z. Su, “Sparse4d v3: Ad- vancing end-to-end 3d detection and tracking,”arXiv preprint arXiv:2311.11722, 2023

    work page arXiv 2023

  22. [22]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. Vora, V . E. Liong, Q. Xu, A. Krishnan, Y. Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” inIEEE Conference on Computer Vision and Pattern Recognition, 2020, pp. 11 618–11 628

    2020

  23. [23]

    Argoverse 2: Next generation datasets for self-driving perception and forecasting,

    B. Wilson, W. Qi, T. Agarwal, J. Lambert, J. Singh, S. Khandelwal, B. Pan, R. Kumar, A. Hartnett, J. Pontes, D. Ramanan, P . Carr, and J. Hays, “Argoverse 2: Next generation datasets for self-driving perception and forecasting,” inAdvances in Neural Information Processing Systems, Aug 2021

    2021

  24. [24]

    Cascade r-cnn: Delving into high quality object detection,

    Z. Cai and N. Vasconcelos, “Cascade r-cnn: Delving into high quality object detection,” inIEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 6154–6162

    2018

  25. [25]

    Libra r-cnn: Towards balanced learning for object detection,

    J. Pang, K. Chen, J. Shi, H. Feng, W. Ouyang, and D. Lin, “Libra r-cnn: Towards balanced learning for object detection,” inIEEE Conference on Computer Vision and Pattern Recognition, 2019, pp. 821–830

    2019

  26. [26]

    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,” inIEEE International Conference on Computer Vision, 2019, pp. 6569–6578

    2019

  27. [27]

    Deformable detr: Deformable transformers for end-to-end object detection,

    X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai, “Deformable detr: Deformable transformers for end-to-end object detection,” in International Conference on Learning Representations, 2020

    2020

  28. [28]

    Conditional detr for fast training convergence,

    D. Meng, X. Chen, Z. Fan, G. Zeng, H. Li, Y. Yuan, L. Sun, and J. Wang, “Conditional detr for fast training convergence,” inIEEE International Conference on Computer Vision, 2021, pp. 3651–3660

    2021

  29. [29]

    Sparse detr: Efficient end- to-end object detection with learnable sparsity,

    B. Roh, J. Shin, W. Shin, and S. Kim, “Sparse detr: Efficient end- to-end object detection with learnable sparsity,” inInternational Conference on Learning Representations, 2021

    2021

  30. [30]

    Dab-detr: Dynamic anchor boxes are better queries for detr,

    S. Liu, F. Li, H. Zhang, X. Yang, X. Qi, H. Su, J. Zhu, and L. Zhang, “Dab-detr: Dynamic anchor boxes are better queries for detr,” in International Conference on Learning Representations, 2022

    2022

  31. [31]

    Dino: Detr with improved denoising anchor boxes for end-to-end object detection,

    H. Zhang, F. Li, S. Liu, L. Zhang, H. Su, J. Zhu, L. M. Ni, and H.- Y. Shum, “Dino: Detr with improved denoising anchor boxes for end-to-end object detection,” inInternational Conference on Learning Representations, 2022

    2022

  32. [32]

    Detrs with collaborative hybrid assignments training,

    Z. Zong, G. Song, and Y. Liu, “Detrs with collaborative hybrid assignments training,” inIEEE International Conference on Computer Vision, 2023, pp. 6748–6758

    2023

  33. [33]

    Bevdet: High-performance multi-camera 3d object detection in bird-eye-view.arXiv preprint arXiv:2112.11790, 2021

    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

    work page arXiv 2021

  34. [34]

    arXiv preprint arXiv:2203.17054 (2022)

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

    work page arXiv 2022

  35. [35]

    Bevformer v2: Adapting modern image backbones to bird’s-eye-view recognition via per- spective supervision,

    C. Yang, Y. Chen, H. Tian, C. Tao, X. Zhu, Z. Zhang, G. Huang, H. Li, Y. Qiao, L. Lu, J. Zhou, and J. Dai, “Bevformer v2: Adapting modern image backbones to bird’s-eye-view recognition via per- spective supervision,” inIEEE Conference on Computer Vision and Pattern Recognition, 2023, pp. 17 830–17 839

    2023

  36. [36]

    Bevstereo: Enhancing depth estimation in multi-view 3d object detection with temporal stereo,

    Y. Li, H. Bao, Z. Ge, J. Yang, J. Sun, and Z. Li, “Bevstereo: Enhancing depth estimation in multi-view 3d object detection with temporal stereo,” inAAAI Conference on Artificial Intelligence, 2023, pp. 1486–1494

    2023

  37. [37]

    Time will tell: New outlooks and a baseline for temporal multi-view 3d object detection,

    J. Park, C. Xu, S. Yang, K. Keutzer, K. Kitani, M. Tomizuka, and W. Zhan, “Time will tell: New outlooks and a baseline for temporal multi-view 3d object detection,” inInternational Conference on Learning Representations, 2023

    2023

  38. [38]

    Exploring recurrent long-term temporal fusion for multi-view 3d perception,

    C. Han, J. Yang, J. Sun, Z. Ge, R. Dong, H. Zhou, W. Mao, Y. Peng, and X. Zhang, “Exploring recurrent long-term temporal fusion for multi-view 3d perception,”IEEE Robotics and Automation Letters, vol. 9, no. 7, pp. 6544–6551, 2024

    2024

  39. [39]

    Temporal enhanced training of multi-view 3d object detector via historical object prediction,

    Z. Zong, D. Jiang, G. Song, Z. Xue, J. Su, H. Li, and Y. Liu, “Temporal enhanced training of multi-view 3d object detector via historical object prediction,” inIEEE International Conference on Computer Vision, 2023

    2023

  40. [40]

    Geobev: Learning geometric bev representation for multi-view 3d object detection,

    J. Zhang, Y. Zhang, Y. Qi, Z. Fu, Q. Liu, and Y. Wang, “Geobev: Learning geometric bev representation for multi-view 3d object detection,” inAAAI Conference on Artificial Intelligence, vol. 39, no. 9, 2025, pp. 9960–9968

    2025

  41. [41]

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

    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 im- ages,” inIEEE International Conference on Computer Vision, 2023, pp. 3262–3272

    2023

  42. [42]

    Detr4d: Direct multi- view 3d object detection with sparse attention,

    Z. Luo, C. Zhou, G. Zhang, and S. Lu, “Detr4d: Direct multi- view 3d object detection with sparse attention,”arXiv preprint arXiv:2212.07849, 2022

    work page arXiv 2022

  43. [43]

    Sparse4d v2: Recurrent temporal fusion with sparse model.arXiv preprint arXiv:2305.14018, 2023

    X. Lin, T. Lin, Z. Pei, L. Huang, and Z. Su, “Sparse4d v2: Recurrent temporal fusion with sparse model,”arXiv preprint arXiv:2305.14018, 2023. SUBMITTED TO IEEE JOURNAL 16

    work page arXiv 2023

  44. [44]

    Dynamicbev: Leveraging dynamic queries and temporal context for 3d object detection,

    J. Yao and Y. Lai, “Dynamicbev: Leveraging dynamic queries and temporal context for 3d object detection,”arXiv preprint arXiv:2310.05989, 2023

    work page arXiv 2023

  45. [45]

    Bridging perspectives: Foundation model guided bev maps for 3d object detection and tracking,

    M. K ¨appeler, ¨Ozg ¨un C ¸ ic ¸ek, D. Cattaneo, C. Gl¨aser, Y. Miron, and A. Valada, “Bridging perspectives: Foundation model guided bev maps for 3d object detection and tracking,” 2025

    2025

  46. [46]

    Vision-based detec- tion and distance estimation of micro unmanned aerial vehicles,

    F. G ¨okc ¸e, G.¨Uc ¸oluk, E. S ¸ahin, and S. Kalkan, “Vision-based detec- tion and distance estimation of micro unmanned aerial vehicles,” Sensors, vol. 15, no. 9, pp. 23 805–23 846, 2015

    2015

  47. [47]

    Disnet: A novel method for distance estimation from monocular camera,

    M. A. Haseeb, J. Guan, D. Ristic-Durrant, and A. Gr ¨aser, “Disnet: A novel method for distance estimation from monocular camera,” 10th Planning, Perception and Navigation for Intelligent Vehicles (PP- NIV18), IROS, 2018

    2018

  48. [48]

    Learn- ing object-specific distance from a monocular image,

    J. Zhu, Y. Fang, H. Abu-Haimed, K. Lien, D. Fu, and J. Gu, “Learn- ing object-specific distance from a monocular image,”CoRR, 2019

    2019

  49. [49]

    R4d: Utilizing reference objects for long-range distance estimation,

    Y. Li, T. Chen, M. Kabkab, R. Yu, L. Jing, Y. You, and H. Zhao, “R4d: Utilizing reference objects for long-range distance estimation,” arXiv preprint arXiv:2206.04831, 2022

    work page arXiv 2022

  50. [50]

    Improving distant 3d object detection using 2d box supervision,

    Z. Yang, Z. Yu, C. Choy, R. Wang, A. Anandkumar, and J. M. Alvarez, “Improving distant 3d object detection using 2d box supervision,” inIEEE Conference on Computer Vision and Pattern Recognition, 2024, pp. 14 853–14 863

    2024

  51. [51]

    Towards long-range 3d object detection for autonomous vehi- cles,

    A. Khoche, L. P . S´anchez, N. Batool, S. S. Mansouri, and P . Jensfelt, “Towards long-range 3d object detection for autonomous vehi- cles,” in2024 IEEE Intelligent Vehicles Symposium. IEEE, 2024, pp. 2206–2212

    2024

  52. [52]

    Sparsefusion: Efficient sparse multi-modal fusion framework for long-range 3d perception,

    Y. Li, H. Li, Z. Huang, H. Chang, and N. Wang, “Sparsefusion: Efficient sparse multi-modal fusion framework for long-range 3d perception,”arXiv preprint arXiv:2403.10036, 2024

    work page arXiv 2024

  53. [53]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inIEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778

    2016

  54. [54]

    An energy and gpu-computation efficient backbone network for real-time object detection,

    Y. Lee, J.-w. Hwang, S. Lee, Y. Bae, and J. Park, “An energy and gpu-computation efficient backbone network for real-time object detection,” inIEEE Conference on Computer Vision and Pattern Recognition, 2019

    2019

  55. [55]

    Dn- detr: Accelerate detr training by introducing query denoising,

    F. Li, H. Zhang, S. Liu, J. Guo, L. M. Ni, and L. Zhang, “Dn- detr: Accelerate detr training by introducing query denoising,” in IEEE Conference on Computer Vision and Pattern Recognition, 2022, pp. 13 619–13 627

    2022

  56. [56]

    Ray denoising: Depth-aware hard negative sampling for multi-view 3d object detection,

    F. Liu, T. Huang, Q. Zhang, H. Yao, C. Zhang, F. Wan, Q. Ye, and Y. Zhou, “Ray denoising: Depth-aware hard negative sampling for multi-view 3d object detection,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 200–217

    2024

  57. [57]

    3d bound- ing box estimation using deep learning and geometry,

    A. Mousavian, D. Anguelov, J. Flynn, and J. Kosecka, “3d bound- ing box estimation using deep learning and geometry,” inIEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 7074–7082

    2017

  58. [58]

    Microsoft coco: Common objects in context,

    T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P . Perona, D. Ramanan, P . Doll´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” inEuropean Conference on Computer Vision. Springer, 2014, pp. 740–755

    2014

  59. [59]

    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

    work page arXiv 1908

  60. [60]

    Corrbev: Multi- view 3d object detection by correlation learning with multi-modal prototypes,

    Z. Xue, M. Guo, H. Fan, S. Zhang, and Z. Zhang, “Corrbev: Multi- view 3d object detection by correlation learning with multi-modal prototypes,” inIEEE Conference on Computer Vision and Pattern Recognition, 2025, pp. 27 413–27 423. SUBMITTED TO IEEE JOURNAL 1 Supplementary Material This supplementary material contains additional details of the main manuscr...

    2025